diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu')
136 files changed, 6292 insertions, 11677 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile index 138cb787d27e..466da5954a68 100644 --- a/drivers/gpu/drm/amd/amdgpu/Makefile +++ b/drivers/gpu/drm/amd/amdgpu/Makefile @@ -53,11 +53,11 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \ amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \ amdgpu_gtt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o amdgpu_atomfirmware.o \ amdgpu_vf_error.o amdgpu_sched.o amdgpu_debugfs.o amdgpu_ids.o \ - amdgpu_gmc.o amdgpu_xgmi.o + amdgpu_gmc.o amdgpu_xgmi.o amdgpu_csa.o # add asic specific block amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \ - ci_smc.o ci_dpm.o dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o + dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o amdgpu-$(CONFIG_DRM_AMDGPU_SI)+= si.o gmc_v6_0.o gfx_v6_0.o si_ih.o si_dma.o dce_v6_0.o si_dpm.o si_smc.o @@ -105,6 +105,7 @@ amdgpu-y += \ # add GFX block amdgpu-y += \ amdgpu_gfx.o \ + amdgpu_rlc.o \ gfx_v8_0.o \ gfx_v9_0.o diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index b0fc116296cb..8d0d7f3dd5fb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -75,11 +75,14 @@ #include "amdgpu_sdma.h" #include "amdgpu_dm.h" #include "amdgpu_virt.h" +#include "amdgpu_csa.h" #include "amdgpu_gart.h" #include "amdgpu_debugfs.h" #include "amdgpu_job.h" #include "amdgpu_bo_list.h" #include "amdgpu_gem.h" +#include "amdgpu_doorbell.h" +#include "amdgpu_amdkfd.h" #define MAX_GPU_INSTANCE 16 @@ -161,6 +164,7 @@ extern int amdgpu_si_support; extern int amdgpu_cik_support; #endif +#define AMDGPU_VM_MAX_NUM_CTX 4096 #define AMDGPU_SG_THRESHOLD (256*1024*1024) #define AMDGPU_DEFAULT_GTT_SIZE_MB 3072ULL /* 3GB by default */ #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000 @@ -360,123 +364,6 @@ int amdgpu_fence_slab_init(void); void amdgpu_fence_slab_fini(void); /* - * GPU doorbell structures, functions & helpers - */ -typedef enum _AMDGPU_DOORBELL_ASSIGNMENT -{ - AMDGPU_DOORBELL_KIQ = 0x000, - AMDGPU_DOORBELL_HIQ = 0x001, - AMDGPU_DOORBELL_DIQ = 0x002, - AMDGPU_DOORBELL_MEC_RING0 = 0x010, - AMDGPU_DOORBELL_MEC_RING1 = 0x011, - AMDGPU_DOORBELL_MEC_RING2 = 0x012, - AMDGPU_DOORBELL_MEC_RING3 = 0x013, - AMDGPU_DOORBELL_MEC_RING4 = 0x014, - AMDGPU_DOORBELL_MEC_RING5 = 0x015, - AMDGPU_DOORBELL_MEC_RING6 = 0x016, - AMDGPU_DOORBELL_MEC_RING7 = 0x017, - AMDGPU_DOORBELL_GFX_RING0 = 0x020, - AMDGPU_DOORBELL_sDMA_ENGINE0 = 0x1E0, - AMDGPU_DOORBELL_sDMA_ENGINE1 = 0x1E1, - AMDGPU_DOORBELL_IH = 0x1E8, - AMDGPU_DOORBELL_MAX_ASSIGNMENT = 0x3FF, - AMDGPU_DOORBELL_INVALID = 0xFFFF -} AMDGPU_DOORBELL_ASSIGNMENT; - -struct amdgpu_doorbell { - /* doorbell mmio */ - resource_size_t base; - resource_size_t size; - u32 __iomem *ptr; - u32 num_doorbells; /* Number of doorbells actually reserved for amdgpu. */ -}; - -/* - * 64bit doorbell, offset are in QWORD, occupy 2KB doorbell space - */ -typedef enum _AMDGPU_DOORBELL64_ASSIGNMENT -{ - /* - * All compute related doorbells: kiq, hiq, diq, traditional compute queue, user queue, should locate in - * a continues range so that programming CP_MEC_DOORBELL_RANGE_LOWER/UPPER can cover this range. - * Compute related doorbells are allocated from 0x00 to 0x8a - */ - - - /* kernel scheduling */ - AMDGPU_DOORBELL64_KIQ = 0x00, - - /* HSA interface queue and debug queue */ - AMDGPU_DOORBELL64_HIQ = 0x01, - AMDGPU_DOORBELL64_DIQ = 0x02, - - /* Compute engines */ - AMDGPU_DOORBELL64_MEC_RING0 = 0x03, - AMDGPU_DOORBELL64_MEC_RING1 = 0x04, - AMDGPU_DOORBELL64_MEC_RING2 = 0x05, - AMDGPU_DOORBELL64_MEC_RING3 = 0x06, - AMDGPU_DOORBELL64_MEC_RING4 = 0x07, - AMDGPU_DOORBELL64_MEC_RING5 = 0x08, - AMDGPU_DOORBELL64_MEC_RING6 = 0x09, - AMDGPU_DOORBELL64_MEC_RING7 = 0x0a, - - /* User queue doorbell range (128 doorbells) */ - AMDGPU_DOORBELL64_USERQUEUE_START = 0x0b, - AMDGPU_DOORBELL64_USERQUEUE_END = 0x8a, - - /* Graphics engine */ - AMDGPU_DOORBELL64_GFX_RING0 = 0x8b, - - /* - * Other graphics doorbells can be allocated here: from 0x8c to 0xdf - * Graphics voltage island aperture 1 - * default non-graphics QWORD index is 0xe0 - 0xFF inclusive - */ - - /* sDMA engines reserved from 0xe0 -oxef */ - AMDGPU_DOORBELL64_sDMA_ENGINE0 = 0xE0, - AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE0 = 0xE1, - AMDGPU_DOORBELL64_sDMA_ENGINE1 = 0xE8, - AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE1 = 0xE9, - - /* For vega10 sriov, the sdma doorbell must be fixed as follow - * to keep the same setting with host driver, or it will - * happen conflicts - */ - AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE0 = 0xF0, - AMDGPU_VEGA10_DOORBELL64_sDMA_HI_PRI_ENGINE0 = 0xF1, - AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE1 = 0xF2, - AMDGPU_VEGA10_DOORBELL64_sDMA_HI_PRI_ENGINE1 = 0xF3, - - /* Interrupt handler */ - AMDGPU_DOORBELL64_IH = 0xF4, /* For legacy interrupt ring buffer */ - AMDGPU_DOORBELL64_IH_RING1 = 0xF5, /* For page migration request log */ - AMDGPU_DOORBELL64_IH_RING2 = 0xF6, /* For page migration translation/invalidation log */ - - /* VCN engine use 32 bits doorbell */ - AMDGPU_DOORBELL64_VCN0_1 = 0xF8, /* lower 32 bits for VNC0 and upper 32 bits for VNC1 */ - AMDGPU_DOORBELL64_VCN2_3 = 0xF9, - AMDGPU_DOORBELL64_VCN4_5 = 0xFA, - AMDGPU_DOORBELL64_VCN6_7 = 0xFB, - - /* overlap the doorbell assignment with VCN as they are mutually exclusive - * VCE engine's doorbell is 32 bit and two VCE ring share one QWORD - */ - AMDGPU_DOORBELL64_UVD_RING0_1 = 0xF8, - AMDGPU_DOORBELL64_UVD_RING2_3 = 0xF9, - AMDGPU_DOORBELL64_UVD_RING4_5 = 0xFA, - AMDGPU_DOORBELL64_UVD_RING6_7 = 0xFB, - - AMDGPU_DOORBELL64_VCE_RING0_1 = 0xFC, - AMDGPU_DOORBELL64_VCE_RING2_3 = 0xFD, - AMDGPU_DOORBELL64_VCE_RING4_5 = 0xFE, - AMDGPU_DOORBELL64_VCE_RING6_7 = 0xFF, - - AMDGPU_DOORBELL64_MAX_ASSIGNMENT = 0xFF, - AMDGPU_DOORBELL64_INVALID = 0xFFFF -} AMDGPU_DOORBELL64_ASSIGNMENT; - -/* * IRQS. */ @@ -524,6 +411,8 @@ struct amdgpu_fpriv { struct amdgpu_ctx_mgr ctx_mgr; }; +int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv); + int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm, unsigned size, struct amdgpu_ib *ib); void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib, @@ -653,6 +542,13 @@ struct amdgpu_asic_funcs { struct amdgpu_ring *ring); /* check if the asic needs a full reset of if soft reset will work */ bool (*need_full_reset)(struct amdgpu_device *adev); + /* initialize doorbell layout for specific asic*/ + void (*init_doorbell_index)(struct amdgpu_device *adev); + /* PCIe bandwidth usage */ + void (*get_pcie_usage)(struct amdgpu_device *adev, uint64_t *count0, + uint64_t *count1); + /* do we need to reset the asic at init time (e.g., kexec) */ + bool (*need_reset_on_init)(struct amdgpu_device *adev); }; /* @@ -745,7 +641,7 @@ struct amdgpu_nbio_funcs { void (*hdp_flush)(struct amdgpu_device *adev, struct amdgpu_ring *ring); u32 (*get_memsize)(struct amdgpu_device *adev); void (*sdma_doorbell_range)(struct amdgpu_device *adev, int instance, - bool use_doorbell, int doorbell_index); + bool use_doorbell, int doorbell_index, int doorbell_size); void (*enable_doorbell_aperture)(struct amdgpu_device *adev, bool enable); void (*enable_doorbell_selfring_aperture)(struct amdgpu_device *adev, @@ -831,7 +727,6 @@ struct amdgpu_device { bool need_dma32; bool need_swiotlb; bool accel_working; - struct work_struct reset_work; struct notifier_block acpi_nb; struct amdgpu_i2c_chan *i2c_bus[AMDGPU_MAX_I2C_BUS]; struct amdgpu_debugfs debugfs[AMDGPU_DEBUGFS_MAX_COMPONENTS]; @@ -976,6 +871,9 @@ struct amdgpu_device { /* GDS */ struct amdgpu_gds gds; + /* KFD */ + struct amdgpu_kfd_dev kfd; + /* display related functionality */ struct amdgpu_display_manager dm; @@ -989,9 +887,6 @@ struct amdgpu_device { atomic64_t visible_pin_size; atomic64_t gart_pin_size; - /* amdkfd interface */ - struct kfd_dev *kfd; - /* soc15 register offset based on ip, instance and segment */ uint32_t *reg_offset[MAX_HWIP][HWIP_MAX_INSTANCE]; @@ -1023,6 +918,10 @@ struct amdgpu_device { unsigned long last_mm_index; bool in_gpu_reset; struct mutex lock_reset; + struct amdgpu_doorbell_index doorbell_index; + + int asic_reset_res; + struct work_struct xgmi_reset_work; }; static inline struct amdgpu_device *amdgpu_ttm_adev(struct ttm_bo_device *bdev) @@ -1047,11 +946,6 @@ uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset); u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg); void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v); -u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index); -void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v); -u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index); -void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v); - bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type); bool amdgpu_device_has_dc_support(struct amdgpu_device *adev); @@ -1113,11 +1007,6 @@ int emu_soc_asic_init(struct amdgpu_device *adev); #define RREG32_IO(reg) amdgpu_io_rreg(adev, (reg)) #define WREG32_IO(reg, v) amdgpu_io_wreg(adev, (reg), (v)) -#define RDOORBELL32(index) amdgpu_mm_rdoorbell(adev, (index)) -#define WDOORBELL32(index, v) amdgpu_mm_wdoorbell(adev, (index), (v)) -#define RDOORBELL64(index) amdgpu_mm_rdoorbell64(adev, (index)) -#define WDOORBELL64(index, v) amdgpu_mm_wdoorbell64(adev, (index), (v)) - #define REG_FIELD_SHIFT(reg, field) reg##__##field##__SHIFT #define REG_FIELD_MASK(reg, field) reg##__##field##_MASK @@ -1159,6 +1048,9 @@ int emu_soc_asic_init(struct amdgpu_device *adev); #define amdgpu_asic_flush_hdp(adev, r) (adev)->asic_funcs->flush_hdp((adev), (r)) #define amdgpu_asic_invalidate_hdp(adev, r) (adev)->asic_funcs->invalidate_hdp((adev), (r)) #define amdgpu_asic_need_full_reset(adev) (adev)->asic_funcs->need_full_reset((adev)) +#define amdgpu_asic_init_doorbell_index(adev) (adev)->asic_funcs->init_doorbell_index((adev)) +#define amdgpu_asic_get_pcie_usage(adev, cnt0, cnt1) ((adev)->asic_funcs->get_pcie_usage((adev), (cnt0), (cnt1))) +#define amdgpu_asic_need_reset_on_init(adev) (adev)->asic_funcs->need_reset_on_init((adev)) /* Common functions */ bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev); @@ -1219,12 +1111,6 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe); long amdgpu_kms_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg); - -/* - * functions used by amdgpu_xgmi.c - */ -int amdgpu_xgmi_add_device(struct amdgpu_device *adev); - /* * functions used by amdgpu_encoder.c */ @@ -1252,6 +1138,9 @@ bool amdgpu_acpi_is_pcie_performance_request_supported(struct amdgpu_device *ade int amdgpu_acpi_pcie_performance_request(struct amdgpu_device *adev, u8 perf_req, bool advertise); int amdgpu_acpi_pcie_notify_device_ready(struct amdgpu_device *adev); + +void amdgpu_acpi_get_backlight_caps(struct amdgpu_device *adev, + struct amdgpu_dm_backlight_caps *caps); #else static inline int amdgpu_acpi_init(struct amdgpu_device *adev) { return 0; } static inline void amdgpu_acpi_fini(struct amdgpu_device *adev) { } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c index 7f0afc526419..4376b17ca594 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c @@ -41,28 +41,21 @@ struct amdgpu_atif_notification_cfg { }; struct amdgpu_atif_notifications { - bool display_switch; - bool expansion_mode_change; bool thermal_state; bool forced_power_state; bool system_power_state; - bool display_conf_change; - bool px_gfx_switch; bool brightness_change; bool dgpu_display_event; + bool gpu_package_power_limit; }; struct amdgpu_atif_functions { bool system_params; bool sbios_requests; - bool select_active_disp; - bool lid_state; - bool get_tv_standard; - bool set_tv_standard; - bool get_panel_expansion_mode; - bool set_panel_expansion_mode; bool temperature_change; - bool graphics_device_types; + bool query_backlight_transfer_characteristics; + bool ready_to_undock; + bool external_gpu_information; }; struct amdgpu_atif { @@ -72,6 +65,7 @@ struct amdgpu_atif { struct amdgpu_atif_functions functions; struct amdgpu_atif_notification_cfg notification_cfg; struct amdgpu_encoder *encoder_for_bl; + struct amdgpu_dm_backlight_caps backlight_caps; }; /* Call the ATIF method @@ -137,15 +131,12 @@ static union acpi_object *amdgpu_atif_call(struct amdgpu_atif *atif, */ static void amdgpu_atif_parse_notification(struct amdgpu_atif_notifications *n, u32 mask) { - n->display_switch = mask & ATIF_DISPLAY_SWITCH_REQUEST_SUPPORTED; - n->expansion_mode_change = mask & ATIF_EXPANSION_MODE_CHANGE_REQUEST_SUPPORTED; n->thermal_state = mask & ATIF_THERMAL_STATE_CHANGE_REQUEST_SUPPORTED; n->forced_power_state = mask & ATIF_FORCED_POWER_STATE_CHANGE_REQUEST_SUPPORTED; n->system_power_state = mask & ATIF_SYSTEM_POWER_SOURCE_CHANGE_REQUEST_SUPPORTED; - n->display_conf_change = mask & ATIF_DISPLAY_CONF_CHANGE_REQUEST_SUPPORTED; - n->px_gfx_switch = mask & ATIF_PX_GFX_SWITCH_REQUEST_SUPPORTED; n->brightness_change = mask & ATIF_PANEL_BRIGHTNESS_CHANGE_REQUEST_SUPPORTED; n->dgpu_display_event = mask & ATIF_DGPU_DISPLAY_EVENT_SUPPORTED; + n->gpu_package_power_limit = mask & ATIF_GPU_PACKAGE_POWER_LIMIT_REQUEST_SUPPORTED; } /** @@ -162,14 +153,11 @@ static void amdgpu_atif_parse_functions(struct amdgpu_atif_functions *f, u32 mas { f->system_params = mask & ATIF_GET_SYSTEM_PARAMETERS_SUPPORTED; f->sbios_requests = mask & ATIF_GET_SYSTEM_BIOS_REQUESTS_SUPPORTED; - f->select_active_disp = mask & ATIF_SELECT_ACTIVE_DISPLAYS_SUPPORTED; - f->lid_state = mask & ATIF_GET_LID_STATE_SUPPORTED; - f->get_tv_standard = mask & ATIF_GET_TV_STANDARD_FROM_CMOS_SUPPORTED; - f->set_tv_standard = mask & ATIF_SET_TV_STANDARD_IN_CMOS_SUPPORTED; - f->get_panel_expansion_mode = mask & ATIF_GET_PANEL_EXPANSION_MODE_FROM_CMOS_SUPPORTED; - f->set_panel_expansion_mode = mask & ATIF_SET_PANEL_EXPANSION_MODE_IN_CMOS_SUPPORTED; f->temperature_change = mask & ATIF_TEMPERATURE_CHANGE_NOTIFICATION_SUPPORTED; - f->graphics_device_types = mask & ATIF_GET_GRAPHICS_DEVICE_TYPES_SUPPORTED; + f->query_backlight_transfer_characteristics = + mask & ATIF_QUERY_BACKLIGHT_TRANSFER_CHARACTERISTICS_SUPPORTED; + f->ready_to_undock = mask & ATIF_READY_TO_UNDOCK_NOTIFICATION_SUPPORTED; + f->external_gpu_information = mask & ATIF_GET_EXTERNAL_GPU_INFORMATION_SUPPORTED; } /** @@ -311,6 +299,65 @@ out: } /** + * amdgpu_atif_query_backlight_caps - get min and max backlight input signal + * + * @handle: acpi handle + * + * Execute the QUERY_BRIGHTNESS_TRANSFER_CHARACTERISTICS ATIF function + * to determine the acceptable range of backlight values + * + * Backlight_caps.caps_valid will be set to true if the query is successful + * + * The input signals are in range 0-255 + * + * This function assumes the display with backlight is the first LCD + * + * Returns 0 on success, error on failure. + */ +static int amdgpu_atif_query_backlight_caps(struct amdgpu_atif *atif) +{ + union acpi_object *info; + struct atif_qbtc_output characteristics; + struct atif_qbtc_arguments arguments; + struct acpi_buffer params; + size_t size; + int err = 0; + + arguments.size = sizeof(arguments); + arguments.requested_display = ATIF_QBTC_REQUEST_LCD1; + + params.length = sizeof(arguments); + params.pointer = (void *)&arguments; + + info = amdgpu_atif_call(atif, + ATIF_FUNCTION_QUERY_BRIGHTNESS_TRANSFER_CHARACTERISTICS, + ¶ms); + if (!info) { + err = -EIO; + goto out; + } + + size = *(u16 *) info->buffer.pointer; + if (size < 10) { + err = -EINVAL; + goto out; + } + + memset(&characteristics, 0, sizeof(characteristics)); + size = min(sizeof(characteristics), size); + memcpy(&characteristics, info->buffer.pointer, size); + + atif->backlight_caps.caps_valid = true; + atif->backlight_caps.min_input_signal = + characteristics.min_input_signal; + atif->backlight_caps.max_input_signal = + characteristics.max_input_signal; +out: + kfree(info); + return err; +} + +/** * amdgpu_atif_get_sbios_requests - get requested sbios event * * @handle: acpi handle @@ -799,6 +846,17 @@ int amdgpu_acpi_init(struct amdgpu_device *adev) } } + if (atif->functions.query_backlight_transfer_characteristics) { + ret = amdgpu_atif_query_backlight_caps(atif); + if (ret) { + DRM_DEBUG_DRIVER("Call to QUERY_BACKLIGHT_TRANSFER_CHARACTERISTICS failed: %d\n", + ret); + atif->backlight_caps.caps_valid = false; + } + } else { + atif->backlight_caps.caps_valid = false; + } + out: adev->acpi_nb.notifier_call = amdgpu_acpi_event; register_acpi_notifier(&adev->acpi_nb); @@ -806,6 +864,18 @@ out: return ret; } +void amdgpu_acpi_get_backlight_caps(struct amdgpu_device *adev, + struct amdgpu_dm_backlight_caps *caps) +{ + if (!adev->atif) { + caps->caps_valid = false; + return; + } + caps->caps_valid = adev->atif->backlight_caps.caps_valid; + caps->min_input_signal = adev->atif->backlight_caps.min_input_signal; + caps->max_input_signal = adev->atif->backlight_caps.max_input_signal; +} + /** * amdgpu_acpi_fini - tear down driver acpi support * @@ -816,6 +886,5 @@ out: void amdgpu_acpi_fini(struct amdgpu_device *adev) { unregister_acpi_notifier(&adev->acpi_nb); - if (adev->atif) - kfree(adev->atif); + kfree(adev->atif); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 1580ec60b89f..fe1d7368c1e6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -26,22 +26,28 @@ #include "amdgpu.h" #include "amdgpu_gfx.h" #include <linux/module.h> - -const struct kgd2kfd_calls *kgd2kfd; +#include <linux/dma-buf.h> static const unsigned int compute_vmid_bitmap = 0xFF00; +/* Total memory size in system memory and all GPU VRAM. Used to + * estimate worst case amount of memory to reserve for page tables + */ +uint64_t amdgpu_amdkfd_total_mem_size; + int amdgpu_amdkfd_init(void) { + struct sysinfo si; int ret; + si_meminfo(&si); + amdgpu_amdkfd_total_mem_size = si.totalram - si.totalhigh; + amdgpu_amdkfd_total_mem_size *= si.mem_unit; + #ifdef CONFIG_HSA_AMD - ret = kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd); - if (ret) - kgd2kfd = NULL; + ret = kgd2kfd_init(); amdgpu_amdkfd_gpuvm_init_mem_limits(); #else - kgd2kfd = NULL; ret = -ENOENT; #endif @@ -50,17 +56,13 @@ int amdgpu_amdkfd_init(void) void amdgpu_amdkfd_fini(void) { - if (kgd2kfd) - kgd2kfd->exit(); + kgd2kfd_exit(); } void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev) { const struct kfd2kgd_calls *kfd2kgd; - if (!kgd2kfd) - return; - switch (adev->asic_type) { #ifdef CONFIG_DRM_AMDGPU_CIK case CHIP_KAVERI: @@ -73,9 +75,11 @@ void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev) case CHIP_FIJI: case CHIP_POLARIS10: case CHIP_POLARIS11: + case CHIP_POLARIS12: kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions(); break; case CHIP_VEGA10: + case CHIP_VEGA12: case CHIP_VEGA20: case CHIP_RAVEN: kfd2kgd = amdgpu_amdkfd_gfx_9_0_get_functions(); @@ -85,8 +89,11 @@ void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev) return; } - adev->kfd = kgd2kfd->probe((struct kgd_dev *)adev, - adev->pdev, kfd2kgd); + adev->kfd.dev = kgd2kfd_probe((struct kgd_dev *)adev, + adev->pdev, kfd2kgd); + + if (adev->kfd.dev) + amdgpu_amdkfd_total_mem_size += adev->gmc.real_vram_size; } /** @@ -124,9 +131,10 @@ static void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev, void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) { - int i, n; + int i; int last_valid_bit; - if (adev->kfd) { + + if (adev->kfd.dev) { struct kgd2kfd_shared_resources gpu_resources = { .compute_vmid_bitmap = compute_vmid_bitmap, .num_pipe_per_mec = adev->gfx.mec.num_pipe_per_mec, @@ -134,7 +142,9 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) .gpuvm_size = min(adev->vm_manager.max_pfn << AMDGPU_GPU_PAGE_SHIFT, AMDGPU_GMC_HOLE_START), - .drm_render_minor = adev->ddev->render->index + .drm_render_minor = adev->ddev->render->index, + .sdma_doorbell_idx = adev->doorbell_index.sdma_engine, + }; /* this is going to have a few of the MSBs set that we need to @@ -144,7 +154,7 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) KGD_MAX_QUEUES); /* remove the KIQ bit as well */ - if (adev->gfx.kiq.ring.ready) + if (adev->gfx.kiq.ring.sched.ready) clear_bit(amdgpu_gfx_queue_to_bit(adev, adev->gfx.kiq.ring.me - 1, adev->gfx.kiq.ring.pipe, @@ -164,78 +174,52 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) &gpu_resources.doorbell_aperture_size, &gpu_resources.doorbell_start_offset); - if (adev->asic_type < CHIP_VEGA10) { - kgd2kfd->device_init(adev->kfd, &gpu_resources); - return; - } - - n = (adev->asic_type < CHIP_VEGA20) ? 2 : 8; - - for (i = 0; i < n; i += 2) { - /* On SOC15 the BIF is involved in routing - * doorbells using the low 12 bits of the - * address. Communicate the assignments to - * KFD. KFD uses two doorbell pages per - * process in case of 64-bit doorbells so we - * can use each doorbell assignment twice. - */ - if (adev->asic_type == CHIP_VEGA10) { - gpu_resources.sdma_doorbell[0][i] = - AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE0 + (i >> 1); - gpu_resources.sdma_doorbell[0][i+1] = - AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE0 + 0x200 + (i >> 1); - gpu_resources.sdma_doorbell[1][i] = - AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE1 + (i >> 1); - gpu_resources.sdma_doorbell[1][i+1] = - AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE1 + 0x200 + (i >> 1); - } else { - gpu_resources.sdma_doorbell[0][i] = - AMDGPU_DOORBELL64_sDMA_ENGINE0 + (i >> 1); - gpu_resources.sdma_doorbell[0][i+1] = - AMDGPU_DOORBELL64_sDMA_ENGINE0 + 0x200 + (i >> 1); - gpu_resources.sdma_doorbell[1][i] = - AMDGPU_DOORBELL64_sDMA_ENGINE1 + (i >> 1); - gpu_resources.sdma_doorbell[1][i+1] = - AMDGPU_DOORBELL64_sDMA_ENGINE1 + 0x200 + (i >> 1); - } - } - /* Doorbells 0x0e0-0ff and 0x2e0-2ff are reserved for - * SDMA, IH and VCN. So don't use them for the CP. + /* Since SOC15, BIF starts to statically use the + * lower 12 bits of doorbell addresses for routing + * based on settings in registers like + * SDMA0_DOORBELL_RANGE etc.. + * In order to route a doorbell to CP engine, the lower + * 12 bits of its address has to be outside the range + * set for SDMA, VCN, and IH blocks. */ - gpu_resources.reserved_doorbell_mask = 0x1e0; - gpu_resources.reserved_doorbell_val = 0x0e0; + if (adev->asic_type >= CHIP_VEGA10) { + gpu_resources.non_cp_doorbells_start = + adev->doorbell_index.first_non_cp; + gpu_resources.non_cp_doorbells_end = + adev->doorbell_index.last_non_cp; + } - kgd2kfd->device_init(adev->kfd, &gpu_resources); + kgd2kfd_device_init(adev->kfd.dev, &gpu_resources); } } void amdgpu_amdkfd_device_fini(struct amdgpu_device *adev) { - if (adev->kfd) { - kgd2kfd->device_exit(adev->kfd); - adev->kfd = NULL; + if (adev->kfd.dev) { + kgd2kfd_device_exit(adev->kfd.dev); + adev->kfd.dev = NULL; } } void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev, const void *ih_ring_entry) { - if (adev->kfd) - kgd2kfd->interrupt(adev->kfd, ih_ring_entry); + if (adev->kfd.dev) + kgd2kfd_interrupt(adev->kfd.dev, ih_ring_entry); } void amdgpu_amdkfd_suspend(struct amdgpu_device *adev) { - if (adev->kfd) - kgd2kfd->suspend(adev->kfd); + if (adev->kfd.dev) + kgd2kfd_suspend(adev->kfd.dev); } int amdgpu_amdkfd_resume(struct amdgpu_device *adev) { int r = 0; - if (adev->kfd) - r = kgd2kfd->resume(adev->kfd); + if (adev->kfd.dev) + r = kgd2kfd_resume(adev->kfd.dev); return r; } @@ -244,8 +228,8 @@ int amdgpu_amdkfd_pre_reset(struct amdgpu_device *adev) { int r = 0; - if (adev->kfd) - r = kgd2kfd->pre_reset(adev->kfd); + if (adev->kfd.dev) + r = kgd2kfd_pre_reset(adev->kfd.dev); return r; } @@ -254,8 +238,8 @@ int amdgpu_amdkfd_post_reset(struct amdgpu_device *adev) { int r = 0; - if (adev->kfd) - r = kgd2kfd->post_reset(adev->kfd); + if (adev->kfd.dev) + r = kgd2kfd_post_reset(adev->kfd.dev); return r; } @@ -268,9 +252,9 @@ void amdgpu_amdkfd_gpu_reset(struct kgd_dev *kgd) amdgpu_device_gpu_recover(adev, NULL); } -int alloc_gtt_mem(struct kgd_dev *kgd, size_t size, - void **mem_obj, uint64_t *gpu_addr, - void **cpu_ptr, bool mqd_gfx9) +int amdgpu_amdkfd_alloc_gtt_mem(struct kgd_dev *kgd, size_t size, + void **mem_obj, uint64_t *gpu_addr, + void **cpu_ptr, bool mqd_gfx9) { struct amdgpu_device *adev = (struct amdgpu_device *)kgd; struct amdgpu_bo *bo = NULL; @@ -340,7 +324,7 @@ allocate_mem_reserve_bo_failed: return r; } -void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj) +void amdgpu_amdkfd_free_gtt_mem(struct kgd_dev *kgd, void *mem_obj) { struct amdgpu_bo *bo = (struct amdgpu_bo *) mem_obj; @@ -351,8 +335,8 @@ void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj) amdgpu_bo_unref(&(bo)); } -void get_local_mem_info(struct kgd_dev *kgd, - struct kfd_local_mem_info *mem_info) +void amdgpu_amdkfd_get_local_mem_info(struct kgd_dev *kgd, + struct kfd_local_mem_info *mem_info) { struct amdgpu_device *adev = (struct amdgpu_device *)kgd; uint64_t address_mask = adev->dev->dma_mask ? ~*adev->dev->dma_mask : @@ -383,7 +367,7 @@ void get_local_mem_info(struct kgd_dev *kgd, mem_info->mem_clk_max = 100; } -uint64_t get_gpu_clock_counter(struct kgd_dev *kgd) +uint64_t amdgpu_amdkfd_get_gpu_clock_counter(struct kgd_dev *kgd) { struct amdgpu_device *adev = (struct amdgpu_device *)kgd; @@ -392,7 +376,7 @@ uint64_t get_gpu_clock_counter(struct kgd_dev *kgd) return 0; } -uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd) +uint32_t amdgpu_amdkfd_get_max_engine_clock_in_mhz(struct kgd_dev *kgd) { struct amdgpu_device *adev = (struct amdgpu_device *)kgd; @@ -405,7 +389,7 @@ uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd) return 100; } -void get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info) +void amdgpu_amdkfd_get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info) { struct amdgpu_device *adev = (struct amdgpu_device *)kgd; struct amdgpu_cu_info acu_info = adev->gfx.cu_info; @@ -428,6 +412,62 @@ void get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info) cu_info->lds_size = acu_info.lds_size; } +int amdgpu_amdkfd_get_dmabuf_info(struct kgd_dev *kgd, int dma_buf_fd, + struct kgd_dev **dma_buf_kgd, + uint64_t *bo_size, void *metadata_buffer, + size_t buffer_size, uint32_t *metadata_size, + uint32_t *flags) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)kgd; + struct dma_buf *dma_buf; + struct drm_gem_object *obj; + struct amdgpu_bo *bo; + uint64_t metadata_flags; + int r = -EINVAL; + + dma_buf = dma_buf_get(dma_buf_fd); + if (IS_ERR(dma_buf)) + return PTR_ERR(dma_buf); + + if (dma_buf->ops != &amdgpu_dmabuf_ops) + /* Can't handle non-graphics buffers */ + goto out_put; + + obj = dma_buf->priv; + if (obj->dev->driver != adev->ddev->driver) + /* Can't handle buffers from different drivers */ + goto out_put; + + adev = obj->dev->dev_private; + bo = gem_to_amdgpu_bo(obj); + if (!(bo->preferred_domains & (AMDGPU_GEM_DOMAIN_VRAM | + AMDGPU_GEM_DOMAIN_GTT))) + /* Only VRAM and GTT BOs are supported */ + goto out_put; + + r = 0; + if (dma_buf_kgd) + *dma_buf_kgd = (struct kgd_dev *)adev; + if (bo_size) + *bo_size = amdgpu_bo_size(bo); + if (metadata_size) + *metadata_size = bo->metadata_size; + if (metadata_buffer) + r = amdgpu_bo_get_metadata(bo, metadata_buffer, buffer_size, + metadata_size, &metadata_flags); + if (flags) { + *flags = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) ? + ALLOC_MEM_FLAGS_VRAM : ALLOC_MEM_FLAGS_GTT; + + if (bo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) + *flags |= ALLOC_MEM_FLAGS_PUBLIC; + } + +out_put: + dma_buf_put(dma_buf); + return r; +} + uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd) { struct amdgpu_device *adev = (struct amdgpu_device *)kgd; @@ -510,7 +550,7 @@ void amdgpu_amdkfd_set_compute_idle(struct kgd_dev *kgd, bool idle) bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid) { - if (adev->kfd) { + if (adev->kfd.dev) { if ((1 << vmid) & compute_vmid_bitmap) return true; } @@ -524,7 +564,7 @@ bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm) return false; } -void amdgpu_amdkfd_unreserve_system_memory_limit(struct amdgpu_bo *bo) +void amdgpu_amdkfd_unreserve_memory_limit(struct amdgpu_bo *bo) { } @@ -557,4 +597,47 @@ struct kfd2kgd_calls *amdgpu_amdkfd_gfx_9_0_get_functions(void) { return NULL; } + +struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, struct pci_dev *pdev, + const struct kfd2kgd_calls *f2g) +{ + return NULL; +} + +bool kgd2kfd_device_init(struct kfd_dev *kfd, + const struct kgd2kfd_shared_resources *gpu_resources) +{ + return false; +} + +void kgd2kfd_device_exit(struct kfd_dev *kfd) +{ +} + +void kgd2kfd_exit(void) +{ +} + +void kgd2kfd_suspend(struct kfd_dev *kfd) +{ +} + +int kgd2kfd_resume(struct kfd_dev *kfd) +{ + return 0; +} + +int kgd2kfd_pre_reset(struct kfd_dev *kfd) +{ + return 0; +} + +int kgd2kfd_post_reset(struct kfd_dev *kfd) +{ + return 0; +} + +void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry) +{ +} #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index 8e0d4f7196b4..0b31a1859023 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -27,14 +27,13 @@ #include <linux/types.h> #include <linux/mm.h> -#include <linux/mmu_context.h> #include <linux/workqueue.h> #include <kgd_kfd_interface.h> #include <drm/ttm/ttm_execbuf_util.h> #include "amdgpu_sync.h" #include "amdgpu_vm.h" -extern const struct kgd2kfd_calls *kgd2kfd; +extern uint64_t amdgpu_amdkfd_total_mem_size; struct amdgpu_device; @@ -77,6 +76,11 @@ struct amdgpu_amdkfd_fence { char timeline_name[TASK_COMM_LEN]; }; +struct amdgpu_kfd_dev { + struct kfd_dev *dev; + uint64_t vram_used; +}; + struct amdgpu_amdkfd_fence *amdgpu_amdkfd_fence_create(u64 context, struct mm_struct *mm); bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm); @@ -134,16 +138,21 @@ int amdgpu_amdkfd_post_reset(struct amdgpu_device *adev); void amdgpu_amdkfd_gpu_reset(struct kgd_dev *kgd); /* Shared API */ -int alloc_gtt_mem(struct kgd_dev *kgd, size_t size, - void **mem_obj, uint64_t *gpu_addr, - void **cpu_ptr, bool mqd_gfx9); -void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj); -void get_local_mem_info(struct kgd_dev *kgd, - struct kfd_local_mem_info *mem_info); -uint64_t get_gpu_clock_counter(struct kgd_dev *kgd); - -uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd); -void get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info); +int amdgpu_amdkfd_alloc_gtt_mem(struct kgd_dev *kgd, size_t size, + void **mem_obj, uint64_t *gpu_addr, + void **cpu_ptr, bool mqd_gfx9); +void amdgpu_amdkfd_free_gtt_mem(struct kgd_dev *kgd, void *mem_obj); +void amdgpu_amdkfd_get_local_mem_info(struct kgd_dev *kgd, + struct kfd_local_mem_info *mem_info); +uint64_t amdgpu_amdkfd_get_gpu_clock_counter(struct kgd_dev *kgd); + +uint32_t amdgpu_amdkfd_get_max_engine_clock_in_mhz(struct kgd_dev *kgd); +void amdgpu_amdkfd_get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info); +int amdgpu_amdkfd_get_dmabuf_info(struct kgd_dev *kgd, int dma_buf_fd, + struct kgd_dev **dmabuf_kgd, + uint64_t *bo_size, void *metadata_buffer, + size_t buffer_size, uint32_t *metadata_size, + uint32_t *flags); uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd); uint64_t amdgpu_amdkfd_get_hive_id(struct kgd_dev *kgd); @@ -195,7 +204,31 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *process_info, int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct kgd_dev *kgd, struct kfd_vm_fault_info *info); +int amdgpu_amdkfd_gpuvm_import_dmabuf(struct kgd_dev *kgd, + struct dma_buf *dmabuf, + uint64_t va, void *vm, + struct kgd_mem **mem, uint64_t *size, + uint64_t *mmap_offset); + void amdgpu_amdkfd_gpuvm_init_mem_limits(void); -void amdgpu_amdkfd_unreserve_system_memory_limit(struct amdgpu_bo *bo); +void amdgpu_amdkfd_unreserve_memory_limit(struct amdgpu_bo *bo); + +/* KGD2KFD callbacks */ +int kgd2kfd_init(void); +void kgd2kfd_exit(void); +struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, struct pci_dev *pdev, + const struct kfd2kgd_calls *f2g); +bool kgd2kfd_device_init(struct kfd_dev *kfd, + const struct kgd2kfd_shared_resources *gpu_resources); +void kgd2kfd_device_exit(struct kfd_dev *kfd); +void kgd2kfd_suspend(struct kfd_dev *kfd); +int kgd2kfd_resume(struct kfd_dev *kfd); +int kgd2kfd_pre_reset(struct kfd_dev *kfd); +int kgd2kfd_post_reset(struct kfd_dev *kfd); +void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry); +int kgd2kfd_quiesce_mm(struct mm_struct *mm); +int kgd2kfd_resume_mm(struct mm_struct *mm); +int kgd2kfd_schedule_evict_and_restore_process(struct mm_struct *mm, + struct dma_fence *fence); #endif /* AMDGPU_AMDKFD_H_INCLUDED */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c index 574c1181ae9a..3107b9575929 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c @@ -122,7 +122,7 @@ static bool amdkfd_fence_enable_signaling(struct dma_fence *f) if (dma_fence_is_signaled(f)) return true; - if (!kgd2kfd->schedule_evict_and_restore_process(fence->mm, f)) + if (!kgd2kfd_schedule_evict_and_restore_process(fence->mm, f)) return true; return false; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c index 244d9834a381..ff7fac7df34b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c @@ -23,6 +23,7 @@ #include <linux/fdtable.h> #include <linux/uaccess.h> #include <linux/firmware.h> +#include <linux/mmu_context.h> #include <drm/drmP.h> #include "amdgpu.h" #include "amdgpu_amdkfd.h" @@ -173,13 +174,6 @@ static int get_tile_config(struct kgd_dev *kgd, } static const struct kfd2kgd_calls kfd2kgd = { - .init_gtt_mem_allocation = alloc_gtt_mem, - .free_gtt_mem = free_gtt_mem, - .get_local_mem_info = get_local_mem_info, - .get_gpu_clock_counter = get_gpu_clock_counter, - .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz, - .alloc_pasid = amdgpu_pasid_alloc, - .free_pasid = amdgpu_pasid_free, .program_sh_mem_settings = kgd_program_sh_mem_settings, .set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping, .init_interrupts = kgd_init_interrupts, @@ -200,28 +194,10 @@ static const struct kfd2kgd_calls kfd2kgd = { .get_fw_version = get_fw_version, .set_scratch_backing_va = set_scratch_backing_va, .get_tile_config = get_tile_config, - .get_cu_info = get_cu_info, - .get_vram_usage = amdgpu_amdkfd_get_vram_usage, - .create_process_vm = amdgpu_amdkfd_gpuvm_create_process_vm, - .acquire_process_vm = amdgpu_amdkfd_gpuvm_acquire_process_vm, - .destroy_process_vm = amdgpu_amdkfd_gpuvm_destroy_process_vm, - .release_process_vm = amdgpu_amdkfd_gpuvm_release_process_vm, - .get_process_page_dir = amdgpu_amdkfd_gpuvm_get_process_page_dir, .set_vm_context_page_table_base = set_vm_context_page_table_base, - .alloc_memory_of_gpu = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu, - .free_memory_of_gpu = amdgpu_amdkfd_gpuvm_free_memory_of_gpu, - .map_memory_to_gpu = amdgpu_amdkfd_gpuvm_map_memory_to_gpu, - .unmap_memory_to_gpu = amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu, - .sync_memory = amdgpu_amdkfd_gpuvm_sync_memory, - .map_gtt_bo_to_kernel = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel, - .restore_process_bos = amdgpu_amdkfd_gpuvm_restore_process_bos, .invalidate_tlbs = invalidate_tlbs, .invalidate_tlbs_vmid = invalidate_tlbs_vmid, - .submit_ib = amdgpu_amdkfd_submit_ib, - .get_vm_fault_info = amdgpu_amdkfd_gpuvm_get_vm_fault_info, .read_vmid_from_vmfault_reg = read_vmid_from_vmfault_reg, - .gpu_recover = amdgpu_amdkfd_gpu_reset, - .set_compute_idle = amdgpu_amdkfd_set_compute_idle }; struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c index 9f149914ad6c..56ea929f524b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c @@ -24,6 +24,7 @@ #include <linux/fdtable.h> #include <linux/uaccess.h> #include <linux/firmware.h> +#include <linux/mmu_context.h> #include <drm/drmP.h> #include "amdgpu.h" #include "amdgpu_amdkfd.h" @@ -128,13 +129,6 @@ static int get_tile_config(struct kgd_dev *kgd, } static const struct kfd2kgd_calls kfd2kgd = { - .init_gtt_mem_allocation = alloc_gtt_mem, - .free_gtt_mem = free_gtt_mem, - .get_local_mem_info = get_local_mem_info, - .get_gpu_clock_counter = get_gpu_clock_counter, - .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz, - .alloc_pasid = amdgpu_pasid_alloc, - .free_pasid = amdgpu_pasid_free, .program_sh_mem_settings = kgd_program_sh_mem_settings, .set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping, .init_interrupts = kgd_init_interrupts, @@ -157,27 +151,9 @@ static const struct kfd2kgd_calls kfd2kgd = { .get_fw_version = get_fw_version, .set_scratch_backing_va = set_scratch_backing_va, .get_tile_config = get_tile_config, - .get_cu_info = get_cu_info, - .get_vram_usage = amdgpu_amdkfd_get_vram_usage, - .create_process_vm = amdgpu_amdkfd_gpuvm_create_process_vm, - .acquire_process_vm = amdgpu_amdkfd_gpuvm_acquire_process_vm, - .destroy_process_vm = amdgpu_amdkfd_gpuvm_destroy_process_vm, - .release_process_vm = amdgpu_amdkfd_gpuvm_release_process_vm, - .get_process_page_dir = amdgpu_amdkfd_gpuvm_get_process_page_dir, .set_vm_context_page_table_base = set_vm_context_page_table_base, - .alloc_memory_of_gpu = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu, - .free_memory_of_gpu = amdgpu_amdkfd_gpuvm_free_memory_of_gpu, - .map_memory_to_gpu = amdgpu_amdkfd_gpuvm_map_memory_to_gpu, - .unmap_memory_to_gpu = amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu, - .sync_memory = amdgpu_amdkfd_gpuvm_sync_memory, - .map_gtt_bo_to_kernel = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel, - .restore_process_bos = amdgpu_amdkfd_gpuvm_restore_process_bos, .invalidate_tlbs = invalidate_tlbs, .invalidate_tlbs_vmid = invalidate_tlbs_vmid, - .submit_ib = amdgpu_amdkfd_submit_ib, - .get_vm_fault_info = amdgpu_amdkfd_gpuvm_get_vm_fault_info, - .gpu_recover = amdgpu_amdkfd_gpu_reset, - .set_compute_idle = amdgpu_amdkfd_set_compute_idle }; struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c index 42cb4c4e0929..5c51d4910650 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c @@ -26,6 +26,7 @@ #include <linux/fdtable.h> #include <linux/uaccess.h> #include <linux/firmware.h> +#include <linux/mmu_context.h> #include <drm/drmP.h> #include "amdgpu.h" #include "amdgpu_amdkfd.h" @@ -46,38 +47,9 @@ #include "v9_structs.h" #include "soc15.h" #include "soc15d.h" +#include "mmhub_v1_0.h" +#include "gfxhub_v1_0.h" -/* HACK: MMHUB and GC both have VM-related register with the same - * names but different offsets. Define the MMHUB register we need here - * with a prefix. A proper solution would be to move the functions - * programming these registers into gfx_v9_0.c and mmhub_v1_0.c - * respectively. - */ -#define mmMMHUB_VM_INVALIDATE_ENG16_REQ 0x06f3 -#define mmMMHUB_VM_INVALIDATE_ENG16_REQ_BASE_IDX 0 - -#define mmMMHUB_VM_INVALIDATE_ENG16_ACK 0x0705 -#define mmMMHUB_VM_INVALIDATE_ENG16_ACK_BASE_IDX 0 - -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32 0x072b -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32_BASE_IDX 0 -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32 0x072c -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32_BASE_IDX 0 - -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32 0x074b -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32_BASE_IDX 0 -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32 0x074c -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32_BASE_IDX 0 - -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32 0x076b -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32_BASE_IDX 0 -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32 0x076c -#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32_BASE_IDX 0 - -#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32 0x0727 -#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32_BASE_IDX 0 -#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32 0x0728 -#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32_BASE_IDX 0 #define V9_PIPE_PER_MEC (4) #define V9_QUEUES_PER_PIPE_MEC (8) @@ -167,13 +139,6 @@ static int amdgpu_amdkfd_get_tile_config(struct kgd_dev *kgd, } static const struct kfd2kgd_calls kfd2kgd = { - .init_gtt_mem_allocation = alloc_gtt_mem, - .free_gtt_mem = free_gtt_mem, - .get_local_mem_info = get_local_mem_info, - .get_gpu_clock_counter = get_gpu_clock_counter, - .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz, - .alloc_pasid = amdgpu_pasid_alloc, - .free_pasid = amdgpu_pasid_free, .program_sh_mem_settings = kgd_program_sh_mem_settings, .set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping, .init_interrupts = kgd_init_interrupts, @@ -196,26 +161,9 @@ static const struct kfd2kgd_calls kfd2kgd = { .get_fw_version = get_fw_version, .set_scratch_backing_va = set_scratch_backing_va, .get_tile_config = amdgpu_amdkfd_get_tile_config, - .get_cu_info = get_cu_info, - .get_vram_usage = amdgpu_amdkfd_get_vram_usage, - .create_process_vm = amdgpu_amdkfd_gpuvm_create_process_vm, - .acquire_process_vm = amdgpu_amdkfd_gpuvm_acquire_process_vm, - .destroy_process_vm = amdgpu_amdkfd_gpuvm_destroy_process_vm, - .release_process_vm = amdgpu_amdkfd_gpuvm_release_process_vm, - .get_process_page_dir = amdgpu_amdkfd_gpuvm_get_process_page_dir, .set_vm_context_page_table_base = set_vm_context_page_table_base, - .alloc_memory_of_gpu = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu, - .free_memory_of_gpu = amdgpu_amdkfd_gpuvm_free_memory_of_gpu, - .map_memory_to_gpu = amdgpu_amdkfd_gpuvm_map_memory_to_gpu, - .unmap_memory_to_gpu = amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu, - .sync_memory = amdgpu_amdkfd_gpuvm_sync_memory, - .map_gtt_bo_to_kernel = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel, - .restore_process_bos = amdgpu_amdkfd_gpuvm_restore_process_bos, .invalidate_tlbs = invalidate_tlbs, .invalidate_tlbs_vmid = invalidate_tlbs_vmid, - .submit_ib = amdgpu_amdkfd_submit_ib, - .gpu_recover = amdgpu_amdkfd_gpu_reset, - .set_compute_idle = amdgpu_amdkfd_set_compute_idle, .get_hive_id = amdgpu_amdkfd_get_hive_id, }; @@ -785,15 +733,6 @@ static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid) { struct amdgpu_device *adev = (struct amdgpu_device *) kgd; - uint32_t req = (1 << vmid) | - (0 << VM_INVALIDATE_ENG16_REQ__FLUSH_TYPE__SHIFT) | /* legacy */ - VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PTES_MASK | - VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE0_MASK | - VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE1_MASK | - VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE2_MASK | - VM_INVALIDATE_ENG16_REQ__INVALIDATE_L1_PTES_MASK; - - mutex_lock(&adev->srbm_mutex); /* Use legacy mode tlb invalidation. * @@ -810,34 +749,7 @@ static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid) * TODO 2: support range-based invalidation, requires kfg2kgd * interface change */ - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ADDR_RANGE_LO32), - 0xffffffff); - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ADDR_RANGE_HI32), - 0x0000001f); - - WREG32(SOC15_REG_OFFSET(MMHUB, 0, - mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32), - 0xffffffff); - WREG32(SOC15_REG_OFFSET(MMHUB, 0, - mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32), - 0x0000001f); - - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_REQ), req); - - WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_INVALIDATE_ENG16_REQ), - req); - - while (!(RREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ACK)) & - (1 << vmid))) - cpu_relax(); - - while (!(RREG32(SOC15_REG_OFFSET(MMHUB, 0, - mmMMHUB_VM_INVALIDATE_ENG16_ACK)) & - (1 << vmid))) - cpu_relax(); - - mutex_unlock(&adev->srbm_mutex); - + amdgpu_gmc_flush_gpu_tlb(adev, vmid, 0); } static int invalidate_tlbs_with_kiq(struct amdgpu_device *adev, uint16_t pasid) @@ -876,7 +788,7 @@ static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid) if (adev->in_gpu_reset) return -EIO; - if (ring->ready) + if (ring->sched.ready) return invalidate_tlbs_with_kiq(adev, pasid); for (vmid = 0; vmid < 16; vmid++) { @@ -1016,7 +928,6 @@ static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid, uint64_t page_table_base) { struct amdgpu_device *adev = get_amdgpu_device(kgd); - uint64_t base = page_table_base | AMDGPU_PTE_VALID; if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) { pr_err("trying to set page table base for wrong VMID %u\n", @@ -1028,25 +939,7 @@ static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid, * now, all processes share the same address space size, like * on GFX8 and older. */ - WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32) + (vmid*2), 0); - WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32) + (vmid*2), 0); - - WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32) + (vmid*2), - lower_32_bits(adev->vm_manager.max_pfn - 1)); - WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32) + (vmid*2), - upper_32_bits(adev->vm_manager.max_pfn - 1)); - - WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32) + (vmid*2), lower_32_bits(base)); - WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32) + (vmid*2), upper_32_bits(base)); - - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32) + (vmid*2), 0); - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32) + (vmid*2), 0); - - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32) + (vmid*2), - lower_32_bits(adev->vm_manager.max_pfn - 1)); - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32) + (vmid*2), - upper_32_bits(adev->vm_manager.max_pfn - 1)); + mmhub_v1_0_setup_vm_pt_regs(adev, vmid, page_table_base); - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32) + (vmid*2), lower_32_bits(base)); - WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32) + (vmid*2), upper_32_bits(base)); + gfxhub_v1_0_setup_vm_pt_regs(adev, vmid, page_table_base); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index df0a059565f9..1921dec3df7a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -25,6 +25,7 @@ #include <linux/list.h> #include <linux/pagemap.h> #include <linux/sched/mm.h> +#include <linux/dma-buf.h> #include <drm/drmP.h> #include "amdgpu_object.h" #include "amdgpu_vm.h" @@ -46,9 +47,9 @@ /* Impose limit on how much memory KFD can use */ static struct { uint64_t max_system_mem_limit; - uint64_t max_userptr_mem_limit; + uint64_t max_ttm_mem_limit; int64_t system_mem_used; - int64_t userptr_mem_used; + int64_t ttm_mem_used; spinlock_t mem_limit_lock; } kfd_mem_limit; @@ -90,8 +91,8 @@ static bool check_if_add_bo_to_vm(struct amdgpu_vm *avm, } /* Set memory usage limits. Current, limits are - * System (kernel) memory - 3/8th System RAM - * Userptr memory - 3/4th System RAM + * System (TTM + userptr) memory - 3/4th System RAM + * TTM memory - 3/8th System RAM */ void amdgpu_amdkfd_gpuvm_init_mem_limits(void) { @@ -103,48 +104,61 @@ void amdgpu_amdkfd_gpuvm_init_mem_limits(void) mem *= si.mem_unit; spin_lock_init(&kfd_mem_limit.mem_limit_lock); - kfd_mem_limit.max_system_mem_limit = (mem >> 1) - (mem >> 3); - kfd_mem_limit.max_userptr_mem_limit = mem - (mem >> 2); - pr_debug("Kernel memory limit %lluM, userptr limit %lluM\n", + kfd_mem_limit.max_system_mem_limit = (mem >> 1) + (mem >> 2); + kfd_mem_limit.max_ttm_mem_limit = (mem >> 1) - (mem >> 3); + pr_debug("Kernel memory limit %lluM, TTM limit %lluM\n", (kfd_mem_limit.max_system_mem_limit >> 20), - (kfd_mem_limit.max_userptr_mem_limit >> 20)); + (kfd_mem_limit.max_ttm_mem_limit >> 20)); } -static int amdgpu_amdkfd_reserve_system_mem_limit(struct amdgpu_device *adev, - uint64_t size, u32 domain) +static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev, + uint64_t size, u32 domain, bool sg) { - size_t acc_size; + size_t acc_size, system_mem_needed, ttm_mem_needed, vram_needed; + uint64_t reserved_for_pt = amdgpu_amdkfd_total_mem_size >> 9; int ret = 0; acc_size = ttm_bo_dma_acc_size(&adev->mman.bdev, size, sizeof(struct amdgpu_bo)); - spin_lock(&kfd_mem_limit.mem_limit_lock); + vram_needed = 0; if (domain == AMDGPU_GEM_DOMAIN_GTT) { - if (kfd_mem_limit.system_mem_used + (acc_size + size) > - kfd_mem_limit.max_system_mem_limit) { - ret = -ENOMEM; - goto err_no_mem; - } - kfd_mem_limit.system_mem_used += (acc_size + size); - } else if (domain == AMDGPU_GEM_DOMAIN_CPU) { - if ((kfd_mem_limit.system_mem_used + acc_size > - kfd_mem_limit.max_system_mem_limit) || - (kfd_mem_limit.userptr_mem_used + (size + acc_size) > - kfd_mem_limit.max_userptr_mem_limit)) { - ret = -ENOMEM; - goto err_no_mem; - } - kfd_mem_limit.system_mem_used += acc_size; - kfd_mem_limit.userptr_mem_used += size; + /* TTM GTT memory */ + system_mem_needed = acc_size + size; + ttm_mem_needed = acc_size + size; + } else if (domain == AMDGPU_GEM_DOMAIN_CPU && !sg) { + /* Userptr */ + system_mem_needed = acc_size + size; + ttm_mem_needed = acc_size; + } else { + /* VRAM and SG */ + system_mem_needed = acc_size; + ttm_mem_needed = acc_size; + if (domain == AMDGPU_GEM_DOMAIN_VRAM) + vram_needed = size; + } + + spin_lock(&kfd_mem_limit.mem_limit_lock); + + if ((kfd_mem_limit.system_mem_used + system_mem_needed > + kfd_mem_limit.max_system_mem_limit) || + (kfd_mem_limit.ttm_mem_used + ttm_mem_needed > + kfd_mem_limit.max_ttm_mem_limit) || + (adev->kfd.vram_used + vram_needed > + adev->gmc.real_vram_size - reserved_for_pt)) { + ret = -ENOMEM; + } else { + kfd_mem_limit.system_mem_used += system_mem_needed; + kfd_mem_limit.ttm_mem_used += ttm_mem_needed; + adev->kfd.vram_used += vram_needed; } -err_no_mem: + spin_unlock(&kfd_mem_limit.mem_limit_lock); return ret; } -static void unreserve_system_mem_limit(struct amdgpu_device *adev, - uint64_t size, u32 domain) +static void unreserve_mem_limit(struct amdgpu_device *adev, + uint64_t size, u32 domain, bool sg) { size_t acc_size; @@ -154,70 +168,61 @@ static void unreserve_system_mem_limit(struct amdgpu_device *adev, spin_lock(&kfd_mem_limit.mem_limit_lock); if (domain == AMDGPU_GEM_DOMAIN_GTT) { kfd_mem_limit.system_mem_used -= (acc_size + size); - } else if (domain == AMDGPU_GEM_DOMAIN_CPU) { + kfd_mem_limit.ttm_mem_used -= (acc_size + size); + } else if (domain == AMDGPU_GEM_DOMAIN_CPU && !sg) { + kfd_mem_limit.system_mem_used -= (acc_size + size); + kfd_mem_limit.ttm_mem_used -= acc_size; + } else { kfd_mem_limit.system_mem_used -= acc_size; - kfd_mem_limit.userptr_mem_used -= size; + kfd_mem_limit.ttm_mem_used -= acc_size; + if (domain == AMDGPU_GEM_DOMAIN_VRAM) { + adev->kfd.vram_used -= size; + WARN_ONCE(adev->kfd.vram_used < 0, + "kfd VRAM memory accounting unbalanced"); + } } WARN_ONCE(kfd_mem_limit.system_mem_used < 0, "kfd system memory accounting unbalanced"); - WARN_ONCE(kfd_mem_limit.userptr_mem_used < 0, - "kfd userptr memory accounting unbalanced"); + WARN_ONCE(kfd_mem_limit.ttm_mem_used < 0, + "kfd TTM memory accounting unbalanced"); spin_unlock(&kfd_mem_limit.mem_limit_lock); } -void amdgpu_amdkfd_unreserve_system_memory_limit(struct amdgpu_bo *bo) +void amdgpu_amdkfd_unreserve_memory_limit(struct amdgpu_bo *bo) { - spin_lock(&kfd_mem_limit.mem_limit_lock); + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); + u32 domain = bo->preferred_domains; + bool sg = (bo->preferred_domains == AMDGPU_GEM_DOMAIN_CPU); if (bo->flags & AMDGPU_AMDKFD_USERPTR_BO) { - kfd_mem_limit.system_mem_used -= bo->tbo.acc_size; - kfd_mem_limit.userptr_mem_used -= amdgpu_bo_size(bo); - } else if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT) { - kfd_mem_limit.system_mem_used -= - (bo->tbo.acc_size + amdgpu_bo_size(bo)); + domain = AMDGPU_GEM_DOMAIN_CPU; + sg = false; } - WARN_ONCE(kfd_mem_limit.system_mem_used < 0, - "kfd system memory accounting unbalanced"); - WARN_ONCE(kfd_mem_limit.userptr_mem_used < 0, - "kfd userptr memory accounting unbalanced"); - spin_unlock(&kfd_mem_limit.mem_limit_lock); + unreserve_mem_limit(adev, amdgpu_bo_size(bo), domain, sg); } -/* amdgpu_amdkfd_remove_eviction_fence - Removes eviction fence(s) from BO's +/* amdgpu_amdkfd_remove_eviction_fence - Removes eviction fence from BO's * reservation object. * * @bo: [IN] Remove eviction fence(s) from this BO - * @ef: [IN] If ef is specified, then this eviction fence is removed if it + * @ef: [IN] This eviction fence is removed if it * is present in the shared list. - * @ef_list: [OUT] Returns list of eviction fences. These fences are removed - * from BO's reservation object shared list. - * @ef_count: [OUT] Number of fences in ef_list. * - * NOTE: If called with ef_list, then amdgpu_amdkfd_add_eviction_fence must be - * called to restore the eviction fences and to avoid memory leak. This is - * useful for shared BOs. * NOTE: Must be called with BO reserved i.e. bo->tbo.resv->lock held. */ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, - struct amdgpu_amdkfd_fence *ef, - struct amdgpu_amdkfd_fence ***ef_list, - unsigned int *ef_count) + struct amdgpu_amdkfd_fence *ef) { struct reservation_object *resv = bo->tbo.resv; struct reservation_object_list *old, *new; unsigned int i, j, k; - if (!ef && !ef_list) + if (!ef) return -EINVAL; - if (ef_list) { - *ef_list = NULL; - *ef_count = 0; - } - old = reservation_object_get_list(resv); if (!old) return 0; @@ -236,8 +241,7 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, f = rcu_dereference_protected(old->shared[i], reservation_object_held(resv)); - if ((ef && f->context == ef->base.context) || - (!ef && to_amdgpu_amdkfd_fence(f))) + if (f->context == ef->base.context) RCU_INIT_POINTER(new->shared[--j], f); else RCU_INIT_POINTER(new->shared[k++], f); @@ -245,21 +249,6 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, new->shared_max = old->shared_max; new->shared_count = k; - if (!ef) { - unsigned int count = old->shared_count - j; - - /* Alloc memory for count number of eviction fence pointers. - * Fill the ef_list array and ef_count - */ - *ef_list = kcalloc(count, sizeof(**ef_list), GFP_KERNEL); - *ef_count = count; - - if (!*ef_list) { - kfree(new); - return -ENOMEM; - } - } - /* Install the new fence list, seqcount provides the barriers */ preempt_disable(); write_seqcount_begin(&resv->seq); @@ -273,46 +262,13 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, f = rcu_dereference_protected(new->shared[i], reservation_object_held(resv)); - if (!ef) - (*ef_list)[k++] = to_amdgpu_amdkfd_fence(f); - else - dma_fence_put(f); + dma_fence_put(f); } kfree_rcu(old, rcu); return 0; } -/* amdgpu_amdkfd_add_eviction_fence - Adds eviction fence(s) back into BO's - * reservation object. - * - * @bo: [IN] Add eviction fences to this BO - * @ef_list: [IN] List of eviction fences to be added - * @ef_count: [IN] Number of fences in ef_list. - * - * NOTE: Must call amdgpu_amdkfd_remove_eviction_fence before calling this - * function. - */ -static void amdgpu_amdkfd_add_eviction_fence(struct amdgpu_bo *bo, - struct amdgpu_amdkfd_fence **ef_list, - unsigned int ef_count) -{ - int i; - - if (!ef_list || !ef_count) - return; - - for (i = 0; i < ef_count; i++) { - amdgpu_bo_fence(bo, &ef_list[i]->base, true); - /* Re-adding the fence takes an additional reference. Drop that - * reference. - */ - dma_fence_put(&ef_list[i]->base); - } - - kfree(ef_list); -} - static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain, bool wait) { @@ -328,18 +284,8 @@ static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain, ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); if (ret) goto validate_fail; - if (wait) { - struct amdgpu_amdkfd_fence **ef_list; - unsigned int ef_count; - - ret = amdgpu_amdkfd_remove_eviction_fence(bo, NULL, &ef_list, - &ef_count); - if (ret) - goto validate_fail; - - ttm_bo_wait(&bo->tbo, false, false); - amdgpu_amdkfd_add_eviction_fence(bo, ef_list, ef_count); - } + if (wait) + amdgpu_bo_sync_wait(bo, AMDGPU_FENCE_OWNER_KFD, false); validate_fail: return ret; @@ -395,23 +341,6 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm) return 0; } -static int sync_vm_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync, - struct dma_fence *f) -{ - int ret = amdgpu_sync_fence(adev, sync, f, false); - - /* Sync objects can't handle multiple GPUs (contexts) updating - * sync->last_vm_update. Fortunately we don't need it for - * KFD's purposes, so we can just drop that fence. - */ - if (sync->last_vm_update) { - dma_fence_put(sync->last_vm_update); - sync->last_vm_update = NULL; - } - - return ret; -} - static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync) { struct amdgpu_bo *pd = vm->root.base.bo; @@ -422,7 +351,7 @@ static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync) if (ret) return ret; - return sync_vm_fence(adev, sync, vm->last_update); + return amdgpu_sync_fence(NULL, sync, vm->last_update, false); } /* add_bo_to_vm - Add a BO to a VM @@ -443,7 +372,6 @@ static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, { int ret; struct kfd_bo_va_list *bo_va_entry; - struct amdgpu_bo *pd = vm->root.base.bo; struct amdgpu_bo *bo = mem->bo; uint64_t va = mem->va; struct list_head *list_bo_va = &mem->bo_va_list; @@ -483,14 +411,8 @@ static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, *p_bo_va_entry = bo_va_entry; /* Allocate new page tables if needed and validate - * them. Clearing of new page tables and validate need to wait - * on move fences. We don't want that to trigger the eviction - * fence, so remove it temporarily. + * them. */ - amdgpu_amdkfd_remove_eviction_fence(pd, - vm->process_info->eviction_fence, - NULL, NULL); - ret = amdgpu_vm_alloc_pts(adev, vm, va, amdgpu_bo_size(bo)); if (ret) { pr_err("Failed to allocate pts, err=%d\n", ret); @@ -503,13 +425,9 @@ static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, goto err_alloc_pts; } - /* Add the eviction fence back */ - amdgpu_bo_fence(pd, &vm->process_info->eviction_fence->base, true); - return 0; err_alloc_pts: - amdgpu_bo_fence(pd, &vm->process_info->eviction_fence->base, true); amdgpu_vm_bo_rmv(adev, bo_va_entry->bo_va); list_del(&bo_va_entry->bo_list); err_vmadd: @@ -536,7 +454,7 @@ static void add_kgd_mem_to_kfd_bo_list(struct kgd_mem *mem, struct amdgpu_bo *bo = mem->bo; INIT_LIST_HEAD(&entry->head); - entry->shared = true; + entry->num_shared = 1; entry->bo = &bo->tbo; mutex_lock(&process_info->lock); if (userptr) @@ -677,7 +595,7 @@ static int reserve_bo_and_vm(struct kgd_mem *mem, ctx->kfd_bo.priority = 0; ctx->kfd_bo.tv.bo = &bo->tbo; - ctx->kfd_bo.tv.shared = true; + ctx->kfd_bo.tv.num_shared = 1; ctx->kfd_bo.user_pages = NULL; list_add(&ctx->kfd_bo.tv.head, &ctx->list); @@ -741,7 +659,7 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem, ctx->kfd_bo.priority = 0; ctx->kfd_bo.tv.bo = &bo->tbo; - ctx->kfd_bo.tv.shared = true; + ctx->kfd_bo.tv.num_shared = 1; ctx->kfd_bo.user_pages = NULL; list_add(&ctx->kfd_bo.tv.head, &ctx->list); @@ -808,25 +726,12 @@ static int unmap_bo_from_gpuvm(struct amdgpu_device *adev, { struct amdgpu_bo_va *bo_va = entry->bo_va; struct amdgpu_vm *vm = bo_va->base.vm; - struct amdgpu_bo *pd = vm->root.base.bo; - /* Remove eviction fence from PD (and thereby from PTs too as - * they share the resv. object). Otherwise during PT update - * job (see amdgpu_vm_bo_update_mapping), eviction fence would - * get added to job->sync object and job execution would - * trigger the eviction fence. - */ - amdgpu_amdkfd_remove_eviction_fence(pd, - vm->process_info->eviction_fence, - NULL, NULL); amdgpu_vm_bo_unmap(adev, bo_va, entry->va); amdgpu_vm_clear_freed(adev, vm, &bo_va->last_pt_update); - /* Add the eviction fence back */ - amdgpu_bo_fence(pd, &vm->process_info->eviction_fence->base, true); - - sync_vm_fence(adev, sync, bo_va->last_pt_update); + amdgpu_sync_fence(NULL, sync, bo_va->last_pt_update, false); return 0; } @@ -851,7 +756,7 @@ static int update_gpuvm_pte(struct amdgpu_device *adev, return ret; } - return sync_vm_fence(adev, sync, bo_va->last_pt_update); + return amdgpu_sync_fence(NULL, sync, bo_va->last_pt_update, false); } static int map_bo_to_gpuvm(struct amdgpu_device *adev, @@ -886,6 +791,24 @@ update_gpuvm_pte_failed: return ret; } +static struct sg_table *create_doorbell_sg(uint64_t addr, uint32_t size) +{ + struct sg_table *sg = kmalloc(sizeof(*sg), GFP_KERNEL); + + if (!sg) + return NULL; + if (sg_alloc_table(sg, 1, GFP_KERNEL)) { + kfree(sg); + return NULL; + } + sg->sgl->dma_address = addr; + sg->sgl->length = size; +#ifdef CONFIG_NEED_SG_DMA_LENGTH + sg->sgl->dma_length = size; +#endif + return sg; +} + static int process_validate_vms(struct amdkfd_process_info *process_info) { struct amdgpu_vm *peer_vm; @@ -901,6 +824,26 @@ static int process_validate_vms(struct amdkfd_process_info *process_info) return 0; } +static int process_sync_pds_resv(struct amdkfd_process_info *process_info, + struct amdgpu_sync *sync) +{ + struct amdgpu_vm *peer_vm; + int ret; + + list_for_each_entry(peer_vm, &process_info->vm_list_head, + vm_list_node) { + struct amdgpu_bo *pd = peer_vm->root.base.bo; + + ret = amdgpu_sync_resv(NULL, + sync, pd->tbo.resv, + AMDGPU_FENCE_OWNER_UNDEFINED, false); + if (ret) + return ret; + } + + return 0; +} + static int process_update_pds(struct amdkfd_process_info *process_info, struct amdgpu_sync *sync) { @@ -963,7 +906,7 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info, pr_err("validate_pt_pd_bos() failed\n"); goto validate_pd_fail; } - ret = ttm_bo_wait(&vm->root.base.bo->tbo, false, false); + amdgpu_bo_sync_wait(vm->root.base.bo, AMDGPU_FENCE_OWNER_KFD, false); if (ret) goto wait_pd_fail; amdgpu_bo_fence(vm->root.base.bo, @@ -1149,6 +1092,8 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( { struct amdgpu_device *adev = get_amdgpu_device(kgd); struct amdgpu_vm *avm = (struct amdgpu_vm *)vm; + enum ttm_bo_type bo_type = ttm_bo_type_device; + struct sg_table *sg = NULL; uint64_t user_addr = 0; struct amdgpu_bo *bo; struct amdgpu_bo_param bp; @@ -1177,13 +1122,25 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( if (!offset || !*offset) return -EINVAL; user_addr = *offset; + } else if (flags & ALLOC_MEM_FLAGS_DOORBELL) { + domain = AMDGPU_GEM_DOMAIN_GTT; + alloc_domain = AMDGPU_GEM_DOMAIN_CPU; + bo_type = ttm_bo_type_sg; + alloc_flags = 0; + if (size > UINT_MAX) + return -EINVAL; + sg = create_doorbell_sg(*offset, size); + if (!sg) + return -ENOMEM; } else { return -EINVAL; } *mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL); - if (!*mem) - return -ENOMEM; + if (!*mem) { + ret = -ENOMEM; + goto err; + } INIT_LIST_HEAD(&(*mem)->bo_va_list); mutex_init(&(*mem)->lock); (*mem)->aql_queue = !!(flags & ALLOC_MEM_FLAGS_AQL_QUEUE_MEM); @@ -1199,7 +1156,8 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( byte_align = (adev->family == AMDGPU_FAMILY_VI && adev->asic_type != CHIP_FIJI && adev->asic_type != CHIP_POLARIS10 && - adev->asic_type != CHIP_POLARIS11) ? + adev->asic_type != CHIP_POLARIS11 && + adev->asic_type != CHIP_POLARIS12) ? VI_BO_SIZE_ALIGN : 1; mapping_flags = AMDGPU_VM_PAGE_READABLE; @@ -1215,10 +1173,10 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( amdgpu_sync_create(&(*mem)->sync); - ret = amdgpu_amdkfd_reserve_system_mem_limit(adev, size, alloc_domain); + ret = amdgpu_amdkfd_reserve_mem_limit(adev, size, alloc_domain, !!sg); if (ret) { pr_debug("Insufficient system memory\n"); - goto err_reserve_system_mem; + goto err_reserve_limit; } pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s\n", @@ -1229,7 +1187,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( bp.byte_align = byte_align; bp.domain = alloc_domain; bp.flags = alloc_flags; - bp.type = ttm_bo_type_device; + bp.type = bo_type; bp.resv = NULL; ret = amdgpu_bo_create(adev, &bp, &bo); if (ret) { @@ -1237,6 +1195,10 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( domain_string(alloc_domain), ret); goto err_bo_create; } + if (bo_type == ttm_bo_type_sg) { + bo->tbo.sg = sg; + bo->tbo.ttm->sg = sg; + } bo->kfd_bo = *mem; (*mem)->bo = bo; if (user_addr) @@ -1266,12 +1228,17 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( allocate_init_user_pages_failed: amdgpu_bo_unref(&bo); /* Don't unreserve system mem limit twice */ - goto err_reserve_system_mem; + goto err_reserve_limit; err_bo_create: - unreserve_system_mem_limit(adev, size, alloc_domain); -err_reserve_system_mem: + unreserve_mem_limit(adev, size, alloc_domain, !!sg); +err_reserve_limit: mutex_destroy(&(*mem)->lock); kfree(*mem); +err: + if (sg) { + sg_free_table(sg); + kfree(sg); + } return ret; } @@ -1326,8 +1293,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( * attached */ amdgpu_amdkfd_remove_eviction_fence(mem->bo, - process_info->eviction_fence, - NULL, NULL); + process_info->eviction_fence); pr_debug("Release VA 0x%llx - 0x%llx\n", mem->va, mem->va + bo_size * (1 + mem->aql_queue)); @@ -1341,6 +1307,14 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( /* Free the sync object */ amdgpu_sync_free(&mem->sync); + /* If the SG is not NULL, it's one we created for a doorbell + * BO. We need to free it. + */ + if (mem->bo->tbo.sg) { + sg_free_table(mem->bo->tbo.sg); + kfree(mem->bo->tbo.sg); + } + /* Free the BO*/ amdgpu_bo_unref(&mem->bo); mutex_destroy(&mem->lock); @@ -1405,7 +1379,8 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( * the queues are still stopped and we can leave mapping for * the next restore worker */ - if (bo->tbo.mem.mem_type == TTM_PL_SYSTEM) + if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) && + bo->tbo.mem.mem_type == TTM_PL_SYSTEM) is_invalid_userptr = true; if (check_if_add_bo_to_vm(avm, mem)) { @@ -1545,8 +1520,7 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu( if (mem->mapped_to_gpu_memory == 0 && !amdgpu_ttm_tt_get_usermm(mem->bo->tbo.ttm) && !mem->bo->pin_count) amdgpu_amdkfd_remove_eviction_fence(mem->bo, - process_info->eviction_fence, - NULL, NULL); + process_info->eviction_fence); unreserve_out: unreserve_bo_and_vms(&ctx, false, false); @@ -1607,7 +1581,7 @@ int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_dev *kgd, } amdgpu_amdkfd_remove_eviction_fence( - bo, mem->process_info->eviction_fence, NULL, NULL); + bo, mem->process_info->eviction_fence); list_del_init(&mem->validate_list.head); if (size) @@ -1642,6 +1616,60 @@ int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct kgd_dev *kgd, return 0; } +int amdgpu_amdkfd_gpuvm_import_dmabuf(struct kgd_dev *kgd, + struct dma_buf *dma_buf, + uint64_t va, void *vm, + struct kgd_mem **mem, uint64_t *size, + uint64_t *mmap_offset) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)kgd; + struct drm_gem_object *obj; + struct amdgpu_bo *bo; + struct amdgpu_vm *avm = (struct amdgpu_vm *)vm; + + if (dma_buf->ops != &amdgpu_dmabuf_ops) + /* Can't handle non-graphics buffers */ + return -EINVAL; + + obj = dma_buf->priv; + if (obj->dev->dev_private != adev) + /* Can't handle buffers from other devices */ + return -EINVAL; + + bo = gem_to_amdgpu_bo(obj); + if (!(bo->preferred_domains & (AMDGPU_GEM_DOMAIN_VRAM | + AMDGPU_GEM_DOMAIN_GTT))) + /* Only VRAM and GTT BOs are supported */ + return -EINVAL; + + *mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL); + if (!*mem) + return -ENOMEM; + + if (size) + *size = amdgpu_bo_size(bo); + + if (mmap_offset) + *mmap_offset = amdgpu_bo_mmap_offset(bo); + + INIT_LIST_HEAD(&(*mem)->bo_va_list); + mutex_init(&(*mem)->lock); + (*mem)->mapping_flags = + AMDGPU_VM_PAGE_READABLE | AMDGPU_VM_PAGE_WRITEABLE | + AMDGPU_VM_PAGE_EXECUTABLE | AMDGPU_VM_MTYPE_NC; + + (*mem)->bo = amdgpu_bo_ref(bo); + (*mem)->va = va; + (*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) ? + AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT; + (*mem)->mapped_to_gpu_memory = 0; + (*mem)->process_info = avm->process_info; + add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, false); + amdgpu_sync_create(&(*mem)->sync); + + return 0; +} + /* Evict a userptr BO by stopping the queues if necessary * * Runs in MMU notifier, may be in RECLAIM_FS context. This means it @@ -1664,7 +1692,7 @@ int amdgpu_amdkfd_evict_userptr(struct kgd_mem *mem, evicted_bos = atomic_inc_return(&process_info->evicted_bos); if (evicted_bos == 1) { /* First eviction, stop the queues */ - r = kgd2kfd->quiesce_mm(mm); + r = kgd2kfd_quiesce_mm(mm); if (r) pr_err("Failed to quiesce KFD\n"); schedule_delayed_work(&process_info->restore_userptr_work, @@ -1808,7 +1836,7 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) validate_list.head) { list_add_tail(&mem->resv_list.head, &resv_list); mem->resv_list.bo = mem->validate_list.bo; - mem->resv_list.shared = mem->validate_list.shared; + mem->resv_list.num_shared = mem->validate_list.num_shared; } /* Reserve all BOs and page tables for validation */ @@ -1819,16 +1847,6 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) amdgpu_sync_create(&sync); - /* Avoid triggering eviction fences when unmapping invalid - * userptr BOs (waits for all fences, doesn't use - * FENCE_OWNER_VM) - */ - list_for_each_entry(peer_vm, &process_info->vm_list_head, - vm_list_node) - amdgpu_amdkfd_remove_eviction_fence(peer_vm->root.base.bo, - process_info->eviction_fence, - NULL, NULL); - ret = process_validate_vms(process_info); if (ret) goto unreserve_out; @@ -1889,10 +1907,6 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) ret = process_update_pds(process_info, &sync); unreserve_out: - list_for_each_entry(peer_vm, &process_info->vm_list_head, - vm_list_node) - amdgpu_bo_fence(peer_vm->root.base.bo, - &process_info->eviction_fence->base, true); ttm_eu_backoff_reservation(&ticket, &resv_list); amdgpu_sync_wait(&sync, false); amdgpu_sync_free(&sync); @@ -1956,7 +1970,7 @@ static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work) evicted_bos) goto unlock_out; evicted_bos = 0; - if (kgd2kfd->resume_mm(mm)) { + if (kgd2kfd_resume_mm(mm)) { pr_err("%s: Failed to resume KFD\n", __func__); /* No recovery from this failure. Probably the CP is * hanging. No point trying again. @@ -2027,7 +2041,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef) list_add_tail(&mem->resv_list.head, &ctx.list); mem->resv_list.bo = mem->validate_list.bo; - mem->resv_list.shared = mem->validate_list.shared; + mem->resv_list.num_shared = mem->validate_list.num_shared; } ret = ttm_eu_reserve_buffers(&ctx.ticket, &ctx.list, @@ -2044,13 +2058,10 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef) if (ret) goto validate_map_fail; - /* Wait for PD/PTs validate to finish */ - /* FIXME: I think this isn't needed */ - list_for_each_entry(peer_vm, &process_info->vm_list_head, - vm_list_node) { - struct amdgpu_bo *bo = peer_vm->root.base.bo; - - ttm_bo_wait(&bo->tbo, false, false); + ret = process_sync_pds_resv(process_info, &sync_obj); + if (ret) { + pr_debug("Memory eviction: Failed to sync to PD BO moving fence. Try again\n"); + goto validate_map_fail; } /* Validate BOs and map them to GPUVM (update VM page tables). */ @@ -2066,7 +2077,11 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef) pr_debug("Memory eviction: Validate BOs failed. Try again\n"); goto validate_map_fail; } - + ret = amdgpu_sync_fence(NULL, &sync_obj, bo->tbo.moving, false); + if (ret) { + pr_debug("Memory eviction: Sync BO fence failed. Try again\n"); + goto validate_map_fail; + } list_for_each_entry(bo_va_entry, &mem->bo_va_list, bo_list) { ret = update_gpuvm_pte((struct amdgpu_device *) @@ -2087,6 +2102,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef) goto validate_map_fail; } + /* Wait for validate and PT updates to finish */ amdgpu_sync_wait(&sync_obj, false); /* Release old eviction fence and create new one, because fence only @@ -2105,10 +2121,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef) process_info->eviction_fence = new_fence; *ef = dma_fence_get(&new_fence->base); - /* Wait for validate to finish and attach new eviction fence */ - list_for_each_entry(mem, &process_info->kfd_bo_list, - validate_list.head) - ttm_bo_wait(&mem->bo->tbo, false, false); + /* Attach new eviction fence to all BOs */ list_for_each_entry(mem, &process_info->kfd_bo_list, validate_list.head) amdgpu_bo_fence(mem->bo, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c index a028661d9e20..92b11de19581 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c @@ -576,6 +576,7 @@ static const struct amdgpu_px_quirk amdgpu_px_quirk_list[] = { { 0x1002, 0x6900, 0x1028, 0x0812, AMDGPU_PX_QUIRK_FORCE_ATPX }, { 0x1002, 0x6900, 0x1028, 0x0813, AMDGPU_PX_QUIRK_FORCE_ATPX }, { 0x1002, 0x6900, 0x1025, 0x125A, AMDGPU_PX_QUIRK_FORCE_ATPX }, + { 0x1002, 0x6900, 0x17AA, 0x3806, AMDGPU_PX_QUIRK_FORCE_ATPX }, { 0, 0, 0, 0, 0 }, }; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c index 14d2982a47cc..5c79da8e1150 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c @@ -118,7 +118,6 @@ int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp, entry->priority = min(info[i].bo_priority, AMDGPU_BO_LIST_MAX_PRIORITY); entry->tv.bo = &bo->tbo; - entry->tv.shared = !bo->prime_shared_count; if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GDS) list->gds_obj = bo; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c index 69ad6ec0a4f3..bf04c12bd324 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c @@ -25,8 +25,8 @@ */ #include <drm/drmP.h> #include <drm/drm_edid.h> -#include <drm/drm_crtc_helper.h> #include <drm/drm_fb_helper.h> +#include <drm/drm_probe_helper.h> #include <drm/amdgpu_drm.h> #include "amdgpu.h" #include "atom.h" diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 0acc8dee2cb8..52a5e4fdc95b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -50,7 +50,8 @@ static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p, bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj)); p->uf_entry.priority = 0; p->uf_entry.tv.bo = &bo->tbo; - p->uf_entry.tv.shared = true; + /* One for TTM and one for the CS job */ + p->uf_entry.tv.num_shared = 2; p->uf_entry.user_pages = NULL; drm_gem_object_put_unlocked(gobj); @@ -213,6 +214,7 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, union drm_amdgpu_cs case AMDGPU_CHUNK_ID_DEPENDENCIES: case AMDGPU_CHUNK_ID_SYNCOBJ_IN: case AMDGPU_CHUNK_ID_SYNCOBJ_OUT: + case AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES: break; default: @@ -598,6 +600,10 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, return r; } + /* One for TTM and one for the CS job */ + amdgpu_bo_list_for_each_entry(e, p->bo_list) + e->tv.num_shared = 2; + amdgpu_bo_list_get_list(p->bo_list, &p->validated); if (p->bo_list->first_userptr != p->bo_list->num_entries) p->mn = amdgpu_mn_get(p->adev, AMDGPU_MN_TYPE_GFX); @@ -717,8 +723,14 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, gws = p->bo_list->gws_obj; oa = p->bo_list->oa_obj; - amdgpu_bo_list_for_each_entry(e, p->bo_list) - e->bo_va = amdgpu_vm_bo_find(vm, ttm_to_amdgpu_bo(e->tv.bo)); + amdgpu_bo_list_for_each_entry(e, p->bo_list) { + struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo); + + /* Make sure we use the exclusive slot for shared BOs */ + if (bo->prime_shared_count) + e->tv.num_shared = 0; + e->bo_va = amdgpu_vm_bo_find(vm, bo); + } if (gds) { p->job->gds_base = amdgpu_bo_gpu_offset(gds) >> PAGE_SHIFT; @@ -955,10 +967,6 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p) if (r) return r; - r = reservation_object_reserve_shared(vm->root.base.bo->tbo.resv); - if (r) - return r; - p->job->vm_pd_addr = amdgpu_gmc_pd_addr(vm->root.base.bo); if (amdgpu_vm_debug) { @@ -1083,6 +1091,15 @@ static int amdgpu_cs_process_fence_dep(struct amdgpu_cs_parser *p, fence = amdgpu_ctx_get_fence(ctx, entity, deps[i].handle); + + if (chunk->chunk_id == AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES) { + struct drm_sched_fence *s_fence = to_drm_sched_fence(fence); + struct dma_fence *old = fence; + + fence = dma_fence_get(&s_fence->scheduled); + dma_fence_put(old); + } + if (IS_ERR(fence)) { r = PTR_ERR(fence); amdgpu_ctx_put(ctx); @@ -1104,7 +1121,7 @@ static int amdgpu_syncobj_lookup_and_add_to_sync(struct amdgpu_cs_parser *p, { int r; struct dma_fence *fence; - r = drm_syncobj_find_fence(p->filp, handle, 0, &fence); + r = drm_syncobj_find_fence(p->filp, handle, 0, 0, &fence); if (r) return r; @@ -1170,7 +1187,8 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, chunk = &p->chunks[i]; - if (chunk->chunk_id == AMDGPU_CHUNK_ID_DEPENDENCIES) { + if (chunk->chunk_id == AMDGPU_CHUNK_ID_DEPENDENCIES || + chunk->chunk_id == AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES) { r = amdgpu_cs_process_fence_dep(p, chunk); if (r) return r; @@ -1193,7 +1211,7 @@ static void amdgpu_cs_post_dependencies(struct amdgpu_cs_parser *p) int i; for (i = 0; i < p->num_post_dep_syncobjs; ++i) - drm_syncobj_replace_fence(p->post_dep_syncobjs[i], 0, p->fence); + drm_syncobj_replace_fence(p->post_dep_syncobjs[i], p->fence); } static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, @@ -1260,8 +1278,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, return 0; error_abort: - dma_fence_put(&job->base.s_fence->finished); - job->base.s_fence = NULL; + drm_sched_job_cleanup(&job->base); amdgpu_mn_unlock(p->mn); error_unlock: @@ -1285,7 +1302,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) r = amdgpu_cs_parser_init(&parser, data); if (r) { - DRM_ERROR("Failed to initialize parser !\n"); + DRM_ERROR("Failed to initialize parser %d!\n", r); goto out; } @@ -1422,6 +1439,9 @@ int amdgpu_cs_fence_to_handle_ioctl(struct drm_device *dev, void *data, if (IS_ERR(fence)) return PTR_ERR(fence); + if (!fence) + fence = dma_fence_get_stub(); + switch (info->in.what) { case AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ: r = drm_syncobj_create(&syncobj, 0, fence); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c new file mode 100644 index 000000000000..7e22be7ca68a --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c @@ -0,0 +1,117 @@ +/* + * Copyright 2016 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + + * * Author: Monk.liu@amd.com + */ + +#include "amdgpu.h" + +uint64_t amdgpu_csa_vaddr(struct amdgpu_device *adev) +{ + uint64_t addr = adev->vm_manager.max_pfn << AMDGPU_GPU_PAGE_SHIFT; + + addr -= AMDGPU_VA_RESERVED_SIZE; + addr = amdgpu_gmc_sign_extend(addr); + + return addr; +} + +int amdgpu_allocate_static_csa(struct amdgpu_device *adev, struct amdgpu_bo **bo, + u32 domain, uint32_t size) +{ + int r; + void *ptr; + + r = amdgpu_bo_create_kernel(adev, size, PAGE_SIZE, + domain, bo, + NULL, &ptr); + if (!*bo) + return -ENOMEM; + + memset(ptr, 0, size); + return 0; +} + +void amdgpu_free_static_csa(struct amdgpu_bo **bo) +{ + amdgpu_bo_free_kernel(bo, NULL, NULL); +} + +/* + * amdgpu_map_static_csa should be called during amdgpu_vm_init + * it maps virtual address amdgpu_csa_vaddr() to this VM, and each command + * submission of GFX should use this virtual address within META_DATA init + * package to support SRIOV gfx preemption. + */ +int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm, + struct amdgpu_bo *bo, struct amdgpu_bo_va **bo_va, + uint64_t csa_addr, uint32_t size) +{ + struct ww_acquire_ctx ticket; + struct list_head list; + struct amdgpu_bo_list_entry pd; + struct ttm_validate_buffer csa_tv; + int r; + + INIT_LIST_HEAD(&list); + INIT_LIST_HEAD(&csa_tv.head); + csa_tv.bo = &bo->tbo; + csa_tv.num_shared = 1; + + list_add(&csa_tv.head, &list); + amdgpu_vm_get_pd_bo(vm, &list, &pd); + + r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL); + if (r) { + DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r); + return r; + } + + *bo_va = amdgpu_vm_bo_add(adev, vm, bo); + if (!*bo_va) { + ttm_eu_backoff_reservation(&ticket, &list); + DRM_ERROR("failed to create bo_va for static CSA\n"); + return -ENOMEM; + } + + r = amdgpu_vm_alloc_pts(adev, (*bo_va)->base.vm, csa_addr, + size); + if (r) { + DRM_ERROR("failed to allocate pts for static CSA, err=%d\n", r); + amdgpu_vm_bo_rmv(adev, *bo_va); + ttm_eu_backoff_reservation(&ticket, &list); + return r; + } + + r = amdgpu_vm_bo_map(adev, *bo_va, csa_addr, 0, size, + AMDGPU_PTE_READABLE | AMDGPU_PTE_WRITEABLE | + AMDGPU_PTE_EXECUTABLE); + + if (r) { + DRM_ERROR("failed to do bo_map on static CSA, err=%d\n", r); + amdgpu_vm_bo_rmv(adev, *bo_va); + ttm_eu_backoff_reservation(&ticket, &list); + return r; + } + + ttm_eu_backoff_reservation(&ticket, &list); + return 0; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.h new file mode 100644 index 000000000000..524b4437a021 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.h @@ -0,0 +1,39 @@ +/* + * Copyright 2016 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + * Author: Monk.liu@amd.com + */ + +#ifndef AMDGPU_CSA_MANAGER_H +#define AMDGPU_CSA_MANAGER_H + +#define AMDGPU_CSA_SIZE (128 * 1024) + +uint32_t amdgpu_get_total_csa_size(struct amdgpu_device *adev); +uint64_t amdgpu_csa_vaddr(struct amdgpu_device *adev); +int amdgpu_allocate_static_csa(struct amdgpu_device *adev, struct amdgpu_bo **bo, + u32 domain, uint32_t size); +int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm, + struct amdgpu_bo *bo, struct amdgpu_bo_va **bo_va, + uint64_t csa_addr, uint32_t size); +void amdgpu_free_static_csa(struct amdgpu_bo **bo); + +#endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 95f4c4139fc6..7b526593eb77 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -124,6 +124,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ring *rings[AMDGPU_MAX_RINGS]; struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS]; unsigned num_rings; + unsigned num_rqs = 0; switch (i) { case AMDGPU_HW_IP_GFX: @@ -166,12 +167,16 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, break; } - for (j = 0; j < num_rings; ++j) - rqs[j] = &rings[j]->sched.sched_rq[priority]; + for (j = 0; j < num_rings; ++j) { + if (!rings[j]->adev) + continue; + + rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority]; + } for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) r = drm_sched_entity_init(&ctx->entities[i][j].entity, - rqs, num_rings, &ctx->guilty); + rqs, num_rqs, &ctx->guilty); if (r) goto error_cleanup_entities; } @@ -248,7 +253,7 @@ static int amdgpu_ctx_alloc(struct amdgpu_device *adev, return -ENOMEM; mutex_lock(&mgr->lock); - r = idr_alloc(&mgr->ctx_handles, ctx, 1, 0, GFP_KERNEL); + r = idr_alloc(&mgr->ctx_handles, ctx, 1, AMDGPU_VM_MAX_NUM_CTX, GFP_KERNEL); if (r < 0) { mutex_unlock(&mgr->lock); kfree(ctx); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c index dd9a4fb9ce39..4ae3ff9a1d4c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c @@ -158,9 +158,6 @@ static int amdgpu_debugfs_process_reg_op(bool read, struct file *f, while (size) { uint32_t value; - if (*pos > adev->rmmio_size) - goto end; - if (read) { value = RREG32(*pos >> 2); r = put_user(value, (uint32_t *)buf); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 30bc345d6fdf..79fb302fb954 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -30,8 +30,8 @@ #include <linux/console.h> #include <linux/slab.h> #include <drm/drmP.h> -#include <drm/drm_crtc_helper.h> #include <drm/drm_atomic_helper.h> +#include <drm/drm_probe_helper.h> #include <drm/amdgpu_drm.h> #include <linux/vgaarb.h> #include <linux/vga_switcheroo.h> @@ -59,6 +59,8 @@ #include "amdgpu_amdkfd.h" #include "amdgpu_pm.h" +#include "amdgpu_xgmi.h" + MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin"); MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin"); MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin"); @@ -513,6 +515,7 @@ void amdgpu_device_pci_config_reset(struct amdgpu_device *adev) */ static int amdgpu_device_doorbell_init(struct amdgpu_device *adev) { + /* No doorbell on SI hardware generation */ if (adev->asic_type < CHIP_BONAIRE) { adev->doorbell.base = 0; @@ -525,15 +528,26 @@ static int amdgpu_device_doorbell_init(struct amdgpu_device *adev) if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET) return -EINVAL; + amdgpu_asic_init_doorbell_index(adev); + /* doorbell bar mapping */ adev->doorbell.base = pci_resource_start(adev->pdev, 2); adev->doorbell.size = pci_resource_len(adev->pdev, 2); adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32), - AMDGPU_DOORBELL_MAX_ASSIGNMENT+1); + adev->doorbell_index.max_assignment+1); if (adev->doorbell.num_doorbells == 0) return -EINVAL; + /* For Vega, reserve and map two pages on doorbell BAR since SDMA + * paging queue doorbell use the second page. The + * AMDGPU_DOORBELL64_MAX_ASSIGNMENT definition assumes all the + * doorbells are in the first page. So with paging queue enabled, + * the max num_doorbells should + 1 page (0x400 in dword) + */ + if (adev->asic_type >= CHIP_VEGA10) + adev->doorbell.num_doorbells += 0x400; + adev->doorbell.ptr = ioremap(adev->doorbell.base, adev->doorbell.num_doorbells * sizeof(u32)); @@ -1631,7 +1645,7 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) if (r) { DRM_ERROR("sw_init of IP block <%s> failed %d\n", adev->ip_blocks[i].version->funcs->name, r); - return r; + goto init_failed; } adev->ip_blocks[i].status.sw = true; @@ -1640,26 +1654,28 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) r = amdgpu_device_vram_scratch_init(adev); if (r) { DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r); - return r; + goto init_failed; } r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev); if (r) { DRM_ERROR("hw_init %d failed %d\n", i, r); - return r; + goto init_failed; } r = amdgpu_device_wb_init(adev); if (r) { DRM_ERROR("amdgpu_device_wb_init failed %d\n", r); - return r; + goto init_failed; } adev->ip_blocks[i].status.hw = true; /* right after GMC hw init, we create CSA */ if (amdgpu_sriov_vf(adev)) { - r = amdgpu_allocate_static_csa(adev); + r = amdgpu_allocate_static_csa(adev, &adev->virt.csa_obj, + AMDGPU_GEM_DOMAIN_VRAM, + AMDGPU_CSA_SIZE); if (r) { DRM_ERROR("allocate CSA failed %d\n", r); - return r; + goto init_failed; } } } @@ -1667,27 +1683,32 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) r = amdgpu_ucode_create_bo(adev); /* create ucode bo when sw_init complete*/ if (r) - return r; + goto init_failed; r = amdgpu_device_ip_hw_init_phase1(adev); if (r) - return r; + goto init_failed; r = amdgpu_device_fw_loading(adev); if (r) - return r; + goto init_failed; r = amdgpu_device_ip_hw_init_phase2(adev); if (r) - return r; + goto init_failed; - amdgpu_xgmi_add_device(adev); + if (adev->gmc.xgmi.num_physical_nodes > 1) + amdgpu_xgmi_add_device(adev); amdgpu_amdkfd_device_init(adev); - if (amdgpu_sriov_vf(adev)) +init_failed: + if (amdgpu_sriov_vf(adev)) { + if (!r) + amdgpu_virt_init_data_exchange(adev); amdgpu_virt_release_full_gpu(adev, true); + } - return 0; + return r; } /** @@ -1848,6 +1869,9 @@ static int amdgpu_device_ip_fini(struct amdgpu_device *adev) { int i, r; + if (adev->gmc.xgmi.num_physical_nodes > 1) + amdgpu_xgmi_remove_device(adev); + amdgpu_amdkfd_device_fini(adev); amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); @@ -1890,7 +1914,7 @@ static int amdgpu_device_ip_fini(struct amdgpu_device *adev) if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { amdgpu_ucode_free_bo(adev); - amdgpu_free_static_csa(adev); + amdgpu_free_static_csa(&adev->virt.csa_obj); amdgpu_device_wb_fini(adev); amdgpu_device_vram_scratch_fini(adev); } @@ -2111,7 +2135,7 @@ static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev) continue; r = block->version->funcs->hw_init(adev); - DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); + DRM_INFO("RE-INIT-early: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); if (r) return r; } @@ -2145,7 +2169,7 @@ static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev) continue; r = block->version->funcs->hw_init(adev); - DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); + DRM_INFO("RE-INIT-late: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); if (r) return r; } @@ -2337,6 +2361,19 @@ bool amdgpu_device_has_dc_support(struct amdgpu_device *adev) return amdgpu_device_asic_has_dc_support(adev->asic_type); } + +static void amdgpu_device_xgmi_reset_func(struct work_struct *__work) +{ + struct amdgpu_device *adev = + container_of(__work, struct amdgpu_device, xgmi_reset_work); + + adev->asic_reset_res = amdgpu_asic_reset(adev); + if (adev->asic_reset_res) + DRM_WARN("ASIC reset failed with err r, %d for drm dev, %s", + adev->asic_reset_res, adev->ddev->unique); +} + + /** * amdgpu_device_init - initialize the driver * @@ -2435,6 +2472,8 @@ int amdgpu_device_init(struct amdgpu_device *adev, INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work, amdgpu_device_delay_enable_gfx_off); + INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func); + adev->gfx.gfx_off_req_count = 1; adev->pm.ac_power = power_supply_is_system_supplied() > 0 ? true : false; @@ -2455,9 +2494,6 @@ int amdgpu_device_init(struct amdgpu_device *adev, DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base); DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size); - /* doorbell bar mapping */ - amdgpu_device_doorbell_init(adev); - /* io port mapping */ for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) { if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) { @@ -2476,6 +2512,9 @@ int amdgpu_device_init(struct amdgpu_device *adev, if (r) return r; + /* doorbell bar mapping and doorbell index init*/ + amdgpu_device_doorbell_init(adev); + /* if we have > 1 VGA cards, then disable the amdgpu VGA resources */ /* this will fail for cards that aren't VGA class devices, just * ignore it */ @@ -2511,6 +2550,17 @@ int amdgpu_device_init(struct amdgpu_device *adev, /* detect if we are with an SRIOV vbios */ amdgpu_device_detect_sriov_bios(adev); + /* check if we need to reset the asic + * E.g., driver was not cleanly unloaded previously, etc. + */ + if (!amdgpu_sriov_vf(adev) && amdgpu_asic_need_reset_on_init(adev)) { + r = amdgpu_asic_reset(adev); + if (r) { + dev_err(adev->dev, "asic reset on init failed\n"); + goto failed; + } + } + /* Post card if necessary */ if (amdgpu_device_need_post(adev)) { if (!adev->bios) { @@ -2575,6 +2625,8 @@ fence_driver_init: } dev_err(adev->dev, "amdgpu_device_ip_init failed\n"); amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0); + if (amdgpu_virt_request_full_gpu(adev, false)) + amdgpu_virt_release_full_gpu(adev, false); goto failed; } @@ -2597,9 +2649,6 @@ fence_driver_init: goto failed; } - if (amdgpu_sriov_vf(adev)) - amdgpu_virt_init_data_exchange(adev); - amdgpu_fbdev_init(adev); r = amdgpu_pm_sysfs_init(adev); @@ -2673,7 +2722,7 @@ void amdgpu_device_fini(struct amdgpu_device *adev) amdgpu_irq_disable_all(adev); if (adev->mode_info.mode_config_initialized){ if (!amdgpu_device_has_dc_support(adev)) - drm_crtc_force_disable_all(adev->ddev); + drm_helper_force_disable_all(adev->ddev); else drm_atomic_helper_shutdown(adev->ddev); } @@ -2763,7 +2812,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon) struct drm_framebuffer *fb = crtc->primary->fb; struct amdgpu_bo *robj; - if (amdgpu_crtc->cursor_bo) { + if (amdgpu_crtc->cursor_bo && !adev->enable_virtual_display) { struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); r = amdgpu_bo_reserve(aobj, true); if (r == 0) { @@ -2871,7 +2920,7 @@ int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon) list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); - if (amdgpu_crtc->cursor_bo) { + if (amdgpu_crtc->cursor_bo && !adev->enable_virtual_display) { struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); r = amdgpu_bo_reserve(aobj, true); if (r == 0) { @@ -3116,6 +3165,7 @@ static int amdgpu_device_recover_vram(struct amdgpu_device *adev) /* No need to recover an evicted BO */ if (shadow->tbo.mem.mem_type != TTM_PL_TT || + shadow->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET || shadow->parent->tbo.mem.mem_type != TTM_PL_VRAM) continue; @@ -3124,11 +3174,16 @@ static int amdgpu_device_recover_vram(struct amdgpu_device *adev) break; if (fence) { - r = dma_fence_wait_timeout(fence, false, tmo); + tmo = dma_fence_wait_timeout(fence, false, tmo); dma_fence_put(fence); fence = next; - if (r <= 0) + if (tmo == 0) { + r = -ETIMEDOUT; break; + } else if (tmo < 0) { + r = tmo; + break; + } } else { fence = next; } @@ -3139,8 +3194,8 @@ static int amdgpu_device_recover_vram(struct amdgpu_device *adev) tmo = dma_fence_wait_timeout(fence, false, tmo); dma_fence_put(fence); - if (r <= 0 || tmo <= 0) { - DRM_ERROR("recover vram bo from shadow failed\n"); + if (r < 0 || tmo <= 0) { + DRM_ERROR("recover vram bo from shadow failed, r is %ld, tmo is %ld\n", r, tmo); return -EIO; } @@ -3148,86 +3203,6 @@ static int amdgpu_device_recover_vram(struct amdgpu_device *adev) return 0; } -/** - * amdgpu_device_reset - reset ASIC/GPU for bare-metal or passthrough - * - * @adev: amdgpu device pointer - * - * attempt to do soft-reset or full-reset and reinitialize Asic - * return 0 means succeeded otherwise failed - */ -static int amdgpu_device_reset(struct amdgpu_device *adev) -{ - bool need_full_reset, vram_lost = 0; - int r; - - need_full_reset = amdgpu_device_ip_need_full_reset(adev); - - if (!need_full_reset) { - amdgpu_device_ip_pre_soft_reset(adev); - r = amdgpu_device_ip_soft_reset(adev); - amdgpu_device_ip_post_soft_reset(adev); - if (r || amdgpu_device_ip_check_soft_reset(adev)) { - DRM_INFO("soft reset failed, will fallback to full reset!\n"); - need_full_reset = true; - } - } - - if (need_full_reset) { - r = amdgpu_device_ip_suspend(adev); - -retry: - r = amdgpu_asic_reset(adev); - /* post card */ - amdgpu_atom_asic_init(adev->mode_info.atom_context); - - if (!r) { - dev_info(adev->dev, "GPU reset succeeded, trying to resume\n"); - r = amdgpu_device_ip_resume_phase1(adev); - if (r) - goto out; - - vram_lost = amdgpu_device_check_vram_lost(adev); - if (vram_lost) { - DRM_ERROR("VRAM is lost!\n"); - atomic_inc(&adev->vram_lost_counter); - } - - r = amdgpu_gtt_mgr_recover( - &adev->mman.bdev.man[TTM_PL_TT]); - if (r) - goto out; - - r = amdgpu_device_fw_loading(adev); - if (r) - return r; - - r = amdgpu_device_ip_resume_phase2(adev); - if (r) - goto out; - - if (vram_lost) - amdgpu_device_fill_reset_magic(adev); - } - } - -out: - if (!r) { - amdgpu_irq_gpu_reset_resume_helper(adev); - r = amdgpu_ib_ring_tests(adev); - if (r) { - dev_err(adev->dev, "ib ring test failed (%d).\n", r); - r = amdgpu_device_ip_suspend(adev); - need_full_reset = true; - goto retry; - } - } - - if (!r) - r = amdgpu_device_recover_vram(adev); - - return r; -} /** * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf @@ -3271,6 +3246,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, r = amdgpu_ib_ring_tests(adev); error: + amdgpu_virt_init_data_exchange(adev); amdgpu_virt_release_full_gpu(adev, true); if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) { atomic_inc(&adev->vram_lost_counter); @@ -3295,40 +3271,46 @@ bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev) return false; } - if (amdgpu_gpu_recovery == 0 || (amdgpu_gpu_recovery == -1 && - !amdgpu_sriov_vf(adev))) { - DRM_INFO("GPU recovery disabled.\n"); - return false; - } + if (amdgpu_gpu_recovery == 0) + goto disabled; - return true; -} + if (amdgpu_sriov_vf(adev)) + return true; -/** - * amdgpu_device_gpu_recover - reset the asic and recover scheduler - * - * @adev: amdgpu device pointer - * @job: which job trigger hang - * - * Attempt to reset the GPU if it has hung (all asics). - * Returns 0 for success or an error on failure. - */ -int amdgpu_device_gpu_recover(struct amdgpu_device *adev, - struct amdgpu_job *job) -{ - int i, r, resched; + if (amdgpu_gpu_recovery == -1) { + switch (adev->asic_type) { + case CHIP_BONAIRE: + case CHIP_HAWAII: + case CHIP_TOPAZ: + case CHIP_TONGA: + case CHIP_FIJI: + case CHIP_POLARIS10: + case CHIP_POLARIS11: + case CHIP_POLARIS12: + case CHIP_VEGAM: + case CHIP_VEGA20: + case CHIP_VEGA10: + case CHIP_VEGA12: + break; + default: + goto disabled; + } + } - dev_info(adev->dev, "GPU reset begin!\n"); + return true; - mutex_lock(&adev->lock_reset); - atomic_inc(&adev->gpu_reset_counter); - adev->in_gpu_reset = 1; +disabled: + DRM_INFO("GPU recovery disabled.\n"); + return false; +} - /* Block kfd */ - amdgpu_amdkfd_pre_reset(adev); - /* block TTM */ - resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev); +static int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, + struct amdgpu_job *job, + bool *need_full_reset_arg) +{ + int i, r = 0; + bool need_full_reset = *need_full_reset_arg; /* block all schedulers and reset given job's ring */ for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { @@ -3337,21 +3319,153 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, if (!ring || !ring->sched.thread) continue; - kthread_park(ring->sched.thread); - - if (job && job->base.sched != &ring->sched) - continue; - - drm_sched_hw_job_reset(&ring->sched, job ? &job->base : NULL); + drm_sched_stop(&ring->sched); /* after all hw jobs are reset, hw fence is meaningless, so force_completion */ amdgpu_fence_driver_force_completion(ring); } - if (amdgpu_sriov_vf(adev)) - r = amdgpu_device_reset_sriov(adev, job ? false : true); - else - r = amdgpu_device_reset(adev); + if(job) + drm_sched_increase_karma(&job->base); + + + + if (!amdgpu_sriov_vf(adev)) { + + if (!need_full_reset) + need_full_reset = amdgpu_device_ip_need_full_reset(adev); + + if (!need_full_reset) { + amdgpu_device_ip_pre_soft_reset(adev); + r = amdgpu_device_ip_soft_reset(adev); + amdgpu_device_ip_post_soft_reset(adev); + if (r || amdgpu_device_ip_check_soft_reset(adev)) { + DRM_INFO("soft reset failed, will fallback to full reset!\n"); + need_full_reset = true; + } + } + + if (need_full_reset) + r = amdgpu_device_ip_suspend(adev); + + *need_full_reset_arg = need_full_reset; + } + + return r; +} + +static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, + struct list_head *device_list_handle, + bool *need_full_reset_arg) +{ + struct amdgpu_device *tmp_adev = NULL; + bool need_full_reset = *need_full_reset_arg, vram_lost = false; + int r = 0; + + /* + * ASIC reset has to be done on all HGMI hive nodes ASAP + * to allow proper links negotiation in FW (within 1 sec) + */ + if (need_full_reset) { + list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { + /* For XGMI run all resets in parallel to speed up the process */ + if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { + if (!queue_work(system_highpri_wq, &tmp_adev->xgmi_reset_work)) + r = -EALREADY; + } else + r = amdgpu_asic_reset(tmp_adev); + + if (r) { + DRM_ERROR("ASIC reset failed with err r, %d for drm dev, %s", + r, tmp_adev->ddev->unique); + break; + } + } + + /* For XGMI wait for all PSP resets to complete before proceed */ + if (!r) { + list_for_each_entry(tmp_adev, device_list_handle, + gmc.xgmi.head) { + if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { + flush_work(&tmp_adev->xgmi_reset_work); + r = tmp_adev->asic_reset_res; + if (r) + break; + } + } + } + } + + + list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { + if (need_full_reset) { + /* post card */ + if (amdgpu_atom_asic_init(tmp_adev->mode_info.atom_context)) + DRM_WARN("asic atom init failed!"); + + if (!r) { + dev_info(tmp_adev->dev, "GPU reset succeeded, trying to resume\n"); + r = amdgpu_device_ip_resume_phase1(tmp_adev); + if (r) + goto out; + + vram_lost = amdgpu_device_check_vram_lost(tmp_adev); + if (vram_lost) { + DRM_ERROR("VRAM is lost!\n"); + atomic_inc(&tmp_adev->vram_lost_counter); + } + + r = amdgpu_gtt_mgr_recover( + &tmp_adev->mman.bdev.man[TTM_PL_TT]); + if (r) + goto out; + + r = amdgpu_device_fw_loading(tmp_adev); + if (r) + return r; + + r = amdgpu_device_ip_resume_phase2(tmp_adev); + if (r) + goto out; + + if (vram_lost) + amdgpu_device_fill_reset_magic(tmp_adev); + + /* Update PSP FW topology after reset */ + if (hive && tmp_adev->gmc.xgmi.num_physical_nodes > 1) + r = amdgpu_xgmi_update_topology(hive, tmp_adev); + } + } + + +out: + if (!r) { + amdgpu_irq_gpu_reset_resume_helper(tmp_adev); + r = amdgpu_ib_ring_tests(tmp_adev); + if (r) { + dev_err(tmp_adev->dev, "ib ring test failed (%d).\n", r); + r = amdgpu_device_ip_suspend(tmp_adev); + need_full_reset = true; + r = -EAGAIN; + goto end; + } + } + + if (!r) + r = amdgpu_device_recover_vram(tmp_adev); + else + tmp_adev->asic_reset_res = r; + } + +end: + *need_full_reset_arg = need_full_reset; + return r; +} + +static void amdgpu_device_post_asic_reset(struct amdgpu_device *adev, + struct amdgpu_job *job) +{ + int i; for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { struct amdgpu_ring *ring = adev->rings[i]; @@ -3359,38 +3473,194 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, if (!ring || !ring->sched.thread) continue; - /* only need recovery sched of the given job's ring - * or all rings (in the case @job is NULL) - * after above amdgpu_reset accomplished - */ - if ((!job || job->base.sched == &ring->sched) && !r) - drm_sched_job_recovery(&ring->sched); + if (!adev->asic_reset_res) + drm_sched_resubmit_jobs(&ring->sched); - kthread_unpark(ring->sched.thread); + drm_sched_start(&ring->sched, !adev->asic_reset_res); } if (!amdgpu_device_has_dc_support(adev)) { drm_helper_resume_force_mode(adev->ddev); } - ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched); + adev->asic_reset_res = 0; +} - if (r) { - /* bad news, how to tell it to userspace ? */ - dev_info(adev->dev, "GPU reset(%d) failed\n", atomic_read(&adev->gpu_reset_counter)); - amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r); - } else { - dev_info(adev->dev, "GPU reset(%d) succeeded!\n",atomic_read(&adev->gpu_reset_counter)); - } +static void amdgpu_device_lock_adev(struct amdgpu_device *adev) +{ + mutex_lock(&adev->lock_reset); + atomic_inc(&adev->gpu_reset_counter); + adev->in_gpu_reset = 1; + /* Block kfd: SRIOV would do it separately */ + if (!amdgpu_sriov_vf(adev)) + amdgpu_amdkfd_pre_reset(adev); +} - /*unlock kfd */ - amdgpu_amdkfd_post_reset(adev); +static void amdgpu_device_unlock_adev(struct amdgpu_device *adev) +{ + /*unlock kfd: SRIOV would do it separately */ + if (!amdgpu_sriov_vf(adev)) + amdgpu_amdkfd_post_reset(adev); amdgpu_vf_error_trans_all(adev); adev->in_gpu_reset = 0; mutex_unlock(&adev->lock_reset); +} + + +/** + * amdgpu_device_gpu_recover - reset the asic and recover scheduler + * + * @adev: amdgpu device pointer + * @job: which job trigger hang + * + * Attempt to reset the GPU if it has hung (all asics). + * Attempt to do soft-reset or full-reset and reinitialize Asic + * Returns 0 for success or an error on failure. + */ + +int amdgpu_device_gpu_recover(struct amdgpu_device *adev, + struct amdgpu_job *job) +{ + int r; + struct amdgpu_hive_info *hive = NULL; + bool need_full_reset = false; + struct amdgpu_device *tmp_adev = NULL; + struct list_head device_list, *device_list_handle = NULL; + + INIT_LIST_HEAD(&device_list); + + dev_info(adev->dev, "GPU reset begin!\n"); + + /* + * In case of XGMI hive disallow concurrent resets to be triggered + * by different nodes. No point also since the one node already executing + * reset will also reset all the other nodes in the hive. + */ + hive = amdgpu_get_xgmi_hive(adev, 0); + if (hive && adev->gmc.xgmi.num_physical_nodes > 1 && + !mutex_trylock(&hive->reset_lock)) + return 0; + + /* Start with adev pre asic reset first for soft reset check.*/ + amdgpu_device_lock_adev(adev); + r = amdgpu_device_pre_asic_reset(adev, + job, + &need_full_reset); + if (r) { + /*TODO Should we stop ?*/ + DRM_ERROR("GPU pre asic reset failed with err, %d for drm dev, %s ", + r, adev->ddev->unique); + adev->asic_reset_res = r; + } + + /* Build list of devices to reset */ + if (need_full_reset && adev->gmc.xgmi.num_physical_nodes > 1) { + if (!hive) { + amdgpu_device_unlock_adev(adev); + return -ENODEV; + } + + /* + * In case we are in XGMI hive mode device reset is done for all the + * nodes in the hive to retrain all XGMI links and hence the reset + * sequence is executed in loop on all nodes. + */ + device_list_handle = &hive->device_list; + } else { + list_add_tail(&adev->gmc.xgmi.head, &device_list); + device_list_handle = &device_list; + } + +retry: /* Rest of adevs pre asic reset from XGMI hive. */ + list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { + + if (tmp_adev == adev) + continue; + + amdgpu_device_lock_adev(tmp_adev); + r = amdgpu_device_pre_asic_reset(tmp_adev, + NULL, + &need_full_reset); + /*TODO Should we stop ?*/ + if (r) { + DRM_ERROR("GPU pre asic reset failed with err, %d for drm dev, %s ", + r, tmp_adev->ddev->unique); + tmp_adev->asic_reset_res = r; + } + } + + /* Actual ASIC resets if needed.*/ + /* TODO Implement XGMI hive reset logic for SRIOV */ + if (amdgpu_sriov_vf(adev)) { + r = amdgpu_device_reset_sriov(adev, job ? false : true); + if (r) + adev->asic_reset_res = r; + } else { + r = amdgpu_do_asic_reset(hive, device_list_handle, &need_full_reset); + if (r && r == -EAGAIN) + goto retry; + } + + /* Post ASIC reset for all devs .*/ + list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { + amdgpu_device_post_asic_reset(tmp_adev, tmp_adev == adev ? job : NULL); + + if (r) { + /* bad news, how to tell it to userspace ? */ + dev_info(tmp_adev->dev, "GPU reset(%d) failed\n", atomic_read(&adev->gpu_reset_counter)); + amdgpu_vf_error_put(tmp_adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r); + } else { + dev_info(tmp_adev->dev, "GPU reset(%d) succeeded!\n", atomic_read(&adev->gpu_reset_counter)); + } + + amdgpu_device_unlock_adev(tmp_adev); + } + + if (hive && adev->gmc.xgmi.num_physical_nodes > 1) + mutex_unlock(&hive->reset_lock); + + if (r) + dev_info(adev->dev, "GPU reset end with ret = %d\n", r); return r; } +static void amdgpu_device_get_min_pci_speed_width(struct amdgpu_device *adev, + enum pci_bus_speed *speed, + enum pcie_link_width *width) +{ + struct pci_dev *pdev = adev->pdev; + enum pci_bus_speed cur_speed; + enum pcie_link_width cur_width; + u32 ret = 1; + + *speed = PCI_SPEED_UNKNOWN; + *width = PCIE_LNK_WIDTH_UNKNOWN; + + while (pdev) { + cur_speed = pcie_get_speed_cap(pdev); + cur_width = pcie_get_width_cap(pdev); + ret = pcie_bandwidth_available(adev->pdev, NULL, + NULL, &cur_width); + if (!ret) + cur_width = PCIE_LNK_WIDTH_RESRV; + + if (cur_speed != PCI_SPEED_UNKNOWN) { + if (*speed == PCI_SPEED_UNKNOWN) + *speed = cur_speed; + else if (cur_speed < *speed) + *speed = cur_speed; + } + + if (cur_width != PCIE_LNK_WIDTH_UNKNOWN) { + if (*width == PCIE_LNK_WIDTH_UNKNOWN) + *width = cur_width; + else if (cur_width < *width) + *width = cur_width; + } + pdev = pci_upstream_bridge(pdev); + } +} + /** * amdgpu_device_get_pcie_info - fence pcie info about the PCIE slot * @@ -3403,8 +3673,8 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev) { struct pci_dev *pdev; - enum pci_bus_speed speed_cap; - enum pcie_link_width link_width; + enum pci_bus_speed speed_cap, platform_speed_cap; + enum pcie_link_width platform_link_width; if (amdgpu_pcie_gen_cap) adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap; @@ -3421,6 +3691,12 @@ static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev) return; } + if (adev->pm.pcie_gen_mask && adev->pm.pcie_mlw_mask) + return; + + amdgpu_device_get_min_pci_speed_width(adev, &platform_speed_cap, + &platform_link_width); + if (adev->pm.pcie_gen_mask == 0) { /* asic caps */ pdev = adev->pdev; @@ -3446,22 +3722,20 @@ static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev) adev->pm.pcie_gen_mask |= CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1; } /* platform caps */ - pdev = adev->ddev->pdev->bus->self; - speed_cap = pcie_get_speed_cap(pdev); - if (speed_cap == PCI_SPEED_UNKNOWN) { + if (platform_speed_cap == PCI_SPEED_UNKNOWN) { adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2); } else { - if (speed_cap == PCIE_SPEED_16_0GT) + if (platform_speed_cap == PCIE_SPEED_16_0GT) adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4); - else if (speed_cap == PCIE_SPEED_8_0GT) + else if (platform_speed_cap == PCIE_SPEED_8_0GT) adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3); - else if (speed_cap == PCIE_SPEED_5_0GT) + else if (platform_speed_cap == PCIE_SPEED_5_0GT) adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2); else @@ -3470,12 +3744,10 @@ static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev) } } if (adev->pm.pcie_mlw_mask == 0) { - pdev = adev->ddev->pdev->bus->self; - link_width = pcie_get_width_cap(pdev); - if (link_width == PCIE_LNK_WIDTH_UNKNOWN) { + if (platform_link_width == PCIE_LNK_WIDTH_UNKNOWN) { adev->pm.pcie_mlw_mask |= AMDGPU_DEFAULT_PCIE_MLW_MASK; } else { - switch (link_width) { + switch (platform_link_width) { case PCIE_LNK_X32: adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c index 686a26de50f9..b083b219b1a9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c @@ -188,10 +188,12 @@ int amdgpu_display_crtc_page_flip_target(struct drm_crtc *crtc, goto cleanup; } - r = amdgpu_bo_pin(new_abo, amdgpu_display_supported_domains(adev)); - if (unlikely(r != 0)) { - DRM_ERROR("failed to pin new abo buffer before flip\n"); - goto unreserve; + if (!adev->enable_virtual_display) { + r = amdgpu_bo_pin(new_abo, amdgpu_display_supported_domains(adev)); + if (unlikely(r != 0)) { + DRM_ERROR("failed to pin new abo buffer before flip\n"); + goto unreserve; + } } r = amdgpu_ttm_alloc_gart(&new_abo->tbo); @@ -211,7 +213,8 @@ int amdgpu_display_crtc_page_flip_target(struct drm_crtc *crtc, amdgpu_bo_get_tiling_flags(new_abo, &tiling_flags); amdgpu_bo_unreserve(new_abo); - work->base = amdgpu_bo_gpu_offset(new_abo); + if (!adev->enable_virtual_display) + work->base = amdgpu_bo_gpu_offset(new_abo); work->target_vblank = target - (uint32_t)drm_crtc_vblank_count(crtc) + amdgpu_get_vblank_counter_kms(dev, work->crtc_id); @@ -242,9 +245,10 @@ pflip_cleanup: goto cleanup; } unpin: - if (unlikely(amdgpu_bo_unpin(new_abo) != 0)) { - DRM_ERROR("failed to unpin new abo in error path\n"); - } + if (!adev->enable_virtual_display) + if (unlikely(amdgpu_bo_unpin(new_abo) != 0)) + DRM_ERROR("failed to unpin new abo in error path\n"); + unreserve: amdgpu_bo_unreserve(new_abo); @@ -631,6 +635,11 @@ int amdgpu_display_modeset_create_props(struct amdgpu_device *adev) drm_property_create_range(adev->ddev, 0, "max bpc", 8, 16); if (!adev->mode_info.max_bpc_property) return -ENOMEM; + adev->mode_info.abm_level_property = + drm_property_create_range(adev->ddev, 0, + "abm level", 0, 4); + if (!adev->mode_info.abm_level_property) + return -ENOMEM; } return 0; @@ -857,7 +866,12 @@ int amdgpu_display_get_crtc_scanoutpos(struct drm_device *dev, /* Inside "upper part" of vblank area? Apply corrective offset if so: */ if (in_vbl && (*vpos >= vbl_start)) { vtotal = mode->crtc_vtotal; - *vpos = *vpos - vtotal; + + /* With variable refresh rate displays the vpos can exceed + * the vtotal value. Clamp to 0 to return -vbl_end instead + * of guessing the remaining number of lines until scanout. + */ + *vpos = (*vpos < vtotal) ? (*vpos - vtotal) : 0; } /* Correct for shifted end of vbl at vbl_end. */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h new file mode 100644 index 000000000000..68959b923f89 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h @@ -0,0 +1,247 @@ +/* + * Copyright 2018 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + */ + +/* + * GPU doorbell structures, functions & helpers + */ +struct amdgpu_doorbell { + /* doorbell mmio */ + resource_size_t base; + resource_size_t size; + u32 __iomem *ptr; + u32 num_doorbells; /* Number of doorbells actually reserved for amdgpu. */ +}; + +/* Reserved doorbells for amdgpu (including multimedia). + * KFD can use all the rest in the 2M doorbell bar. + * For asic before vega10, doorbell is 32-bit, so the + * index/offset is in dword. For vega10 and after, doorbell + * can be 64-bit, so the index defined is in qword. + */ +struct amdgpu_doorbell_index { + uint32_t kiq; + uint32_t mec_ring0; + uint32_t mec_ring1; + uint32_t mec_ring2; + uint32_t mec_ring3; + uint32_t mec_ring4; + uint32_t mec_ring5; + uint32_t mec_ring6; + uint32_t mec_ring7; + uint32_t userqueue_start; + uint32_t userqueue_end; + uint32_t gfx_ring0; + uint32_t sdma_engine[8]; + uint32_t ih; + union { + struct { + uint32_t vcn_ring0_1; + uint32_t vcn_ring2_3; + uint32_t vcn_ring4_5; + uint32_t vcn_ring6_7; + } vcn; + struct { + uint32_t uvd_ring0_1; + uint32_t uvd_ring2_3; + uint32_t uvd_ring4_5; + uint32_t uvd_ring6_7; + uint32_t vce_ring0_1; + uint32_t vce_ring2_3; + uint32_t vce_ring4_5; + uint32_t vce_ring6_7; + } uvd_vce; + }; + uint32_t first_non_cp; + uint32_t last_non_cp; + uint32_t max_assignment; + /* Per engine SDMA doorbell size in dword */ + uint32_t sdma_doorbell_range; +}; + +typedef enum _AMDGPU_DOORBELL_ASSIGNMENT +{ + AMDGPU_DOORBELL_KIQ = 0x000, + AMDGPU_DOORBELL_HIQ = 0x001, + AMDGPU_DOORBELL_DIQ = 0x002, + AMDGPU_DOORBELL_MEC_RING0 = 0x010, + AMDGPU_DOORBELL_MEC_RING1 = 0x011, + AMDGPU_DOORBELL_MEC_RING2 = 0x012, + AMDGPU_DOORBELL_MEC_RING3 = 0x013, + AMDGPU_DOORBELL_MEC_RING4 = 0x014, + AMDGPU_DOORBELL_MEC_RING5 = 0x015, + AMDGPU_DOORBELL_MEC_RING6 = 0x016, + AMDGPU_DOORBELL_MEC_RING7 = 0x017, + AMDGPU_DOORBELL_GFX_RING0 = 0x020, + AMDGPU_DOORBELL_sDMA_ENGINE0 = 0x1E0, + AMDGPU_DOORBELL_sDMA_ENGINE1 = 0x1E1, + AMDGPU_DOORBELL_IH = 0x1E8, + AMDGPU_DOORBELL_MAX_ASSIGNMENT = 0x3FF, + AMDGPU_DOORBELL_INVALID = 0xFFFF +} AMDGPU_DOORBELL_ASSIGNMENT; + +typedef enum _AMDGPU_VEGA20_DOORBELL_ASSIGNMENT +{ + /* Compute + GFX: 0~255 */ + AMDGPU_VEGA20_DOORBELL_KIQ = 0x000, + AMDGPU_VEGA20_DOORBELL_HIQ = 0x001, + AMDGPU_VEGA20_DOORBELL_DIQ = 0x002, + AMDGPU_VEGA20_DOORBELL_MEC_RING0 = 0x003, + AMDGPU_VEGA20_DOORBELL_MEC_RING1 = 0x004, + AMDGPU_VEGA20_DOORBELL_MEC_RING2 = 0x005, + AMDGPU_VEGA20_DOORBELL_MEC_RING3 = 0x006, + AMDGPU_VEGA20_DOORBELL_MEC_RING4 = 0x007, + AMDGPU_VEGA20_DOORBELL_MEC_RING5 = 0x008, + AMDGPU_VEGA20_DOORBELL_MEC_RING6 = 0x009, + AMDGPU_VEGA20_DOORBELL_MEC_RING7 = 0x00A, + AMDGPU_VEGA20_DOORBELL_USERQUEUE_START = 0x00B, + AMDGPU_VEGA20_DOORBELL_USERQUEUE_END = 0x08A, + AMDGPU_VEGA20_DOORBELL_GFX_RING0 = 0x08B, + /* SDMA:256~335*/ + AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0 = 0x100, + AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE1 = 0x10A, + AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE2 = 0x114, + AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE3 = 0x11E, + AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE4 = 0x128, + AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE5 = 0x132, + AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE6 = 0x13C, + AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE7 = 0x146, + /* IH: 376~391 */ + AMDGPU_VEGA20_DOORBELL_IH = 0x178, + /* MMSCH: 392~407 + * overlap the doorbell assignment with VCN as they are mutually exclusive + * VCE engine's doorbell is 32 bit and two VCE ring share one QWORD + */ + AMDGPU_VEGA20_DOORBELL64_VCN0_1 = 0x188, /* lower 32 bits for VNC0 and upper 32 bits for VNC1 */ + AMDGPU_VEGA20_DOORBELL64_VCN2_3 = 0x189, + AMDGPU_VEGA20_DOORBELL64_VCN4_5 = 0x18A, + AMDGPU_VEGA20_DOORBELL64_VCN6_7 = 0x18B, + + AMDGPU_VEGA20_DOORBELL64_UVD_RING0_1 = 0x188, + AMDGPU_VEGA20_DOORBELL64_UVD_RING2_3 = 0x189, + AMDGPU_VEGA20_DOORBELL64_UVD_RING4_5 = 0x18A, + AMDGPU_VEGA20_DOORBELL64_UVD_RING6_7 = 0x18B, + + AMDGPU_VEGA20_DOORBELL64_VCE_RING0_1 = 0x18C, + AMDGPU_VEGA20_DOORBELL64_VCE_RING2_3 = 0x18D, + AMDGPU_VEGA20_DOORBELL64_VCE_RING4_5 = 0x18E, + AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7 = 0x18F, + + AMDGPU_VEGA20_DOORBELL64_FIRST_NON_CP = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0, + AMDGPU_VEGA20_DOORBELL64_LAST_NON_CP = AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7, + + AMDGPU_VEGA20_DOORBELL_MAX_ASSIGNMENT = 0x18F, + AMDGPU_VEGA20_DOORBELL_INVALID = 0xFFFF +} AMDGPU_VEGA20_DOORBELL_ASSIGNMENT; + +/* + * 64bit doorbell, offset are in QWORD, occupy 2KB doorbell space + */ +typedef enum _AMDGPU_DOORBELL64_ASSIGNMENT +{ + /* + * All compute related doorbells: kiq, hiq, diq, traditional compute queue, user queue, should locate in + * a continues range so that programming CP_MEC_DOORBELL_RANGE_LOWER/UPPER can cover this range. + * Compute related doorbells are allocated from 0x00 to 0x8a + */ + + + /* kernel scheduling */ + AMDGPU_DOORBELL64_KIQ = 0x00, + + /* HSA interface queue and debug queue */ + AMDGPU_DOORBELL64_HIQ = 0x01, + AMDGPU_DOORBELL64_DIQ = 0x02, + + /* Compute engines */ + AMDGPU_DOORBELL64_MEC_RING0 = 0x03, + AMDGPU_DOORBELL64_MEC_RING1 = 0x04, + AMDGPU_DOORBELL64_MEC_RING2 = 0x05, + AMDGPU_DOORBELL64_MEC_RING3 = 0x06, + AMDGPU_DOORBELL64_MEC_RING4 = 0x07, + AMDGPU_DOORBELL64_MEC_RING5 = 0x08, + AMDGPU_DOORBELL64_MEC_RING6 = 0x09, + AMDGPU_DOORBELL64_MEC_RING7 = 0x0a, + + /* User queue doorbell range (128 doorbells) */ + AMDGPU_DOORBELL64_USERQUEUE_START = 0x0b, + AMDGPU_DOORBELL64_USERQUEUE_END = 0x8a, + + /* Graphics engine */ + AMDGPU_DOORBELL64_GFX_RING0 = 0x8b, + + /* + * Other graphics doorbells can be allocated here: from 0x8c to 0xdf + * Graphics voltage island aperture 1 + * default non-graphics QWORD index is 0xe0 - 0xFF inclusive + */ + + /* For vega10 sriov, the sdma doorbell must be fixed as follow + * to keep the same setting with host driver, or it will + * happen conflicts + */ + AMDGPU_DOORBELL64_sDMA_ENGINE0 = 0xF0, + AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE0 = 0xF1, + AMDGPU_DOORBELL64_sDMA_ENGINE1 = 0xF2, + AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE1 = 0xF3, + + /* Interrupt handler */ + AMDGPU_DOORBELL64_IH = 0xF4, /* For legacy interrupt ring buffer */ + AMDGPU_DOORBELL64_IH_RING1 = 0xF5, /* For page migration request log */ + AMDGPU_DOORBELL64_IH_RING2 = 0xF6, /* For page migration translation/invalidation log */ + + /* VCN engine use 32 bits doorbell */ + AMDGPU_DOORBELL64_VCN0_1 = 0xF8, /* lower 32 bits for VNC0 and upper 32 bits for VNC1 */ + AMDGPU_DOORBELL64_VCN2_3 = 0xF9, + AMDGPU_DOORBELL64_VCN4_5 = 0xFA, + AMDGPU_DOORBELL64_VCN6_7 = 0xFB, + + /* overlap the doorbell assignment with VCN as they are mutually exclusive + * VCE engine's doorbell is 32 bit and two VCE ring share one QWORD + */ + AMDGPU_DOORBELL64_UVD_RING0_1 = 0xF8, + AMDGPU_DOORBELL64_UVD_RING2_3 = 0xF9, + AMDGPU_DOORBELL64_UVD_RING4_5 = 0xFA, + AMDGPU_DOORBELL64_UVD_RING6_7 = 0xFB, + + AMDGPU_DOORBELL64_VCE_RING0_1 = 0xFC, + AMDGPU_DOORBELL64_VCE_RING2_3 = 0xFD, + AMDGPU_DOORBELL64_VCE_RING4_5 = 0xFE, + AMDGPU_DOORBELL64_VCE_RING6_7 = 0xFF, + + AMDGPU_DOORBELL64_FIRST_NON_CP = AMDGPU_DOORBELL64_sDMA_ENGINE0, + AMDGPU_DOORBELL64_LAST_NON_CP = AMDGPU_DOORBELL64_VCE_RING6_7, + + AMDGPU_DOORBELL64_MAX_ASSIGNMENT = 0xFF, + AMDGPU_DOORBELL64_INVALID = 0xFFFF +} AMDGPU_DOORBELL64_ASSIGNMENT; + +u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index); +void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v); +u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index); +void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v); + +#define RDOORBELL32(index) amdgpu_mm_rdoorbell(adev, (index)) +#define WDOORBELL32(index, v) amdgpu_mm_wdoorbell(adev, (index), (v)) +#define RDOORBELL64(index) amdgpu_mm_rdoorbell64(adev, (index)) +#define WDOORBELL64(index, v) amdgpu_mm_wdoorbell64(adev, (index), (v)) + diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c index 1c4595562f8f..344967df3137 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c @@ -184,61 +184,6 @@ u32 amdgpu_dpm_get_vrefresh(struct amdgpu_device *adev) return vrefresh; } -void amdgpu_calculate_u_and_p(u32 i, u32 r_c, u32 p_b, - u32 *p, u32 *u) -{ - u32 b_c = 0; - u32 i_c; - u32 tmp; - - i_c = (i * r_c) / 100; - tmp = i_c >> p_b; - - while (tmp) { - b_c++; - tmp >>= 1; - } - - *u = (b_c + 1) / 2; - *p = i_c / (1 << (2 * (*u))); -} - -int amdgpu_calculate_at(u32 t, u32 h, u32 fh, u32 fl, u32 *tl, u32 *th) -{ - u32 k, a, ah, al; - u32 t1; - - if ((fl == 0) || (fh == 0) || (fl > fh)) - return -EINVAL; - - k = (100 * fh) / fl; - t1 = (t * (k - 100)); - a = (1000 * (100 * h + t1)) / (10000 + (t1 / 100)); - a = (a + 5) / 10; - ah = ((a * t) + 5000) / 10000; - al = a - ah; - - *th = t - ah; - *tl = t + al; - - return 0; -} - -bool amdgpu_is_uvd_state(u32 class, u32 class2) -{ - if (class & ATOM_PPLIB_CLASSIFICATION_UVDSTATE) - return true; - if (class & ATOM_PPLIB_CLASSIFICATION_HD2STATE) - return true; - if (class & ATOM_PPLIB_CLASSIFICATION_HDSTATE) - return true; - if (class & ATOM_PPLIB_CLASSIFICATION_SDSTATE) - return true; - if (class2 & ATOM_PPLIB_CLASSIFICATION2_MVC) - return true; - return false; -} - bool amdgpu_is_internal_thermal_sensor(enum amdgpu_int_thermal_type sensor) { switch (sensor) { @@ -949,39 +894,6 @@ enum amdgpu_pcie_gen amdgpu_get_pcie_gen_support(struct amdgpu_device *adev, return AMDGPU_PCIE_GEN1; } -u16 amdgpu_get_pcie_lane_support(struct amdgpu_device *adev, - u16 asic_lanes, - u16 default_lanes) -{ - switch (asic_lanes) { - case 0: - default: - return default_lanes; - case 1: - return 1; - case 2: - return 2; - case 4: - return 4; - case 8: - return 8; - case 12: - return 12; - case 16: - return 16; - } -} - -u8 amdgpu_encode_pci_lane_width(u32 lanes) -{ - u8 encoded_lanes[] = { 0, 1, 2, 0, 3, 0, 0, 0, 4, 0, 0, 0, 5, 0, 0, 0, 6 }; - - if (lanes > 16) - return 0; - - return encoded_lanes[lanes]; -} - struct amd_vce_state* amdgpu_get_vce_clock_state(void *handle, u32 idx) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h index f972cd156795..e871e022c129 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h @@ -364,6 +364,14 @@ enum amdgpu_pcie_gen { ((adev)->powerplay.pp_funcs->enable_mgpu_fan_boost(\ (adev)->powerplay.pp_handle)) +#define amdgpu_dpm_get_ppfeature_status(adev, buf) \ + ((adev)->powerplay.pp_funcs->get_ppfeature_status(\ + (adev)->powerplay.pp_handle, (buf))) + +#define amdgpu_dpm_set_ppfeature_status(adev, ppfeatures) \ + ((adev)->powerplay.pp_funcs->set_ppfeature_status(\ + (adev)->powerplay.pp_handle, (ppfeatures))) + struct amdgpu_dpm { struct amdgpu_ps *ps; /* number of valid power states */ @@ -478,10 +486,6 @@ void amdgpu_dpm_print_ps_status(struct amdgpu_device *adev, u32 amdgpu_dpm_get_vblank_time(struct amdgpu_device *adev); u32 amdgpu_dpm_get_vrefresh(struct amdgpu_device *adev); void amdgpu_dpm_get_active_displays(struct amdgpu_device *adev); -bool amdgpu_is_uvd_state(u32 class, u32 class2); -void amdgpu_calculate_u_and_p(u32 i, u32 r_c, u32 p_b, - u32 *p, u32 *u); -int amdgpu_calculate_at(u32 t, u32 h, u32 fh, u32 fl, u32 *tl, u32 *th); bool amdgpu_is_internal_thermal_sensor(enum amdgpu_int_thermal_type sensor); @@ -497,11 +501,6 @@ enum amdgpu_pcie_gen amdgpu_get_pcie_gen_support(struct amdgpu_device *adev, enum amdgpu_pcie_gen asic_gen, enum amdgpu_pcie_gen default_gen); -u16 amdgpu_get_pcie_lane_support(struct amdgpu_device *adev, - u16 asic_lanes, - u16 default_lanes); -u8 amdgpu_encode_pci_lane_width(u32 lanes); - struct amd_vce_state* amdgpu_get_vce_clock_state(void *handle, u32 idx); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 74b611e8a1b1..7419ea8a388b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -32,7 +32,7 @@ #include <linux/module.h> #include <linux/pm_runtime.h> #include <linux/vga_switcheroo.h> -#include <drm/drm_crtc_helper.h> +#include <drm/drm_probe_helper.h> #include "amdgpu.h" #include "amdgpu_irq.h" @@ -71,9 +71,12 @@ * - 3.25.0 - Add support for sensor query info (stable pstate sclk/mclk). * - 3.26.0 - GFX9: Process AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE. * - 3.27.0 - Add new chunk to to AMDGPU_CS to enable BO_LIST creation. + * - 3.28.0 - Add AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES + * - 3.29.0 - Add AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID + * - 3.30.0 - Add AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE. */ #define KMS_DRIVER_MAJOR 3 -#define KMS_DRIVER_MINOR 27 +#define KMS_DRIVER_MINOR 30 #define KMS_DRIVER_PATCHLEVEL 0 int amdgpu_vram_limit = 0; @@ -454,9 +457,10 @@ module_param_named(cntl_sb_buf_per_se, amdgpu_cntl_sb_buf_per_se, int, 0444); /** * DOC: param_buf_per_se (int) - * Override the size of Off-Chip Pramater Cache per Shader Engine in Byte. The default is 0 (depending on gfx). + * Override the size of Off-Chip Parameter Cache per Shader Engine in Byte. + * The default is 0 (depending on gfx). */ -MODULE_PARM_DESC(param_buf_per_se, "the size of Off-Chip Pramater Cache per Shader Engine (default depending on gfx)"); +MODULE_PARM_DESC(param_buf_per_se, "the size of Off-Chip Parameter Cache per Shader Engine (default depending on gfx)"); module_param_named(param_buf_per_se, amdgpu_param_buf_per_se, int, 0444); /** @@ -864,6 +868,7 @@ static const struct pci_device_id pciidlist[] = { /* VEGAM */ {0x1002, 0x694C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGAM}, {0x1002, 0x694E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGAM}, + {0x1002, 0x694F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGAM}, /* Vega 10 */ {0x1002, 0x6860, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10}, {0x1002, 0x6861, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10}, @@ -1174,6 +1179,22 @@ static const struct file_operations amdgpu_driver_kms_fops = { #endif }; +int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv) +{ + struct drm_file *file; + + if (!filp) + return -EINVAL; + + if (filp->f_op != &amdgpu_driver_kms_fops) { + return -EINVAL; + } + + file = filp->private_data; + *fpriv = file->driver_priv; + return 0; +} + static bool amdgpu_get_crtc_scanout_position(struct drm_device *dev, unsigned int pipe, bool in_vblank_irq, int *vpos, int *hpos, @@ -1187,7 +1208,7 @@ amdgpu_get_crtc_scanout_position(struct drm_device *dev, unsigned int pipe, static struct drm_driver kms_driver = { .driver_features = DRIVER_USE_AGP | DRIVER_ATOMIC | - DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | DRIVER_GEM | + DRIVER_GEM | DRIVER_PRIME | DRIVER_RENDER | DRIVER_MODESET | DRIVER_SYNCOBJ, .load = amdgpu_driver_load_kms, .open = amdgpu_driver_open_kms, @@ -1227,9 +1248,6 @@ static struct drm_driver kms_driver = { .patchlevel = KMS_DRIVER_PATCHLEVEL, }; -static struct drm_driver *driver; -static struct pci_driver *pdriver; - static struct pci_driver amdgpu_kms_pci_driver = { .name = DRIVER_NAME, .id_table = pciidlist, @@ -1259,16 +1277,14 @@ static int __init amdgpu_init(void) goto error_fence; DRM_INFO("amdgpu kernel modesetting enabled.\n"); - driver = &kms_driver; - pdriver = &amdgpu_kms_pci_driver; - driver->num_ioctls = amdgpu_max_kms_ioctl; + kms_driver.num_ioctls = amdgpu_max_kms_ioctl; amdgpu_register_atpx_handler(); /* Ignore KFD init failures. Normal when CONFIG_HSA_AMD is not set. */ amdgpu_amdkfd_init(); /* let modprobe override vga console setting */ - return pci_register_driver(pdriver); + return pci_register_driver(&amdgpu_kms_pci_driver); error_fence: amdgpu_sync_fini(); @@ -1280,7 +1296,7 @@ error_sync: static void __exit amdgpu_exit(void) { amdgpu_amdkfd_fini(); - pci_unregister_driver(pdriver); + pci_unregister_driver(&amdgpu_kms_pci_driver); amdgpu_unregister_atpx_handler(); amdgpu_sync_fini(); amdgpu_fence_slab_fini(); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 5448cf27654e..ee47c11e92ce 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -398,9 +398,9 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring, ring->fence_drv.irq_type = irq_type; ring->fence_drv.initialized = true; - dev_dbg(adev->dev, "fence driver on ring %d use gpu addr 0x%016llx, " - "cpu addr 0x%p\n", ring->idx, - ring->fence_drv.gpu_addr, ring->fence_drv.cpu_addr); + DRM_DEV_DEBUG(adev->dev, "fence driver on ring %s use gpu addr " + "0x%016llx, cpu addr 0x%p\n", ring->name, + ring->fence_drv.gpu_addr, ring->fence_drv.cpu_addr); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c index 11fea28f8ad3..6d11e1721147 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c @@ -248,7 +248,7 @@ int amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, } mb(); amdgpu_asic_flush_hdp(adev, NULL); - amdgpu_gmc_flush_gpu_tlb(adev, 0); + amdgpu_gmc_flush_gpu_tlb(adev, 0, 0); return 0; } @@ -259,6 +259,8 @@ int amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, * @offset: offset into the GPU's gart aperture * @pages: number of pages to bind * @dma_addr: DMA addresses of pages + * @flags: page table entry flags + * @dst: CPU address of the gart table * * Map the dma_addresses into GART entries (all asics). * Returns 0 for success, -EINVAL for failure. @@ -331,7 +333,7 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset, mb(); amdgpu_asic_flush_hdp(adev, NULL); - amdgpu_gmc_flush_gpu_tlb(adev, 0); + amdgpu_gmc_flush_gpu_tlb(adev, 0, 0); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h index 9ff62887e4e3..afa2e2877d87 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h @@ -41,6 +41,7 @@ struct amdgpu_bo; struct amdgpu_gart { struct amdgpu_bo *bo; + /* CPU kmapped address of gart table */ void *ptr; unsigned num_gpu_pages; unsigned num_cpu_pages; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h index ecbcefe49a98..f89f5734d985 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h @@ -37,6 +37,8 @@ struct amdgpu_gds { struct amdgpu_gds_asic_info mem; struct amdgpu_gds_asic_info gws; struct amdgpu_gds_asic_info oa; + uint32_t gds_compute_max_wave_id; + /* At present, GDS, GWS and OA resources for gfx (graphics) * is always pre-allocated and available for graphics operation. * Such resource is shared between all gfx clients. diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 7b3d1ebda9df..d21dd2f369da 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -54,10 +54,6 @@ int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size, memset(&bp, 0, sizeof(bp)); *obj = NULL; - /* At least align on page size */ - if (alignment < PAGE_SIZE) { - alignment = PAGE_SIZE; - } bp.size = size; bp.byte_align = alignment; @@ -169,7 +165,7 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj, INIT_LIST_HEAD(&duplicates); tv.bo = &bo->tbo; - tv.shared = true; + tv.num_shared = 1; list_add(&tv.head, &list); amdgpu_vm_get_pd_bo(vm, &list, &vm_pd); @@ -244,9 +240,6 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, return -EINVAL; } flags |= AMDGPU_GEM_CREATE_NO_CPU_ACCESS; - /* GDS allocations must be DW aligned */ - if (args->in.domains & AMDGPU_GEM_DOMAIN_GDS) - size = ALIGN(size, 4); } if (flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID) { @@ -604,7 +597,10 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, return -ENOENT; abo = gem_to_amdgpu_bo(gobj); tv.bo = &abo->tbo; - tv.shared = !!(abo->flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID); + if (abo->flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID) + tv.num_shared = 1; + else + tv.num_shared = 0; list_add(&tv.head, &list); } else { gobj = NULL; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h index d63daba9b17c..f1ddfc50bcc7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h @@ -54,6 +54,8 @@ void *amdgpu_gem_prime_vmap(struct drm_gem_object *obj); void amdgpu_gem_prime_vunmap(struct drm_gem_object *obj, void *vaddr); int amdgpu_gem_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma); +extern const struct dma_buf_ops amdgpu_dmabuf_ops; + /* * GEM objects. */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index 1a656b8657f7..97a60da62004 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -25,6 +25,7 @@ #include <drm/drmP.h> #include "amdgpu.h" #include "amdgpu_gfx.h" +#include "amdgpu_rlc.h" /* delay 0.1 second to enable gfx off feature */ #define GFX_OFF_DELAY_ENABLE msecs_to_jiffies(100) @@ -249,7 +250,7 @@ int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, ring->adev = NULL; ring->ring_obj = NULL; ring->use_doorbell = true; - ring->doorbell_index = AMDGPU_DOORBELL_KIQ; + ring->doorbell_index = adev->doorbell_index.kiq; r = amdgpu_gfx_kiq_acquire(adev, ring); if (r) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h index b61b5c11aead..f790e15bcd08 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h @@ -29,6 +29,7 @@ */ #include "clearstate_defs.h" #include "amdgpu_ring.h" +#include "amdgpu_rlc.h" /* GFX current status */ #define AMDGPU_GFX_NORMAL_MODE 0x00000000L @@ -37,59 +38,6 @@ #define AMDGPU_GFX_CG_DISABLED_MODE 0x00000004L #define AMDGPU_GFX_LBPW_DISABLED_MODE 0x00000008L - -struct amdgpu_rlc_funcs { - void (*enter_safe_mode)(struct amdgpu_device *adev); - void (*exit_safe_mode)(struct amdgpu_device *adev); -}; - -struct amdgpu_rlc { - /* for power gating */ - struct amdgpu_bo *save_restore_obj; - uint64_t save_restore_gpu_addr; - volatile uint32_t *sr_ptr; - const u32 *reg_list; - u32 reg_list_size; - /* for clear state */ - struct amdgpu_bo *clear_state_obj; - uint64_t clear_state_gpu_addr; - volatile uint32_t *cs_ptr; - const struct cs_section_def *cs_data; - u32 clear_state_size; - /* for cp tables */ - struct amdgpu_bo *cp_table_obj; - uint64_t cp_table_gpu_addr; - volatile uint32_t *cp_table_ptr; - u32 cp_table_size; - - /* safe mode for updating CG/PG state */ - bool in_safe_mode; - const struct amdgpu_rlc_funcs *funcs; - - /* for firmware data */ - u32 save_and_restore_offset; - u32 clear_state_descriptor_offset; - u32 avail_scratch_ram_locations; - u32 reg_restore_list_size; - u32 reg_list_format_start; - u32 reg_list_format_separate_start; - u32 starting_offsets_start; - u32 reg_list_format_size_bytes; - u32 reg_list_size_bytes; - u32 reg_list_format_direct_reg_list_length; - u32 save_restore_list_cntl_size_bytes; - u32 save_restore_list_gpm_size_bytes; - u32 save_restore_list_srm_size_bytes; - - u32 *register_list_format; - u32 *register_restore; - u8 *save_restore_list_cntl; - u8 *save_restore_list_gpm; - u8 *save_restore_list_srm; - - bool is_rlc_v2_1; -}; - #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES struct amdgpu_mec { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h index 6fa7ef446e46..81e6070d255b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h @@ -64,7 +64,7 @@ struct amdgpu_vmhub { struct amdgpu_gmc_funcs { /* flush the vm tlb via mmio */ void (*flush_gpu_tlb)(struct amdgpu_device *adev, - uint32_t vmid); + uint32_t vmid, uint32_t flush_type); /* flush the vm tlb via ring */ uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid, uint64_t pd_addr); @@ -89,7 +89,7 @@ struct amdgpu_gmc_funcs { struct amdgpu_xgmi { /* from psp */ - u64 device_id; + u64 node_id; u64 hive_id; /* fixed per family */ u64 node_segment_size; @@ -99,6 +99,7 @@ struct amdgpu_xgmi { unsigned num_physical_nodes; /* gpu list in the same hive */ struct list_head head; + bool supported; }; struct amdgpu_gmc { @@ -151,7 +152,7 @@ struct amdgpu_gmc { struct amdgpu_xgmi xgmi; }; -#define amdgpu_gmc_flush_gpu_tlb(adev, vmid) (adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid)) +#define amdgpu_gmc_flush_gpu_tlb(adev, vmid, type) (adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (type)) #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr)) #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid)) #define amdgpu_gmc_set_pte_pde(adev, pt, idx, addr, flags) (adev)->gmc.gmc_funcs->set_pte_pde((adev), (pt), (idx), (addr), (flags)) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index b8963b725dfa..fe393a46f881 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -35,6 +35,7 @@ #include "amdgpu_trace.h" #define AMDGPU_IB_TEST_TIMEOUT msecs_to_jiffies(1000) +#define AMDGPU_IB_TEST_GFX_XGMI_TIMEOUT msecs_to_jiffies(2000) /* * IB @@ -146,7 +147,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, fence_ctx = 0; } - if (!ring->ready) { + if (!ring->sched.ready) { dev_err(adev->dev, "couldn't schedule ib on ring <%s>\n", ring->name); return -EINVAL; } @@ -202,12 +203,12 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, amdgpu_asic_flush_hdp(adev, ring); } + if (need_ctx_switch) + status |= AMDGPU_HAVE_CTX_SWITCH; + skip_preamble = ring->current_ctx == fence_ctx; if (job && ring->funcs->emit_cntxcntl) { - if (need_ctx_switch) - status |= AMDGPU_HAVE_CTX_SWITCH; status |= job->preamble_status; - amdgpu_ring_emit_cntxcntl(ring, status); } @@ -221,9 +222,8 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, !amdgpu_sriov_vf(adev)) /* for SRIOV preemption, Preamble CE ib must be inserted anyway */ continue; - amdgpu_ring_emit_ib(ring, ib, job ? job->vmid : 0, - need_ctx_switch); - need_ctx_switch = false; + amdgpu_ring_emit_ib(ring, job, ib, status); + status &= ~AMDGPU_HAVE_CTX_SWITCH; } if (ring->funcs->emit_tmz) @@ -345,21 +345,18 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev) * cost waiting for it coming back under RUNTIME only */ tmo_gfx = 8 * AMDGPU_IB_TEST_TIMEOUT; + } else if (adev->gmc.xgmi.hive_id) { + tmo_gfx = AMDGPU_IB_TEST_GFX_XGMI_TIMEOUT; } - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + for (i = 0; i < adev->num_rings; ++i) { struct amdgpu_ring *ring = adev->rings[i]; long tmo; - if (!ring || !ring->ready) - continue; - - /* skip IB tests for KIQ in general for the below reasons: - * 1. We never submit IBs to the KIQ - * 2. KIQ doesn't use the EOP interrupts, - * we use some other CP interrupt. + /* KIQ rings don't have an IB test because we never submit IBs + * to them and they have no interrupt support. */ - if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ) + if (!ring->sched.ready || !ring->funcs->test_ib) continue; /* MM engine need more time */ @@ -374,20 +371,23 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev) tmo = tmo_gfx; r = amdgpu_ring_test_ib(ring, tmo); - if (r) { - ring->ready = false; - - if (ring == &adev->gfx.gfx_ring[0]) { - /* oh, oh, that's really bad */ - DRM_ERROR("amdgpu: failed testing IB on GFX ring (%d).\n", r); - adev->accel_working = false; - return r; - - } else { - /* still not good, but we can live with it */ - DRM_ERROR("amdgpu: failed testing IB on ring %d (%d).\n", i, r); - ret = r; - } + if (!r) { + DRM_DEV_DEBUG(adev->dev, "ib test on %s succeeded\n", + ring->name); + continue; + } + + ring->sched.ready = false; + DRM_DEV_ERROR(adev->dev, "IB test failed on %s (%d).\n", + ring->name, r); + + if (ring == &adev->gfx.gfx_ring[0]) { + /* oh, oh, that's really bad */ + adev->accel_working = false; + return r; + + } else { + ret = r; } } return ret; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c index 8af67f649660..1c50be3ab8a9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c @@ -52,6 +52,8 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, ih->use_bus_addr = use_bus_addr; if (use_bus_addr) { + dma_addr_t dma_addr; + if (ih->ring) return 0; @@ -59,21 +61,26 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, * add them to the end of the ring allocation. */ ih->ring = dma_alloc_coherent(adev->dev, ih->ring_size + 8, - &ih->rb_dma_addr, GFP_KERNEL); + &dma_addr, GFP_KERNEL); if (ih->ring == NULL) return -ENOMEM; memset((void *)ih->ring, 0, ih->ring_size + 8); - ih->wptr_offs = (ih->ring_size / 4) + 0; - ih->rptr_offs = (ih->ring_size / 4) + 1; + ih->gpu_addr = dma_addr; + ih->wptr_addr = dma_addr + ih->ring_size; + ih->wptr_cpu = &ih->ring[ih->ring_size / 4]; + ih->rptr_addr = dma_addr + ih->ring_size + 4; + ih->rptr_cpu = &ih->ring[(ih->ring_size / 4) + 1]; } else { - r = amdgpu_device_wb_get(adev, &ih->wptr_offs); + unsigned wptr_offs, rptr_offs; + + r = amdgpu_device_wb_get(adev, &wptr_offs); if (r) return r; - r = amdgpu_device_wb_get(adev, &ih->rptr_offs); + r = amdgpu_device_wb_get(adev, &rptr_offs); if (r) { - amdgpu_device_wb_free(adev, ih->wptr_offs); + amdgpu_device_wb_free(adev, wptr_offs); return r; } @@ -82,10 +89,15 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, &ih->ring_obj, &ih->gpu_addr, (void **)&ih->ring); if (r) { - amdgpu_device_wb_free(adev, ih->rptr_offs); - amdgpu_device_wb_free(adev, ih->wptr_offs); + amdgpu_device_wb_free(adev, rptr_offs); + amdgpu_device_wb_free(adev, wptr_offs); return r; } + + ih->wptr_addr = adev->wb.gpu_addr + wptr_offs * 4; + ih->wptr_cpu = &adev->wb.wb[wptr_offs]; + ih->rptr_addr = adev->wb.gpu_addr + rptr_offs * 4; + ih->rptr_cpu = &adev->wb.wb[rptr_offs]; } return 0; } @@ -109,13 +121,13 @@ void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) * add them to the end of the ring allocation. */ dma_free_coherent(adev->dev, ih->ring_size + 8, - (void *)ih->ring, ih->rb_dma_addr); + (void *)ih->ring, ih->gpu_addr); ih->ring = NULL; } else { amdgpu_bo_free_kernel(&ih->ring_obj, &ih->gpu_addr, (void **)&ih->ring); - amdgpu_device_wb_free(adev, ih->wptr_offs); - amdgpu_device_wb_free(adev, ih->rptr_offs); + amdgpu_device_wb_free(adev, (ih->wptr_addr - ih->gpu_addr) / 4); + amdgpu_device_wb_free(adev, (ih->rptr_addr - ih->gpu_addr) / 4); } } @@ -128,16 +140,14 @@ void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) * Interrupt hander (VI), walk the IH ring. * Returns irq process return code. */ -int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, - void (*callback)(struct amdgpu_device *adev, - struct amdgpu_ih_ring *ih)) +int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) { u32 wptr; if (!ih->enabled || adev->shutdown) return IRQ_NONE; - wptr = amdgpu_ih_get_wptr(adev); + wptr = amdgpu_ih_get_wptr(adev, ih); restart_ih: /* is somebody else already processing irqs? */ @@ -150,15 +160,15 @@ restart_ih: rmb(); while (ih->rptr != wptr) { - callback(adev, ih); + amdgpu_irq_dispatch(adev, ih); ih->rptr &= ih->ptr_mask; } - amdgpu_ih_set_rptr(adev); + amdgpu_ih_set_rptr(adev, ih); atomic_set(&ih->lock, 0); /* make sure wptr hasn't changed while processing */ - wptr = amdgpu_ih_get_wptr(adev); + wptr = amdgpu_ih_get_wptr(adev, ih); if (wptr != ih->rptr) goto restart_ih; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h index 9ce8c93ec19b..113a1ba13d4a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h @@ -31,42 +31,44 @@ struct amdgpu_iv_entry; * R6xx+ IH ring */ struct amdgpu_ih_ring { - struct amdgpu_bo *ring_obj; - volatile uint32_t *ring; - unsigned rptr; unsigned ring_size; - uint64_t gpu_addr; uint32_t ptr_mask; - atomic_t lock; - bool enabled; - unsigned wptr_offs; - unsigned rptr_offs; u32 doorbell_index; bool use_doorbell; bool use_bus_addr; - dma_addr_t rb_dma_addr; /* only used when use_bus_addr = true */ + + struct amdgpu_bo *ring_obj; + volatile uint32_t *ring; + uint64_t gpu_addr; + + uint64_t wptr_addr; + volatile uint32_t *wptr_cpu; + + uint64_t rptr_addr; + volatile uint32_t *rptr_cpu; + + bool enabled; + unsigned rptr; + atomic_t lock; }; /* provided by the ih block */ struct amdgpu_ih_funcs { /* ring read/write ptr handling, called from interrupt context */ - u32 (*get_wptr)(struct amdgpu_device *adev); - bool (*prescreen_iv)(struct amdgpu_device *adev); - void (*decode_iv)(struct amdgpu_device *adev, + u32 (*get_wptr)(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih); + void (*decode_iv)(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, struct amdgpu_iv_entry *entry); - void (*set_rptr)(struct amdgpu_device *adev); + void (*set_rptr)(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih); }; -#define amdgpu_ih_get_wptr(adev) (adev)->irq.ih_funcs->get_wptr((adev)) -#define amdgpu_ih_prescreen_iv(adev) (adev)->irq.ih_funcs->prescreen_iv((adev)) -#define amdgpu_ih_decode_iv(adev, iv) (adev)->irq.ih_funcs->decode_iv((adev), (iv)) -#define amdgpu_ih_set_rptr(adev) (adev)->irq.ih_funcs->set_rptr((adev)) +#define amdgpu_ih_get_wptr(adev, ih) (adev)->irq.ih_funcs->get_wptr((adev), (ih)) +#define amdgpu_ih_decode_iv(adev, iv) \ + (adev)->irq.ih_funcs->decode_iv((adev), (ih), (iv)) +#define amdgpu_ih_set_rptr(adev, ih) (adev)->irq.ih_funcs->set_rptr((adev), (ih)) int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, unsigned ring_size, bool use_bus_addr); void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih); -int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, - void (*callback)(struct amdgpu_device *adev, - struct amdgpu_ih_ring *ih)); +int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c index 52c17f6219a7..af4c3b1af322 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c @@ -94,23 +94,6 @@ static void amdgpu_hotplug_work_func(struct work_struct *work) } /** - * amdgpu_irq_reset_work_func - execute GPU reset - * - * @work: work struct pointer - * - * Execute scheduled GPU reset (Cayman+). - * This function is called when the IRQ handler thinks we need a GPU reset. - */ -static void amdgpu_irq_reset_work_func(struct work_struct *work) -{ - struct amdgpu_device *adev = container_of(work, struct amdgpu_device, - reset_work); - - if (!amdgpu_sriov_vf(adev) && amdgpu_device_should_recover_gpu(adev)) - amdgpu_device_gpu_recover(adev, NULL); -} - -/** * amdgpu_irq_disable_all - disable *all* interrupts * * @adev: amdgpu device pointer @@ -148,34 +131,6 @@ void amdgpu_irq_disable_all(struct amdgpu_device *adev) } /** - * amdgpu_irq_callback - callback from the IH ring - * - * @adev: amdgpu device pointer - * @ih: amdgpu ih ring - * - * Callback from IH ring processing to handle the entry at the current position - * and advance the read pointer. - */ -static void amdgpu_irq_callback(struct amdgpu_device *adev, - struct amdgpu_ih_ring *ih) -{ - u32 ring_index = ih->rptr >> 2; - struct amdgpu_iv_entry entry; - - /* Prescreening of high-frequency interrupts */ - if (!amdgpu_ih_prescreen_iv(adev)) - return; - - /* Before dispatching irq to IP blocks, send it to amdkfd */ - amdgpu_amdkfd_interrupt(adev, (const void *) &ih->ring[ring_index]); - - entry.iv_entry = (const uint32_t *)&ih->ring[ring_index]; - amdgpu_ih_decode_iv(adev, &entry); - - amdgpu_irq_dispatch(adev, &entry); -} - -/** * amdgpu_irq_handler - IRQ handler * * @irq: IRQ number (unused) @@ -192,13 +147,43 @@ irqreturn_t amdgpu_irq_handler(int irq, void *arg) struct amdgpu_device *adev = dev->dev_private; irqreturn_t ret; - ret = amdgpu_ih_process(adev, &adev->irq.ih, amdgpu_irq_callback); + ret = amdgpu_ih_process(adev, &adev->irq.ih); if (ret == IRQ_HANDLED) pm_runtime_mark_last_busy(dev->dev); return ret; } /** + * amdgpu_irq_handle_ih1 - kick of processing for IH1 + * + * @work: work structure in struct amdgpu_irq + * + * Kick of processing IH ring 1. + */ +static void amdgpu_irq_handle_ih1(struct work_struct *work) +{ + struct amdgpu_device *adev = container_of(work, struct amdgpu_device, + irq.ih1_work); + + amdgpu_ih_process(adev, &adev->irq.ih1); +} + +/** + * amdgpu_irq_handle_ih2 - kick of processing for IH2 + * + * @work: work structure in struct amdgpu_irq + * + * Kick of processing IH ring 2. + */ +static void amdgpu_irq_handle_ih2(struct work_struct *work) +{ + struct amdgpu_device *adev = container_of(work, struct amdgpu_device, + irq.ih2_work); + + amdgpu_ih_process(adev, &adev->irq.ih2); +} + +/** * amdgpu_msi_ok - check whether MSI functionality is enabled * * @adev: amdgpu device pointer (unused) @@ -262,7 +247,8 @@ int amdgpu_irq_init(struct amdgpu_device *adev) amdgpu_hotplug_work_func); } - INIT_WORK(&adev->reset_work, amdgpu_irq_reset_work_func); + INIT_WORK(&adev->irq.ih1_work, amdgpu_irq_handle_ih1); + INIT_WORK(&adev->irq.ih2_work, amdgpu_irq_handle_ih2); adev->irq.installed = true; r = drm_irq_install(adev->ddev, adev->ddev->pdev->irq); @@ -270,7 +256,6 @@ int amdgpu_irq_init(struct amdgpu_device *adev) adev->irq.installed = false; if (!amdgpu_device_has_dc_support(adev)) flush_work(&adev->hotplug_work); - cancel_work_sync(&adev->reset_work); return r; } adev->ddev->max_vblank_count = 0x00ffffff; @@ -299,7 +284,6 @@ void amdgpu_irq_fini(struct amdgpu_device *adev) pci_disable_msi(adev->pdev); if (!amdgpu_device_has_dc_support(adev)) flush_work(&adev->hotplug_work); - cancel_work_sync(&adev->reset_work); } for (i = 0; i < AMDGPU_IRQ_CLIENTID_MAX; ++i) { @@ -387,44 +371,50 @@ int amdgpu_irq_add_id(struct amdgpu_device *adev, * Dispatches IRQ to IP blocks. */ void amdgpu_irq_dispatch(struct amdgpu_device *adev, - struct amdgpu_iv_entry *entry) + struct amdgpu_ih_ring *ih) { - unsigned client_id = entry->client_id; - unsigned src_id = entry->src_id; + u32 ring_index = ih->rptr >> 2; + struct amdgpu_iv_entry entry; + unsigned client_id, src_id; struct amdgpu_irq_src *src; + bool handled = false; int r; - trace_amdgpu_iv(entry); + entry.iv_entry = (const uint32_t *)&ih->ring[ring_index]; + amdgpu_ih_decode_iv(adev, &entry); + + trace_amdgpu_iv(ih - &adev->irq.ih, &entry); + + client_id = entry.client_id; + src_id = entry.src_id; if (client_id >= AMDGPU_IRQ_CLIENTID_MAX) { DRM_DEBUG("Invalid client_id in IV: %d\n", client_id); - return; - } - if (src_id >= AMDGPU_MAX_IRQ_SRC_ID) { + } else if (src_id >= AMDGPU_MAX_IRQ_SRC_ID) { DRM_DEBUG("Invalid src_id in IV: %d\n", src_id); - return; - } - if (adev->irq.virq[src_id]) { + } else if (adev->irq.virq[src_id]) { generic_handle_irq(irq_find_mapping(adev->irq.domain, src_id)); - } else { - if (!adev->irq.client[client_id].sources) { - DRM_DEBUG("Unregistered interrupt client_id: %d src_id: %d\n", - client_id, src_id); - return; - } - src = adev->irq.client[client_id].sources[src_id]; - if (!src) { - DRM_DEBUG("Unhandled interrupt src_id: %d\n", src_id); - return; - } + } else if (!adev->irq.client[client_id].sources) { + DRM_DEBUG("Unregistered interrupt client_id: %d src_id: %d\n", + client_id, src_id); - r = src->funcs->process(adev, src, entry); - if (r) + } else if ((src = adev->irq.client[client_id].sources[src_id])) { + r = src->funcs->process(adev, src, &entry); + if (r < 0) DRM_ERROR("error processing interrupt (%d)\n", r); + else if (r) + handled = true; + + } else { + DRM_DEBUG("Unhandled interrupt src_id: %d\n", src_id); } + + /* Send it to amdkfd as well if it isn't already handled */ + if (!handled) + amdgpu_amdkfd_interrupt(adev, entry.iv_entry); } /** diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h index f6ce171cb8aa..c718e94a55c9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h @@ -87,9 +87,11 @@ struct amdgpu_irq { /* status, etc. */ bool msi_enabled; /* msi enabled */ - /* interrupt ring */ - struct amdgpu_ih_ring ih; - const struct amdgpu_ih_funcs *ih_funcs; + /* interrupt rings */ + struct amdgpu_ih_ring ih, ih1, ih2; + const struct amdgpu_ih_funcs *ih_funcs; + struct work_struct ih1_work, ih2_work; + struct amdgpu_irq_src self_irq; /* gen irq stuff */ struct irq_domain *domain; /* GPU irq controller domain */ @@ -106,7 +108,7 @@ int amdgpu_irq_add_id(struct amdgpu_device *adev, unsigned client_id, unsigned src_id, struct amdgpu_irq_src *source); void amdgpu_irq_dispatch(struct amdgpu_device *adev, - struct amdgpu_iv_entry *entry); + struct amdgpu_ih_ring *ih); int amdgpu_irq_update(struct amdgpu_device *adev, struct amdgpu_irq_src *src, unsigned type); int amdgpu_irq_get(struct amdgpu_device *adev, struct amdgpu_irq_src *src, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 755f733bf0d9..0a17fb1af204 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -32,6 +32,9 @@ static void amdgpu_job_timedout(struct drm_sched_job *s_job) { struct amdgpu_ring *ring = to_amdgpu_ring(s_job->sched); struct amdgpu_job *job = to_amdgpu_job(s_job); + struct amdgpu_task_info ti; + + memset(&ti, 0, sizeof(struct amdgpu_task_info)); if (amdgpu_ring_soft_recovery(ring, job->vmid, s_job->s_fence->parent)) { DRM_ERROR("ring %s timeout, but soft recovered\n", @@ -39,9 +42,12 @@ static void amdgpu_job_timedout(struct drm_sched_job *s_job) return; } + amdgpu_vm_get_task_info(ring->adev, job->pasid, &ti); DRM_ERROR("ring %s timeout, signaled seq=%u, emitted seq=%u\n", job->base.sched->name, atomic_read(&ring->fence_drv.last_seq), ring->fence_drv.sync_seq); + DRM_ERROR("Process information: process %s pid %d thread %s pid %d\n", + ti.process_name, ti.tgid, ti.task_name, ti.pid); if (amdgpu_device_should_recover_gpu(ring->adev)) amdgpu_device_gpu_recover(ring->adev, job); @@ -112,6 +118,8 @@ static void amdgpu_job_free_cb(struct drm_sched_job *s_job) struct amdgpu_ring *ring = to_amdgpu_ring(s_job->sched); struct amdgpu_job *job = to_amdgpu_job(s_job); + drm_sched_job_cleanup(s_job); + amdgpu_ring_priority_put(ring, s_job->s_priority); dma_fence_put(job->fence); amdgpu_sync_free(&job->sync); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h index 57cfe78a262b..e1b46a6703de 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h @@ -33,6 +33,8 @@ #define to_amdgpu_job(sched_job) \ container_of((sched_job), struct amdgpu_job, base) +#define AMDGPU_JOB_GET_VMID(job) ((job) ? (job)->vmid : 0) + struct amdgpu_fence; struct amdgpu_job { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index 8f3d44e5e787..e860412043bb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -207,11 +207,12 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags) if (!r) { acpi_status = amdgpu_acpi_init(adev); if (acpi_status) - dev_dbg(&dev->pdev->dev, + dev_dbg(&dev->pdev->dev, "Error during ACPI methods call\n"); } if (amdgpu_device_is_px(dev)) { + dev_pm_set_driver_flags(dev->dev, DPM_FLAG_NEVER_SKIP); pm_runtime_use_autosuspend(dev->dev); pm_runtime_set_autosuspend_delay(dev->dev, 5000); pm_runtime_set_active(dev->dev); @@ -336,7 +337,7 @@ static int amdgpu_hw_ip_info(struct amdgpu_device *adev, case AMDGPU_HW_IP_GFX: type = AMD_IP_BLOCK_TYPE_GFX; for (i = 0; i < adev->gfx.num_gfx_rings; i++) - if (adev->gfx.gfx_ring[i].ready) + if (adev->gfx.gfx_ring[i].sched.ready) ++num_rings; ib_start_alignment = 32; ib_size_alignment = 32; @@ -344,7 +345,7 @@ static int amdgpu_hw_ip_info(struct amdgpu_device *adev, case AMDGPU_HW_IP_COMPUTE: type = AMD_IP_BLOCK_TYPE_GFX; for (i = 0; i < adev->gfx.num_compute_rings; i++) - if (adev->gfx.compute_ring[i].ready) + if (adev->gfx.compute_ring[i].sched.ready) ++num_rings; ib_start_alignment = 32; ib_size_alignment = 32; @@ -352,7 +353,7 @@ static int amdgpu_hw_ip_info(struct amdgpu_device *adev, case AMDGPU_HW_IP_DMA: type = AMD_IP_BLOCK_TYPE_SDMA; for (i = 0; i < adev->sdma.num_instances; i++) - if (adev->sdma.instance[i].ring.ready) + if (adev->sdma.instance[i].ring.sched.ready) ++num_rings; ib_start_alignment = 256; ib_size_alignment = 4; @@ -363,7 +364,7 @@ static int amdgpu_hw_ip_info(struct amdgpu_device *adev, if (adev->uvd.harvest_config & (1 << i)) continue; - if (adev->uvd.inst[i].ring.ready) + if (adev->uvd.inst[i].ring.sched.ready) ++num_rings; } ib_start_alignment = 64; @@ -372,7 +373,7 @@ static int amdgpu_hw_ip_info(struct amdgpu_device *adev, case AMDGPU_HW_IP_VCE: type = AMD_IP_BLOCK_TYPE_VCE; for (i = 0; i < adev->vce.num_rings; i++) - if (adev->vce.ring[i].ready) + if (adev->vce.ring[i].sched.ready) ++num_rings; ib_start_alignment = 4; ib_size_alignment = 1; @@ -384,7 +385,7 @@ static int amdgpu_hw_ip_info(struct amdgpu_device *adev, continue; for (j = 0; j < adev->uvd.num_enc_rings; j++) - if (adev->uvd.inst[i].ring_enc[j].ready) + if (adev->uvd.inst[i].ring_enc[j].sched.ready) ++num_rings; } ib_start_alignment = 64; @@ -392,7 +393,7 @@ static int amdgpu_hw_ip_info(struct amdgpu_device *adev, break; case AMDGPU_HW_IP_VCN_DEC: type = AMD_IP_BLOCK_TYPE_VCN; - if (adev->vcn.ring_dec.ready) + if (adev->vcn.ring_dec.sched.ready) ++num_rings; ib_start_alignment = 16; ib_size_alignment = 16; @@ -400,14 +401,14 @@ static int amdgpu_hw_ip_info(struct amdgpu_device *adev, case AMDGPU_HW_IP_VCN_ENC: type = AMD_IP_BLOCK_TYPE_VCN; for (i = 0; i < adev->vcn.num_enc_rings; i++) - if (adev->vcn.ring_enc[i].ready) + if (adev->vcn.ring_enc[i].sched.ready) ++num_rings; ib_start_alignment = 64; ib_size_alignment = 1; break; case AMDGPU_HW_IP_VCN_JPEG: type = AMD_IP_BLOCK_TYPE_VCN; - if (adev->vcn.ring_jpeg.ready) + if (adev->vcn.ring_jpeg.sched.ready) ++num_rings; ib_start_alignment = 16; ib_size_alignment = 16; @@ -978,7 +979,10 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv) } if (amdgpu_sriov_vf(adev)) { - r = amdgpu_map_static_csa(adev, &fpriv->vm, &fpriv->csa_va); + uint64_t csa_addr = amdgpu_csa_vaddr(adev) & AMDGPU_GMC_HOLE_MASK; + + r = amdgpu_map_static_csa(adev, &fpriv->vm, adev->virt.csa_obj, + &fpriv->csa_va, csa_addr, AMDGPU_CSA_SIZE); if (r) goto error_vm; } @@ -1048,8 +1052,8 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev, pasid = fpriv->vm.pasid; pd = amdgpu_bo_ref(fpriv->vm.root.base.bo); - amdgpu_vm_fini(adev, &fpriv->vm); amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr); + amdgpu_vm_fini(adev, &fpriv->vm); if (pasid) amdgpu_pasid_free_delayed(pd->tbo.resv, pasid); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c index e55508b39496..3e6823fdd939 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c @@ -238,44 +238,40 @@ static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node, * amdgpu_mn_invalidate_range_start_gfx - callback to notify about mm change * * @mn: our notifier - * @mm: the mm this callback is about - * @start: start of updated range - * @end: end of updated range + * @range: mmu notifier context * * Block for operations on BOs to finish and mark pages as accessed and * potentially dirty. */ static int amdgpu_mn_invalidate_range_start_gfx(struct mmu_notifier *mn, - struct mm_struct *mm, - unsigned long start, - unsigned long end, - bool blockable) + const struct mmu_notifier_range *range) { struct amdgpu_mn *amn = container_of(mn, struct amdgpu_mn, mn); struct interval_tree_node *it; + unsigned long end; /* notification is exclusive, but interval is inclusive */ - end -= 1; + end = range->end - 1; /* TODO we should be able to split locking for interval tree and * amdgpu_mn_invalidate_node */ - if (amdgpu_mn_read_lock(amn, blockable)) + if (amdgpu_mn_read_lock(amn, range->blockable)) return -EAGAIN; - it = interval_tree_iter_first(&amn->objects, start, end); + it = interval_tree_iter_first(&amn->objects, range->start, end); while (it) { struct amdgpu_mn_node *node; - if (!blockable) { + if (!range->blockable) { amdgpu_mn_read_unlock(amn); return -EAGAIN; } node = container_of(it, struct amdgpu_mn_node, it); - it = interval_tree_iter_next(it, start, end); + it = interval_tree_iter_next(it, range->start, end); - amdgpu_mn_invalidate_node(node, start, end); + amdgpu_mn_invalidate_node(node, range->start, end); } return 0; @@ -294,39 +290,38 @@ static int amdgpu_mn_invalidate_range_start_gfx(struct mmu_notifier *mn, * are restorted in amdgpu_mn_invalidate_range_end_hsa. */ static int amdgpu_mn_invalidate_range_start_hsa(struct mmu_notifier *mn, - struct mm_struct *mm, - unsigned long start, - unsigned long end, - bool blockable) + const struct mmu_notifier_range *range) { struct amdgpu_mn *amn = container_of(mn, struct amdgpu_mn, mn); struct interval_tree_node *it; + unsigned long end; /* notification is exclusive, but interval is inclusive */ - end -= 1; + end = range->end - 1; - if (amdgpu_mn_read_lock(amn, blockable)) + if (amdgpu_mn_read_lock(amn, range->blockable)) return -EAGAIN; - it = interval_tree_iter_first(&amn->objects, start, end); + it = interval_tree_iter_first(&amn->objects, range->start, end); while (it) { struct amdgpu_mn_node *node; struct amdgpu_bo *bo; - if (!blockable) { + if (!range->blockable) { amdgpu_mn_read_unlock(amn); return -EAGAIN; } node = container_of(it, struct amdgpu_mn_node, it); - it = interval_tree_iter_next(it, start, end); + it = interval_tree_iter_next(it, range->start, end); list_for_each_entry(bo, &node->bos, mn_list) { struct kgd_mem *mem = bo->kfd_bo; if (amdgpu_ttm_tt_affect_userptr(bo->tbo.ttm, - start, end)) - amdgpu_amdkfd_evict_userptr(mem, mm); + range->start, + end)) + amdgpu_amdkfd_evict_userptr(mem, range->mm); } } @@ -344,9 +339,7 @@ static int amdgpu_mn_invalidate_range_start_hsa(struct mmu_notifier *mn, * Release the lock again to allow new command submissions. */ static void amdgpu_mn_invalidate_range_end(struct mmu_notifier *mn, - struct mm_struct *mm, - unsigned long start, - unsigned long end) + const struct mmu_notifier_range *range) { struct amdgpu_mn *amn = container_of(mn, struct amdgpu_mn, mn); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h index d1b4d9b6aae0..889e443eeee7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h @@ -38,7 +38,7 @@ #include <drm/drm_crtc_helper.h> #include <drm/drm_fb_helper.h> #include <drm/drm_plane_helper.h> -#include <drm/drm_fb_helper.h> +#include <drm/drm_probe_helper.h> #include <linux/i2c.h> #include <linux/i2c-algo-bit.h> #include <linux/hrtimer.h> @@ -57,7 +57,6 @@ struct amdgpu_hpd; #define to_amdgpu_connector(x) container_of(x, struct amdgpu_connector, base) #define to_amdgpu_encoder(x) container_of(x, struct amdgpu_encoder, base) #define to_amdgpu_framebuffer(x) container_of(x, struct amdgpu_framebuffer, base) -#define to_amdgpu_plane(x) container_of(x, struct amdgpu_plane, base) #define to_dm_plane_state(x) container_of(x, struct dm_plane_state, base); @@ -295,13 +294,6 @@ struct amdgpu_display_funcs { uint16_t connector_object_id, struct amdgpu_hpd *hpd, struct amdgpu_router *router); - /* it is used to enter or exit into free sync mode */ - int (*notify_freesync)(struct drm_device *dev, void *data, - struct drm_file *filp); - /* it is used to allow enablement of freesync mode */ - int (*set_freesync_property)(struct drm_connector *connector, - struct drm_property *property, - uint64_t val); }; @@ -325,7 +317,7 @@ struct amdgpu_mode_info { struct card_info *atom_card_info; bool mode_config_initialized; struct amdgpu_crtc *crtcs[AMDGPU_MAX_CRTCS]; - struct amdgpu_plane *planes[AMDGPU_MAX_PLANES]; + struct drm_plane *planes[AMDGPU_MAX_PLANES]; struct amdgpu_afmt *afmt[AMDGPU_MAX_AFMT_BLOCKS]; /* DVI-I properties */ struct drm_property *coherent_mode_property; @@ -341,6 +333,8 @@ struct amdgpu_mode_info { struct drm_property *dither_property; /* maximum number of bits per channel for monitor color */ struct drm_property *max_bpc_property; + /* Adaptive Backlight Modulation (power feature) */ + struct drm_property *abm_level_property; /* hardcoded DFP edid from BIOS */ struct edid *bios_hardcoded_edid; int bios_hardcoded_edid_size; @@ -412,6 +406,7 @@ struct amdgpu_crtc { struct amdgpu_flip_work *pflip_works; enum amdgpu_flip_status pflip_status; int deferred_flip_completion; + u64 last_flip_vblank; /* pll sharing */ struct amdgpu_atom_ss ss; bool ss_enabled; @@ -436,11 +431,6 @@ struct amdgpu_crtc { struct drm_pending_vblank_event *event; }; -struct amdgpu_plane { - struct drm_plane base; - enum drm_plane_type plane_type; -}; - struct amdgpu_encoder_atom_dig { bool linkb; /* atom dig */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 904014dc5915..ec9e45004bff 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -81,7 +81,7 @@ static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo) amdgpu_bo_subtract_pin_size(bo); if (bo->kfd_bo) - amdgpu_amdkfd_unreserve_system_memory_limit(bo); + amdgpu_amdkfd_unreserve_memory_limit(bo); amdgpu_bo_kunmap(bo); @@ -426,12 +426,20 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, size_t acc_size; int r; - page_align = roundup(bp->byte_align, PAGE_SIZE) >> PAGE_SHIFT; - if (bp->domain & (AMDGPU_GEM_DOMAIN_GDS | AMDGPU_GEM_DOMAIN_GWS | - AMDGPU_GEM_DOMAIN_OA)) + /* Note that GDS/GWS/OA allocates 1 page per byte/resource. */ + if (bp->domain & (AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA)) { + /* GWS and OA don't need any alignment. */ + page_align = bp->byte_align; size <<= PAGE_SHIFT; - else + } else if (bp->domain & AMDGPU_GEM_DOMAIN_GDS) { + /* Both size and alignment must be a multiple of 4. */ + page_align = ALIGN(bp->byte_align, 4); + size = ALIGN(size, 4) << PAGE_SHIFT; + } else { + /* Memory should be aligned at least to a page size. */ + page_align = ALIGN(bp->byte_align, PAGE_SIZE) >> PAGE_SHIFT; size = ALIGN(size, PAGE_SIZE); + } if (!amdgpu_bo_validate_size(adev, size, bp->domain)) return -ENOMEM; @@ -608,53 +616,6 @@ int amdgpu_bo_create(struct amdgpu_device *adev, } /** - * amdgpu_bo_backup_to_shadow - Backs up an &amdgpu_bo buffer object - * @adev: amdgpu device object - * @ring: amdgpu_ring for the engine handling the buffer operations - * @bo: &amdgpu_bo buffer to be backed up - * @resv: reservation object with embedded fence - * @fence: dma_fence associated with the operation - * @direct: whether to submit the job directly - * - * Copies an &amdgpu_bo buffer object to its shadow object. - * Not used for now. - * - * Returns: - * 0 for success or a negative error code on failure. - */ -int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev, - struct amdgpu_ring *ring, - struct amdgpu_bo *bo, - struct reservation_object *resv, - struct dma_fence **fence, - bool direct) - -{ - struct amdgpu_bo *shadow = bo->shadow; - uint64_t bo_addr, shadow_addr; - int r; - - if (!shadow) - return -EINVAL; - - bo_addr = amdgpu_bo_gpu_offset(bo); - shadow_addr = amdgpu_bo_gpu_offset(bo->shadow); - - r = reservation_object_reserve_shared(bo->tbo.resv); - if (r) - goto err; - - r = amdgpu_copy_buffer(ring, bo_addr, shadow_addr, - amdgpu_bo_size(bo), resv, fence, - direct, false); - if (!r) - amdgpu_bo_fence(bo, *fence, true); - -err: - return r; -} - -/** * amdgpu_bo_validate - validate an &amdgpu_bo buffer object * @bo: pointer to the buffer object * @@ -959,7 +920,7 @@ int amdgpu_bo_unpin(struct amdgpu_bo *bo) struct ttm_operation_ctx ctx = { false, false }; int r, i; - if (!bo->pin_count) { + if (WARN_ON_ONCE(!bo->pin_count)) { dev_warn(adev->dev, "%p unpin not necessary\n", bo); return 0; } @@ -1324,6 +1285,30 @@ void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence, } /** + * amdgpu_sync_wait_resv - Wait for BO reservation fences + * + * @bo: buffer object + * @owner: fence owner + * @intr: Whether the wait is interruptible + * + * Returns: + * 0 on success, errno otherwise. + */ +int amdgpu_bo_sync_wait(struct amdgpu_bo *bo, void *owner, bool intr) +{ + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); + struct amdgpu_sync sync; + int r; + + amdgpu_sync_create(&sync); + amdgpu_sync_resv(adev, &sync, bo->tbo.resv, owner, false); + r = amdgpu_sync_wait(&sync, intr); + amdgpu_sync_free(&sync); + + return r; +} + +/** * amdgpu_bo_gpu_offset - return GPU offset of bo * @bo: amdgpu object for which we query the offset * diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h index 7d3312d0da11..220a6a7b1bc1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h @@ -266,12 +266,8 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo, int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo); void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence, bool shared); +int amdgpu_bo_sync_wait(struct amdgpu_bo *bo, void *owner, bool intr); u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo); -int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev, - struct amdgpu_ring *ring, - struct amdgpu_bo *bo, - struct reservation_object *resv, - struct dma_fence **fence, bool direct); int amdgpu_bo_validate(struct amdgpu_bo *bo); int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow, struct dma_fence **fence); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c index 59cc678de8c1..a7adb7b6bd98 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c @@ -33,6 +33,8 @@ #include <linux/hwmon.h> #include <linux/hwmon-sysfs.h> #include <linux/nospec.h> +#include "hwmgr.h" +#define WIDTH_4K 3840 static int amdgpu_debugfs_pm_init(struct amdgpu_device *adev); @@ -624,11 +626,71 @@ static ssize_t amdgpu_get_pp_od_clk_voltage(struct device *dev, } /** - * DOC: pp_dpm_sclk pp_dpm_mclk pp_dpm_pcie + * DOC: ppfeatures + * + * The amdgpu driver provides a sysfs API for adjusting what powerplay + * features to be enabled. The file ppfeatures is used for this. And + * this is only available for Vega10 and later dGPUs. + * + * Reading back the file will show you the followings: + * - Current ppfeature masks + * - List of the all supported powerplay features with their naming, + * bitmasks and enablement status('Y'/'N' means "enabled"/"disabled"). + * + * To manually enable or disable a specific feature, just set or clear + * the corresponding bit from original ppfeature masks and input the + * new ppfeature masks. + */ +static ssize_t amdgpu_set_ppfeature_status(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + uint64_t featuremask; + int ret; + + ret = kstrtou64(buf, 0, &featuremask); + if (ret) + return -EINVAL; + + pr_debug("featuremask = 0x%llx\n", featuremask); + + if (adev->powerplay.pp_funcs->set_ppfeature_status) { + ret = amdgpu_dpm_set_ppfeature_status(adev, featuremask); + if (ret) + return -EINVAL; + } + + return count; +} + +static ssize_t amdgpu_get_ppfeature_status(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + + if (adev->powerplay.pp_funcs->get_ppfeature_status) + return amdgpu_dpm_get_ppfeature_status(adev, buf); + + return snprintf(buf, PAGE_SIZE, "\n"); +} + +/** + * DOC: pp_dpm_sclk pp_dpm_mclk pp_dpm_socclk pp_dpm_fclk pp_dpm_dcefclk + * pp_dpm_pcie * * The amdgpu driver provides a sysfs API for adjusting what power levels * are enabled for a given power state. The files pp_dpm_sclk, pp_dpm_mclk, - * and pp_dpm_pcie are used for this. + * pp_dpm_socclk, pp_dpm_fclk, pp_dpm_dcefclk and pp_dpm_pcie are used for + * this. + * + * pp_dpm_socclk and pp_dpm_dcefclk interfaces are only available for + * Vega10 and later ASICs. + * pp_dpm_fclk interface is only available for Vega20 and later ASICs. * * Reading back the files will show you the available power levels within * the power state and the clock information for those levels. @@ -638,6 +700,8 @@ static ssize_t amdgpu_get_pp_od_clk_voltage(struct device *dev, * Secondly,Enter a new value for each level by inputing a string that * contains " echo xx xx xx > pp_dpm_sclk/mclk/pcie" * E.g., echo 4 5 6 to > pp_dpm_sclk will enable sclk levels 4, 5, and 6. + * + * NOTE: change to the dcefclk max dpm level is not supported now */ static ssize_t amdgpu_get_pp_dpm_sclk(struct device *dev, @@ -748,6 +812,114 @@ static ssize_t amdgpu_set_pp_dpm_mclk(struct device *dev, return count; } +static ssize_t amdgpu_get_pp_dpm_socclk(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + + if (adev->powerplay.pp_funcs->print_clock_levels) + return amdgpu_dpm_print_clock_levels(adev, PP_SOCCLK, buf); + else + return snprintf(buf, PAGE_SIZE, "\n"); +} + +static ssize_t amdgpu_set_pp_dpm_socclk(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + int ret; + uint32_t mask = 0; + + ret = amdgpu_read_mask(buf, count, &mask); + if (ret) + return ret; + + if (adev->powerplay.pp_funcs->force_clock_level) + ret = amdgpu_dpm_force_clock_level(adev, PP_SOCCLK, mask); + + if (ret) + return -EINVAL; + + return count; +} + +static ssize_t amdgpu_get_pp_dpm_fclk(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + + if (adev->powerplay.pp_funcs->print_clock_levels) + return amdgpu_dpm_print_clock_levels(adev, PP_FCLK, buf); + else + return snprintf(buf, PAGE_SIZE, "\n"); +} + +static ssize_t amdgpu_set_pp_dpm_fclk(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + int ret; + uint32_t mask = 0; + + ret = amdgpu_read_mask(buf, count, &mask); + if (ret) + return ret; + + if (adev->powerplay.pp_funcs->force_clock_level) + ret = amdgpu_dpm_force_clock_level(adev, PP_FCLK, mask); + + if (ret) + return -EINVAL; + + return count; +} + +static ssize_t amdgpu_get_pp_dpm_dcefclk(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + + if (adev->powerplay.pp_funcs->print_clock_levels) + return amdgpu_dpm_print_clock_levels(adev, PP_DCEFCLK, buf); + else + return snprintf(buf, PAGE_SIZE, "\n"); +} + +static ssize_t amdgpu_set_pp_dpm_dcefclk(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + int ret; + uint32_t mask = 0; + + ret = amdgpu_read_mask(buf, count, &mask); + if (ret) + return ret; + + if (adev->powerplay.pp_funcs->force_clock_level) + ret = amdgpu_dpm_force_clock_level(adev, PP_DCEFCLK, mask); + + if (ret) + return -EINVAL; + + return count; +} + static ssize_t amdgpu_get_pp_dpm_pcie(struct device *dev, struct device_attribute *attr, char *buf) @@ -988,6 +1160,31 @@ static ssize_t amdgpu_get_busy_percent(struct device *dev, return snprintf(buf, PAGE_SIZE, "%d\n", value); } +/** + * DOC: pcie_bw + * + * The amdgpu driver provides a sysfs API for estimating how much data + * has been received and sent by the GPU in the last second through PCIe. + * The file pcie_bw is used for this. + * The Perf counters count the number of received and sent messages and return + * those values, as well as the maximum payload size of a PCIe packet (mps). + * Note that it is not possible to easily and quickly obtain the size of each + * packet transmitted, so we output the max payload size (mps) to allow for + * quick estimation of the PCIe bandwidth usage + */ +static ssize_t amdgpu_get_pcie_bw(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + uint64_t count0, count1; + + amdgpu_asic_get_pcie_usage(adev, &count0, &count1); + return snprintf(buf, PAGE_SIZE, "%llu %llu %i\n", + count0, count1, pcie_get_mps(adev->pdev)); +} + static DEVICE_ATTR(power_dpm_state, S_IRUGO | S_IWUSR, amdgpu_get_dpm_state, amdgpu_set_dpm_state); static DEVICE_ATTR(power_dpm_force_performance_level, S_IRUGO | S_IWUSR, amdgpu_get_dpm_forced_performance_level, @@ -1006,6 +1203,15 @@ static DEVICE_ATTR(pp_dpm_sclk, S_IRUGO | S_IWUSR, static DEVICE_ATTR(pp_dpm_mclk, S_IRUGO | S_IWUSR, amdgpu_get_pp_dpm_mclk, amdgpu_set_pp_dpm_mclk); +static DEVICE_ATTR(pp_dpm_socclk, S_IRUGO | S_IWUSR, + amdgpu_get_pp_dpm_socclk, + amdgpu_set_pp_dpm_socclk); +static DEVICE_ATTR(pp_dpm_fclk, S_IRUGO | S_IWUSR, + amdgpu_get_pp_dpm_fclk, + amdgpu_set_pp_dpm_fclk); +static DEVICE_ATTR(pp_dpm_dcefclk, S_IRUGO | S_IWUSR, + amdgpu_get_pp_dpm_dcefclk, + amdgpu_set_pp_dpm_dcefclk); static DEVICE_ATTR(pp_dpm_pcie, S_IRUGO | S_IWUSR, amdgpu_get_pp_dpm_pcie, amdgpu_set_pp_dpm_pcie); @@ -1023,6 +1229,10 @@ static DEVICE_ATTR(pp_od_clk_voltage, S_IRUGO | S_IWUSR, amdgpu_set_pp_od_clk_voltage); static DEVICE_ATTR(gpu_busy_percent, S_IRUGO, amdgpu_get_busy_percent, NULL); +static DEVICE_ATTR(pcie_bw, S_IRUGO, amdgpu_get_pcie_bw, NULL); +static DEVICE_ATTR(ppfeatures, S_IRUGO | S_IWUSR, + amdgpu_get_ppfeature_status, + amdgpu_set_ppfeature_status); static ssize_t amdgpu_hwmon_show_temp(struct device *dev, struct device_attribute *attr, @@ -1514,6 +1724,75 @@ static ssize_t amdgpu_hwmon_set_power_cap(struct device *dev, return count; } +static ssize_t amdgpu_hwmon_show_sclk(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct amdgpu_device *adev = dev_get_drvdata(dev); + struct drm_device *ddev = adev->ddev; + uint32_t sclk; + int r, size = sizeof(sclk); + + /* Can't get voltage when the card is off */ + if ((adev->flags & AMD_IS_PX) && + (ddev->switch_power_state != DRM_SWITCH_POWER_ON)) + return -EINVAL; + + /* sanity check PP is enabled */ + if (!(adev->powerplay.pp_funcs && + adev->powerplay.pp_funcs->read_sensor)) + return -EINVAL; + + /* get the sclk */ + r = amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GFX_SCLK, + (void *)&sclk, &size); + if (r) + return r; + + return snprintf(buf, PAGE_SIZE, "%d\n", sclk * 10 * 1000); +} + +static ssize_t amdgpu_hwmon_show_sclk_label(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + return snprintf(buf, PAGE_SIZE, "sclk\n"); +} + +static ssize_t amdgpu_hwmon_show_mclk(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct amdgpu_device *adev = dev_get_drvdata(dev); + struct drm_device *ddev = adev->ddev; + uint32_t mclk; + int r, size = sizeof(mclk); + + /* Can't get voltage when the card is off */ + if ((adev->flags & AMD_IS_PX) && + (ddev->switch_power_state != DRM_SWITCH_POWER_ON)) + return -EINVAL; + + /* sanity check PP is enabled */ + if (!(adev->powerplay.pp_funcs && + adev->powerplay.pp_funcs->read_sensor)) + return -EINVAL; + + /* get the sclk */ + r = amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GFX_MCLK, + (void *)&mclk, &size); + if (r) + return r; + + return snprintf(buf, PAGE_SIZE, "%d\n", mclk * 10 * 1000); +} + +static ssize_t amdgpu_hwmon_show_mclk_label(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + return snprintf(buf, PAGE_SIZE, "mclk\n"); +} /** * DOC: hwmon @@ -1530,6 +1809,10 @@ static ssize_t amdgpu_hwmon_set_power_cap(struct device *dev, * * - GPU fan * + * - GPU gfx/compute engine clock + * + * - GPU memory clock (dGPU only) + * * hwmon interfaces for GPU temperature: * * - temp1_input: the on die GPU temperature in millidegrees Celsius @@ -1574,6 +1857,12 @@ static ssize_t amdgpu_hwmon_set_power_cap(struct device *dev, * * - fan[1-*]_enable: Enable or disable the sensors.1: Enable 0: Disable * + * hwmon interfaces for GPU clocks: + * + * - freq1_input: the gfx/compute clock in hertz + * + * - freq2_input: the memory clock in hertz + * * You can use hwmon tools like sensors to view this information on your system. * */ @@ -1598,6 +1887,10 @@ static SENSOR_DEVICE_ATTR(power1_average, S_IRUGO, amdgpu_hwmon_show_power_avg, static SENSOR_DEVICE_ATTR(power1_cap_max, S_IRUGO, amdgpu_hwmon_show_power_cap_max, NULL, 0); static SENSOR_DEVICE_ATTR(power1_cap_min, S_IRUGO, amdgpu_hwmon_show_power_cap_min, NULL, 0); static SENSOR_DEVICE_ATTR(power1_cap, S_IRUGO | S_IWUSR, amdgpu_hwmon_show_power_cap, amdgpu_hwmon_set_power_cap, 0); +static SENSOR_DEVICE_ATTR(freq1_input, S_IRUGO, amdgpu_hwmon_show_sclk, NULL, 0); +static SENSOR_DEVICE_ATTR(freq1_label, S_IRUGO, amdgpu_hwmon_show_sclk_label, NULL, 0); +static SENSOR_DEVICE_ATTR(freq2_input, S_IRUGO, amdgpu_hwmon_show_mclk, NULL, 0); +static SENSOR_DEVICE_ATTR(freq2_label, S_IRUGO, amdgpu_hwmon_show_mclk_label, NULL, 0); static struct attribute *hwmon_attributes[] = { &sensor_dev_attr_temp1_input.dev_attr.attr, @@ -1620,6 +1913,10 @@ static struct attribute *hwmon_attributes[] = { &sensor_dev_attr_power1_cap_max.dev_attr.attr, &sensor_dev_attr_power1_cap_min.dev_attr.attr, &sensor_dev_attr_power1_cap.dev_attr.attr, + &sensor_dev_attr_freq1_input.dev_attr.attr, + &sensor_dev_attr_freq1_label.dev_attr.attr, + &sensor_dev_attr_freq2_input.dev_attr.attr, + &sensor_dev_attr_freq2_label.dev_attr.attr, NULL }; @@ -1642,6 +1939,19 @@ static umode_t hwmon_attributes_visible(struct kobject *kobj, attr == &sensor_dev_attr_fan1_enable.dev_attr.attr)) return 0; + /* Skip fan attributes on APU */ + if ((adev->flags & AMD_IS_APU) && + (attr == &sensor_dev_attr_pwm1.dev_attr.attr || + attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr || + attr == &sensor_dev_attr_pwm1_max.dev_attr.attr || + attr == &sensor_dev_attr_pwm1_min.dev_attr.attr || + attr == &sensor_dev_attr_fan1_input.dev_attr.attr || + attr == &sensor_dev_attr_fan1_min.dev_attr.attr || + attr == &sensor_dev_attr_fan1_max.dev_attr.attr || + attr == &sensor_dev_attr_fan1_target.dev_attr.attr || + attr == &sensor_dev_attr_fan1_enable.dev_attr.attr)) + return 0; + /* Skip limit attributes if DPM is not enabled */ if (!adev->pm.dpm_enabled && (attr == &sensor_dev_attr_temp1_crit.dev_attr.attr || @@ -1671,7 +1981,8 @@ static umode_t hwmon_attributes_visible(struct kobject *kobj, effective_mode &= ~S_IWUSR; if ((adev->flags & AMD_IS_APU) && - (attr == &sensor_dev_attr_power1_cap_max.dev_attr.attr || + (attr == &sensor_dev_attr_power1_average.dev_attr.attr || + attr == &sensor_dev_attr_power1_cap_max.dev_attr.attr || attr == &sensor_dev_attr_power1_cap_min.dev_attr.attr|| attr == &sensor_dev_attr_power1_cap.dev_attr.attr)) return 0; @@ -1697,6 +2008,12 @@ static umode_t hwmon_attributes_visible(struct kobject *kobj, attr == &sensor_dev_attr_in1_label.dev_attr.attr)) return 0; + /* no mclk on APUs */ + if ((adev->flags & AMD_IS_APU) && + (attr == &sensor_dev_attr_freq2_input.dev_attr.attr || + attr == &sensor_dev_attr_freq2_label.dev_attr.attr)) + return 0; + return effective_mode; } @@ -1956,6 +2273,17 @@ void amdgpu_dpm_enable_uvd(struct amdgpu_device *adev, bool enable) amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_UVD, !enable); mutex_unlock(&adev->pm.mutex); } + /* enable/disable Low Memory PState for UVD (4k videos) */ + if (adev->asic_type == CHIP_STONEY && + adev->uvd.decode_image_width >= WIDTH_4K) { + struct pp_hwmgr *hwmgr = adev->powerplay.pp_handle; + + if (hwmgr && hwmgr->hwmgr_func && + hwmgr->hwmgr_func->update_nbdpm_pstate) + hwmgr->hwmgr_func->update_nbdpm_pstate(hwmgr, + !enable, + true); + } } void amdgpu_dpm_enable_vce(struct amdgpu_device *adev, bool enable) @@ -1982,6 +2310,7 @@ void amdgpu_pm_print_power_states(struct amdgpu_device *adev) int amdgpu_pm_sysfs_init(struct amdgpu_device *adev) { + struct pp_hwmgr *hwmgr = adev->powerplay.pp_handle; int ret; if (adev->pm.sysfs_initialized) @@ -2043,6 +2372,25 @@ int amdgpu_pm_sysfs_init(struct amdgpu_device *adev) DRM_ERROR("failed to create device file pp_dpm_mclk\n"); return ret; } + if (adev->asic_type >= CHIP_VEGA10) { + ret = device_create_file(adev->dev, &dev_attr_pp_dpm_socclk); + if (ret) { + DRM_ERROR("failed to create device file pp_dpm_socclk\n"); + return ret; + } + ret = device_create_file(adev->dev, &dev_attr_pp_dpm_dcefclk); + if (ret) { + DRM_ERROR("failed to create device file pp_dpm_dcefclk\n"); + return ret; + } + } + if (adev->asic_type >= CHIP_VEGA20) { + ret = device_create_file(adev->dev, &dev_attr_pp_dpm_fclk); + if (ret) { + DRM_ERROR("failed to create device file pp_dpm_fclk\n"); + return ret; + } + } ret = device_create_file(adev->dev, &dev_attr_pp_dpm_pcie); if (ret) { DRM_ERROR("failed to create device file pp_dpm_pcie\n"); @@ -2065,12 +2413,14 @@ int amdgpu_pm_sysfs_init(struct amdgpu_device *adev) "pp_power_profile_mode\n"); return ret; } - ret = device_create_file(adev->dev, - &dev_attr_pp_od_clk_voltage); - if (ret) { - DRM_ERROR("failed to create device file " - "pp_od_clk_voltage\n"); - return ret; + if (hwmgr->od_enabled) { + ret = device_create_file(adev->dev, + &dev_attr_pp_od_clk_voltage); + if (ret) { + DRM_ERROR("failed to create device file " + "pp_od_clk_voltage\n"); + return ret; + } } ret = device_create_file(adev->dev, &dev_attr_gpu_busy_percent); @@ -2079,12 +2429,31 @@ int amdgpu_pm_sysfs_init(struct amdgpu_device *adev) "gpu_busy_level\n"); return ret; } + /* PCIe Perf counters won't work on APU nodes */ + if (!(adev->flags & AMD_IS_APU)) { + ret = device_create_file(adev->dev, &dev_attr_pcie_bw); + if (ret) { + DRM_ERROR("failed to create device file pcie_bw\n"); + return ret; + } + } ret = amdgpu_debugfs_pm_init(adev); if (ret) { DRM_ERROR("Failed to register debugfs file for dpm!\n"); return ret; } + if ((adev->asic_type >= CHIP_VEGA10) && + !(adev->flags & AMD_IS_APU)) { + ret = device_create_file(adev->dev, + &dev_attr_ppfeatures); + if (ret) { + DRM_ERROR("failed to create device file " + "ppfeatures\n"); + return ret; + } + } + adev->pm.sysfs_initialized = true; return 0; @@ -2092,6 +2461,8 @@ int amdgpu_pm_sysfs_init(struct amdgpu_device *adev) void amdgpu_pm_sysfs_fini(struct amdgpu_device *adev) { + struct pp_hwmgr *hwmgr = adev->powerplay.pp_handle; + if (adev->pm.dpm_enabled == 0) return; @@ -2107,14 +2478,26 @@ void amdgpu_pm_sysfs_fini(struct amdgpu_device *adev) device_remove_file(adev->dev, &dev_attr_pp_dpm_sclk); device_remove_file(adev->dev, &dev_attr_pp_dpm_mclk); + if (adev->asic_type >= CHIP_VEGA10) { + device_remove_file(adev->dev, &dev_attr_pp_dpm_socclk); + device_remove_file(adev->dev, &dev_attr_pp_dpm_dcefclk); + } device_remove_file(adev->dev, &dev_attr_pp_dpm_pcie); + if (adev->asic_type >= CHIP_VEGA20) + device_remove_file(adev->dev, &dev_attr_pp_dpm_fclk); device_remove_file(adev->dev, &dev_attr_pp_sclk_od); device_remove_file(adev->dev, &dev_attr_pp_mclk_od); device_remove_file(adev->dev, &dev_attr_pp_power_profile_mode); - device_remove_file(adev->dev, - &dev_attr_pp_od_clk_voltage); + if (hwmgr->od_enabled) + device_remove_file(adev->dev, + &dev_attr_pp_od_clk_voltage); device_remove_file(adev->dev, &dev_attr_gpu_busy_percent); + if (!(adev->flags & AMD_IS_APU)) + device_remove_file(adev->dev, &dev_attr_pcie_bw); + if ((adev->asic_type >= CHIP_VEGA10) && + !(adev->flags & AMD_IS_APU)) + device_remove_file(adev->dev, &dev_attr_ppfeatures); } void amdgpu_pm_compute_clocks(struct amdgpu_device *adev) @@ -2129,7 +2512,7 @@ void amdgpu_pm_compute_clocks(struct amdgpu_device *adev) for (i = 0; i < AMDGPU_MAX_RINGS; i++) { struct amdgpu_ring *ring = adev->rings[i]; - if (ring && ring->ready) + if (ring && ring->sched.ready) amdgpu_fence_wait_empty(ring); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c index e45e929aaab5..a38e0fb4a6fe 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c @@ -38,8 +38,7 @@ #include "amdgpu_gem.h" #include <drm/amdgpu_drm.h> #include <linux/dma-buf.h> - -static const struct dma_buf_ops amdgpu_dmabuf_ops; +#include <linux/dma-fence-array.h> /** * amdgpu_gem_prime_get_sg_table - &drm_driver.gem_prime_get_sg_table @@ -189,6 +188,48 @@ error: return ERR_PTR(ret); } +static int +__reservation_object_make_exclusive(struct reservation_object *obj) +{ + struct dma_fence **fences; + unsigned int count; + int r; + + if (!reservation_object_get_list(obj)) /* no shared fences to convert */ + return 0; + + r = reservation_object_get_fences_rcu(obj, NULL, &count, &fences); + if (r) + return r; + + if (count == 0) { + /* Now that was unexpected. */ + } else if (count == 1) { + reservation_object_add_excl_fence(obj, fences[0]); + dma_fence_put(fences[0]); + kfree(fences); + } else { + struct dma_fence_array *array; + + array = dma_fence_array_create(count, fences, + dma_fence_context_alloc(1), 0, + false); + if (!array) + goto err_fences_put; + + reservation_object_add_excl_fence(obj, &array->base); + dma_fence_put(&array->base); + } + + return 0; + +err_fences_put: + while (count--) + dma_fence_put(fences[count]); + kfree(fences); + return -ENOMEM; +} + /** * amdgpu_gem_map_attach - &dma_buf_ops.attach implementation * @dma_buf: Shared DMA buffer @@ -220,16 +261,16 @@ static int amdgpu_gem_map_attach(struct dma_buf *dma_buf, if (attach->dev->driver != adev->dev->driver) { /* - * Wait for all shared fences to complete before we switch to future - * use of exclusive fence on this prime shared bo. + * We only create shared fences for internal use, but importers + * of the dmabuf rely on exclusive fences for implicitly + * tracking write hazards. As any of the current fences may + * correspond to a write, we need to convert all existing + * fences on the reservation object into a single exclusive + * fence. */ - r = reservation_object_wait_timeout_rcu(bo->tbo.resv, - true, false, - MAX_SCHEDULE_TIMEOUT); - if (unlikely(r < 0)) { - DRM_DEBUG_PRIME("Fence wait failed: %li\n", r); + r = __reservation_object_make_exclusive(bo->tbo.resv); + if (r) goto error_unreserve; - } } /* pin buffer into GTT */ @@ -332,15 +373,13 @@ static int amdgpu_gem_begin_cpu_access(struct dma_buf *dma_buf, return ret; } -static const struct dma_buf_ops amdgpu_dmabuf_ops = { +const struct dma_buf_ops amdgpu_dmabuf_ops = { .attach = amdgpu_gem_map_attach, .detach = amdgpu_gem_map_detach, .map_dma_buf = drm_gem_map_dma_buf, .unmap_dma_buf = drm_gem_unmap_dma_buf, .release = drm_gem_dmabuf_release, .begin_cpu_access = amdgpu_gem_begin_cpu_access, - .map = drm_gem_dmabuf_kmap, - .unmap = drm_gem_dmabuf_kunmap, .mmap = drm_gem_dmabuf_mmap, .vmap = drm_gem_dmabuf_vmap, .vunmap = drm_gem_dmabuf_vunmap, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c index 25d2f3e757f1..3091488cd8cc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c @@ -67,9 +67,6 @@ static int psp_sw_init(void *handle) psp->adev = adev; - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) - return 0; - ret = psp_init_microcode(psp); if (ret) { DRM_ERROR("Failed to load psp firmware!\n"); @@ -83,13 +80,14 @@ static int psp_sw_fini(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) - return 0; - release_firmware(adev->psp.sos_fw); adev->psp.sos_fw = NULL; release_firmware(adev->psp.asd_fw); adev->psp.asd_fw = NULL; + if (adev->psp.ta_fw) { + release_firmware(adev->psp.ta_fw); + adev->psp.ta_fw = NULL; + } return 0; } @@ -118,29 +116,44 @@ int psp_wait_for(struct psp_context *psp, uint32_t reg_index, static int psp_cmd_submit_buf(struct psp_context *psp, struct amdgpu_firmware_info *ucode, - struct psp_gfx_cmd_resp *cmd, uint64_t fence_mc_addr, - int index) + struct psp_gfx_cmd_resp *cmd, uint64_t fence_mc_addr) { int ret; + int index; memset(psp->cmd_buf_mem, 0, PSP_CMD_BUFFER_SIZE); memcpy(psp->cmd_buf_mem, cmd, sizeof(struct psp_gfx_cmd_resp)); + index = atomic_inc_return(&psp->fence_value); ret = psp_cmd_submit(psp, ucode, psp->cmd_buf_mc_addr, fence_mc_addr, index); + if (ret) { + atomic_dec(&psp->fence_value); + return ret; + } - while (*((unsigned int *)psp->fence_buf) != index) { + while (*((unsigned int *)psp->fence_buf) != index) msleep(1); - } - /* the status field must be 0 after FW is loaded */ - if (ucode && psp->cmd_buf_mem->resp.status) { - DRM_ERROR("failed loading with status (%d) and ucode id (%d)\n", - psp->cmd_buf_mem->resp.status, ucode->ucode_id); - return -EINVAL; + /* In some cases, psp response status is not 0 even there is no + * problem while the command is submitted. Some version of PSP FW + * doesn't write 0 to that field. + * So here we would like to only print a warning instead of an error + * during psp initialization to avoid breaking hw_init and it doesn't + * return -EINVAL. + */ + if (psp->cmd_buf_mem->resp.status) { + if (ucode) + DRM_WARN("failed to load ucode id (%d) ", + ucode->ucode_id); + DRM_WARN("psp command failed and response status is (%d)\n", + psp->cmd_buf_mem->resp.status); } + /* get xGMI session id from response buffer */ + cmd->resp.session_id = psp->cmd_buf_mem->resp.session_id; + if (ucode) { ucode->tmr_mc_addr_lo = psp->cmd_buf_mem->resp.fw_addr_lo; ucode->tmr_mc_addr_hi = psp->cmd_buf_mem->resp.fw_addr_hi; @@ -149,10 +162,14 @@ psp_cmd_submit_buf(struct psp_context *psp, return ret; } -static void psp_prep_tmr_cmd_buf(struct psp_gfx_cmd_resp *cmd, +static void psp_prep_tmr_cmd_buf(struct psp_context *psp, + struct psp_gfx_cmd_resp *cmd, uint64_t tmr_mc, uint32_t size) { - cmd->cmd_id = GFX_CMD_ID_SETUP_TMR; + if (psp_support_vmr_ring(psp)) + cmd->cmd_id = GFX_CMD_ID_SETUP_VMR; + else + cmd->cmd_id = GFX_CMD_ID_SETUP_TMR; cmd->cmd.cmd_setup_tmr.buf_phy_addr_lo = lower_32_bits(tmr_mc); cmd->cmd.cmd_setup_tmr.buf_phy_addr_hi = upper_32_bits(tmr_mc); cmd->cmd.cmd_setup_tmr.buf_size = size; @@ -186,12 +203,12 @@ static int psp_tmr_load(struct psp_context *psp) if (!cmd) return -ENOMEM; - psp_prep_tmr_cmd_buf(cmd, psp->tmr_mc_addr, PSP_TMR_SIZE); + psp_prep_tmr_cmd_buf(psp, cmd, psp->tmr_mc_addr, PSP_TMR_SIZE); DRM_INFO("reserve 0x%x from 0x%llx for PSP TMR SIZE\n", PSP_TMR_SIZE, psp->tmr_mc_addr); ret = psp_cmd_submit_buf(psp, NULL, cmd, - psp->fence_buf_mc_addr, 1); + psp->fence_buf_mc_addr); if (ret) goto failed; @@ -258,13 +275,197 @@ static int psp_asd_load(struct psp_context *psp) psp->asd_ucode_size, PSP_ASD_SHARED_MEM_SIZE); ret = psp_cmd_submit_buf(psp, NULL, cmd, - psp->fence_buf_mc_addr, 2); + psp->fence_buf_mc_addr); kfree(cmd); return ret; } +static void psp_prep_xgmi_ta_load_cmd_buf(struct psp_gfx_cmd_resp *cmd, + uint64_t xgmi_ta_mc, uint64_t xgmi_mc_shared, + uint32_t xgmi_ta_size, uint32_t shared_size) +{ + cmd->cmd_id = GFX_CMD_ID_LOAD_TA; + cmd->cmd.cmd_load_ta.app_phy_addr_lo = lower_32_bits(xgmi_ta_mc); + cmd->cmd.cmd_load_ta.app_phy_addr_hi = upper_32_bits(xgmi_ta_mc); + cmd->cmd.cmd_load_ta.app_len = xgmi_ta_size; + + cmd->cmd.cmd_load_ta.cmd_buf_phy_addr_lo = lower_32_bits(xgmi_mc_shared); + cmd->cmd.cmd_load_ta.cmd_buf_phy_addr_hi = upper_32_bits(xgmi_mc_shared); + cmd->cmd.cmd_load_ta.cmd_buf_len = shared_size; +} + +static int psp_xgmi_init_shared_buf(struct psp_context *psp) +{ + int ret; + + /* + * Allocate 16k memory aligned to 4k from Frame Buffer (local + * physical) for xgmi ta <-> Driver + */ + ret = amdgpu_bo_create_kernel(psp->adev, PSP_XGMI_SHARED_MEM_SIZE, + PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, + &psp->xgmi_context.xgmi_shared_bo, + &psp->xgmi_context.xgmi_shared_mc_addr, + &psp->xgmi_context.xgmi_shared_buf); + + return ret; +} + +static int psp_xgmi_load(struct psp_context *psp) +{ + int ret; + struct psp_gfx_cmd_resp *cmd; + + /* + * TODO: bypass the loading in sriov for now + */ + if (amdgpu_sriov_vf(psp->adev)) + return 0; + + cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL); + if (!cmd) + return -ENOMEM; + + memset(psp->fw_pri_buf, 0, PSP_1_MEG); + memcpy(psp->fw_pri_buf, psp->ta_xgmi_start_addr, psp->ta_xgmi_ucode_size); + + psp_prep_xgmi_ta_load_cmd_buf(cmd, psp->fw_pri_mc_addr, + psp->xgmi_context.xgmi_shared_mc_addr, + psp->ta_xgmi_ucode_size, PSP_XGMI_SHARED_MEM_SIZE); + + ret = psp_cmd_submit_buf(psp, NULL, cmd, + psp->fence_buf_mc_addr); + + if (!ret) { + psp->xgmi_context.initialized = 1; + psp->xgmi_context.session_id = cmd->resp.session_id; + } + + kfree(cmd); + + return ret; +} + +static void psp_prep_xgmi_ta_unload_cmd_buf(struct psp_gfx_cmd_resp *cmd, + uint32_t xgmi_session_id) +{ + cmd->cmd_id = GFX_CMD_ID_UNLOAD_TA; + cmd->cmd.cmd_unload_ta.session_id = xgmi_session_id; +} + +static int psp_xgmi_unload(struct psp_context *psp) +{ + int ret; + struct psp_gfx_cmd_resp *cmd; + + /* + * TODO: bypass the unloading in sriov for now + */ + if (amdgpu_sriov_vf(psp->adev)) + return 0; + + cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL); + if (!cmd) + return -ENOMEM; + + psp_prep_xgmi_ta_unload_cmd_buf(cmd, psp->xgmi_context.session_id); + + ret = psp_cmd_submit_buf(psp, NULL, cmd, + psp->fence_buf_mc_addr); + + kfree(cmd); + + return ret; +} + +static void psp_prep_xgmi_ta_invoke_cmd_buf(struct psp_gfx_cmd_resp *cmd, + uint32_t ta_cmd_id, + uint32_t xgmi_session_id) +{ + cmd->cmd_id = GFX_CMD_ID_INVOKE_CMD; + cmd->cmd.cmd_invoke_cmd.session_id = xgmi_session_id; + cmd->cmd.cmd_invoke_cmd.ta_cmd_id = ta_cmd_id; + /* Note: cmd_invoke_cmd.buf is not used for now */ +} + +int psp_xgmi_invoke(struct psp_context *psp, uint32_t ta_cmd_id) +{ + int ret; + struct psp_gfx_cmd_resp *cmd; + + /* + * TODO: bypass the loading in sriov for now + */ + if (amdgpu_sriov_vf(psp->adev)) + return 0; + + cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL); + if (!cmd) + return -ENOMEM; + + psp_prep_xgmi_ta_invoke_cmd_buf(cmd, ta_cmd_id, + psp->xgmi_context.session_id); + + ret = psp_cmd_submit_buf(psp, NULL, cmd, + psp->fence_buf_mc_addr); + + kfree(cmd); + + return ret; +} + +static int psp_xgmi_terminate(struct psp_context *psp) +{ + int ret; + + if (!psp->xgmi_context.initialized) + return 0; + + ret = psp_xgmi_unload(psp); + if (ret) + return ret; + + psp->xgmi_context.initialized = 0; + + /* free xgmi shared memory */ + amdgpu_bo_free_kernel(&psp->xgmi_context.xgmi_shared_bo, + &psp->xgmi_context.xgmi_shared_mc_addr, + &psp->xgmi_context.xgmi_shared_buf); + + return 0; +} + +static int psp_xgmi_initialize(struct psp_context *psp) +{ + struct ta_xgmi_shared_memory *xgmi_cmd; + int ret; + + if (!psp->adev->psp.ta_fw) + return -ENOENT; + + if (!psp->xgmi_context.initialized) { + ret = psp_xgmi_init_shared_buf(psp); + if (ret) + return ret; + } + + /* Load XGMI TA */ + ret = psp_xgmi_load(psp); + if (ret) + return ret; + + /* Initialize XGMI session */ + xgmi_cmd = (struct ta_xgmi_shared_memory *)(psp->xgmi_context.xgmi_shared_buf); + memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory)); + xgmi_cmd->cmd_id = TA_COMMAND_XGMI__INITIALIZE; + + ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id); + + return ret; +} + static int psp_hw_start(struct psp_context *psp) { struct amdgpu_device *adev = psp->adev; @@ -292,9 +493,110 @@ static int psp_hw_start(struct psp_context *psp) if (ret) return ret; + if (adev->gmc.xgmi.num_physical_nodes > 1) { + ret = psp_xgmi_initialize(psp); + /* Warning the XGMI seesion initialize failure + * Instead of stop driver initialization + */ + if (ret) + dev_err(psp->adev->dev, + "XGMI: Failed to initialize XGMI session\n"); + } + return 0; +} + +static int psp_get_fw_type(struct amdgpu_firmware_info *ucode, + enum psp_gfx_fw_type *type) +{ + switch (ucode->ucode_id) { + case AMDGPU_UCODE_ID_SDMA0: + *type = GFX_FW_TYPE_SDMA0; + break; + case AMDGPU_UCODE_ID_SDMA1: + *type = GFX_FW_TYPE_SDMA1; + break; + case AMDGPU_UCODE_ID_CP_CE: + *type = GFX_FW_TYPE_CP_CE; + break; + case AMDGPU_UCODE_ID_CP_PFP: + *type = GFX_FW_TYPE_CP_PFP; + break; + case AMDGPU_UCODE_ID_CP_ME: + *type = GFX_FW_TYPE_CP_ME; + break; + case AMDGPU_UCODE_ID_CP_MEC1: + *type = GFX_FW_TYPE_CP_MEC; + break; + case AMDGPU_UCODE_ID_CP_MEC1_JT: + *type = GFX_FW_TYPE_CP_MEC_ME1; + break; + case AMDGPU_UCODE_ID_CP_MEC2: + *type = GFX_FW_TYPE_CP_MEC; + break; + case AMDGPU_UCODE_ID_CP_MEC2_JT: + *type = GFX_FW_TYPE_CP_MEC_ME2; + break; + case AMDGPU_UCODE_ID_RLC_G: + *type = GFX_FW_TYPE_RLC_G; + break; + case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL: + *type = GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_CNTL; + break; + case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM: + *type = GFX_FW_TYPE_RLC_RESTORE_LIST_GPM_MEM; + break; + case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM: + *type = GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_MEM; + break; + case AMDGPU_UCODE_ID_SMC: + *type = GFX_FW_TYPE_SMU; + break; + case AMDGPU_UCODE_ID_UVD: + *type = GFX_FW_TYPE_UVD; + break; + case AMDGPU_UCODE_ID_UVD1: + *type = GFX_FW_TYPE_UVD1; + break; + case AMDGPU_UCODE_ID_VCE: + *type = GFX_FW_TYPE_VCE; + break; + case AMDGPU_UCODE_ID_VCN: + *type = GFX_FW_TYPE_VCN; + break; + case AMDGPU_UCODE_ID_DMCU_ERAM: + *type = GFX_FW_TYPE_DMCU_ERAM; + break; + case AMDGPU_UCODE_ID_DMCU_INTV: + *type = GFX_FW_TYPE_DMCU_ISR; + break; + case AMDGPU_UCODE_ID_MAXIMUM: + default: + return -EINVAL; + } + return 0; } +static int psp_prep_load_ip_fw_cmd_buf(struct amdgpu_firmware_info *ucode, + struct psp_gfx_cmd_resp *cmd) +{ + int ret; + uint64_t fw_mem_mc_addr = ucode->mc_addr; + + memset(cmd, 0, sizeof(struct psp_gfx_cmd_resp)); + + cmd->cmd_id = GFX_CMD_ID_LOAD_IP_FW; + cmd->cmd.cmd_load_ip_fw.fw_phy_addr_lo = lower_32_bits(fw_mem_mc_addr); + cmd->cmd.cmd_load_ip_fw.fw_phy_addr_hi = upper_32_bits(fw_mem_mc_addr); + cmd->cmd.cmd_load_ip_fw.fw_size = ucode->ucode_size; + + ret = psp_get_fw_type(ucode, &cmd->cmd.cmd_load_ip_fw.fw_type); + if (ret) + DRM_ERROR("Unknown firmware type\n"); + + return ret; +} + static int psp_np_fw_load(struct psp_context *psp) { int i, ret; @@ -316,12 +618,12 @@ static int psp_np_fw_load(struct psp_context *psp) /*skip ucode loading in SRIOV VF */ continue; - ret = psp_prep_cmd_buf(ucode, psp->cmd); + ret = psp_prep_load_ip_fw_cmd_buf(ucode, psp->cmd); if (ret) return ret; ret = psp_cmd_submit_buf(psp, ucode, psp->cmd, - psp->fence_buf_mc_addr, i + 3); + psp->fence_buf_mc_addr); if (ret) return ret; @@ -340,8 +642,10 @@ static int psp_load_fw(struct amdgpu_device *adev) int ret; struct psp_context *psp = &adev->psp; - if (amdgpu_sriov_vf(adev) && adev->in_gpu_reset != 0) + if (amdgpu_sriov_vf(adev) && adev->in_gpu_reset) { + psp_ring_stop(psp, PSP_RING_TYPE__KM); /* should not destroy ring, only stop */ goto skip_memalloc; + } psp->cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL); if (!psp->cmd) @@ -416,10 +720,6 @@ static int psp_hw_init(void *handle) int ret; struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) - return 0; - mutex_lock(&adev->firmware.mutex); /* * This sequence is just used on hw_init only once, no need on @@ -449,8 +749,9 @@ static int psp_hw_fini(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; struct psp_context *psp = &adev->psp; - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) - return 0; + if (adev->gmc.xgmi.num_physical_nodes > 1 && + psp->xgmi_context.initialized == 1) + psp_xgmi_terminate(psp); psp_ring_destroy(psp, PSP_RING_TYPE__KM); @@ -476,8 +777,14 @@ static int psp_suspend(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; struct psp_context *psp = &adev->psp; - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) - return 0; + if (adev->gmc.xgmi.num_physical_nodes > 1 && + psp->xgmi_context.initialized == 1) { + ret = psp_xgmi_terminate(psp); + if (ret) { + DRM_ERROR("Failed to terminate xgmi ta\n"); + return ret; + } + } ret = psp_ring_stop(psp, PSP_RING_TYPE__KM); if (ret) { @@ -494,9 +801,6 @@ static int psp_resume(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; struct psp_context *psp = &adev->psp; - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) - return 0; - DRM_INFO("PSP is resuming...\n"); mutex_lock(&adev->firmware.mutex); @@ -532,11 +836,6 @@ static bool psp_check_fw_loading_status(struct amdgpu_device *adev, { struct amdgpu_firmware_info *ucode = NULL; - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { - DRM_INFO("firmware is not loaded by PSP\n"); - return true; - } - if (!adev->firmware.fw_size) return false; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h index 8b8720e9c3f0..2ef98cc755d6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h @@ -27,14 +27,17 @@ #include "amdgpu.h" #include "psp_gfx_if.h" +#include "ta_xgmi_if.h" #define PSP_FENCE_BUFFER_SIZE 0x1000 #define PSP_CMD_BUFFER_SIZE 0x1000 -#define PSP_ASD_SHARED_MEM_SIZE 0x4000 +#define PSP_ASD_SHARED_MEM_SIZE 0x4000 +#define PSP_XGMI_SHARED_MEM_SIZE 0x4000 #define PSP_1_MEG 0x100000 #define PSP_TMR_SIZE 0x400000 struct psp_context; +struct psp_xgmi_node_info; struct psp_xgmi_topology_info; enum psp_ring_type @@ -62,8 +65,6 @@ struct psp_funcs int (*init_microcode)(struct psp_context *psp); int (*bootloader_load_sysdrv)(struct psp_context *psp); int (*bootloader_load_sos)(struct psp_context *psp); - int (*prep_cmd_buf)(struct amdgpu_firmware_info *ucode, - struct psp_gfx_cmd_resp *cmd); int (*ring_init)(struct psp_context *psp, enum psp_ring_type ring_type); int (*ring_create)(struct psp_context *psp, enum psp_ring_type ring_type); @@ -80,12 +81,21 @@ struct psp_funcs enum AMDGPU_UCODE_ID ucode_type); bool (*smu_reload_quirk)(struct psp_context *psp); int (*mode1_reset)(struct psp_context *psp); - uint64_t (*xgmi_get_device_id)(struct psp_context *psp); - uint64_t (*xgmi_get_hive_id)(struct psp_context *psp); + int (*xgmi_get_node_id)(struct psp_context *psp, uint64_t *node_id); + int (*xgmi_get_hive_id)(struct psp_context *psp, uint64_t *hive_id); int (*xgmi_get_topology_info)(struct psp_context *psp, int number_devices, - struct psp_xgmi_topology_info *topology); + struct psp_xgmi_topology_info *topology); int (*xgmi_set_topology_info)(struct psp_context *psp, int number_devices, - struct psp_xgmi_topology_info *topology); + struct psp_xgmi_topology_info *topology); + bool (*support_vmr_ring)(struct psp_context *psp); +}; + +struct psp_xgmi_context { + uint8_t initialized; + uint32_t session_id; + struct amdgpu_bo *xgmi_shared_bo; + uint64_t xgmi_shared_mc_addr; + void *xgmi_shared_buf; }; struct psp_context @@ -96,7 +106,7 @@ struct psp_context const struct psp_funcs *funcs; - /* fence buffer */ + /* firmware buffer */ struct amdgpu_bo *fw_pri_bo; uint64_t fw_pri_mc_addr; void *fw_pri_buf; @@ -134,6 +144,16 @@ struct psp_context struct amdgpu_bo *cmd_buf_bo; uint64_t cmd_buf_mc_addr; struct psp_gfx_cmd_resp *cmd_buf_mem; + + /* fence value associated with cmd buffer */ + atomic_t fence_value; + + /* xgmi ta firmware and buffer */ + const struct firmware *ta_fw; + uint32_t ta_xgmi_ucode_version; + uint32_t ta_xgmi_ucode_size; + uint8_t *ta_xgmi_start_addr; + struct psp_xgmi_context xgmi_context; }; struct amdgpu_psp_funcs { @@ -141,24 +161,19 @@ struct amdgpu_psp_funcs { enum AMDGPU_UCODE_ID); }; +#define AMDGPU_XGMI_MAX_CONNECTED_NODES 64 +struct psp_xgmi_node_info { + uint64_t node_id; + uint8_t num_hops; + uint8_t is_sharing_enabled; + enum ta_xgmi_assigned_sdma_engine sdma_engine; +}; + struct psp_xgmi_topology_info { - /* Generated by PSP to identify the GPU instance within xgmi connection */ - uint64_t device_id; - /* - * If all bits set to 0 , driver indicates it wants to retrieve the xgmi - * connection vector topology, but not access enable the connections - * if some or all bits are set to 1, driver indicates it want to retrieve the - * current xgmi topology and access enable the link to GPU[i] associated - * with the bit position in the vector. - * On return,: bits indicated which xgmi links are present/active depending - * on the value passed in. The relative bit offset for the relative GPU index - * within the hive is always marked active. - */ - uint32_t connection_mask; - uint32_t reserved; /* must be 0 */ + uint32_t num_nodes; + struct psp_xgmi_node_info nodes[AMDGPU_XGMI_MAX_CONNECTED_NODES]; }; -#define psp_prep_cmd_buf(ucode, type) (psp)->funcs->prep_cmd_buf((ucode), (type)) #define psp_ring_init(psp, type) (psp)->funcs->ring_init((psp), (type)) #define psp_ring_create(psp, type) (psp)->funcs->ring_create((psp), (type)) #define psp_ring_stop(psp, type) (psp)->funcs->ring_stop((psp), (type)) @@ -175,12 +190,14 @@ struct psp_xgmi_topology_info { ((psp)->funcs->bootloader_load_sos ? (psp)->funcs->bootloader_load_sos((psp)) : 0) #define psp_smu_reload_quirk(psp) \ ((psp)->funcs->smu_reload_quirk ? (psp)->funcs->smu_reload_quirk((psp)) : false) +#define psp_support_vmr_ring(psp) \ + ((psp)->funcs->support_vmr_ring ? (psp)->funcs->support_vmr_ring((psp)) : false) #define psp_mode1_reset(psp) \ ((psp)->funcs->mode1_reset ? (psp)->funcs->mode1_reset((psp)) : false) -#define psp_xgmi_get_device_id(psp) \ - ((psp)->funcs->xgmi_get_device_id ? (psp)->funcs->xgmi_get_device_id((psp)) : 0) -#define psp_xgmi_get_hive_id(psp) \ - ((psp)->funcs->xgmi_get_hive_id ? (psp)->funcs->xgmi_get_hive_id((psp)) : 0) +#define psp_xgmi_get_node_id(psp, node_id) \ + ((psp)->funcs->xgmi_get_node_id ? (psp)->funcs->xgmi_get_node_id((psp), (node_id)) : -EINVAL) +#define psp_xgmi_get_hive_id(psp, hive_id) \ + ((psp)->funcs->xgmi_get_hive_id ? (psp)->funcs->xgmi_get_hive_id((psp), (hive_id)) : -EINVAL) #define psp_xgmi_get_topology_info(psp, num_device, topology) \ ((psp)->funcs->xgmi_get_topology_info ? \ (psp)->funcs->xgmi_get_topology_info((psp), (num_device), (topology)) : -EINVAL) @@ -199,6 +216,7 @@ extern int psp_wait_for(struct psp_context *psp, uint32_t reg_index, extern const struct amdgpu_ip_block_version psp_v10_0_ip_block; int psp_gpu_reset(struct amdgpu_device *adev); +int psp_xgmi_invoke(struct psp_context *psp, uint32_t ta_cmd_id); extern const struct amdgpu_ip_block_version psp_v11_0_ip_block; #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c index b70e85ec147d..335a0edf114b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c @@ -338,7 +338,7 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, */ void amdgpu_ring_fini(struct amdgpu_ring *ring) { - ring->ready = false; + ring->sched.ready = false; /* Not to finish a ring which is not initialized */ if (!(ring->adev) || !(ring->adev->rings[ring->idx])) @@ -397,7 +397,7 @@ bool amdgpu_ring_soft_recovery(struct amdgpu_ring *ring, unsigned int vmid, { ktime_t deadline = ktime_add_us(ktime_get(), 10000); - if (!ring->funcs->soft_recovery) + if (!ring->funcs->soft_recovery || !fence) return false; atomic_inc(&ring->adev->gpu_reset_counter); @@ -500,3 +500,29 @@ static void amdgpu_debugfs_ring_fini(struct amdgpu_ring *ring) debugfs_remove(ring->ent); #endif } + +/** + * amdgpu_ring_test_helper - tests ring and set sched readiness status + * + * @ring: ring to try the recovery on + * + * Tests ring and set sched readiness status + * + * Returns 0 on success, error on failure. + */ +int amdgpu_ring_test_helper(struct amdgpu_ring *ring) +{ + struct amdgpu_device *adev = ring->adev; + int r; + + r = amdgpu_ring_test_ring(ring); + if (r) + DRM_DEV_ERROR(adev->dev, "ring %s test failed (%d)\n", + ring->name, r); + else + DRM_DEV_DEBUG(adev->dev, "ring test on %s succeeded\n", + ring->name); + + ring->sched.ready = !r; + return r; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h index 4caa301ce454..d7fae2676269 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h @@ -29,7 +29,7 @@ #include <drm/drm_print.h> /* max number of rings */ -#define AMDGPU_MAX_RINGS 21 +#define AMDGPU_MAX_RINGS 23 #define AMDGPU_MAX_GFX_RINGS 1 #define AMDGPU_MAX_COMPUTE_RINGS 8 #define AMDGPU_MAX_VCE_RINGS 3 @@ -129,8 +129,9 @@ struct amdgpu_ring_funcs { unsigned emit_ib_size; /* command emit functions */ void (*emit_ib)(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch); + uint32_t flags); void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr, uint64_t seq, unsigned flags); void (*emit_pipeline_sync)(struct amdgpu_ring *ring); @@ -189,7 +190,6 @@ struct amdgpu_ring { uint64_t gpu_addr; uint64_t ptr_mask; uint32_t buf_mask; - bool ready; u32 idx; u32 me; u32 pipe; @@ -229,7 +229,7 @@ struct amdgpu_ring { #define amdgpu_ring_get_rptr(r) (r)->funcs->get_rptr((r)) #define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r)) #define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r)) -#define amdgpu_ring_emit_ib(r, ib, vmid, c) (r)->funcs->emit_ib((r), (ib), (vmid), (c)) +#define amdgpu_ring_emit_ib(r, job, ib, flags) ((r)->funcs->emit_ib((r), (job), (ib), (flags))) #define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r)) #define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr)) #define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags)) @@ -313,4 +313,6 @@ static inline void amdgpu_ring_write_multiple(struct amdgpu_ring *ring, ring->count_dw -= count_dw; } +int amdgpu_ring_test_helper(struct amdgpu_ring *ring); + #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.c new file mode 100644 index 000000000000..c8793e6cc3c5 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.c @@ -0,0 +1,282 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * Copyright 2008 Red Hat Inc. + * Copyright 2009 Jerome Glisse. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + */ +#include <linux/firmware.h> +#include "amdgpu.h" +#include "amdgpu_gfx.h" +#include "amdgpu_rlc.h" + +/** + * amdgpu_gfx_rlc_enter_safe_mode - Set RLC into safe mode + * + * @adev: amdgpu_device pointer + * + * Set RLC enter into safe mode if RLC is enabled and haven't in safe mode. + */ +void amdgpu_gfx_rlc_enter_safe_mode(struct amdgpu_device *adev) +{ + if (adev->gfx.rlc.in_safe_mode) + return; + + /* if RLC is not enabled, do nothing */ + if (!adev->gfx.rlc.funcs->is_rlc_enabled(adev)) + return; + + if (adev->cg_flags & + (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG | + AMD_CG_SUPPORT_GFX_3D_CGCG)) { + adev->gfx.rlc.funcs->set_safe_mode(adev); + adev->gfx.rlc.in_safe_mode = true; + } +} + +/** + * amdgpu_gfx_rlc_exit_safe_mode - Set RLC out of safe mode + * + * @adev: amdgpu_device pointer + * + * Set RLC exit safe mode if RLC is enabled and have entered into safe mode. + */ +void amdgpu_gfx_rlc_exit_safe_mode(struct amdgpu_device *adev) +{ + if (!(adev->gfx.rlc.in_safe_mode)) + return; + + /* if RLC is not enabled, do nothing */ + if (!adev->gfx.rlc.funcs->is_rlc_enabled(adev)) + return; + + if (adev->cg_flags & + (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG | + AMD_CG_SUPPORT_GFX_3D_CGCG)) { + adev->gfx.rlc.funcs->unset_safe_mode(adev); + adev->gfx.rlc.in_safe_mode = false; + } +} + +/** + * amdgpu_gfx_rlc_init_sr - Init save restore block + * + * @adev: amdgpu_device pointer + * @dws: the size of save restore block + * + * Allocate and setup value to save restore block of rlc. + * Returns 0 on succeess or negative error code if allocate failed. + */ +int amdgpu_gfx_rlc_init_sr(struct amdgpu_device *adev, u32 dws) +{ + const u32 *src_ptr; + volatile u32 *dst_ptr; + u32 i; + int r; + + /* allocate save restore block */ + r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.save_restore_obj, + &adev->gfx.rlc.save_restore_gpu_addr, + (void **)&adev->gfx.rlc.sr_ptr); + if (r) { + dev_warn(adev->dev, "(%d) create RLC sr bo failed\n", r); + amdgpu_gfx_rlc_fini(adev); + return r; + } + + /* write the sr buffer */ + src_ptr = adev->gfx.rlc.reg_list; + dst_ptr = adev->gfx.rlc.sr_ptr; + for (i = 0; i < adev->gfx.rlc.reg_list_size; i++) + dst_ptr[i] = cpu_to_le32(src_ptr[i]); + amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj); + + return 0; +} + +/** + * amdgpu_gfx_rlc_init_csb - Init clear state block + * + * @adev: amdgpu_device pointer + * + * Allocate and setup value to clear state block of rlc. + * Returns 0 on succeess or negative error code if allocate failed. + */ +int amdgpu_gfx_rlc_init_csb(struct amdgpu_device *adev) +{ + volatile u32 *dst_ptr; + u32 dws; + int r; + + /* allocate clear state block */ + adev->gfx.rlc.clear_state_size = dws = adev->gfx.rlc.funcs->get_csb_size(adev); + r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.clear_state_obj, + &adev->gfx.rlc.clear_state_gpu_addr, + (void **)&adev->gfx.rlc.cs_ptr); + if (r) { + dev_err(adev->dev, "(%d) failed to create rlc csb bo\n", r); + amdgpu_gfx_rlc_fini(adev); + return r; + } + + /* set up the cs buffer */ + dst_ptr = adev->gfx.rlc.cs_ptr; + adev->gfx.rlc.funcs->get_csb_buffer(adev, dst_ptr); + amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); + amdgpu_bo_unpin(adev->gfx.rlc.clear_state_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); + + return 0; +} + +/** + * amdgpu_gfx_rlc_init_cpt - Init cp table + * + * @adev: amdgpu_device pointer + * + * Allocate and setup value to cp table of rlc. + * Returns 0 on succeess or negative error code if allocate failed. + */ +int amdgpu_gfx_rlc_init_cpt(struct amdgpu_device *adev) +{ + int r; + + r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size, + PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.cp_table_obj, + &adev->gfx.rlc.cp_table_gpu_addr, + (void **)&adev->gfx.rlc.cp_table_ptr); + if (r) { + dev_err(adev->dev, "(%d) failed to create cp table bo\n", r); + amdgpu_gfx_rlc_fini(adev); + return r; + } + + /* set up the cp table */ + amdgpu_gfx_rlc_setup_cp_table(adev); + amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj); + + return 0; +} + +/** + * amdgpu_gfx_rlc_setup_cp_table - setup cp the buffer of cp table + * + * @adev: amdgpu_device pointer + * + * Write cp firmware data into cp table. + */ +void amdgpu_gfx_rlc_setup_cp_table(struct amdgpu_device *adev) +{ + const __le32 *fw_data; + volatile u32 *dst_ptr; + int me, i, max_me; + u32 bo_offset = 0; + u32 table_offset, table_size; + + max_me = adev->gfx.rlc.funcs->get_cp_table_num(adev); + + /* write the cp table buffer */ + dst_ptr = adev->gfx.rlc.cp_table_ptr; + for (me = 0; me < max_me; me++) { + if (me == 0) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; + fw_data = (const __le32 *) + (adev->gfx.ce_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 1) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; + fw_data = (const __le32 *) + (adev->gfx.pfp_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 2) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; + fw_data = (const __le32 *) + (adev->gfx.me_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 3) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; + fw_data = (const __le32 *) + (adev->gfx.mec_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 4) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; + fw_data = (const __le32 *) + (adev->gfx.mec2_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } + + for (i = 0; i < table_size; i ++) { + dst_ptr[bo_offset + i] = + cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); + } + + bo_offset += table_size; + } +} + +/** + * amdgpu_gfx_rlc_fini - Free BO which used for RLC + * + * @adev: amdgpu_device pointer + * + * Free three BO which is used for rlc_save_restore_block, rlc_clear_state_block + * and rlc_jump_table_block. + */ +void amdgpu_gfx_rlc_fini(struct amdgpu_device *adev) +{ + /* save restore block */ + if (adev->gfx.rlc.save_restore_obj) { + amdgpu_bo_free_kernel(&adev->gfx.rlc.save_restore_obj, + &adev->gfx.rlc.save_restore_gpu_addr, + (void **)&adev->gfx.rlc.sr_ptr); + } + + /* clear state block */ + amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, + &adev->gfx.rlc.clear_state_gpu_addr, + (void **)&adev->gfx.rlc.cs_ptr); + + /* jump table block */ + amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, + &adev->gfx.rlc.cp_table_gpu_addr, + (void **)&adev->gfx.rlc.cp_table_ptr); +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.h new file mode 100644 index 000000000000..49a8ab52113b --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.h @@ -0,0 +1,98 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef __AMDGPU_RLC_H__ +#define __AMDGPU_RLC_H__ + +#include "clearstate_defs.h" + +struct amdgpu_rlc_funcs { + bool (*is_rlc_enabled)(struct amdgpu_device *adev); + void (*set_safe_mode)(struct amdgpu_device *adev); + void (*unset_safe_mode)(struct amdgpu_device *adev); + int (*init)(struct amdgpu_device *adev); + u32 (*get_csb_size)(struct amdgpu_device *adev); + void (*get_csb_buffer)(struct amdgpu_device *adev, volatile u32 *buffer); + int (*get_cp_table_num)(struct amdgpu_device *adev); + int (*resume)(struct amdgpu_device *adev); + void (*stop)(struct amdgpu_device *adev); + void (*reset)(struct amdgpu_device *adev); + void (*start)(struct amdgpu_device *adev); +}; + +struct amdgpu_rlc { + /* for power gating */ + struct amdgpu_bo *save_restore_obj; + uint64_t save_restore_gpu_addr; + volatile uint32_t *sr_ptr; + const u32 *reg_list; + u32 reg_list_size; + /* for clear state */ + struct amdgpu_bo *clear_state_obj; + uint64_t clear_state_gpu_addr; + volatile uint32_t *cs_ptr; + const struct cs_section_def *cs_data; + u32 clear_state_size; + /* for cp tables */ + struct amdgpu_bo *cp_table_obj; + uint64_t cp_table_gpu_addr; + volatile uint32_t *cp_table_ptr; + u32 cp_table_size; + + /* safe mode for updating CG/PG state */ + bool in_safe_mode; + const struct amdgpu_rlc_funcs *funcs; + + /* for firmware data */ + u32 save_and_restore_offset; + u32 clear_state_descriptor_offset; + u32 avail_scratch_ram_locations; + u32 reg_restore_list_size; + u32 reg_list_format_start; + u32 reg_list_format_separate_start; + u32 starting_offsets_start; + u32 reg_list_format_size_bytes; + u32 reg_list_size_bytes; + u32 reg_list_format_direct_reg_list_length; + u32 save_restore_list_cntl_size_bytes; + u32 save_restore_list_gpm_size_bytes; + u32 save_restore_list_srm_size_bytes; + + u32 *register_list_format; + u32 *register_restore; + u8 *save_restore_list_cntl; + u8 *save_restore_list_gpm; + u8 *save_restore_list_srm; + + bool is_rlc_v2_1; +}; + +void amdgpu_gfx_rlc_enter_safe_mode(struct amdgpu_device *adev); +void amdgpu_gfx_rlc_exit_safe_mode(struct amdgpu_device *adev); +int amdgpu_gfx_rlc_init_sr(struct amdgpu_device *adev, u32 dws); +int amdgpu_gfx_rlc_init_csb(struct amdgpu_device *adev); +int amdgpu_gfx_rlc_init_cpt(struct amdgpu_device *adev); +void amdgpu_gfx_rlc_setup_cp_table(struct amdgpu_device *adev); +void amdgpu_gfx_rlc_fini(struct amdgpu_device *adev); + +#endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c index 12f2bf97611f..bfaf5c6323be 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c @@ -388,7 +388,7 @@ void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager, soffset, eoffset, eoffset - soffset); if (i->fence) - seq_printf(m, " protected by 0x%08x on context %llu", + seq_printf(m, " protected by 0x%016llx on context %llu", i->fence->seqno, i->fence->context); seq_printf(m, "\n"); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c index 1cafe8d83a4d..0767a93e4d91 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c @@ -54,16 +54,20 @@ static int amdgpu_sched_process_priority_override(struct amdgpu_device *adev, enum drm_sched_priority priority) { struct file *filp = fget(fd); - struct drm_file *file; struct amdgpu_fpriv *fpriv; struct amdgpu_ctx *ctx; uint32_t id; + int r; if (!filp) return -EINVAL; - file = filp->private_data; - fpriv = file->driver_priv; + r = amdgpu_file_to_fpriv(filp, &fpriv); + if (r) { + fput(filp); + return r; + } + idr_for_each_entry(&fpriv->ctx_mgr.ctx_handles, ctx, id) amdgpu_ctx_priority_override(ctx, priority); @@ -72,6 +76,39 @@ static int amdgpu_sched_process_priority_override(struct amdgpu_device *adev, return 0; } +static int amdgpu_sched_context_priority_override(struct amdgpu_device *adev, + int fd, + unsigned ctx_id, + enum drm_sched_priority priority) +{ + struct file *filp = fget(fd); + struct amdgpu_fpriv *fpriv; + struct amdgpu_ctx *ctx; + int r; + + if (!filp) + return -EINVAL; + + r = amdgpu_file_to_fpriv(filp, &fpriv); + if (r) { + fput(filp); + return r; + } + + ctx = amdgpu_ctx_get(fpriv, ctx_id); + + if (!ctx) { + fput(filp); + return -EINVAL; + } + + amdgpu_ctx_priority_override(ctx, priority); + amdgpu_ctx_put(ctx); + fput(filp); + + return 0; +} + int amdgpu_sched_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) { @@ -81,7 +118,7 @@ int amdgpu_sched_ioctl(struct drm_device *dev, void *data, int r; priority = amdgpu_to_sched_priority(args->in.priority); - if (args->in.flags || priority == DRM_SCHED_PRIORITY_INVALID) + if (priority == DRM_SCHED_PRIORITY_INVALID) return -EINVAL; switch (args->in.op) { @@ -90,6 +127,12 @@ int amdgpu_sched_ioctl(struct drm_device *dev, void *data, args->in.fd, priority); break; + case AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE: + r = amdgpu_sched_context_priority_override(adev, + args->in.fd, + args->in.ctx_id, + priority); + break; default: DRM_ERROR("Invalid sched op specified: %d\n", args->in.op); r = -EINVAL; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c index bc9244b429ef..115bb0c99b0f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c @@ -28,17 +28,31 @@ * GPU SDMA IP block helpers function. */ -struct amdgpu_sdma_instance * amdgpu_get_sdma_instance(struct amdgpu_ring *ring) +struct amdgpu_sdma_instance *amdgpu_sdma_get_instance_from_ring(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; int i; for (i = 0; i < adev->sdma.num_instances; i++) - if (&adev->sdma.instance[i].ring == ring) - break; + if (ring == &adev->sdma.instance[i].ring || + ring == &adev->sdma.instance[i].page) + return &adev->sdma.instance[i]; - if (i < AMDGPU_MAX_SDMA_INSTANCES) - return &adev->sdma.instance[i]; - else - return NULL; + return NULL; +} + +int amdgpu_sdma_get_index_from_ring(struct amdgpu_ring *ring, uint32_t *index) +{ + struct amdgpu_device *adev = ring->adev; + int i; + + for (i = 0; i < adev->sdma.num_instances; i++) { + if (ring == &adev->sdma.instance[i].ring || + ring == &adev->sdma.instance[i].page) { + *index = i; + return 0; + } + } + + return -EINVAL; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h index 500113ec65ca..16b1a6ae5ba6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h @@ -41,6 +41,7 @@ struct amdgpu_sdma_instance { uint32_t feature_version; struct amdgpu_ring ring; + struct amdgpu_ring page; bool burst_nop; }; @@ -50,6 +51,7 @@ struct amdgpu_sdma { struct amdgpu_irq_src illegal_inst_irq; int num_instances; uint32_t srbm_soft_reset; + bool has_page_queue; }; /* @@ -92,6 +94,7 @@ struct amdgpu_buffer_funcs { #define amdgpu_emit_fill_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_fill_buffer((ib), (s), (d), (b)) struct amdgpu_sdma_instance * -amdgpu_get_sdma_instance(struct amdgpu_ring *ring); +amdgpu_sdma_get_instance_from_ring(struct amdgpu_ring *ring); +int amdgpu_sdma_get_index_from_ring(struct amdgpu_ring *ring, uint32_t *index); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index e9bf70e2ac51..d3ca2424b5fe 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -76,9 +76,10 @@ TRACE_EVENT(amdgpu_mm_wreg, ); TRACE_EVENT(amdgpu_iv, - TP_PROTO(struct amdgpu_iv_entry *iv), - TP_ARGS(iv), + TP_PROTO(unsigned ih, struct amdgpu_iv_entry *iv), + TP_ARGS(ih, iv), TP_STRUCT__entry( + __field(unsigned, ih) __field(unsigned, client_id) __field(unsigned, src_id) __field(unsigned, ring_id) @@ -90,6 +91,7 @@ TRACE_EVENT(amdgpu_iv, __array(unsigned, src_data, 4) ), TP_fast_assign( + __entry->ih = ih; __entry->client_id = iv->client_id; __entry->src_id = iv->src_id; __entry->ring_id = iv->ring_id; @@ -103,8 +105,9 @@ TRACE_EVENT(amdgpu_iv, __entry->src_data[2] = iv->src_data[2]; __entry->src_data[3] = iv->src_data[3]; ), - TP_printk("client_id:%u src_id:%u ring:%u vmid:%u timestamp: %llu pasid:%u src_data: %08x %08x %08x %08x", - __entry->client_id, __entry->src_id, + TP_printk("ih:%u client_id:%u src_id:%u ring:%u vmid:%u " + "timestamp: %llu pasid:%u src_data: %08x %08x %08x %08x", + __entry->ih, __entry->client_id, __entry->src_id, __entry->ring_id, __entry->vmid, __entry->timestamp, __entry->pasid, __entry->src_data[0], __entry->src_data[1], @@ -218,6 +221,7 @@ TRACE_EVENT(amdgpu_vm_grab_id, TP_ARGS(vm, ring, job), TP_STRUCT__entry( __field(u32, pasid) + __string(ring, ring->name) __field(u32, ring) __field(u32, vmid) __field(u32, vm_hub) @@ -227,14 +231,14 @@ TRACE_EVENT(amdgpu_vm_grab_id, TP_fast_assign( __entry->pasid = vm->pasid; - __entry->ring = ring->idx; + __assign_str(ring, ring->name) __entry->vmid = job->vmid; __entry->vm_hub = ring->funcs->vmhub, __entry->pd_addr = job->vm_pd_addr; __entry->needs_flush = job->vm_needs_flush; ), - TP_printk("pasid=%d, ring=%u, id=%u, hub=%u, pd_addr=%010Lx needs_flush=%u", - __entry->pasid, __entry->ring, __entry->vmid, + TP_printk("pasid=%d, ring=%s, id=%u, hub=%u, pd_addr=%010Lx needs_flush=%u", + __entry->pasid, __get_str(ring), __entry->vmid, __entry->vm_hub, __entry->pd_addr, __entry->needs_flush) ); @@ -366,20 +370,20 @@ TRACE_EVENT(amdgpu_vm_flush, uint64_t pd_addr), TP_ARGS(ring, vmid, pd_addr), TP_STRUCT__entry( - __field(u32, ring) + __string(ring, ring->name) __field(u32, vmid) __field(u32, vm_hub) __field(u64, pd_addr) ), TP_fast_assign( - __entry->ring = ring->idx; + __assign_str(ring, ring->name) __entry->vmid = vmid; __entry->vm_hub = ring->funcs->vmhub; __entry->pd_addr = pd_addr; ), - TP_printk("ring=%u, id=%u, hub=%u, pd_addr=%010Lx", - __entry->ring, __entry->vmid, + TP_printk("ring=%s, id=%u, hub=%u, pd_addr=%010Lx", + __get_str(ring), __entry->vmid, __entry->vm_hub,__entry->pd_addr) ); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index a44fc12ae1f9..73e71e61dc99 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -61,100 +61,6 @@ static int amdgpu_map_buffer(struct ttm_buffer_object *bo, static int amdgpu_ttm_debugfs_init(struct amdgpu_device *adev); static void amdgpu_ttm_debugfs_fini(struct amdgpu_device *adev); -/* - * Global memory. - */ - -/** - * amdgpu_ttm_mem_global_init - Initialize and acquire reference to - * memory object - * - * @ref: Object for initialization. - * - * This is called by drm_global_item_ref() when an object is being - * initialized. - */ -static int amdgpu_ttm_mem_global_init(struct drm_global_reference *ref) -{ - return ttm_mem_global_init(ref->object); -} - -/** - * amdgpu_ttm_mem_global_release - Drop reference to a memory object - * - * @ref: Object being removed - * - * This is called by drm_global_item_unref() when an object is being - * released. - */ -static void amdgpu_ttm_mem_global_release(struct drm_global_reference *ref) -{ - ttm_mem_global_release(ref->object); -} - -/** - * amdgpu_ttm_global_init - Initialize global TTM memory reference structures. - * - * @adev: AMDGPU device for which the global structures need to be registered. - * - * This is called as part of the AMDGPU ttm init from amdgpu_ttm_init() - * during bring up. - */ -static int amdgpu_ttm_global_init(struct amdgpu_device *adev) -{ - struct drm_global_reference *global_ref; - int r; - - /* ensure reference is false in case init fails */ - adev->mman.mem_global_referenced = false; - - global_ref = &adev->mman.mem_global_ref; - global_ref->global_type = DRM_GLOBAL_TTM_MEM; - global_ref->size = sizeof(struct ttm_mem_global); - global_ref->init = &amdgpu_ttm_mem_global_init; - global_ref->release = &amdgpu_ttm_mem_global_release; - r = drm_global_item_ref(global_ref); - if (r) { - DRM_ERROR("Failed setting up TTM memory accounting " - "subsystem.\n"); - goto error_mem; - } - - adev->mman.bo_global_ref.mem_glob = - adev->mman.mem_global_ref.object; - global_ref = &adev->mman.bo_global_ref.ref; - global_ref->global_type = DRM_GLOBAL_TTM_BO; - global_ref->size = sizeof(struct ttm_bo_global); - global_ref->init = &ttm_bo_global_init; - global_ref->release = &ttm_bo_global_release; - r = drm_global_item_ref(global_ref); - if (r) { - DRM_ERROR("Failed setting up TTM BO subsystem.\n"); - goto error_bo; - } - - mutex_init(&adev->mman.gtt_window_lock); - - adev->mman.mem_global_referenced = true; - - return 0; - -error_bo: - drm_global_item_unref(&adev->mman.mem_global_ref); -error_mem: - return r; -} - -static void amdgpu_ttm_global_fini(struct amdgpu_device *adev) -{ - if (adev->mman.mem_global_referenced) { - mutex_destroy(&adev->mman.gtt_window_lock); - drm_global_item_unref(&adev->mman.bo_global_ref.ref); - drm_global_item_unref(&adev->mman.mem_global_ref); - adev->mman.mem_global_referenced = false; - } -} - static int amdgpu_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags) { return 0; @@ -1640,7 +1546,8 @@ static struct ttm_bo_driver amdgpu_bo_driver = { .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_free = &amdgpu_ttm_io_mem_free, .io_mem_pfn = amdgpu_ttm_io_mem_pfn, - .access_memory = &amdgpu_ttm_access_memory + .access_memory = &amdgpu_ttm_access_memory, + .del_from_lru_notify = &amdgpu_vm_del_from_lru_notify }; /* @@ -1758,14 +1665,10 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) int r; u64 vis_vram_limit; - /* initialize global references for vram/gtt */ - r = amdgpu_ttm_global_init(adev); - if (r) { - return r; - } + mutex_init(&adev->mman.gtt_window_lock); + /* No others user of address space so set it to 0 */ r = ttm_bo_device_init(&adev->mman.bdev, - adev->mman.bo_global_ref.ref.object, &amdgpu_bo_driver, adev->ddev->anon_inode->i_mapping, DRM_FILE_PAGE_OFFSET, @@ -1853,7 +1756,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) } r = amdgpu_bo_create_kernel(adev, adev->gds.mem.gfx_partition_size, - PAGE_SIZE, AMDGPU_GEM_DOMAIN_GDS, + 4, AMDGPU_GEM_DOMAIN_GDS, &adev->gds.gds_gfx_bo, NULL, NULL); if (r) return r; @@ -1866,7 +1769,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) } r = amdgpu_bo_create_kernel(adev, adev->gds.gws.gfx_partition_size, - PAGE_SIZE, AMDGPU_GEM_DOMAIN_GWS, + 1, AMDGPU_GEM_DOMAIN_GWS, &adev->gds.gws_gfx_bo, NULL, NULL); if (r) return r; @@ -1879,7 +1782,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) } r = amdgpu_bo_create_kernel(adev, adev->gds.oa.gfx_partition_size, - PAGE_SIZE, AMDGPU_GEM_DOMAIN_OA, + 1, AMDGPU_GEM_DOMAIN_OA, &adev->gds.oa_gfx_bo, NULL, NULL); if (r) return r; @@ -1922,7 +1825,6 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev) ttm_bo_clean_mm(&adev->mman.bdev, AMDGPU_PL_GWS); ttm_bo_clean_mm(&adev->mman.bdev, AMDGPU_PL_OA); ttm_bo_device_release(&adev->mman.bdev); - amdgpu_ttm_global_fini(adev); adev->mman.initialized = false; DRM_INFO("amdgpu: ttm finalized\n"); } @@ -2069,7 +1971,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, unsigned i; int r; - if (direct_submit && !ring->ready) { + if (direct_submit && !ring->sched.ready) { DRM_ERROR("Trying to move memory with ring turned off.\n"); return -EINVAL; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h index fe8f276e9811..b5b2d101f7db 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h @@ -39,8 +39,6 @@ #define AMDGPU_GTT_NUM_TRANSFER_WINDOWS 2 struct amdgpu_mman { - struct ttm_bo_global_ref bo_global_ref; - struct drm_global_reference mem_global_ref; struct ttm_bo_device bdev; bool mem_global_referenced; bool initialized; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.h index aa6641b944a0..7ac25a1c7853 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.h @@ -58,6 +58,17 @@ struct psp_firmware_header_v1_0 { }; /* version_major=1, version_minor=0 */ +struct ta_firmware_header_v1_0 { + struct common_firmware_header header; + uint32_t ta_xgmi_ucode_version; + uint32_t ta_xgmi_offset_bytes; + uint32_t ta_xgmi_size_bytes; + uint32_t ta_ras_ucode_version; + uint32_t ta_ras_offset_bytes; + uint32_t ta_ras_size_bytes; +}; + +/* version_major=1, version_minor=0 */ struct gfx_firmware_header_v1_0 { struct common_firmware_header header; uint32_t ucode_feature_version; @@ -170,6 +181,7 @@ union amdgpu_firmware_header { struct mc_firmware_header_v1_0 mc; struct smc_firmware_header_v1_0 smc; struct psp_firmware_header_v1_0 psp; + struct ta_firmware_header_v1_0 ta; struct gfx_firmware_header_v1_0 gfx; struct rlc_firmware_header_v1_0 rlc; struct rlc_firmware_header_v2_0 rlc_v2_0; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c index e5a6db6beab7..4e5d13e41f6a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c @@ -692,6 +692,8 @@ static int amdgpu_uvd_cs_msg_decode(struct amdgpu_device *adev, uint32_t *msg, buf_sizes[0x1] = dpb_size; buf_sizes[0x2] = image_size; buf_sizes[0x4] = min_ctx_size; + /* store image width to adjust nb memory pstate */ + adev->uvd.decode_image_width = width; return 0; } @@ -1243,30 +1245,20 @@ int amdgpu_uvd_ring_test_ib(struct amdgpu_ring *ring, long timeout) { struct dma_fence *fence; long r; - uint32_t ip_instance = ring->me; r = amdgpu_uvd_get_create_msg(ring, 1, NULL); - if (r) { - DRM_ERROR("amdgpu: (%d)failed to get create msg (%ld).\n", ip_instance, r); + if (r) goto error; - } r = amdgpu_uvd_get_destroy_msg(ring, 1, true, &fence); - if (r) { - DRM_ERROR("amdgpu: (%d)failed to get destroy ib (%ld).\n", ip_instance, r); + if (r) goto error; - } r = dma_fence_wait_timeout(fence, false, timeout); - if (r == 0) { - DRM_ERROR("amdgpu: (%d)IB test timed out.\n", ip_instance); + if (r == 0) r = -ETIMEDOUT; - } else if (r < 0) { - DRM_ERROR("amdgpu: (%d)fence wait failed (%ld).\n", ip_instance, r); - } else { - DRM_DEBUG("ib test on (%d)ring %d succeeded\n", ip_instance, ring->idx); + else if (r > 0) r = 0; - } dma_fence_put(fence); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h index a3ab1a41060f..5eb63288d157 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h @@ -65,6 +65,8 @@ struct amdgpu_uvd { struct drm_sched_entity entity; struct delayed_work idle_work; unsigned harvest_config; + /* store image width to adjust nb memory state */ + unsigned decode_image_width; }; int amdgpu_uvd_sw_init(struct amdgpu_device *adev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c index 5f3f54073818..c021b114c8a4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c @@ -1032,8 +1032,10 @@ out: * @ib: the IB to execute * */ -void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) +void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { amdgpu_ring_write(ring, VCE_CMD_IB); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); @@ -1079,11 +1081,9 @@ int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring) return 0; r = amdgpu_ring_alloc(ring, 16); - if (r) { - DRM_ERROR("amdgpu: vce failed to lock ring %d (%d).\n", - ring->idx, r); + if (r) return r; - } + amdgpu_ring_write(ring, VCE_CMD_END); amdgpu_ring_commit(ring); @@ -1093,14 +1093,8 @@ int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed\n", - ring->idx); + if (i >= timeout) r = -ETIMEDOUT; - } return r; } @@ -1121,27 +1115,19 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout) return 0; r = amdgpu_vce_get_create_msg(ring, 1, NULL); - if (r) { - DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r); + if (r) goto error; - } r = amdgpu_vce_get_destroy_msg(ring, 1, true, &fence); - if (r) { - DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r); + if (r) goto error; - } r = dma_fence_wait_timeout(fence, false, timeout); - if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out.\n"); + if (r == 0) r = -ETIMEDOUT; - } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); - } else { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + else if (r > 0) r = 0; - } + error: dma_fence_put(fence); return r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h index a1f209eed4c4..30ea54dd9117 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h @@ -65,8 +65,8 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp); int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx); int amdgpu_vce_ring_parse_cs_vm(struct amdgpu_cs_parser *p, uint32_t ib_idx); -void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch); +void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_job *job, + struct amdgpu_ib *ib, uint32_t flags); void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, unsigned flags); int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c index 27da13df2f11..ecf6f96df2ad 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c @@ -262,7 +262,7 @@ static int amdgpu_vcn_pause_dpg_mode(struct amdgpu_device *adev, ring = &adev->vcn.ring_dec; WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, - RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2)); + RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); @@ -322,7 +322,7 @@ static int amdgpu_vcn_pause_dpg_mode(struct amdgpu_device *adev, ring = &adev->vcn.ring_dec; WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, - RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2)); + RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); @@ -396,16 +396,26 @@ void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring) if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { struct dpg_pause_state new_state; + unsigned int fences = 0; + unsigned int i; - if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) + for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + fences += amdgpu_fence_count_emitted(&adev->vcn.ring_enc[i]); + } + if (fences) new_state.fw_based = VCN_DPG_STATE__PAUSE; else - new_state.fw_based = adev->vcn.pause_state.fw_based; + new_state.fw_based = VCN_DPG_STATE__UNPAUSE; - if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG) + if (amdgpu_fence_count_emitted(&adev->vcn.ring_jpeg)) new_state.jpeg = VCN_DPG_STATE__PAUSE; else - new_state.jpeg = adev->vcn.pause_state.jpeg; + new_state.jpeg = VCN_DPG_STATE__UNPAUSE; + + if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) + new_state.fw_based = VCN_DPG_STATE__PAUSE; + else if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG) + new_state.jpeg = VCN_DPG_STATE__PAUSE; amdgpu_vcn_pause_dpg_mode(adev, &new_state); } @@ -425,11 +435,9 @@ int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring) WREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9), 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", - ring->idx, r); + if (r) return r; - } + amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9), 0)); amdgpu_ring_write(ring, 0xDEADBEEF); @@ -441,14 +449,9 @@ int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + return r; } @@ -570,30 +573,20 @@ int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_vcn_dec_get_create_msg(ring, 1, NULL); - if (r) { - DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r); + if (r) goto error; - } r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &fence); - if (r) { - DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r); + if (r) goto error; - } r = dma_fence_wait_timeout(fence, false, timeout); - if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out.\n"); + if (r == 0) r = -ETIMEDOUT; - } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); - } else { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + else if (r > 0) r = 0; - } dma_fence_put(fence); - error: return r; } @@ -606,11 +599,9 @@ int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring) int r; r = amdgpu_ring_alloc(ring, 16); - if (r) { - DRM_ERROR("amdgpu: vcn enc failed to lock ring %d (%d).\n", - ring->idx, r); + if (r) return r; - } + amdgpu_ring_write(ring, VCN_ENC_CMD_END); amdgpu_ring_commit(ring); @@ -620,14 +611,8 @@ int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed\n", - ring->idx); + if (i >= adev->usec_timeout) r = -ETIMEDOUT; - } return r; } @@ -742,27 +727,19 @@ int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_vcn_enc_get_create_msg(ring, 1, NULL); - if (r) { - DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r); + if (r) goto error; - } r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &fence); - if (r) { - DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r); + if (r) goto error; - } r = dma_fence_wait_timeout(fence, false, timeout); - if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out.\n"); + if (r == 0) r = -ETIMEDOUT; - } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); - } else { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + else if (r > 0) r = 0; - } + error: dma_fence_put(fence); return r; @@ -778,11 +755,8 @@ int amdgpu_vcn_jpeg_ring_test_ring(struct amdgpu_ring *ring) WREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9), 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", - ring->idx, r); + if (r) return r; - } amdgpu_ring_write(ring, PACKETJ(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9), 0, 0, 0)); @@ -796,14 +770,8 @@ int amdgpu_vcn_jpeg_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; return r; } @@ -856,21 +824,18 @@ int amdgpu_vcn_jpeg_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r = 0; r = amdgpu_vcn_jpeg_set_reg(ring, 1, &fence); - if (r) { - DRM_ERROR("amdgpu: failed to set jpeg register (%ld).\n", r); + if (r) goto error; - } r = dma_fence_wait_timeout(fence, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out.\n"); r = -ETIMEDOUT; goto error; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto error; - } else + } else { r = 0; + } for (i = 0; i < adev->usec_timeout; i++) { tmp = RREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9)); @@ -879,15 +844,10 @@ int amdgpu_vcn_jpeg_ring_test_ib(struct amdgpu_ring *ring, long timeout) DRM_UDELAY(1); } - if (i < adev->usec_timeout) - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); - else { - DRM_ERROR("ib test failed (0x%08X)\n", tmp); - r = -EINVAL; - } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; dma_fence_put(fence); - error: return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c index f2f358aa0597..462a04e0f5e6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c @@ -23,16 +23,6 @@ #include "amdgpu.h" -uint64_t amdgpu_csa_vaddr(struct amdgpu_device *adev) -{ - uint64_t addr = adev->vm_manager.max_pfn << AMDGPU_GPU_PAGE_SHIFT; - - addr -= AMDGPU_VA_RESERVED_SIZE; - addr = amdgpu_gmc_sign_extend(addr); - - return addr; -} - bool amdgpu_virt_mmio_blocked(struct amdgpu_device *adev) { /* By now all MMIO pages except mailbox are blocked */ @@ -41,88 +31,6 @@ bool amdgpu_virt_mmio_blocked(struct amdgpu_device *adev) return RREG32_NO_KIQ(0xc040) == 0xffffffff; } -int amdgpu_allocate_static_csa(struct amdgpu_device *adev) -{ - int r; - void *ptr; - - r = amdgpu_bo_create_kernel(adev, AMDGPU_CSA_SIZE, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, &adev->virt.csa_obj, - &adev->virt.csa_vmid0_addr, &ptr); - if (r) - return r; - - memset(ptr, 0, AMDGPU_CSA_SIZE); - return 0; -} - -void amdgpu_free_static_csa(struct amdgpu_device *adev) { - amdgpu_bo_free_kernel(&adev->virt.csa_obj, - &adev->virt.csa_vmid0_addr, - NULL); -} - -/* - * amdgpu_map_static_csa should be called during amdgpu_vm_init - * it maps virtual address amdgpu_csa_vaddr() to this VM, and each command - * submission of GFX should use this virtual address within META_DATA init - * package to support SRIOV gfx preemption. - */ -int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm, - struct amdgpu_bo_va **bo_va) -{ - uint64_t csa_addr = amdgpu_csa_vaddr(adev) & AMDGPU_GMC_HOLE_MASK; - struct ww_acquire_ctx ticket; - struct list_head list; - struct amdgpu_bo_list_entry pd; - struct ttm_validate_buffer csa_tv; - int r; - - INIT_LIST_HEAD(&list); - INIT_LIST_HEAD(&csa_tv.head); - csa_tv.bo = &adev->virt.csa_obj->tbo; - csa_tv.shared = true; - - list_add(&csa_tv.head, &list); - amdgpu_vm_get_pd_bo(vm, &list, &pd); - - r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL); - if (r) { - DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r); - return r; - } - - *bo_va = amdgpu_vm_bo_add(adev, vm, adev->virt.csa_obj); - if (!*bo_va) { - ttm_eu_backoff_reservation(&ticket, &list); - DRM_ERROR("failed to create bo_va for static CSA\n"); - return -ENOMEM; - } - - r = amdgpu_vm_alloc_pts(adev, (*bo_va)->base.vm, csa_addr, - AMDGPU_CSA_SIZE); - if (r) { - DRM_ERROR("failed to allocate pts for static CSA, err=%d\n", r); - amdgpu_vm_bo_rmv(adev, *bo_va); - ttm_eu_backoff_reservation(&ticket, &list); - return r; - } - - r = amdgpu_vm_bo_map(adev, *bo_va, csa_addr, 0, AMDGPU_CSA_SIZE, - AMDGPU_PTE_READABLE | AMDGPU_PTE_WRITEABLE | - AMDGPU_PTE_EXECUTABLE); - - if (r) { - DRM_ERROR("failed to do bo_map on static CSA, err=%d\n", r); - amdgpu_vm_bo_rmv(adev, *bo_va); - ttm_eu_backoff_reservation(&ticket, &list); - return r; - } - - ttm_eu_backoff_reservation(&ticket, &list); - return 0; -} - void amdgpu_virt_init_setting(struct amdgpu_device *adev) { /* enable virtual display */ @@ -162,9 +70,7 @@ uint32_t amdgpu_virt_kiq_rreg(struct amdgpu_device *adev, uint32_t reg) if (r < 1 && (adev->in_gpu_reset || in_interrupt())) goto failed_kiq_read; - if (in_interrupt()) - might_sleep(); - + might_sleep(); while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) { msleep(MAX_KIQ_REG_BAILOUT_INTERVAL); r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); @@ -210,9 +116,7 @@ void amdgpu_virt_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v) if (r < 1 && (adev->in_gpu_reset || in_interrupt())) goto failed_kiq_write; - if (in_interrupt()) - might_sleep(); - + might_sleep(); while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) { msleep(MAX_KIQ_REG_BAILOUT_INTERVAL); @@ -228,6 +132,46 @@ failed_kiq_write: pr_err("failed to write reg:%x\n", reg); } +void amdgpu_virt_kiq_reg_write_reg_wait(struct amdgpu_device *adev, + uint32_t reg0, uint32_t reg1, + uint32_t ref, uint32_t mask) +{ + struct amdgpu_kiq *kiq = &adev->gfx.kiq; + struct amdgpu_ring *ring = &kiq->ring; + signed long r, cnt = 0; + unsigned long flags; + uint32_t seq; + + spin_lock_irqsave(&kiq->ring_lock, flags); + amdgpu_ring_alloc(ring, 32); + amdgpu_ring_emit_reg_write_reg_wait(ring, reg0, reg1, + ref, mask); + amdgpu_fence_emit_polling(ring, &seq); + amdgpu_ring_commit(ring); + spin_unlock_irqrestore(&kiq->ring_lock, flags); + + r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); + + /* don't wait anymore for IRQ context */ + if (r < 1 && in_interrupt()) + goto failed_kiq; + + might_sleep(); + while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) { + + msleep(MAX_KIQ_REG_BAILOUT_INTERVAL); + r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); + } + + if (cnt > MAX_KIQ_REG_TRY) + goto failed_kiq; + + return; + +failed_kiq: + pr_err("failed to write reg %x wait reg %x\n", reg0, reg1); +} + /** * amdgpu_virt_request_full_gpu() - request full gpu access * @amdgpu: amdgpu device. @@ -390,7 +334,7 @@ void amdgpu_virt_init_data_exchange(struct amdgpu_device *adev) if (adev->fw_vram_usage.va != NULL) { adev->virt.fw_reserve.p_pf2vf = - (struct amdgim_pf2vf_info_header *)( + (struct amd_sriov_msg_pf2vf_info_header *)( adev->fw_vram_usage.va + AMDGIM_DATAEXCHANGE_OFFSET); AMDGPU_FW_VRAM_PF2VF_READ(adev, header.size, &pf2vf_size); AMDGPU_FW_VRAM_PF2VF_READ(adev, checksum, &checksum); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h index 880ac113a3a9..722deefc0a7e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h @@ -63,8 +63,8 @@ struct amdgpu_virt_ops { * Firmware Reserve Frame buffer */ struct amdgpu_virt_fw_reserve { - struct amdgim_pf2vf_info_header *p_pf2vf; - struct amdgim_vf2pf_info_header *p_vf2pf; + struct amd_sriov_msg_pf2vf_info_header *p_pf2vf; + struct amd_sriov_msg_vf2pf_info_header *p_vf2pf; unsigned int checksum_key; }; /* @@ -85,15 +85,17 @@ enum AMDGIM_FEATURE_FLAG { AMDGIM_FEATURE_GIM_FLR_VRAMLOST = 0x4, }; -struct amdgim_pf2vf_info_header { +struct amd_sriov_msg_pf2vf_info_header { /* the total structure size in byte. */ uint32_t size; /* version of this structure, written by the GIM */ uint32_t version; + /* reserved */ + uint32_t reserved[2]; } __aligned(4); struct amdgim_pf2vf_info_v1 { /* header contains size and version */ - struct amdgim_pf2vf_info_header header; + struct amd_sriov_msg_pf2vf_info_header header; /* max_width * max_height */ unsigned int uvd_enc_max_pixels_count; /* 16x16 pixels/sec, codec independent */ @@ -112,7 +114,7 @@ struct amdgim_pf2vf_info_v1 { struct amdgim_pf2vf_info_v2 { /* header contains size and version */ - struct amdgim_pf2vf_info_header header; + struct amd_sriov_msg_pf2vf_info_header header; /* use private key from mailbox 2 to create chueksum */ uint32_t checksum; /* The features flags of the GIM driver supports. */ @@ -137,20 +139,22 @@ struct amdgim_pf2vf_info_v2 { uint64_t vcefw_kboffset; /* VCE FW size in KB */ uint32_t vcefw_ksize; - uint32_t reserved[AMDGIM_GET_STRUCTURE_RESERVED_SIZE(256, 0, 0, (9 + sizeof(struct amdgim_pf2vf_info_header)/sizeof(uint32_t)), 3)]; + uint32_t reserved[AMDGIM_GET_STRUCTURE_RESERVED_SIZE(256, 0, 0, (9 + sizeof(struct amd_sriov_msg_pf2vf_info_header)/sizeof(uint32_t)), 3)]; } __aligned(4); -struct amdgim_vf2pf_info_header { +struct amd_sriov_msg_vf2pf_info_header { /* the total structure size in byte. */ uint32_t size; /*version of this structure, written by the guest */ uint32_t version; + /* reserved */ + uint32_t reserved[2]; } __aligned(4); struct amdgim_vf2pf_info_v1 { /* header contains size and version */ - struct amdgim_vf2pf_info_header header; + struct amd_sriov_msg_vf2pf_info_header header; /* driver version */ char driver_version[64]; /* driver certification, 1=WHQL, 0=None */ @@ -180,7 +184,7 @@ struct amdgim_vf2pf_info_v1 { struct amdgim_vf2pf_info_v2 { /* header contains size and version */ - struct amdgim_vf2pf_info_header header; + struct amd_sriov_msg_vf2pf_info_header header; uint32_t checksum; /* driver version */ uint8_t driver_version[64]; @@ -206,7 +210,7 @@ struct amdgim_vf2pf_info_v2 { uint32_t uvd_enc_usage; /* guest uvd engine usage percentage. 0xffff means N/A. */ uint32_t uvd_enc_health; - uint32_t reserved[AMDGIM_GET_STRUCTURE_RESERVED_SIZE(256, 64, 0, (12 + sizeof(struct amdgim_vf2pf_info_header)/sizeof(uint32_t)), 0)]; + uint32_t reserved[AMDGIM_GET_STRUCTURE_RESERVED_SIZE(256, 64, 0, (12 + sizeof(struct amd_sriov_msg_vf2pf_info_header)/sizeof(uint32_t)), 0)]; } __aligned(4); #define AMDGPU_FW_VRAM_VF2PF_VER 2 @@ -238,7 +242,6 @@ typedef struct amdgim_vf2pf_info_v2 amdgim_vf2pf_info ; struct amdgpu_virt { uint32_t caps; struct amdgpu_bo *csa_obj; - uint64_t csa_vmid0_addr; bool chained_ib_support; uint32_t reg_val_offs; struct amdgpu_irq_src ack_irq; @@ -251,8 +254,6 @@ struct amdgpu_virt { uint32_t gim_feature; }; -#define AMDGPU_CSA_SIZE (8 * 1024) - #define amdgpu_sriov_enabled(adev) \ ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_ENABLE_IOV) @@ -277,17 +278,13 @@ static inline bool is_virtual_machine(void) #endif } -struct amdgpu_vm; - -uint64_t amdgpu_csa_vaddr(struct amdgpu_device *adev); bool amdgpu_virt_mmio_blocked(struct amdgpu_device *adev); -int amdgpu_allocate_static_csa(struct amdgpu_device *adev); -int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm, - struct amdgpu_bo_va **bo_va); -void amdgpu_free_static_csa(struct amdgpu_device *adev); void amdgpu_virt_init_setting(struct amdgpu_device *adev); uint32_t amdgpu_virt_kiq_rreg(struct amdgpu_device *adev, uint32_t reg); void amdgpu_virt_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v); +void amdgpu_virt_kiq_reg_write_reg_wait(struct amdgpu_device *adev, + uint32_t reg0, uint32_t rreg1, + uint32_t ref, uint32_t mask); int amdgpu_virt_request_full_gpu(struct amdgpu_device *adev, bool init); int amdgpu_virt_release_full_gpu(struct amdgpu_device *adev, bool init); int amdgpu_virt_reset_gpu(struct amdgpu_device *adev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 0877ff9a9594..16fcb56c232b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -107,14 +107,6 @@ struct amdgpu_pte_update_params { * DMA addresses to use for mapping, used during VM update by CPU */ dma_addr_t *pages_addr; - - /** - * @kptr: - * - * Kernel pointer of PD/PT BO that needs to be updated, - * used during VM update by CPU - */ - void *kptr; }; /** @@ -617,11 +609,34 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm, { entry->priority = 0; entry->tv.bo = &vm->root.base.bo->tbo; - entry->tv.shared = true; + /* One for the VM updates, one for TTM and one for the CS job */ + entry->tv.num_shared = 3; entry->user_pages = NULL; list_add(&entry->tv.head, validated); } +void amdgpu_vm_del_from_lru_notify(struct ttm_buffer_object *bo) +{ + struct amdgpu_bo *abo; + struct amdgpu_vm_bo_base *bo_base; + + if (!amdgpu_bo_is_amdgpu_bo(bo)) + return; + + if (bo->mem.placement & TTM_PL_FLAG_NO_EVICT) + return; + + abo = ttm_to_amdgpu_bo(bo); + if (!abo->parent) + return; + for (bo_base = abo->vm_bo; bo_base; bo_base = bo_base->next) { + struct amdgpu_vm *vm = bo_base->vm; + + if (abo->tbo.resv == vm->root.base.bo->tbo.resv) + vm->bulk_moveable = false; + } + +} /** * amdgpu_vm_move_to_lru_tail - move all BOs to the end of LRU * @@ -637,12 +652,14 @@ void amdgpu_vm_move_to_lru_tail(struct amdgpu_device *adev, struct ttm_bo_global *glob = adev->mman.bdev.glob; struct amdgpu_vm_bo_base *bo_base; +#if 0 if (vm->bulk_moveable) { spin_lock(&glob->lru_lock); ttm_bo_bulk_move_lru_tail(&vm->lru_bulk_move); spin_unlock(&glob->lru_lock); return; } +#endif memset(&vm->lru_bulk_move, 0, sizeof(vm->lru_bulk_move)); @@ -773,10 +790,6 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, ring = container_of(vm->entity.rq->sched, struct amdgpu_ring, sched); - r = reservation_object_reserve_shared(bo->tbo.resv); - if (r) - return r; - r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); if (r) goto error; @@ -802,15 +815,22 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, addr += ats_entries * 8; } - if (entries) + if (entries) { + uint64_t value = 0; + + /* Workaround for fault priority problem on GMC9 */ + if (level == AMDGPU_VM_PTB && adev->asic_type >= CHIP_VEGA10) + value = AMDGPU_PTE_EXECUTABLE; + amdgpu_vm_set_pte_pde(adev, &job->ibs[0], addr, 0, - entries, 0, 0); + entries, 0, value); + } amdgpu_ring_pad_ib(ring, &job->ibs[0]); WARN_ON(job->ibs[0].length_dw > 64); r = amdgpu_sync_resv(adev, &job->sync, bo->tbo.resv, - AMDGPU_FENCE_OWNER_UNDEFINED, false); + AMDGPU_FENCE_OWNER_KFD, false); if (r) goto error_free; @@ -850,9 +870,6 @@ static void amdgpu_vm_bo_param(struct amdgpu_device *adev, struct amdgpu_vm *vm, bp->size = amdgpu_vm_bo_size(adev, level); bp->byte_align = AMDGPU_GPU_PAGE_SIZE; bp->domain = AMDGPU_GEM_DOMAIN_VRAM; - if (bp->size <= PAGE_SIZE && adev->asic_type >= CHIP_VEGA10 && - adev->flags & AMD_IS_APU) - bp->domain |= AMDGPU_GEM_DOMAIN_GTT; bp->domain = amdgpu_bo_get_preferred_pin_domain(adev, bp->domain); bp->flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS | AMDGPU_GEM_CREATE_CPU_GTT_USWC; @@ -932,10 +949,6 @@ int amdgpu_vm_alloc_pts(struct amdgpu_device *adev, if (r) return r; - r = amdgpu_vm_clear_bo(adev, vm, pt, cursor.level, ats); - if (r) - goto error_free_pt; - if (vm->use_cpu_for_update) { r = amdgpu_bo_kmap(pt, NULL); if (r) @@ -948,6 +961,10 @@ int amdgpu_vm_alloc_pts(struct amdgpu_device *adev, pt->parent = amdgpu_bo_ref(cursor.parent->base.bo); amdgpu_vm_bo_base_init(&entry->base, vm, pt); + + r = amdgpu_vm_clear_bo(adev, vm, pt, cursor.level, ats); + if (r) + goto error_free_pt; } return 0; @@ -1317,31 +1334,6 @@ static void amdgpu_vm_cpu_set_ptes(struct amdgpu_pte_update_params *params, } } - -/** - * amdgpu_vm_wait_pd - Wait for PT BOs to be free. - * - * @adev: amdgpu_device pointer - * @vm: related vm - * @owner: fence owner - * - * Returns: - * 0 on success, errno otherwise. - */ -static int amdgpu_vm_wait_pd(struct amdgpu_device *adev, struct amdgpu_vm *vm, - void *owner) -{ - struct amdgpu_sync sync; - int r; - - amdgpu_sync_create(&sync); - amdgpu_sync_resv(adev, &sync, vm->root.base.bo->tbo.resv, owner, false); - r = amdgpu_sync_wait(&sync, true); - amdgpu_sync_free(&sync); - - return r; -} - /** * amdgpu_vm_update_func - helper to call update function * @@ -1436,7 +1428,8 @@ restart: params.adev = adev; if (vm->use_cpu_for_update) { - r = amdgpu_vm_wait_pd(adev, vm, AMDGPU_FENCE_OWNER_VM); + r = amdgpu_bo_sync_wait(vm->root.base.bo, + AMDGPU_FENCE_OWNER_VM, true); if (unlikely(r)) return r; @@ -1509,20 +1502,27 @@ error: } /** - * amdgpu_vm_update_huge - figure out parameters for PTE updates + * amdgpu_vm_update_flags - figure out flags for PTE updates * * Make sure to set the right flags for the PTEs at the desired level. */ -static void amdgpu_vm_update_huge(struct amdgpu_pte_update_params *params, - struct amdgpu_bo *bo, unsigned level, - uint64_t pe, uint64_t addr, - unsigned count, uint32_t incr, - uint64_t flags) +static void amdgpu_vm_update_flags(struct amdgpu_pte_update_params *params, + struct amdgpu_bo *bo, unsigned level, + uint64_t pe, uint64_t addr, + unsigned count, uint32_t incr, + uint64_t flags) { if (level != AMDGPU_VM_PTB) { flags |= AMDGPU_PDE_PTE; amdgpu_gmc_get_vm_pde(params->adev, level, &addr, &flags); + + } else if (params->adev->asic_type >= CHIP_VEGA10 && + !(flags & AMDGPU_PTE_VALID) && + !(flags & AMDGPU_PTE_PRT)) { + + /* Workaround for fault priority problem on GMC9 */ + flags |= AMDGPU_PTE_EXECUTABLE; } amdgpu_vm_update_func(params, bo, pe, addr, count, incr, flags); @@ -1679,9 +1679,9 @@ static int amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, uint64_t upd_end = min(entry_end, frag_end); unsigned nptes = (upd_end - frag_start) >> shift; - amdgpu_vm_update_huge(params, pt, cursor.level, - pe_start, dst, nptes, incr, - flags | AMDGPU_PTE_FRAG(frag)); + amdgpu_vm_update_flags(params, pt, cursor.level, + pe_start, dst, nptes, incr, + flags | AMDGPU_PTE_FRAG(frag)); pe_start += nptes * 8; dst += (uint64_t)nptes * AMDGPU_GPU_PAGE_SIZE << shift; @@ -1750,22 +1750,29 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, params.adev = adev; params.vm = vm; - /* sync to everything on unmapping */ + /* sync to everything except eviction fences on unmapping */ if (!(flags & AMDGPU_PTE_VALID)) - owner = AMDGPU_FENCE_OWNER_UNDEFINED; + owner = AMDGPU_FENCE_OWNER_KFD; if (vm->use_cpu_for_update) { /* params.src is used as flag to indicate system Memory */ if (pages_addr) params.src = ~0; - /* Wait for PT BOs to be free. PTs share the same resv. object + /* Wait for PT BOs to be idle. PTs share the same resv. object * as the root PD BO */ - r = amdgpu_vm_wait_pd(adev, vm, owner); + r = amdgpu_bo_sync_wait(vm->root.base.bo, owner, true); if (unlikely(r)) return r; + /* Wait for any BO move to be completed */ + if (exclusive) { + r = dma_fence_wait(exclusive, true); + if (unlikely(r)) + return r; + } + params.func = amdgpu_vm_cpu_set_ptes; params.pages_addr = pages_addr; return amdgpu_vm_update_ptes(¶ms, start, last + 1, @@ -1779,13 +1786,12 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, /* * reserve space for two commands every (1 << BLOCK_SIZE) * entries or 2k dwords (whatever is smaller) - * - * The second command is for the shadow pagetables. */ + ncmds = ((nptes >> min(adev->vm_manager.block_size, 11u)) + 1); + + /* The second command is for the shadow pagetables. */ if (vm->root.base.bo->shadow) - ncmds = ((nptes >> min(adev->vm_manager.block_size, 11u)) + 1) * 2; - else - ncmds = ((nptes >> min(adev->vm_manager.block_size, 11u)) + 1); + ncmds *= 2; /* padding, etc. */ ndw = 64; @@ -1804,10 +1810,11 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, ndw += ncmds * 10; /* extra commands for begin/end fragments */ + ncmds = 2 * adev->vm_manager.fragment_size; if (vm->root.base.bo->shadow) - ndw += 2 * 10 * adev->vm_manager.fragment_size * 2; - else - ndw += 2 * 10 * adev->vm_manager.fragment_size; + ncmds *= 2; + + ndw += 10 * ncmds; params.func = amdgpu_vm_do_set_ptes; } @@ -1844,10 +1851,6 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, if (r) goto error_free; - r = reservation_object_reserve_shared(vm->root.base.bo->tbo.resv); - if (r) - goto error_free; - r = amdgpu_vm_update_ptes(¶ms, start, last + 1, addr, flags); if (r) goto error_free; @@ -3013,7 +3016,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, } DRM_DEBUG_DRIVER("VM update mode is %s\n", vm->use_cpu_for_update ? "CPU" : "SDMA"); - WARN_ONCE((vm->use_cpu_for_update & !amdgpu_gmc_vram_full_visible(&adev->gmc)), + WARN_ONCE((vm->use_cpu_for_update && !amdgpu_gmc_vram_full_visible(&adev->gmc)), "CPU update of VM recommended only for large BAR system\n"); vm->last_update = NULL; @@ -3028,13 +3031,18 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, if (r) goto error_free_root; + r = reservation_object_reserve_shared(root->tbo.resv, 1); + if (r) + goto error_unreserve; + + amdgpu_vm_bo_base_init(&vm->root.base, vm, root); + r = amdgpu_vm_clear_bo(adev, vm, root, adev->vm_manager.root_level, vm->pte_support_ats); if (r) goto error_unreserve; - amdgpu_vm_bo_base_init(&vm->root.base, vm, root); amdgpu_bo_unreserve(vm->root.base.bo); if (pasid) { @@ -3057,7 +3065,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, } INIT_KFIFO(vm->faults); - vm->fault_credit = 16; return 0; @@ -3140,7 +3147,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm, uns vm->pte_support_ats = pte_support_ats; DRM_DEBUG_DRIVER("VM update mode is %s\n", vm->use_cpu_for_update ? "CPU" : "SDMA"); - WARN_ONCE((vm->use_cpu_for_update & !amdgpu_gmc_vram_full_visible(&adev->gmc)), + WARN_ONCE((vm->use_cpu_for_update && !amdgpu_gmc_vram_full_visible(&adev->gmc)), "CPU update of VM recommended only for large BAR system\n"); if (vm->pasid) { @@ -3270,42 +3277,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) } /** - * amdgpu_vm_pasid_fault_credit - Check fault credit for given PASID - * - * @adev: amdgpu_device pointer - * @pasid: PASID do identify the VM - * - * This function is expected to be called in interrupt context. - * - * Returns: - * True if there was fault credit, false otherwise - */ -bool amdgpu_vm_pasid_fault_credit(struct amdgpu_device *adev, - unsigned int pasid) -{ - struct amdgpu_vm *vm; - - spin_lock(&adev->vm_manager.pasid_lock); - vm = idr_find(&adev->vm_manager.pasid_idr, pasid); - if (!vm) { - /* VM not found, can't track fault credit */ - spin_unlock(&adev->vm_manager.pasid_lock); - return true; - } - - /* No lock needed. only accessed by IRQ handler */ - if (!vm->fault_credit) { - /* Too many faults in this VM */ - spin_unlock(&adev->vm_manager.pasid_lock); - return false; - } - - vm->fault_credit--; - spin_unlock(&adev->vm_manager.pasid_lock); - return true; -} - -/** * amdgpu_vm_manager_init - init the VM manager * * @adev: amdgpu_device pointer @@ -3406,14 +3377,15 @@ void amdgpu_vm_get_task_info(struct amdgpu_device *adev, unsigned int pasid, struct amdgpu_task_info *task_info) { struct amdgpu_vm *vm; + unsigned long flags; - spin_lock(&adev->vm_manager.pasid_lock); + spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags); vm = idr_find(&adev->vm_manager.pasid_idr, pasid); if (vm) *task_info = vm->task_info; - spin_unlock(&adev->vm_manager.pasid_lock); + spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags); } /** diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h index 2a8898d19c8b..81ff8177f092 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h @@ -229,9 +229,6 @@ struct amdgpu_vm { /* Up to 128 pending retry page faults */ DECLARE_KFIFO(faults, u64, 128); - /* Limit non-retry fault storms */ - unsigned int fault_credit; - /* Points to the KFD process VM info */ struct amdkfd_process_info *process_info; @@ -299,8 +296,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm, unsigned int pasid); void amdgpu_vm_release_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm); void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm); -bool amdgpu_vm_pasid_fault_credit(struct amdgpu_device *adev, - unsigned int pasid); void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm, struct list_head *validated, struct amdgpu_bo_list_entry *entry); @@ -368,4 +363,6 @@ int amdgpu_vm_add_fault(struct amdgpu_retryfault_hashtable *fault_hash, u64 key) void amdgpu_vm_clear_fault(struct amdgpu_retryfault_hashtable *fault_hash, u64 key); +void amdgpu_vm_del_from_lru_notify(struct ttm_buffer_object *bo); + #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c index 897afbb348c1..407dd16cc35c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c @@ -23,7 +23,7 @@ */ #include <linux/list.h> #include "amdgpu.h" -#include "amdgpu_psp.h" +#include "amdgpu_xgmi.h" static DEFINE_MUTEX(xgmi_mutex); @@ -31,89 +31,154 @@ static DEFINE_MUTEX(xgmi_mutex); #define AMDGPU_MAX_XGMI_HIVE 8 #define AMDGPU_MAX_XGMI_DEVICE_PER_HIVE 4 -struct amdgpu_hive_info { - uint64_t hive_id; - struct list_head device_list; -}; - static struct amdgpu_hive_info xgmi_hives[AMDGPU_MAX_XGMI_HIVE]; static unsigned hive_count = 0; -static struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev) + +void *amdgpu_xgmi_hive_try_lock(struct amdgpu_hive_info *hive) +{ + return &hive->device_list; +} + +struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev, int lock) { int i; struct amdgpu_hive_info *tmp; if (!adev->gmc.xgmi.hive_id) return NULL; + + mutex_lock(&xgmi_mutex); + for (i = 0 ; i < hive_count; ++i) { tmp = &xgmi_hives[i]; - if (tmp->hive_id == adev->gmc.xgmi.hive_id) + if (tmp->hive_id == adev->gmc.xgmi.hive_id) { + if (lock) + mutex_lock(&tmp->hive_lock); + mutex_unlock(&xgmi_mutex); return tmp; + } } - if (i >= AMDGPU_MAX_XGMI_HIVE) + if (i >= AMDGPU_MAX_XGMI_HIVE) { + mutex_unlock(&xgmi_mutex); return NULL; + } /* initialize new hive if not exist */ tmp = &xgmi_hives[hive_count++]; tmp->hive_id = adev->gmc.xgmi.hive_id; INIT_LIST_HEAD(&tmp->device_list); + mutex_init(&tmp->hive_lock); + mutex_init(&tmp->reset_lock); + if (lock) + mutex_lock(&tmp->hive_lock); + + mutex_unlock(&xgmi_mutex); + return tmp; } +int amdgpu_xgmi_update_topology(struct amdgpu_hive_info *hive, struct amdgpu_device *adev) +{ + int ret = -EINVAL; + + /* Each psp need to set the latest topology */ + ret = psp_xgmi_set_topology_info(&adev->psp, + hive->number_devices, + &hive->topology_info); + if (ret) + dev_err(adev->dev, + "XGMI: Set topology failure on device %llx, hive %llx, ret %d", + adev->gmc.xgmi.node_id, + adev->gmc.xgmi.hive_id, ret); + + return ret; +} + int amdgpu_xgmi_add_device(struct amdgpu_device *adev) { - struct psp_xgmi_topology_info tmp_topology[AMDGPU_MAX_XGMI_DEVICE_PER_HIVE]; + struct psp_xgmi_topology_info *hive_topology; struct amdgpu_hive_info *hive; struct amdgpu_xgmi *entry; - struct amdgpu_device *tmp_adev; + struct amdgpu_device *tmp_adev = NULL; int count = 0, ret = -EINVAL; - if ((adev->asic_type < CHIP_VEGA20) || - (adev->flags & AMD_IS_APU) ) + if (!adev->gmc.xgmi.supported) return 0; - adev->gmc.xgmi.device_id = psp_xgmi_get_device_id(&adev->psp); - adev->gmc.xgmi.hive_id = psp_xgmi_get_hive_id(&adev->psp); - - memset(&tmp_topology[0], 0, sizeof(tmp_topology)); - mutex_lock(&xgmi_mutex); - hive = amdgpu_get_xgmi_hive(adev); - if (!hive) - goto exit; - list_add_tail(&adev->gmc.xgmi.head, &hive->device_list); - list_for_each_entry(entry, &hive->device_list, head) - tmp_topology[count++].device_id = entry->device_id; + ret = psp_xgmi_get_node_id(&adev->psp, &adev->gmc.xgmi.node_id); + if (ret) { + dev_err(adev->dev, + "XGMI: Failed to get node id\n"); + return ret; + } - ret = psp_xgmi_get_topology_info(&adev->psp, count, tmp_topology); + ret = psp_xgmi_get_hive_id(&adev->psp, &adev->gmc.xgmi.hive_id); if (ret) { dev_err(adev->dev, - "XGMI: Get topology failure on device %llx, hive %llx, ret %d", - adev->gmc.xgmi.device_id, - adev->gmc.xgmi.hive_id, ret); + "XGMI: Failed to get hive id\n"); + return ret; + } + + hive = amdgpu_get_xgmi_hive(adev, 1); + if (!hive) { + ret = -EINVAL; + dev_err(adev->dev, + "XGMI: node 0x%llx, can not match hive 0x%llx in the hive list.\n", + adev->gmc.xgmi.node_id, adev->gmc.xgmi.hive_id); goto exit; } - /* Each psp need to set the latest topology */ + + hive_topology = &hive->topology_info; + + list_add_tail(&adev->gmc.xgmi.head, &hive->device_list); + list_for_each_entry(entry, &hive->device_list, head) + hive_topology->nodes[count++].node_id = entry->node_id; + hive->number_devices = count; + + /* Each psp need to get the latest topology */ list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) { - ret = psp_xgmi_set_topology_info(&tmp_adev->psp, count, tmp_topology); + ret = psp_xgmi_get_topology_info(&tmp_adev->psp, count, hive_topology); if (ret) { dev_err(tmp_adev->dev, - "XGMI: Set topology failure on device %llx, hive %llx, ret %d", - tmp_adev->gmc.xgmi.device_id, + "XGMI: Get topology failure on device %llx, hive %llx, ret %d", + tmp_adev->gmc.xgmi.node_id, tmp_adev->gmc.xgmi.hive_id, ret); - /* To do : continue with some node failed or disable the whole hive */ + /* To do : continue with some node failed or disable the whole hive */ break; } } - if (!ret) - dev_info(adev->dev, "XGMI: Add node %d to hive 0x%llx.\n", - adev->gmc.xgmi.physical_node_id, - adev->gmc.xgmi.hive_id); + list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) { + ret = amdgpu_xgmi_update_topology(hive, tmp_adev); + if (ret) + break; + } + + dev_info(adev->dev, "XGMI: Add node %d, hive 0x%llx.\n", + adev->gmc.xgmi.physical_node_id, adev->gmc.xgmi.hive_id); + + mutex_unlock(&hive->hive_lock); exit: - mutex_unlock(&xgmi_mutex); return ret; } +void amdgpu_xgmi_remove_device(struct amdgpu_device *adev) +{ + struct amdgpu_hive_info *hive; + + if (!adev->gmc.xgmi.supported) + return; + + hive = amdgpu_get_xgmi_hive(adev, 1); + if (!hive) + return; + if (!(hive->number_devices--)) { + mutex_destroy(&hive->hive_lock); + mutex_destroy(&hive->reset_lock); + } else { + mutex_unlock(&hive->hive_lock); + } +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.h new file mode 100644 index 000000000000..14bc60664159 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.h @@ -0,0 +1,41 @@ +/* + * Copyright 2016 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + */ +#ifndef __AMDGPU_XGMI_H__ +#define __AMDGPU_XGMI_H__ + +#include "amdgpu_psp.h" + +struct amdgpu_hive_info { + uint64_t hive_id; + struct list_head device_list; + struct psp_xgmi_topology_info topology_info; + int number_devices; + struct mutex hive_lock, + reset_lock; +}; + +struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev, int lock); +int amdgpu_xgmi_update_topology(struct amdgpu_hive_info *hive, struct amdgpu_device *adev); +int amdgpu_xgmi_add_device(struct amdgpu_device *adev); +void amdgpu_xgmi_remove_device(struct amdgpu_device *adev); + +#endif diff --git a/drivers/gpu/drm/amd/amdgpu/atom.c b/drivers/gpu/drm/amd/amdgpu/atom.c index e9934de1b9cf..dd30f4e61a8c 100644 --- a/drivers/gpu/drm/amd/amdgpu/atom.c +++ b/drivers/gpu/drm/amd/amdgpu/atom.c @@ -27,6 +27,8 @@ #include <linux/slab.h> #include <asm/unaligned.h> +#include <drm/drm_util.h> + #define ATOM_DEBUG #include "atom.h" diff --git a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c deleted file mode 100644 index 79220a91abe3..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c +++ /dev/null @@ -1,6844 +0,0 @@ -/* - * Copyright 2013 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#include <linux/firmware.h> -#include <drm/drmP.h> -#include "amdgpu.h" -#include "amdgpu_pm.h" -#include "amdgpu_ucode.h" -#include "cikd.h" -#include "amdgpu_dpm.h" -#include "ci_dpm.h" -#include "gfx_v7_0.h" -#include "atom.h" -#include "amd_pcie.h" -#include <linux/seq_file.h> - -#include "smu/smu_7_0_1_d.h" -#include "smu/smu_7_0_1_sh_mask.h" - -#include "dce/dce_8_0_d.h" -#include "dce/dce_8_0_sh_mask.h" - -#include "bif/bif_4_1_d.h" -#include "bif/bif_4_1_sh_mask.h" - -#include "gca/gfx_7_2_d.h" -#include "gca/gfx_7_2_sh_mask.h" - -#include "gmc/gmc_7_1_d.h" -#include "gmc/gmc_7_1_sh_mask.h" - -MODULE_FIRMWARE("amdgpu/bonaire_smc.bin"); -MODULE_FIRMWARE("amdgpu/bonaire_k_smc.bin"); -MODULE_FIRMWARE("amdgpu/hawaii_smc.bin"); -MODULE_FIRMWARE("amdgpu/hawaii_k_smc.bin"); - -#define MC_CG_ARB_FREQ_F0 0x0a -#define MC_CG_ARB_FREQ_F1 0x0b -#define MC_CG_ARB_FREQ_F2 0x0c -#define MC_CG_ARB_FREQ_F3 0x0d - -#define SMC_RAM_END 0x40000 - -#define VOLTAGE_SCALE 4 -#define VOLTAGE_VID_OFFSET_SCALE1 625 -#define VOLTAGE_VID_OFFSET_SCALE2 100 - -static const struct amd_pm_funcs ci_dpm_funcs; - -static const struct ci_pt_defaults defaults_hawaii_xt = -{ - 1, 0xF, 0xFD, 0x19, 5, 0x14, 0, 0xB0000, - { 0x2E, 0x00, 0x00, 0x88, 0x00, 0x00, 0x72, 0x60, 0x51, 0xA7, 0x79, 0x6B, 0x90, 0xBD, 0x79 }, - { 0x217, 0x217, 0x217, 0x242, 0x242, 0x242, 0x269, 0x269, 0x269, 0x2A1, 0x2A1, 0x2A1, 0x2C9, 0x2C9, 0x2C9 } -}; - -static const struct ci_pt_defaults defaults_hawaii_pro = -{ - 1, 0xF, 0xFD, 0x19, 5, 0x14, 0, 0x65062, - { 0x2E, 0x00, 0x00, 0x88, 0x00, 0x00, 0x72, 0x60, 0x51, 0xA7, 0x79, 0x6B, 0x90, 0xBD, 0x79 }, - { 0x217, 0x217, 0x217, 0x242, 0x242, 0x242, 0x269, 0x269, 0x269, 0x2A1, 0x2A1, 0x2A1, 0x2C9, 0x2C9, 0x2C9 } -}; - -static const struct ci_pt_defaults defaults_bonaire_xt = -{ - 1, 0xF, 0xFD, 0x19, 5, 45, 0, 0xB0000, - { 0x79, 0x253, 0x25D, 0xAE, 0x72, 0x80, 0x83, 0x86, 0x6F, 0xC8, 0xC9, 0xC9, 0x2F, 0x4D, 0x61 }, - { 0x17C, 0x172, 0x180, 0x1BC, 0x1B3, 0x1BD, 0x206, 0x200, 0x203, 0x25D, 0x25A, 0x255, 0x2C3, 0x2C5, 0x2B4 } -}; - -#if 0 -static const struct ci_pt_defaults defaults_bonaire_pro = -{ - 1, 0xF, 0xFD, 0x19, 5, 45, 0, 0x65062, - { 0x8C, 0x23F, 0x244, 0xA6, 0x83, 0x85, 0x86, 0x86, 0x83, 0xDB, 0xDB, 0xDA, 0x67, 0x60, 0x5F }, - { 0x187, 0x193, 0x193, 0x1C7, 0x1D1, 0x1D1, 0x210, 0x219, 0x219, 0x266, 0x26C, 0x26C, 0x2C9, 0x2CB, 0x2CB } -}; -#endif - -static const struct ci_pt_defaults defaults_saturn_xt = -{ - 1, 0xF, 0xFD, 0x19, 5, 55, 0, 0x70000, - { 0x8C, 0x247, 0x249, 0xA6, 0x80, 0x81, 0x8B, 0x89, 0x86, 0xC9, 0xCA, 0xC9, 0x4D, 0x4D, 0x4D }, - { 0x187, 0x187, 0x187, 0x1C7, 0x1C7, 0x1C7, 0x210, 0x210, 0x210, 0x266, 0x266, 0x266, 0x2C9, 0x2C9, 0x2C9 } -}; - -#if 0 -static const struct ci_pt_defaults defaults_saturn_pro = -{ - 1, 0xF, 0xFD, 0x19, 5, 55, 0, 0x30000, - { 0x96, 0x21D, 0x23B, 0xA1, 0x85, 0x87, 0x83, 0x84, 0x81, 0xE6, 0xE6, 0xE6, 0x71, 0x6A, 0x6A }, - { 0x193, 0x19E, 0x19E, 0x1D2, 0x1DC, 0x1DC, 0x21A, 0x223, 0x223, 0x26E, 0x27E, 0x274, 0x2CF, 0x2D2, 0x2D2 } -}; -#endif - -static const struct ci_pt_config_reg didt_config_ci[] = -{ - { 0x10, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x10, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x10, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x10, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x11, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x11, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x11, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x11, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x12, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x12, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x12, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x12, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x2, 0x00003fff, 0, 0x4, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x2, 0x03ff0000, 16, 0x80, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x2, 0x78000000, 27, 0x3, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x1, 0x0000ffff, 0, 0x3FFF, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x1, 0xffff0000, 16, 0x3FFF, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x0, 0x00000001, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x30, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x30, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x30, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x30, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x31, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x31, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x31, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x31, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x32, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x32, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x32, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x32, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x22, 0x00003fff, 0, 0x4, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x22, 0x03ff0000, 16, 0x80, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x22, 0x78000000, 27, 0x3, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x21, 0x0000ffff, 0, 0x3FFF, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x21, 0xffff0000, 16, 0x3FFF, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x20, 0x00000001, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x50, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x50, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x50, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x50, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x51, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x51, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x51, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x51, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x52, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x52, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x52, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x52, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x42, 0x00003fff, 0, 0x4, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x42, 0x03ff0000, 16, 0x80, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x42, 0x78000000, 27, 0x3, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x41, 0x0000ffff, 0, 0x3FFF, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x41, 0xffff0000, 16, 0x3FFF, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x40, 0x00000001, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x70, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x70, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x70, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x70, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x71, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x71, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x71, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x71, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x72, 0x000000ff, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x72, 0x0000ff00, 8, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x72, 0x00ff0000, 16, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x72, 0xff000000, 24, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x62, 0x00003fff, 0, 0x4, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x62, 0x03ff0000, 16, 0x80, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x62, 0x78000000, 27, 0x3, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x61, 0x0000ffff, 0, 0x3FFF, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x61, 0xffff0000, 16, 0x3FFF, CISLANDS_CONFIGREG_DIDT_IND }, - { 0x60, 0x00000001, 0, 0x0, CISLANDS_CONFIGREG_DIDT_IND }, - { 0xFFFFFFFF } -}; - -static u8 ci_get_memory_module_index(struct amdgpu_device *adev) -{ - return (u8) ((RREG32(mmBIOS_SCRATCH_4) >> 16) & 0xff); -} - -#define MC_CG_ARB_FREQ_F0 0x0a -#define MC_CG_ARB_FREQ_F1 0x0b -#define MC_CG_ARB_FREQ_F2 0x0c -#define MC_CG_ARB_FREQ_F3 0x0d - -static int ci_copy_and_switch_arb_sets(struct amdgpu_device *adev, - u32 arb_freq_src, u32 arb_freq_dest) -{ - u32 mc_arb_dram_timing; - u32 mc_arb_dram_timing2; - u32 burst_time; - u32 mc_cg_config; - - switch (arb_freq_src) { - case MC_CG_ARB_FREQ_F0: - mc_arb_dram_timing = RREG32(mmMC_ARB_DRAM_TIMING); - mc_arb_dram_timing2 = RREG32(mmMC_ARB_DRAM_TIMING2); - burst_time = (RREG32(mmMC_ARB_BURST_TIME) & MC_ARB_BURST_TIME__STATE0_MASK) >> - MC_ARB_BURST_TIME__STATE0__SHIFT; - break; - case MC_CG_ARB_FREQ_F1: - mc_arb_dram_timing = RREG32(mmMC_ARB_DRAM_TIMING_1); - mc_arb_dram_timing2 = RREG32(mmMC_ARB_DRAM_TIMING2_1); - burst_time = (RREG32(mmMC_ARB_BURST_TIME) & MC_ARB_BURST_TIME__STATE1_MASK) >> - MC_ARB_BURST_TIME__STATE1__SHIFT; - break; - default: - return -EINVAL; - } - - switch (arb_freq_dest) { - case MC_CG_ARB_FREQ_F0: - WREG32(mmMC_ARB_DRAM_TIMING, mc_arb_dram_timing); - WREG32(mmMC_ARB_DRAM_TIMING2, mc_arb_dram_timing2); - WREG32_P(mmMC_ARB_BURST_TIME, (burst_time << MC_ARB_BURST_TIME__STATE0__SHIFT), - ~MC_ARB_BURST_TIME__STATE0_MASK); - break; - case MC_CG_ARB_FREQ_F1: - WREG32(mmMC_ARB_DRAM_TIMING_1, mc_arb_dram_timing); - WREG32(mmMC_ARB_DRAM_TIMING2_1, mc_arb_dram_timing2); - WREG32_P(mmMC_ARB_BURST_TIME, (burst_time << MC_ARB_BURST_TIME__STATE1__SHIFT), - ~MC_ARB_BURST_TIME__STATE1_MASK); - break; - default: - return -EINVAL; - } - - mc_cg_config = RREG32(mmMC_CG_CONFIG) | 0x0000000F; - WREG32(mmMC_CG_CONFIG, mc_cg_config); - WREG32_P(mmMC_ARB_CG, (arb_freq_dest) << MC_ARB_CG__CG_ARB_REQ__SHIFT, - ~MC_ARB_CG__CG_ARB_REQ_MASK); - - return 0; -} - -static u8 ci_get_ddr3_mclk_frequency_ratio(u32 memory_clock) -{ - u8 mc_para_index; - - if (memory_clock < 10000) - mc_para_index = 0; - else if (memory_clock >= 80000) - mc_para_index = 0x0f; - else - mc_para_index = (u8)((memory_clock - 10000) / 5000 + 1); - return mc_para_index; -} - -static u8 ci_get_mclk_frequency_ratio(u32 memory_clock, bool strobe_mode) -{ - u8 mc_para_index; - - if (strobe_mode) { - if (memory_clock < 12500) - mc_para_index = 0x00; - else if (memory_clock > 47500) - mc_para_index = 0x0f; - else - mc_para_index = (u8)((memory_clock - 10000) / 2500); - } else { - if (memory_clock < 65000) - mc_para_index = 0x00; - else if (memory_clock > 135000) - mc_para_index = 0x0f; - else - mc_para_index = (u8)((memory_clock - 60000) / 5000); - } - return mc_para_index; -} - -static void ci_trim_voltage_table_to_fit_state_table(struct amdgpu_device *adev, - u32 max_voltage_steps, - struct atom_voltage_table *voltage_table) -{ - unsigned int i, diff; - - if (voltage_table->count <= max_voltage_steps) - return; - - diff = voltage_table->count - max_voltage_steps; - - for (i = 0; i < max_voltage_steps; i++) - voltage_table->entries[i] = voltage_table->entries[i + diff]; - - voltage_table->count = max_voltage_steps; -} - -static int ci_get_std_voltage_value_sidd(struct amdgpu_device *adev, - struct atom_voltage_table_entry *voltage_table, - u16 *std_voltage_hi_sidd, u16 *std_voltage_lo_sidd); -static int ci_set_power_limit(struct amdgpu_device *adev, u32 n); -static int ci_set_overdrive_target_tdp(struct amdgpu_device *adev, - u32 target_tdp); -static int ci_update_uvd_dpm(struct amdgpu_device *adev, bool gate); -static void ci_dpm_set_irq_funcs(struct amdgpu_device *adev); - -static PPSMC_Result amdgpu_ci_send_msg_to_smc_with_parameter(struct amdgpu_device *adev, - PPSMC_Msg msg, u32 parameter); -static void ci_thermal_start_smc_fan_control(struct amdgpu_device *adev); -static void ci_fan_ctrl_set_default_mode(struct amdgpu_device *adev); - -static struct ci_power_info *ci_get_pi(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = adev->pm.dpm.priv; - - return pi; -} - -static struct ci_ps *ci_get_ps(struct amdgpu_ps *rps) -{ - struct ci_ps *ps = rps->ps_priv; - - return ps; -} - -static void ci_initialize_powertune_defaults(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - switch (adev->pdev->device) { - case 0x6649: - case 0x6650: - case 0x6651: - case 0x6658: - case 0x665C: - case 0x665D: - default: - pi->powertune_defaults = &defaults_bonaire_xt; - break; - case 0x6640: - case 0x6641: - case 0x6646: - case 0x6647: - pi->powertune_defaults = &defaults_saturn_xt; - break; - case 0x67B8: - case 0x67B0: - pi->powertune_defaults = &defaults_hawaii_xt; - break; - case 0x67BA: - case 0x67B1: - pi->powertune_defaults = &defaults_hawaii_pro; - break; - case 0x67A0: - case 0x67A1: - case 0x67A2: - case 0x67A8: - case 0x67A9: - case 0x67AA: - case 0x67B9: - case 0x67BE: - pi->powertune_defaults = &defaults_bonaire_xt; - break; - } - - pi->dte_tj_offset = 0; - - pi->caps_power_containment = true; - pi->caps_cac = false; - pi->caps_sq_ramping = false; - pi->caps_db_ramping = false; - pi->caps_td_ramping = false; - pi->caps_tcp_ramping = false; - - if (pi->caps_power_containment) { - pi->caps_cac = true; - if (adev->asic_type == CHIP_HAWAII) - pi->enable_bapm_feature = false; - else - pi->enable_bapm_feature = true; - pi->enable_tdc_limit_feature = true; - pi->enable_pkg_pwr_tracking_feature = true; - } -} - -static u8 ci_convert_to_vid(u16 vddc) -{ - return (6200 - (vddc * VOLTAGE_SCALE)) / 25; -} - -static int ci_populate_bapm_vddc_vid_sidd(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u8 *hi_vid = pi->smc_powertune_table.BapmVddCVidHiSidd; - u8 *lo_vid = pi->smc_powertune_table.BapmVddCVidLoSidd; - u8 *hi2_vid = pi->smc_powertune_table.BapmVddCVidHiSidd2; - u32 i; - - if (adev->pm.dpm.dyn_state.cac_leakage_table.entries == NULL) - return -EINVAL; - if (adev->pm.dpm.dyn_state.cac_leakage_table.count > 8) - return -EINVAL; - if (adev->pm.dpm.dyn_state.cac_leakage_table.count != - adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.count) - return -EINVAL; - - for (i = 0; i < adev->pm.dpm.dyn_state.cac_leakage_table.count; i++) { - if (adev->pm.dpm.platform_caps & ATOM_PP_PLATFORM_CAP_EVV) { - lo_vid[i] = ci_convert_to_vid(adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].vddc1); - hi_vid[i] = ci_convert_to_vid(adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].vddc2); - hi2_vid[i] = ci_convert_to_vid(adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].vddc3); - } else { - lo_vid[i] = ci_convert_to_vid(adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].vddc); - hi_vid[i] = ci_convert_to_vid((u16)adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].leakage); - } - } - return 0; -} - -static int ci_populate_vddc_vid(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u8 *vid = pi->smc_powertune_table.VddCVid; - u32 i; - - if (pi->vddc_voltage_table.count > 8) - return -EINVAL; - - for (i = 0; i < pi->vddc_voltage_table.count; i++) - vid[i] = ci_convert_to_vid(pi->vddc_voltage_table.entries[i].value); - - return 0; -} - -static int ci_populate_svi_load_line(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - const struct ci_pt_defaults *pt_defaults = pi->powertune_defaults; - - pi->smc_powertune_table.SviLoadLineEn = pt_defaults->svi_load_line_en; - pi->smc_powertune_table.SviLoadLineVddC = pt_defaults->svi_load_line_vddc; - pi->smc_powertune_table.SviLoadLineTrimVddC = 3; - pi->smc_powertune_table.SviLoadLineOffsetVddC = 0; - - return 0; -} - -static int ci_populate_tdc_limit(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - const struct ci_pt_defaults *pt_defaults = pi->powertune_defaults; - u16 tdc_limit; - - tdc_limit = adev->pm.dpm.dyn_state.cac_tdp_table->tdc * 256; - pi->smc_powertune_table.TDC_VDDC_PkgLimit = cpu_to_be16(tdc_limit); - pi->smc_powertune_table.TDC_VDDC_ThrottleReleaseLimitPerc = - pt_defaults->tdc_vddc_throttle_release_limit_perc; - pi->smc_powertune_table.TDC_MAWt = pt_defaults->tdc_mawt; - - return 0; -} - -static int ci_populate_dw8(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - const struct ci_pt_defaults *pt_defaults = pi->powertune_defaults; - int ret; - - ret = amdgpu_ci_read_smc_sram_dword(adev, - SMU7_FIRMWARE_HEADER_LOCATION + - offsetof(SMU7_Firmware_Header, PmFuseTable) + - offsetof(SMU7_Discrete_PmFuses, TdcWaterfallCtl), - (u32 *)&pi->smc_powertune_table.TdcWaterfallCtl, - pi->sram_end); - if (ret) - return -EINVAL; - else - pi->smc_powertune_table.TdcWaterfallCtl = pt_defaults->tdc_waterfall_ctl; - - return 0; -} - -static int ci_populate_fuzzy_fan(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if ((adev->pm.dpm.fan.fan_output_sensitivity & (1 << 15)) || - (adev->pm.dpm.fan.fan_output_sensitivity == 0)) - adev->pm.dpm.fan.fan_output_sensitivity = - adev->pm.dpm.fan.default_fan_output_sensitivity; - - pi->smc_powertune_table.FuzzyFan_PwmSetDelta = - cpu_to_be16(adev->pm.dpm.fan.fan_output_sensitivity); - - return 0; -} - -static int ci_min_max_v_gnbl_pm_lid_from_bapm_vddc(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u8 *hi_vid = pi->smc_powertune_table.BapmVddCVidHiSidd; - u8 *lo_vid = pi->smc_powertune_table.BapmVddCVidLoSidd; - int i, min, max; - - min = max = hi_vid[0]; - for (i = 0; i < 8; i++) { - if (0 != hi_vid[i]) { - if (min > hi_vid[i]) - min = hi_vid[i]; - if (max < hi_vid[i]) - max = hi_vid[i]; - } - - if (0 != lo_vid[i]) { - if (min > lo_vid[i]) - min = lo_vid[i]; - if (max < lo_vid[i]) - max = lo_vid[i]; - } - } - - if ((min == 0) || (max == 0)) - return -EINVAL; - pi->smc_powertune_table.GnbLPMLMaxVid = (u8)max; - pi->smc_powertune_table.GnbLPMLMinVid = (u8)min; - - return 0; -} - -static int ci_populate_bapm_vddc_base_leakage_sidd(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u16 hi_sidd = pi->smc_powertune_table.BapmVddCBaseLeakageHiSidd; - u16 lo_sidd = pi->smc_powertune_table.BapmVddCBaseLeakageLoSidd; - struct amdgpu_cac_tdp_table *cac_tdp_table = - adev->pm.dpm.dyn_state.cac_tdp_table; - - hi_sidd = cac_tdp_table->high_cac_leakage / 100 * 256; - lo_sidd = cac_tdp_table->low_cac_leakage / 100 * 256; - - pi->smc_powertune_table.BapmVddCBaseLeakageHiSidd = cpu_to_be16(hi_sidd); - pi->smc_powertune_table.BapmVddCBaseLeakageLoSidd = cpu_to_be16(lo_sidd); - - return 0; -} - -static int ci_populate_bapm_parameters_in_dpm_table(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - const struct ci_pt_defaults *pt_defaults = pi->powertune_defaults; - SMU7_Discrete_DpmTable *dpm_table = &pi->smc_state_table; - struct amdgpu_cac_tdp_table *cac_tdp_table = - adev->pm.dpm.dyn_state.cac_tdp_table; - struct amdgpu_ppm_table *ppm = adev->pm.dpm.dyn_state.ppm_table; - int i, j, k; - const u16 *def1; - const u16 *def2; - - dpm_table->DefaultTdp = cac_tdp_table->tdp * 256; - dpm_table->TargetTdp = cac_tdp_table->configurable_tdp * 256; - - dpm_table->DTETjOffset = (u8)pi->dte_tj_offset; - dpm_table->GpuTjMax = - (u8)(pi->thermal_temp_setting.temperature_high / 1000); - dpm_table->GpuTjHyst = 8; - - dpm_table->DTEAmbientTempBase = pt_defaults->dte_ambient_temp_base; - - if (ppm) { - dpm_table->PPM_PkgPwrLimit = cpu_to_be16((u16)ppm->dgpu_tdp * 256 / 1000); - dpm_table->PPM_TemperatureLimit = cpu_to_be16((u16)ppm->tj_max * 256); - } else { - dpm_table->PPM_PkgPwrLimit = cpu_to_be16(0); - dpm_table->PPM_TemperatureLimit = cpu_to_be16(0); - } - - dpm_table->BAPM_TEMP_GRADIENT = cpu_to_be32(pt_defaults->bapm_temp_gradient); - def1 = pt_defaults->bapmti_r; - def2 = pt_defaults->bapmti_rc; - - for (i = 0; i < SMU7_DTE_ITERATIONS; i++) { - for (j = 0; j < SMU7_DTE_SOURCES; j++) { - for (k = 0; k < SMU7_DTE_SINKS; k++) { - dpm_table->BAPMTI_R[i][j][k] = cpu_to_be16(*def1); - dpm_table->BAPMTI_RC[i][j][k] = cpu_to_be16(*def2); - def1++; - def2++; - } - } - } - - return 0; -} - -static int ci_populate_pm_base(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 pm_fuse_table_offset; - int ret; - - if (pi->caps_power_containment) { - ret = amdgpu_ci_read_smc_sram_dword(adev, - SMU7_FIRMWARE_HEADER_LOCATION + - offsetof(SMU7_Firmware_Header, PmFuseTable), - &pm_fuse_table_offset, pi->sram_end); - if (ret) - return ret; - ret = ci_populate_bapm_vddc_vid_sidd(adev); - if (ret) - return ret; - ret = ci_populate_vddc_vid(adev); - if (ret) - return ret; - ret = ci_populate_svi_load_line(adev); - if (ret) - return ret; - ret = ci_populate_tdc_limit(adev); - if (ret) - return ret; - ret = ci_populate_dw8(adev); - if (ret) - return ret; - ret = ci_populate_fuzzy_fan(adev); - if (ret) - return ret; - ret = ci_min_max_v_gnbl_pm_lid_from_bapm_vddc(adev); - if (ret) - return ret; - ret = ci_populate_bapm_vddc_base_leakage_sidd(adev); - if (ret) - return ret; - ret = amdgpu_ci_copy_bytes_to_smc(adev, pm_fuse_table_offset, - (u8 *)&pi->smc_powertune_table, - sizeof(SMU7_Discrete_PmFuses), pi->sram_end); - if (ret) - return ret; - } - - return 0; -} - -static void ci_do_enable_didt(struct amdgpu_device *adev, const bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 data; - - if (pi->caps_sq_ramping) { - data = RREG32_DIDT(ixDIDT_SQ_CTRL0); - if (enable) - data |= DIDT_SQ_CTRL0__DIDT_CTRL_EN_MASK; - else - data &= ~DIDT_SQ_CTRL0__DIDT_CTRL_EN_MASK; - WREG32_DIDT(ixDIDT_SQ_CTRL0, data); - } - - if (pi->caps_db_ramping) { - data = RREG32_DIDT(ixDIDT_DB_CTRL0); - if (enable) - data |= DIDT_DB_CTRL0__DIDT_CTRL_EN_MASK; - else - data &= ~DIDT_DB_CTRL0__DIDT_CTRL_EN_MASK; - WREG32_DIDT(ixDIDT_DB_CTRL0, data); - } - - if (pi->caps_td_ramping) { - data = RREG32_DIDT(ixDIDT_TD_CTRL0); - if (enable) - data |= DIDT_TD_CTRL0__DIDT_CTRL_EN_MASK; - else - data &= ~DIDT_TD_CTRL0__DIDT_CTRL_EN_MASK; - WREG32_DIDT(ixDIDT_TD_CTRL0, data); - } - - if (pi->caps_tcp_ramping) { - data = RREG32_DIDT(ixDIDT_TCP_CTRL0); - if (enable) - data |= DIDT_TCP_CTRL0__DIDT_CTRL_EN_MASK; - else - data &= ~DIDT_TCP_CTRL0__DIDT_CTRL_EN_MASK; - WREG32_DIDT(ixDIDT_TCP_CTRL0, data); - } -} - -static int ci_program_pt_config_registers(struct amdgpu_device *adev, - const struct ci_pt_config_reg *cac_config_regs) -{ - const struct ci_pt_config_reg *config_regs = cac_config_regs; - u32 data; - u32 cache = 0; - - if (config_regs == NULL) - return -EINVAL; - - while (config_regs->offset != 0xFFFFFFFF) { - if (config_regs->type == CISLANDS_CONFIGREG_CACHE) { - cache |= ((config_regs->value << config_regs->shift) & config_regs->mask); - } else { - switch (config_regs->type) { - case CISLANDS_CONFIGREG_SMC_IND: - data = RREG32_SMC(config_regs->offset); - break; - case CISLANDS_CONFIGREG_DIDT_IND: - data = RREG32_DIDT(config_regs->offset); - break; - default: - data = RREG32(config_regs->offset); - break; - } - - data &= ~config_regs->mask; - data |= ((config_regs->value << config_regs->shift) & config_regs->mask); - data |= cache; - - switch (config_regs->type) { - case CISLANDS_CONFIGREG_SMC_IND: - WREG32_SMC(config_regs->offset, data); - break; - case CISLANDS_CONFIGREG_DIDT_IND: - WREG32_DIDT(config_regs->offset, data); - break; - default: - WREG32(config_regs->offset, data); - break; - } - cache = 0; - } - config_regs++; - } - return 0; -} - -static int ci_enable_didt(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - int ret; - - if (pi->caps_sq_ramping || pi->caps_db_ramping || - pi->caps_td_ramping || pi->caps_tcp_ramping) { - adev->gfx.rlc.funcs->enter_safe_mode(adev); - - if (enable) { - ret = ci_program_pt_config_registers(adev, didt_config_ci); - if (ret) { - adev->gfx.rlc.funcs->exit_safe_mode(adev); - return ret; - } - } - - ci_do_enable_didt(adev, enable); - - adev->gfx.rlc.funcs->exit_safe_mode(adev); - } - - return 0; -} - -static int ci_enable_power_containment(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result smc_result; - int ret = 0; - - if (enable) { - pi->power_containment_features = 0; - if (pi->caps_power_containment) { - if (pi->enable_bapm_feature) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_EnableDTE); - if (smc_result != PPSMC_Result_OK) - ret = -EINVAL; - else - pi->power_containment_features |= POWERCONTAINMENT_FEATURE_BAPM; - } - - if (pi->enable_tdc_limit_feature) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_TDCLimitEnable); - if (smc_result != PPSMC_Result_OK) - ret = -EINVAL; - else - pi->power_containment_features |= POWERCONTAINMENT_FEATURE_TDCLimit; - } - - if (pi->enable_pkg_pwr_tracking_feature) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_PkgPwrLimitEnable); - if (smc_result != PPSMC_Result_OK) { - ret = -EINVAL; - } else { - struct amdgpu_cac_tdp_table *cac_tdp_table = - adev->pm.dpm.dyn_state.cac_tdp_table; - u32 default_pwr_limit = - (u32)(cac_tdp_table->maximum_power_delivery_limit * 256); - - pi->power_containment_features |= POWERCONTAINMENT_FEATURE_PkgPwrLimit; - - ci_set_power_limit(adev, default_pwr_limit); - } - } - } - } else { - if (pi->caps_power_containment && pi->power_containment_features) { - if (pi->power_containment_features & POWERCONTAINMENT_FEATURE_TDCLimit) - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_TDCLimitDisable); - - if (pi->power_containment_features & POWERCONTAINMENT_FEATURE_BAPM) - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_DisableDTE); - - if (pi->power_containment_features & POWERCONTAINMENT_FEATURE_PkgPwrLimit) - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_PkgPwrLimitDisable); - pi->power_containment_features = 0; - } - } - - return ret; -} - -static int ci_enable_smc_cac(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result smc_result; - int ret = 0; - - if (pi->caps_cac) { - if (enable) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_EnableCac); - if (smc_result != PPSMC_Result_OK) { - ret = -EINVAL; - pi->cac_enabled = false; - } else { - pi->cac_enabled = true; - } - } else if (pi->cac_enabled) { - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_DisableCac); - pi->cac_enabled = false; - } - } - - return ret; -} - -static int ci_enable_thermal_based_sclk_dpm(struct amdgpu_device *adev, - bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result smc_result = PPSMC_Result_OK; - - if (pi->thermal_sclk_dpm_enabled) { - if (enable) - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_ENABLE_THERMAL_DPM); - else - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_DISABLE_THERMAL_DPM); - } - - if (smc_result == PPSMC_Result_OK) - return 0; - else - return -EINVAL; -} - -static int ci_power_control_set_level(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_cac_tdp_table *cac_tdp_table = - adev->pm.dpm.dyn_state.cac_tdp_table; - s32 adjust_percent; - s32 target_tdp; - int ret = 0; - bool adjust_polarity = false; /* ??? */ - - if (pi->caps_power_containment) { - adjust_percent = adjust_polarity ? - adev->pm.dpm.tdp_adjustment : (-1 * adev->pm.dpm.tdp_adjustment); - target_tdp = ((100 + adjust_percent) * - (s32)cac_tdp_table->configurable_tdp) / 100; - - ret = ci_set_overdrive_target_tdp(adev, (u32)target_tdp); - } - - return ret; -} - -static void ci_dpm_powergate_uvd(void *handle, bool gate) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - - pi->uvd_power_gated = gate; - - if (gate) { - /* stop the UVD block */ - amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD, - AMD_PG_STATE_GATE); - ci_update_uvd_dpm(adev, gate); - } else { - amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD, - AMD_PG_STATE_UNGATE); - ci_update_uvd_dpm(adev, gate); - } -} - -static bool ci_dpm_vblank_too_short(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - u32 vblank_time = amdgpu_dpm_get_vblank_time(adev); - u32 switch_limit = adev->gmc.vram_type == AMDGPU_VRAM_TYPE_GDDR5 ? 450 : 300; - - /* disable mclk switching if the refresh is >120Hz, even if the - * blanking period would allow it - */ - if (amdgpu_dpm_get_vrefresh(adev) > 120) - return true; - - if (vblank_time < switch_limit) - return true; - else - return false; - -} - -static void ci_apply_state_adjust_rules(struct amdgpu_device *adev, - struct amdgpu_ps *rps) -{ - struct ci_ps *ps = ci_get_ps(rps); - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_clock_and_voltage_limits *max_limits; - bool disable_mclk_switching; - u32 sclk, mclk; - int i; - - if (rps->vce_active) { - rps->evclk = adev->pm.dpm.vce_states[adev->pm.dpm.vce_level].evclk; - rps->ecclk = adev->pm.dpm.vce_states[adev->pm.dpm.vce_level].ecclk; - } else { - rps->evclk = 0; - rps->ecclk = 0; - } - - if ((adev->pm.dpm.new_active_crtc_count > 1) || - ci_dpm_vblank_too_short(adev)) - disable_mclk_switching = true; - else - disable_mclk_switching = false; - - if ((rps->class & ATOM_PPLIB_CLASSIFICATION_UI_MASK) == ATOM_PPLIB_CLASSIFICATION_UI_BATTERY) - pi->battery_state = true; - else - pi->battery_state = false; - - if (adev->pm.ac_power) - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_ac; - else - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_dc; - - if (adev->pm.ac_power == false) { - for (i = 0; i < ps->performance_level_count; i++) { - if (ps->performance_levels[i].mclk > max_limits->mclk) - ps->performance_levels[i].mclk = max_limits->mclk; - if (ps->performance_levels[i].sclk > max_limits->sclk) - ps->performance_levels[i].sclk = max_limits->sclk; - } - } - - /* XXX validate the min clocks required for display */ - - if (disable_mclk_switching) { - mclk = ps->performance_levels[ps->performance_level_count - 1].mclk; - sclk = ps->performance_levels[0].sclk; - } else { - mclk = ps->performance_levels[0].mclk; - sclk = ps->performance_levels[0].sclk; - } - - if (adev->pm.pm_display_cfg.min_core_set_clock > sclk) - sclk = adev->pm.pm_display_cfg.min_core_set_clock; - - if (adev->pm.pm_display_cfg.min_mem_set_clock > mclk) - mclk = adev->pm.pm_display_cfg.min_mem_set_clock; - - if (rps->vce_active) { - if (sclk < adev->pm.dpm.vce_states[adev->pm.dpm.vce_level].sclk) - sclk = adev->pm.dpm.vce_states[adev->pm.dpm.vce_level].sclk; - if (mclk < adev->pm.dpm.vce_states[adev->pm.dpm.vce_level].mclk) - mclk = adev->pm.dpm.vce_states[adev->pm.dpm.vce_level].mclk; - } - - ps->performance_levels[0].sclk = sclk; - ps->performance_levels[0].mclk = mclk; - - if (ps->performance_levels[1].sclk < ps->performance_levels[0].sclk) - ps->performance_levels[1].sclk = ps->performance_levels[0].sclk; - - if (disable_mclk_switching) { - if (ps->performance_levels[0].mclk < ps->performance_levels[1].mclk) - ps->performance_levels[0].mclk = ps->performance_levels[1].mclk; - } else { - if (ps->performance_levels[1].mclk < ps->performance_levels[0].mclk) - ps->performance_levels[1].mclk = ps->performance_levels[0].mclk; - } -} - -static int ci_thermal_set_temperature_range(struct amdgpu_device *adev, - int min_temp, int max_temp) -{ - int low_temp = 0 * 1000; - int high_temp = 255 * 1000; - u32 tmp; - - if (low_temp < min_temp) - low_temp = min_temp; - if (high_temp > max_temp) - high_temp = max_temp; - if (high_temp < low_temp) { - DRM_ERROR("invalid thermal range: %d - %d\n", low_temp, high_temp); - return -EINVAL; - } - - tmp = RREG32_SMC(ixCG_THERMAL_INT); - tmp &= ~(CG_THERMAL_INT__DIG_THERM_INTH_MASK | CG_THERMAL_INT__DIG_THERM_INTL_MASK); - tmp |= ((high_temp / 1000) << CG_THERMAL_INT__DIG_THERM_INTH__SHIFT) | - ((low_temp / 1000)) << CG_THERMAL_INT__DIG_THERM_INTL__SHIFT; - WREG32_SMC(ixCG_THERMAL_INT, tmp); - -#if 0 - /* XXX: need to figure out how to handle this properly */ - tmp = RREG32_SMC(ixCG_THERMAL_CTRL); - tmp &= DIG_THERM_DPM_MASK; - tmp |= DIG_THERM_DPM(high_temp / 1000); - WREG32_SMC(ixCG_THERMAL_CTRL, tmp); -#endif - - adev->pm.dpm.thermal.min_temp = low_temp; - adev->pm.dpm.thermal.max_temp = high_temp; - return 0; -} - -static int ci_thermal_enable_alert(struct amdgpu_device *adev, - bool enable) -{ - u32 thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - PPSMC_Result result; - - if (enable) { - thermal_int &= ~(CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK | - CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK); - WREG32_SMC(ixCG_THERMAL_INT, thermal_int); - result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_Thermal_Cntl_Enable); - if (result != PPSMC_Result_OK) { - DRM_DEBUG_KMS("Could not enable thermal interrupts.\n"); - return -EINVAL; - } - } else { - thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK | - CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; - WREG32_SMC(ixCG_THERMAL_INT, thermal_int); - result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_Thermal_Cntl_Disable); - if (result != PPSMC_Result_OK) { - DRM_DEBUG_KMS("Could not disable thermal interrupts.\n"); - return -EINVAL; - } - } - - return 0; -} - -static void ci_fan_ctrl_set_static_mode(struct amdgpu_device *adev, u32 mode) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 tmp; - - if (pi->fan_ctrl_is_in_default_mode) { - tmp = (RREG32_SMC(ixCG_FDO_CTRL2) & CG_FDO_CTRL2__FDO_PWM_MODE_MASK) - >> CG_FDO_CTRL2__FDO_PWM_MODE__SHIFT; - pi->fan_ctrl_default_mode = tmp; - tmp = (RREG32_SMC(ixCG_FDO_CTRL2) & CG_FDO_CTRL2__TMIN_MASK) - >> CG_FDO_CTRL2__TMIN__SHIFT; - pi->t_min = tmp; - pi->fan_ctrl_is_in_default_mode = false; - } - - tmp = RREG32_SMC(ixCG_FDO_CTRL2) & ~CG_FDO_CTRL2__TMIN_MASK; - tmp |= 0 << CG_FDO_CTRL2__TMIN__SHIFT; - WREG32_SMC(ixCG_FDO_CTRL2, tmp); - - tmp = RREG32_SMC(ixCG_FDO_CTRL2) & ~CG_FDO_CTRL2__FDO_PWM_MODE_MASK; - tmp |= mode << CG_FDO_CTRL2__FDO_PWM_MODE__SHIFT; - WREG32_SMC(ixCG_FDO_CTRL2, tmp); -} - -static int ci_thermal_setup_fan_table(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - SMU7_Discrete_FanTable fan_table = { FDO_MODE_HARDWARE }; - u32 duty100; - u32 t_diff1, t_diff2, pwm_diff1, pwm_diff2; - u16 fdo_min, slope1, slope2; - u32 reference_clock, tmp; - int ret; - u64 tmp64; - - if (!pi->fan_table_start) { - adev->pm.dpm.fan.ucode_fan_control = false; - return 0; - } - - duty100 = (RREG32_SMC(ixCG_FDO_CTRL1) & CG_FDO_CTRL1__FMAX_DUTY100_MASK) - >> CG_FDO_CTRL1__FMAX_DUTY100__SHIFT; - - if (duty100 == 0) { - adev->pm.dpm.fan.ucode_fan_control = false; - return 0; - } - - tmp64 = (u64)adev->pm.dpm.fan.pwm_min * duty100; - do_div(tmp64, 10000); - fdo_min = (u16)tmp64; - - t_diff1 = adev->pm.dpm.fan.t_med - adev->pm.dpm.fan.t_min; - t_diff2 = adev->pm.dpm.fan.t_high - adev->pm.dpm.fan.t_med; - - pwm_diff1 = adev->pm.dpm.fan.pwm_med - adev->pm.dpm.fan.pwm_min; - pwm_diff2 = adev->pm.dpm.fan.pwm_high - adev->pm.dpm.fan.pwm_med; - - slope1 = (u16)((50 + ((16 * duty100 * pwm_diff1) / t_diff1)) / 100); - slope2 = (u16)((50 + ((16 * duty100 * pwm_diff2) / t_diff2)) / 100); - - fan_table.TempMin = cpu_to_be16((50 + adev->pm.dpm.fan.t_min) / 100); - fan_table.TempMed = cpu_to_be16((50 + adev->pm.dpm.fan.t_med) / 100); - fan_table.TempMax = cpu_to_be16((50 + adev->pm.dpm.fan.t_max) / 100); - - fan_table.Slope1 = cpu_to_be16(slope1); - fan_table.Slope2 = cpu_to_be16(slope2); - - fan_table.FdoMin = cpu_to_be16(fdo_min); - - fan_table.HystDown = cpu_to_be16(adev->pm.dpm.fan.t_hyst); - - fan_table.HystUp = cpu_to_be16(1); - - fan_table.HystSlope = cpu_to_be16(1); - - fan_table.TempRespLim = cpu_to_be16(5); - - reference_clock = amdgpu_asic_get_xclk(adev); - - fan_table.RefreshPeriod = cpu_to_be32((adev->pm.dpm.fan.cycle_delay * - reference_clock) / 1600); - - fan_table.FdoMax = cpu_to_be16((u16)duty100); - - tmp = (RREG32_SMC(ixCG_MULT_THERMAL_CTRL) & CG_MULT_THERMAL_CTRL__TEMP_SEL_MASK) - >> CG_MULT_THERMAL_CTRL__TEMP_SEL__SHIFT; - fan_table.TempSrc = (uint8_t)tmp; - - ret = amdgpu_ci_copy_bytes_to_smc(adev, - pi->fan_table_start, - (u8 *)(&fan_table), - sizeof(fan_table), - pi->sram_end); - - if (ret) { - DRM_ERROR("Failed to load fan table to the SMC."); - adev->pm.dpm.fan.ucode_fan_control = false; - } - - return 0; -} - -static int ci_fan_ctrl_start_smc_fan_control(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result ret; - - if (pi->caps_od_fuzzy_fan_control_support) { - ret = amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_StartFanControl, - FAN_CONTROL_FUZZY); - if (ret != PPSMC_Result_OK) - return -EINVAL; - ret = amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_SetFanPwmMax, - adev->pm.dpm.fan.default_max_fan_pwm); - if (ret != PPSMC_Result_OK) - return -EINVAL; - } else { - ret = amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_StartFanControl, - FAN_CONTROL_TABLE); - if (ret != PPSMC_Result_OK) - return -EINVAL; - } - - pi->fan_is_controlled_by_smc = true; - return 0; -} - - -static int ci_fan_ctrl_stop_smc_fan_control(struct amdgpu_device *adev) -{ - PPSMC_Result ret; - struct ci_power_info *pi = ci_get_pi(adev); - - ret = amdgpu_ci_send_msg_to_smc(adev, PPSMC_StopFanControl); - if (ret == PPSMC_Result_OK) { - pi->fan_is_controlled_by_smc = false; - return 0; - } else { - return -EINVAL; - } -} - -static int ci_dpm_get_fan_speed_percent(void *handle, - u32 *speed) -{ - u32 duty, duty100; - u64 tmp64; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - if (adev->pm.no_fan) - return -ENOENT; - - duty100 = (RREG32_SMC(ixCG_FDO_CTRL1) & CG_FDO_CTRL1__FMAX_DUTY100_MASK) - >> CG_FDO_CTRL1__FMAX_DUTY100__SHIFT; - duty = (RREG32_SMC(ixCG_THERMAL_STATUS) & CG_THERMAL_STATUS__FDO_PWM_DUTY_MASK) - >> CG_THERMAL_STATUS__FDO_PWM_DUTY__SHIFT; - - if (duty100 == 0) - return -EINVAL; - - tmp64 = (u64)duty * 100; - do_div(tmp64, duty100); - *speed = (u32)tmp64; - - if (*speed > 100) - *speed = 100; - - return 0; -} - -static int ci_dpm_set_fan_speed_percent(void *handle, - u32 speed) -{ - u32 tmp; - u32 duty, duty100; - u64 tmp64; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - - if (adev->pm.no_fan) - return -ENOENT; - - if (pi->fan_is_controlled_by_smc) - return -EINVAL; - - if (speed > 100) - return -EINVAL; - - duty100 = (RREG32_SMC(ixCG_FDO_CTRL1) & CG_FDO_CTRL1__FMAX_DUTY100_MASK) - >> CG_FDO_CTRL1__FMAX_DUTY100__SHIFT; - - if (duty100 == 0) - return -EINVAL; - - tmp64 = (u64)speed * duty100; - do_div(tmp64, 100); - duty = (u32)tmp64; - - tmp = RREG32_SMC(ixCG_FDO_CTRL0) & ~CG_FDO_CTRL0__FDO_STATIC_DUTY_MASK; - tmp |= duty << CG_FDO_CTRL0__FDO_STATIC_DUTY__SHIFT; - WREG32_SMC(ixCG_FDO_CTRL0, tmp); - - return 0; -} - -static void ci_dpm_set_fan_control_mode(void *handle, u32 mode) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - switch (mode) { - case AMD_FAN_CTRL_NONE: - if (adev->pm.dpm.fan.ucode_fan_control) - ci_fan_ctrl_stop_smc_fan_control(adev); - ci_dpm_set_fan_speed_percent(adev, 100); - break; - case AMD_FAN_CTRL_MANUAL: - if (adev->pm.dpm.fan.ucode_fan_control) - ci_fan_ctrl_stop_smc_fan_control(adev); - break; - case AMD_FAN_CTRL_AUTO: - if (adev->pm.dpm.fan.ucode_fan_control) - ci_thermal_start_smc_fan_control(adev); - break; - default: - break; - } -} - -static u32 ci_dpm_get_fan_control_mode(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - - if (pi->fan_is_controlled_by_smc) - return AMD_FAN_CTRL_AUTO; - else - return AMD_FAN_CTRL_MANUAL; -} - -#if 0 -static int ci_fan_ctrl_get_fan_speed_rpm(struct amdgpu_device *adev, - u32 *speed) -{ - u32 tach_period; - u32 xclk = amdgpu_asic_get_xclk(adev); - - if (adev->pm.no_fan) - return -ENOENT; - - if (adev->pm.fan_pulses_per_revolution == 0) - return -ENOENT; - - tach_period = (RREG32_SMC(ixCG_TACH_STATUS) & CG_TACH_STATUS__TACH_PERIOD_MASK) - >> CG_TACH_STATUS__TACH_PERIOD__SHIFT; - if (tach_period == 0) - return -ENOENT; - - *speed = 60 * xclk * 10000 / tach_period; - - return 0; -} - -static int ci_fan_ctrl_set_fan_speed_rpm(struct amdgpu_device *adev, - u32 speed) -{ - u32 tach_period, tmp; - u32 xclk = amdgpu_asic_get_xclk(adev); - - if (adev->pm.no_fan) - return -ENOENT; - - if (adev->pm.fan_pulses_per_revolution == 0) - return -ENOENT; - - if ((speed < adev->pm.fan_min_rpm) || - (speed > adev->pm.fan_max_rpm)) - return -EINVAL; - - if (adev->pm.dpm.fan.ucode_fan_control) - ci_fan_ctrl_stop_smc_fan_control(adev); - - tach_period = 60 * xclk * 10000 / (8 * speed); - tmp = RREG32_SMC(ixCG_TACH_CTRL) & ~CG_TACH_CTRL__TARGET_PERIOD_MASK; - tmp |= tach_period << CG_TACH_CTRL__TARGET_PERIOD__SHIFT; - WREG32_SMC(CG_TACH_CTRL, tmp); - - ci_fan_ctrl_set_static_mode(adev, FDO_PWM_MODE_STATIC_RPM); - - return 0; -} -#endif - -static void ci_fan_ctrl_set_default_mode(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 tmp; - - if (!pi->fan_ctrl_is_in_default_mode) { - tmp = RREG32_SMC(ixCG_FDO_CTRL2) & ~CG_FDO_CTRL2__FDO_PWM_MODE_MASK; - tmp |= pi->fan_ctrl_default_mode << CG_FDO_CTRL2__FDO_PWM_MODE__SHIFT; - WREG32_SMC(ixCG_FDO_CTRL2, tmp); - - tmp = RREG32_SMC(ixCG_FDO_CTRL2) & ~CG_FDO_CTRL2__TMIN_MASK; - tmp |= pi->t_min << CG_FDO_CTRL2__TMIN__SHIFT; - WREG32_SMC(ixCG_FDO_CTRL2, tmp); - pi->fan_ctrl_is_in_default_mode = true; - } -} - -static void ci_thermal_start_smc_fan_control(struct amdgpu_device *adev) -{ - if (adev->pm.dpm.fan.ucode_fan_control) { - ci_fan_ctrl_start_smc_fan_control(adev); - ci_fan_ctrl_set_static_mode(adev, FDO_PWM_MODE_STATIC); - } -} - -static void ci_thermal_initialize(struct amdgpu_device *adev) -{ - u32 tmp; - - if (adev->pm.fan_pulses_per_revolution) { - tmp = RREG32_SMC(ixCG_TACH_CTRL) & ~CG_TACH_CTRL__EDGE_PER_REV_MASK; - tmp |= (adev->pm.fan_pulses_per_revolution - 1) - << CG_TACH_CTRL__EDGE_PER_REV__SHIFT; - WREG32_SMC(ixCG_TACH_CTRL, tmp); - } - - tmp = RREG32_SMC(ixCG_FDO_CTRL2) & ~CG_FDO_CTRL2__TACH_PWM_RESP_RATE_MASK; - tmp |= 0x28 << CG_FDO_CTRL2__TACH_PWM_RESP_RATE__SHIFT; - WREG32_SMC(ixCG_FDO_CTRL2, tmp); -} - -static int ci_thermal_start_thermal_controller(struct amdgpu_device *adev) -{ - int ret; - - ci_thermal_initialize(adev); - ret = ci_thermal_set_temperature_range(adev, CISLANDS_TEMP_RANGE_MIN, CISLANDS_TEMP_RANGE_MAX); - if (ret) - return ret; - ret = ci_thermal_enable_alert(adev, true); - if (ret) - return ret; - if (adev->pm.dpm.fan.ucode_fan_control) { - ret = ci_thermal_setup_fan_table(adev); - if (ret) - return ret; - ci_thermal_start_smc_fan_control(adev); - } - - return 0; -} - -static void ci_thermal_stop_thermal_controller(struct amdgpu_device *adev) -{ - if (!adev->pm.no_fan) - ci_fan_ctrl_set_default_mode(adev); -} - -static int ci_read_smc_soft_register(struct amdgpu_device *adev, - u16 reg_offset, u32 *value) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - return amdgpu_ci_read_smc_sram_dword(adev, - pi->soft_regs_start + reg_offset, - value, pi->sram_end); -} - -static int ci_write_smc_soft_register(struct amdgpu_device *adev, - u16 reg_offset, u32 value) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - return amdgpu_ci_write_smc_sram_dword(adev, - pi->soft_regs_start + reg_offset, - value, pi->sram_end); -} - -static void ci_init_fps_limits(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - SMU7_Discrete_DpmTable *table = &pi->smc_state_table; - - if (pi->caps_fps) { - u16 tmp; - - tmp = 45; - table->FpsHighT = cpu_to_be16(tmp); - - tmp = 30; - table->FpsLowT = cpu_to_be16(tmp); - } -} - -static int ci_update_sclk_t(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - int ret = 0; - u32 low_sclk_interrupt_t = 0; - - if (pi->caps_sclk_throttle_low_notification) { - low_sclk_interrupt_t = cpu_to_be32(pi->low_sclk_interrupt_t); - - ret = amdgpu_ci_copy_bytes_to_smc(adev, - pi->dpm_table_start + - offsetof(SMU7_Discrete_DpmTable, LowSclkInterruptT), - (u8 *)&low_sclk_interrupt_t, - sizeof(u32), pi->sram_end); - - } - - return ret; -} - -static void ci_get_leakage_voltages(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u16 leakage_id, virtual_voltage_id; - u16 vddc, vddci; - int i; - - pi->vddc_leakage.count = 0; - pi->vddci_leakage.count = 0; - - if (adev->pm.dpm.platform_caps & ATOM_PP_PLATFORM_CAP_EVV) { - for (i = 0; i < CISLANDS_MAX_LEAKAGE_COUNT; i++) { - virtual_voltage_id = ATOM_VIRTUAL_VOLTAGE_ID0 + i; - if (amdgpu_atombios_get_voltage_evv(adev, virtual_voltage_id, &vddc) != 0) - continue; - if (vddc != 0 && vddc != virtual_voltage_id) { - pi->vddc_leakage.actual_voltage[pi->vddc_leakage.count] = vddc; - pi->vddc_leakage.leakage_id[pi->vddc_leakage.count] = virtual_voltage_id; - pi->vddc_leakage.count++; - } - } - } else if (amdgpu_atombios_get_leakage_id_from_vbios(adev, &leakage_id) == 0) { - for (i = 0; i < CISLANDS_MAX_LEAKAGE_COUNT; i++) { - virtual_voltage_id = ATOM_VIRTUAL_VOLTAGE_ID0 + i; - if (amdgpu_atombios_get_leakage_vddc_based_on_leakage_params(adev, &vddc, &vddci, - virtual_voltage_id, - leakage_id) == 0) { - if (vddc != 0 && vddc != virtual_voltage_id) { - pi->vddc_leakage.actual_voltage[pi->vddc_leakage.count] = vddc; - pi->vddc_leakage.leakage_id[pi->vddc_leakage.count] = virtual_voltage_id; - pi->vddc_leakage.count++; - } - if (vddci != 0 && vddci != virtual_voltage_id) { - pi->vddci_leakage.actual_voltage[pi->vddci_leakage.count] = vddci; - pi->vddci_leakage.leakage_id[pi->vddci_leakage.count] = virtual_voltage_id; - pi->vddci_leakage.count++; - } - } - } - } -} - -static void ci_set_dpm_event_sources(struct amdgpu_device *adev, u32 sources) -{ - struct ci_power_info *pi = ci_get_pi(adev); - bool want_thermal_protection; - enum amdgpu_dpm_event_src dpm_event_src; - u32 tmp; - - switch (sources) { - case 0: - default: - want_thermal_protection = false; - break; - case (1 << AMDGPU_DPM_AUTO_THROTTLE_SRC_THERMAL): - want_thermal_protection = true; - dpm_event_src = AMDGPU_DPM_EVENT_SRC_DIGITAL; - break; - case (1 << AMDGPU_DPM_AUTO_THROTTLE_SRC_EXTERNAL): - want_thermal_protection = true; - dpm_event_src = AMDGPU_DPM_EVENT_SRC_EXTERNAL; - break; - case ((1 << AMDGPU_DPM_AUTO_THROTTLE_SRC_EXTERNAL) | - (1 << AMDGPU_DPM_AUTO_THROTTLE_SRC_THERMAL)): - want_thermal_protection = true; - dpm_event_src = AMDGPU_DPM_EVENT_SRC_DIGIAL_OR_EXTERNAL; - break; - } - - if (want_thermal_protection) { -#if 0 - /* XXX: need to figure out how to handle this properly */ - tmp = RREG32_SMC(ixCG_THERMAL_CTRL); - tmp &= DPM_EVENT_SRC_MASK; - tmp |= DPM_EVENT_SRC(dpm_event_src); - WREG32_SMC(ixCG_THERMAL_CTRL, tmp); -#endif - - tmp = RREG32_SMC(ixGENERAL_PWRMGT); - if (pi->thermal_protection) - tmp &= ~GENERAL_PWRMGT__THERMAL_PROTECTION_DIS_MASK; - else - tmp |= GENERAL_PWRMGT__THERMAL_PROTECTION_DIS_MASK; - WREG32_SMC(ixGENERAL_PWRMGT, tmp); - } else { - tmp = RREG32_SMC(ixGENERAL_PWRMGT); - tmp |= GENERAL_PWRMGT__THERMAL_PROTECTION_DIS_MASK; - WREG32_SMC(ixGENERAL_PWRMGT, tmp); - } -} - -static void ci_enable_auto_throttle_source(struct amdgpu_device *adev, - enum amdgpu_dpm_auto_throttle_src source, - bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (enable) { - if (!(pi->active_auto_throttle_sources & (1 << source))) { - pi->active_auto_throttle_sources |= 1 << source; - ci_set_dpm_event_sources(adev, pi->active_auto_throttle_sources); - } - } else { - if (pi->active_auto_throttle_sources & (1 << source)) { - pi->active_auto_throttle_sources &= ~(1 << source); - ci_set_dpm_event_sources(adev, pi->active_auto_throttle_sources); - } - } -} - -static void ci_enable_vr_hot_gpio_interrupt(struct amdgpu_device *adev) -{ - if (adev->pm.dpm.platform_caps & ATOM_PP_PLATFORM_CAP_REGULATOR_HOT) - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_EnableVRHotGPIOInterrupt); -} - -static int ci_unfreeze_sclk_mclk_dpm(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result smc_result; - - if (!pi->need_update_smu7_dpm_table) - return 0; - - if ((!pi->sclk_dpm_key_disabled) && - (pi->need_update_smu7_dpm_table & (DPMTABLE_OD_UPDATE_SCLK | DPMTABLE_UPDATE_SCLK))) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_SCLKDPM_UnfreezeLevel); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - if ((!pi->mclk_dpm_key_disabled) && - (pi->need_update_smu7_dpm_table & DPMTABLE_OD_UPDATE_MCLK)) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_MCLKDPM_UnfreezeLevel); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - pi->need_update_smu7_dpm_table = 0; - return 0; -} - -static int ci_enable_sclk_mclk_dpm(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result smc_result; - - if (enable) { - if (!pi->sclk_dpm_key_disabled) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_DPM_Enable); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - if (!pi->mclk_dpm_key_disabled) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_MCLKDPM_Enable); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - - WREG32_P(mmMC_SEQ_CNTL_3, MC_SEQ_CNTL_3__CAC_EN_MASK, - ~MC_SEQ_CNTL_3__CAC_EN_MASK); - - WREG32_SMC(ixLCAC_MC0_CNTL, 0x05); - WREG32_SMC(ixLCAC_MC1_CNTL, 0x05); - WREG32_SMC(ixLCAC_CPL_CNTL, 0x100005); - - udelay(10); - - WREG32_SMC(ixLCAC_MC0_CNTL, 0x400005); - WREG32_SMC(ixLCAC_MC1_CNTL, 0x400005); - WREG32_SMC(ixLCAC_CPL_CNTL, 0x500005); - } - } else { - if (!pi->sclk_dpm_key_disabled) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_DPM_Disable); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - if (!pi->mclk_dpm_key_disabled) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_MCLKDPM_Disable); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - } - - return 0; -} - -static int ci_start_dpm(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result smc_result; - int ret; - u32 tmp; - - tmp = RREG32_SMC(ixGENERAL_PWRMGT); - tmp |= GENERAL_PWRMGT__GLOBAL_PWRMGT_EN_MASK; - WREG32_SMC(ixGENERAL_PWRMGT, tmp); - - tmp = RREG32_SMC(ixSCLK_PWRMGT_CNTL); - tmp |= SCLK_PWRMGT_CNTL__DYNAMIC_PM_EN_MASK; - WREG32_SMC(ixSCLK_PWRMGT_CNTL, tmp); - - ci_write_smc_soft_register(adev, offsetof(SMU7_SoftRegisters, VoltageChangeTimeout), 0x1000); - - WREG32_P(mmBIF_LNCNT_RESET, 0, ~BIF_LNCNT_RESET__RESET_LNCNT_EN_MASK); - - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_Voltage_Cntl_Enable); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - - ret = ci_enable_sclk_mclk_dpm(adev, true); - if (ret) - return ret; - - if (!pi->pcie_dpm_key_disabled) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_PCIeDPM_Enable); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - return 0; -} - -static int ci_freeze_sclk_mclk_dpm(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result smc_result; - - if (!pi->need_update_smu7_dpm_table) - return 0; - - if ((!pi->sclk_dpm_key_disabled) && - (pi->need_update_smu7_dpm_table & (DPMTABLE_OD_UPDATE_SCLK | DPMTABLE_UPDATE_SCLK))) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_SCLKDPM_FreezeLevel); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - if ((!pi->mclk_dpm_key_disabled) && - (pi->need_update_smu7_dpm_table & DPMTABLE_OD_UPDATE_MCLK)) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_MCLKDPM_FreezeLevel); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - return 0; -} - -static int ci_stop_dpm(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result smc_result; - int ret; - u32 tmp; - - tmp = RREG32_SMC(ixGENERAL_PWRMGT); - tmp &= ~GENERAL_PWRMGT__GLOBAL_PWRMGT_EN_MASK; - WREG32_SMC(ixGENERAL_PWRMGT, tmp); - - tmp = RREG32_SMC(ixSCLK_PWRMGT_CNTL); - tmp &= ~SCLK_PWRMGT_CNTL__DYNAMIC_PM_EN_MASK; - WREG32_SMC(ixSCLK_PWRMGT_CNTL, tmp); - - if (!pi->pcie_dpm_key_disabled) { - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_PCIeDPM_Disable); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - ret = ci_enable_sclk_mclk_dpm(adev, false); - if (ret) - return ret; - - smc_result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_Voltage_Cntl_Disable); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - - return 0; -} - -static void ci_enable_sclk_control(struct amdgpu_device *adev, bool enable) -{ - u32 tmp = RREG32_SMC(ixSCLK_PWRMGT_CNTL); - - if (enable) - tmp &= ~SCLK_PWRMGT_CNTL__SCLK_PWRMGT_OFF_MASK; - else - tmp |= SCLK_PWRMGT_CNTL__SCLK_PWRMGT_OFF_MASK; - WREG32_SMC(ixSCLK_PWRMGT_CNTL, tmp); -} - -#if 0 -static int ci_notify_hw_of_power_source(struct amdgpu_device *adev, - bool ac_power) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_cac_tdp_table *cac_tdp_table = - adev->pm.dpm.dyn_state.cac_tdp_table; - u32 power_limit; - - if (ac_power) - power_limit = (u32)(cac_tdp_table->maximum_power_delivery_limit * 256); - else - power_limit = (u32)(cac_tdp_table->battery_power_limit * 256); - - ci_set_power_limit(adev, power_limit); - - if (pi->caps_automatic_dc_transition) { - if (ac_power) - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_RunningOnAC); - else - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_Remove_DC_Clamp); - } - - return 0; -} -#endif - -static PPSMC_Result amdgpu_ci_send_msg_to_smc_with_parameter(struct amdgpu_device *adev, - PPSMC_Msg msg, u32 parameter) -{ - WREG32(mmSMC_MSG_ARG_0, parameter); - return amdgpu_ci_send_msg_to_smc(adev, msg); -} - -static PPSMC_Result amdgpu_ci_send_msg_to_smc_return_parameter(struct amdgpu_device *adev, - PPSMC_Msg msg, u32 *parameter) -{ - PPSMC_Result smc_result; - - smc_result = amdgpu_ci_send_msg_to_smc(adev, msg); - - if ((smc_result == PPSMC_Result_OK) && parameter) - *parameter = RREG32(mmSMC_MSG_ARG_0); - - return smc_result; -} - -static int ci_dpm_force_state_sclk(struct amdgpu_device *adev, u32 n) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (!pi->sclk_dpm_key_disabled) { - PPSMC_Result smc_result = - amdgpu_ci_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_SCLKDPM_SetEnabledMask, 1 << n); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - return 0; -} - -static int ci_dpm_force_state_mclk(struct amdgpu_device *adev, u32 n) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (!pi->mclk_dpm_key_disabled) { - PPSMC_Result smc_result = - amdgpu_ci_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_MCLKDPM_SetEnabledMask, 1 << n); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - return 0; -} - -static int ci_dpm_force_state_pcie(struct amdgpu_device *adev, u32 n) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (!pi->pcie_dpm_key_disabled) { - PPSMC_Result smc_result = - amdgpu_ci_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_PCIeDPM_ForceLevel, n); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - return 0; -} - -static int ci_set_power_limit(struct amdgpu_device *adev, u32 n) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (pi->power_containment_features & POWERCONTAINMENT_FEATURE_PkgPwrLimit) { - PPSMC_Result smc_result = - amdgpu_ci_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_PkgPwrSetLimit, n); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - - return 0; -} - -static int ci_set_overdrive_target_tdp(struct amdgpu_device *adev, - u32 target_tdp) -{ - PPSMC_Result smc_result = - amdgpu_ci_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_OverDriveSetTargetTdp, target_tdp); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - return 0; -} - -#if 0 -static int ci_set_boot_state(struct amdgpu_device *adev) -{ - return ci_enable_sclk_mclk_dpm(adev, false); -} -#endif - -static u32 ci_get_average_sclk_freq(struct amdgpu_device *adev) -{ - u32 sclk_freq; - PPSMC_Result smc_result = - amdgpu_ci_send_msg_to_smc_return_parameter(adev, - PPSMC_MSG_API_GetSclkFrequency, - &sclk_freq); - if (smc_result != PPSMC_Result_OK) - sclk_freq = 0; - - return sclk_freq; -} - -static u32 ci_get_average_mclk_freq(struct amdgpu_device *adev) -{ - u32 mclk_freq; - PPSMC_Result smc_result = - amdgpu_ci_send_msg_to_smc_return_parameter(adev, - PPSMC_MSG_API_GetMclkFrequency, - &mclk_freq); - if (smc_result != PPSMC_Result_OK) - mclk_freq = 0; - - return mclk_freq; -} - -static void ci_dpm_start_smc(struct amdgpu_device *adev) -{ - int i; - - amdgpu_ci_program_jump_on_start(adev); - amdgpu_ci_start_smc_clock(adev); - amdgpu_ci_start_smc(adev); - for (i = 0; i < adev->usec_timeout; i++) { - if (RREG32_SMC(ixFIRMWARE_FLAGS) & FIRMWARE_FLAGS__INTERRUPTS_ENABLED_MASK) - break; - } -} - -static void ci_dpm_stop_smc(struct amdgpu_device *adev) -{ - amdgpu_ci_reset_smc(adev); - amdgpu_ci_stop_smc_clock(adev); -} - -static int ci_process_firmware_header(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 tmp; - int ret; - - ret = amdgpu_ci_read_smc_sram_dword(adev, - SMU7_FIRMWARE_HEADER_LOCATION + - offsetof(SMU7_Firmware_Header, DpmTable), - &tmp, pi->sram_end); - if (ret) - return ret; - - pi->dpm_table_start = tmp; - - ret = amdgpu_ci_read_smc_sram_dword(adev, - SMU7_FIRMWARE_HEADER_LOCATION + - offsetof(SMU7_Firmware_Header, SoftRegisters), - &tmp, pi->sram_end); - if (ret) - return ret; - - pi->soft_regs_start = tmp; - - ret = amdgpu_ci_read_smc_sram_dword(adev, - SMU7_FIRMWARE_HEADER_LOCATION + - offsetof(SMU7_Firmware_Header, mcRegisterTable), - &tmp, pi->sram_end); - if (ret) - return ret; - - pi->mc_reg_table_start = tmp; - - ret = amdgpu_ci_read_smc_sram_dword(adev, - SMU7_FIRMWARE_HEADER_LOCATION + - offsetof(SMU7_Firmware_Header, FanTable), - &tmp, pi->sram_end); - if (ret) - return ret; - - pi->fan_table_start = tmp; - - ret = amdgpu_ci_read_smc_sram_dword(adev, - SMU7_FIRMWARE_HEADER_LOCATION + - offsetof(SMU7_Firmware_Header, mcArbDramTimingTable), - &tmp, pi->sram_end); - if (ret) - return ret; - - pi->arb_table_start = tmp; - - return 0; -} - -static void ci_read_clock_registers(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - pi->clock_registers.cg_spll_func_cntl = - RREG32_SMC(ixCG_SPLL_FUNC_CNTL); - pi->clock_registers.cg_spll_func_cntl_2 = - RREG32_SMC(ixCG_SPLL_FUNC_CNTL_2); - pi->clock_registers.cg_spll_func_cntl_3 = - RREG32_SMC(ixCG_SPLL_FUNC_CNTL_3); - pi->clock_registers.cg_spll_func_cntl_4 = - RREG32_SMC(ixCG_SPLL_FUNC_CNTL_4); - pi->clock_registers.cg_spll_spread_spectrum = - RREG32_SMC(ixCG_SPLL_SPREAD_SPECTRUM); - pi->clock_registers.cg_spll_spread_spectrum_2 = - RREG32_SMC(ixCG_SPLL_SPREAD_SPECTRUM_2); - pi->clock_registers.dll_cntl = RREG32(mmDLL_CNTL); - pi->clock_registers.mclk_pwrmgt_cntl = RREG32(mmMCLK_PWRMGT_CNTL); - pi->clock_registers.mpll_ad_func_cntl = RREG32(mmMPLL_AD_FUNC_CNTL); - pi->clock_registers.mpll_dq_func_cntl = RREG32(mmMPLL_DQ_FUNC_CNTL); - pi->clock_registers.mpll_func_cntl = RREG32(mmMPLL_FUNC_CNTL); - pi->clock_registers.mpll_func_cntl_1 = RREG32(mmMPLL_FUNC_CNTL_1); - pi->clock_registers.mpll_func_cntl_2 = RREG32(mmMPLL_FUNC_CNTL_2); - pi->clock_registers.mpll_ss1 = RREG32(mmMPLL_SS1); - pi->clock_registers.mpll_ss2 = RREG32(mmMPLL_SS2); -} - -static void ci_init_sclk_t(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - pi->low_sclk_interrupt_t = 0; -} - -static void ci_enable_thermal_protection(struct amdgpu_device *adev, - bool enable) -{ - u32 tmp = RREG32_SMC(ixGENERAL_PWRMGT); - - if (enable) - tmp &= ~GENERAL_PWRMGT__THERMAL_PROTECTION_DIS_MASK; - else - tmp |= GENERAL_PWRMGT__THERMAL_PROTECTION_DIS_MASK; - WREG32_SMC(ixGENERAL_PWRMGT, tmp); -} - -static void ci_enable_acpi_power_management(struct amdgpu_device *adev) -{ - u32 tmp = RREG32_SMC(ixGENERAL_PWRMGT); - - tmp |= GENERAL_PWRMGT__STATIC_PM_EN_MASK; - - WREG32_SMC(ixGENERAL_PWRMGT, tmp); -} - -#if 0 -static int ci_enter_ulp_state(struct amdgpu_device *adev) -{ - - WREG32(mmSMC_MESSAGE_0, PPSMC_MSG_SwitchToMinimumPower); - - udelay(25000); - - return 0; -} - -static int ci_exit_ulp_state(struct amdgpu_device *adev) -{ - int i; - - WREG32(mmSMC_MESSAGE_0, PPSMC_MSG_ResumeFromMinimumPower); - - udelay(7000); - - for (i = 0; i < adev->usec_timeout; i++) { - if (RREG32(mmSMC_RESP_0) == 1) - break; - udelay(1000); - } - - return 0; -} -#endif - -static int ci_notify_smc_display_change(struct amdgpu_device *adev, - bool has_display) -{ - PPSMC_Msg msg = has_display ? PPSMC_MSG_HasDisplay : PPSMC_MSG_NoDisplay; - - return (amdgpu_ci_send_msg_to_smc(adev, msg) == PPSMC_Result_OK) ? 0 : -EINVAL; -} - -static int ci_enable_ds_master_switch(struct amdgpu_device *adev, - bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (enable) { - if (pi->caps_sclk_ds) { - if (amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_MASTER_DeepSleep_ON) != PPSMC_Result_OK) - return -EINVAL; - } else { - if (amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_MASTER_DeepSleep_OFF) != PPSMC_Result_OK) - return -EINVAL; - } - } else { - if (pi->caps_sclk_ds) { - if (amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_MASTER_DeepSleep_OFF) != PPSMC_Result_OK) - return -EINVAL; - } - } - - return 0; -} - -static void ci_program_display_gap(struct amdgpu_device *adev) -{ - u32 tmp = RREG32_SMC(ixCG_DISPLAY_GAP_CNTL); - u32 pre_vbi_time_in_us; - u32 frame_time_in_us; - u32 ref_clock = adev->clock.spll.reference_freq; - u32 refresh_rate = amdgpu_dpm_get_vrefresh(adev); - u32 vblank_time = amdgpu_dpm_get_vblank_time(adev); - - tmp &= ~CG_DISPLAY_GAP_CNTL__DISP_GAP_MASK; - if (adev->pm.dpm.new_active_crtc_count > 0) - tmp |= (AMDGPU_PM_DISPLAY_GAP_VBLANK_OR_WM << CG_DISPLAY_GAP_CNTL__DISP_GAP__SHIFT); - else - tmp |= (AMDGPU_PM_DISPLAY_GAP_IGNORE << CG_DISPLAY_GAP_CNTL__DISP_GAP__SHIFT); - WREG32_SMC(ixCG_DISPLAY_GAP_CNTL, tmp); - - if (refresh_rate == 0) - refresh_rate = 60; - if (vblank_time == 0xffffffff) - vblank_time = 500; - frame_time_in_us = 1000000 / refresh_rate; - pre_vbi_time_in_us = - frame_time_in_us - 200 - vblank_time; - tmp = pre_vbi_time_in_us * (ref_clock / 100); - - WREG32_SMC(ixCG_DISPLAY_GAP_CNTL2, tmp); - ci_write_smc_soft_register(adev, offsetof(SMU7_SoftRegisters, PreVBlankGap), 0x64); - ci_write_smc_soft_register(adev, offsetof(SMU7_SoftRegisters, VBlankTimeout), (frame_time_in_us - pre_vbi_time_in_us)); - - - ci_notify_smc_display_change(adev, (adev->pm.dpm.new_active_crtc_count == 1)); - -} - -static void ci_enable_spread_spectrum(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 tmp; - - if (enable) { - if (pi->caps_sclk_ss_support) { - tmp = RREG32_SMC(ixGENERAL_PWRMGT); - tmp |= GENERAL_PWRMGT__DYN_SPREAD_SPECTRUM_EN_MASK; - WREG32_SMC(ixGENERAL_PWRMGT, tmp); - } - } else { - tmp = RREG32_SMC(ixCG_SPLL_SPREAD_SPECTRUM); - tmp &= ~CG_SPLL_SPREAD_SPECTRUM__SSEN_MASK; - WREG32_SMC(ixCG_SPLL_SPREAD_SPECTRUM, tmp); - - tmp = RREG32_SMC(ixGENERAL_PWRMGT); - tmp &= ~GENERAL_PWRMGT__DYN_SPREAD_SPECTRUM_EN_MASK; - WREG32_SMC(ixGENERAL_PWRMGT, tmp); - } -} - -static void ci_program_sstp(struct amdgpu_device *adev) -{ - WREG32_SMC(ixCG_STATIC_SCREEN_PARAMETER, - ((CISLANDS_SSTU_DFLT << CG_STATIC_SCREEN_PARAMETER__STATIC_SCREEN_THRESHOLD_UNIT__SHIFT) | - (CISLANDS_SST_DFLT << CG_STATIC_SCREEN_PARAMETER__STATIC_SCREEN_THRESHOLD__SHIFT))); -} - -static void ci_enable_display_gap(struct amdgpu_device *adev) -{ - u32 tmp = RREG32_SMC(ixCG_DISPLAY_GAP_CNTL); - - tmp &= ~(CG_DISPLAY_GAP_CNTL__DISP_GAP_MASK | - CG_DISPLAY_GAP_CNTL__DISP_GAP_MCHG_MASK); - tmp |= ((AMDGPU_PM_DISPLAY_GAP_IGNORE << CG_DISPLAY_GAP_CNTL__DISP_GAP__SHIFT) | - (AMDGPU_PM_DISPLAY_GAP_VBLANK << CG_DISPLAY_GAP_CNTL__DISP_GAP_MCHG__SHIFT)); - - WREG32_SMC(ixCG_DISPLAY_GAP_CNTL, tmp); -} - -static void ci_program_vc(struct amdgpu_device *adev) -{ - u32 tmp; - - tmp = RREG32_SMC(ixSCLK_PWRMGT_CNTL); - tmp &= ~(SCLK_PWRMGT_CNTL__RESET_SCLK_CNT_MASK | SCLK_PWRMGT_CNTL__RESET_BUSY_CNT_MASK); - WREG32_SMC(ixSCLK_PWRMGT_CNTL, tmp); - - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_0, CISLANDS_VRC_DFLT0); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_1, CISLANDS_VRC_DFLT1); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_2, CISLANDS_VRC_DFLT2); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_3, CISLANDS_VRC_DFLT3); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_4, CISLANDS_VRC_DFLT4); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_5, CISLANDS_VRC_DFLT5); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_6, CISLANDS_VRC_DFLT6); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_7, CISLANDS_VRC_DFLT7); -} - -static void ci_clear_vc(struct amdgpu_device *adev) -{ - u32 tmp; - - tmp = RREG32_SMC(ixSCLK_PWRMGT_CNTL); - tmp |= (SCLK_PWRMGT_CNTL__RESET_SCLK_CNT_MASK | SCLK_PWRMGT_CNTL__RESET_BUSY_CNT_MASK); - WREG32_SMC(ixSCLK_PWRMGT_CNTL, tmp); - - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_0, 0); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_1, 0); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_2, 0); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_3, 0); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_4, 0); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_5, 0); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_6, 0); - WREG32_SMC(ixCG_FREQ_TRAN_VOTING_7, 0); -} - -static int ci_upload_firmware(struct amdgpu_device *adev) -{ - int i, ret; - - if (amdgpu_ci_is_smc_running(adev)) { - DRM_INFO("smc is running, no need to load smc firmware\n"); - return 0; - } - - for (i = 0; i < adev->usec_timeout; i++) { - if (RREG32_SMC(ixRCU_UC_EVENTS) & RCU_UC_EVENTS__boot_seq_done_MASK) - break; - } - WREG32_SMC(ixSMC_SYSCON_MISC_CNTL, 1); - - amdgpu_ci_stop_smc_clock(adev); - amdgpu_ci_reset_smc(adev); - - ret = amdgpu_ci_load_smc_ucode(adev, SMC_RAM_END); - - return ret; - -} - -static int ci_get_svi2_voltage_table(struct amdgpu_device *adev, - struct amdgpu_clock_voltage_dependency_table *voltage_dependency_table, - struct atom_voltage_table *voltage_table) -{ - u32 i; - - if (voltage_dependency_table == NULL) - return -EINVAL; - - voltage_table->mask_low = 0; - voltage_table->phase_delay = 0; - - voltage_table->count = voltage_dependency_table->count; - for (i = 0; i < voltage_table->count; i++) { - voltage_table->entries[i].value = voltage_dependency_table->entries[i].v; - voltage_table->entries[i].smio_low = 0; - } - - return 0; -} - -static int ci_construct_voltage_tables(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - int ret; - - if (pi->voltage_control == CISLANDS_VOLTAGE_CONTROL_BY_GPIO) { - ret = amdgpu_atombios_get_voltage_table(adev, VOLTAGE_TYPE_VDDC, - VOLTAGE_OBJ_GPIO_LUT, - &pi->vddc_voltage_table); - if (ret) - return ret; - } else if (pi->voltage_control == CISLANDS_VOLTAGE_CONTROL_BY_SVID2) { - ret = ci_get_svi2_voltage_table(adev, - &adev->pm.dpm.dyn_state.vddc_dependency_on_mclk, - &pi->vddc_voltage_table); - if (ret) - return ret; - } - - if (pi->vddc_voltage_table.count > SMU7_MAX_LEVELS_VDDC) - ci_trim_voltage_table_to_fit_state_table(adev, SMU7_MAX_LEVELS_VDDC, - &pi->vddc_voltage_table); - - if (pi->vddci_control == CISLANDS_VOLTAGE_CONTROL_BY_GPIO) { - ret = amdgpu_atombios_get_voltage_table(adev, VOLTAGE_TYPE_VDDCI, - VOLTAGE_OBJ_GPIO_LUT, - &pi->vddci_voltage_table); - if (ret) - return ret; - } else if (pi->vddci_control == CISLANDS_VOLTAGE_CONTROL_BY_SVID2) { - ret = ci_get_svi2_voltage_table(adev, - &adev->pm.dpm.dyn_state.vddci_dependency_on_mclk, - &pi->vddci_voltage_table); - if (ret) - return ret; - } - - if (pi->vddci_voltage_table.count > SMU7_MAX_LEVELS_VDDCI) - ci_trim_voltage_table_to_fit_state_table(adev, SMU7_MAX_LEVELS_VDDCI, - &pi->vddci_voltage_table); - - if (pi->mvdd_control == CISLANDS_VOLTAGE_CONTROL_BY_GPIO) { - ret = amdgpu_atombios_get_voltage_table(adev, VOLTAGE_TYPE_MVDDC, - VOLTAGE_OBJ_GPIO_LUT, - &pi->mvdd_voltage_table); - if (ret) - return ret; - } else if (pi->mvdd_control == CISLANDS_VOLTAGE_CONTROL_BY_SVID2) { - ret = ci_get_svi2_voltage_table(adev, - &adev->pm.dpm.dyn_state.mvdd_dependency_on_mclk, - &pi->mvdd_voltage_table); - if (ret) - return ret; - } - - if (pi->mvdd_voltage_table.count > SMU7_MAX_LEVELS_MVDD) - ci_trim_voltage_table_to_fit_state_table(adev, SMU7_MAX_LEVELS_MVDD, - &pi->mvdd_voltage_table); - - return 0; -} - -static void ci_populate_smc_voltage_table(struct amdgpu_device *adev, - struct atom_voltage_table_entry *voltage_table, - SMU7_Discrete_VoltageLevel *smc_voltage_table) -{ - int ret; - - ret = ci_get_std_voltage_value_sidd(adev, voltage_table, - &smc_voltage_table->StdVoltageHiSidd, - &smc_voltage_table->StdVoltageLoSidd); - - if (ret) { - smc_voltage_table->StdVoltageHiSidd = voltage_table->value * VOLTAGE_SCALE; - smc_voltage_table->StdVoltageLoSidd = voltage_table->value * VOLTAGE_SCALE; - } - - smc_voltage_table->Voltage = cpu_to_be16(voltage_table->value * VOLTAGE_SCALE); - smc_voltage_table->StdVoltageHiSidd = - cpu_to_be16(smc_voltage_table->StdVoltageHiSidd); - smc_voltage_table->StdVoltageLoSidd = - cpu_to_be16(smc_voltage_table->StdVoltageLoSidd); -} - -static int ci_populate_smc_vddc_table(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - struct ci_power_info *pi = ci_get_pi(adev); - unsigned int count; - - table->VddcLevelCount = pi->vddc_voltage_table.count; - for (count = 0; count < table->VddcLevelCount; count++) { - ci_populate_smc_voltage_table(adev, - &pi->vddc_voltage_table.entries[count], - &table->VddcLevel[count]); - - if (pi->voltage_control == CISLANDS_VOLTAGE_CONTROL_BY_GPIO) - table->VddcLevel[count].Smio |= - pi->vddc_voltage_table.entries[count].smio_low; - else - table->VddcLevel[count].Smio = 0; - } - table->VddcLevelCount = cpu_to_be32(table->VddcLevelCount); - - return 0; -} - -static int ci_populate_smc_vddci_table(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - unsigned int count; - struct ci_power_info *pi = ci_get_pi(adev); - - table->VddciLevelCount = pi->vddci_voltage_table.count; - for (count = 0; count < table->VddciLevelCount; count++) { - ci_populate_smc_voltage_table(adev, - &pi->vddci_voltage_table.entries[count], - &table->VddciLevel[count]); - - if (pi->vddci_control == CISLANDS_VOLTAGE_CONTROL_BY_GPIO) - table->VddciLevel[count].Smio |= - pi->vddci_voltage_table.entries[count].smio_low; - else - table->VddciLevel[count].Smio = 0; - } - table->VddciLevelCount = cpu_to_be32(table->VddciLevelCount); - - return 0; -} - -static int ci_populate_smc_mvdd_table(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - struct ci_power_info *pi = ci_get_pi(adev); - unsigned int count; - - table->MvddLevelCount = pi->mvdd_voltage_table.count; - for (count = 0; count < table->MvddLevelCount; count++) { - ci_populate_smc_voltage_table(adev, - &pi->mvdd_voltage_table.entries[count], - &table->MvddLevel[count]); - - if (pi->mvdd_control == CISLANDS_VOLTAGE_CONTROL_BY_GPIO) - table->MvddLevel[count].Smio |= - pi->mvdd_voltage_table.entries[count].smio_low; - else - table->MvddLevel[count].Smio = 0; - } - table->MvddLevelCount = cpu_to_be32(table->MvddLevelCount); - - return 0; -} - -static int ci_populate_smc_voltage_tables(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - int ret; - - ret = ci_populate_smc_vddc_table(adev, table); - if (ret) - return ret; - - ret = ci_populate_smc_vddci_table(adev, table); - if (ret) - return ret; - - ret = ci_populate_smc_mvdd_table(adev, table); - if (ret) - return ret; - - return 0; -} - -static int ci_populate_mvdd_value(struct amdgpu_device *adev, u32 mclk, - SMU7_Discrete_VoltageLevel *voltage) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 i = 0; - - if (pi->mvdd_control != CISLANDS_VOLTAGE_CONTROL_NONE) { - for (i = 0; i < adev->pm.dpm.dyn_state.mvdd_dependency_on_mclk.count; i++) { - if (mclk <= adev->pm.dpm.dyn_state.mvdd_dependency_on_mclk.entries[i].clk) { - voltage->Voltage = pi->mvdd_voltage_table.entries[i].value; - break; - } - } - - if (i >= adev->pm.dpm.dyn_state.mvdd_dependency_on_mclk.count) - return -EINVAL; - } - - return -EINVAL; -} - -static int ci_get_std_voltage_value_sidd(struct amdgpu_device *adev, - struct atom_voltage_table_entry *voltage_table, - u16 *std_voltage_hi_sidd, u16 *std_voltage_lo_sidd) -{ - u16 v_index, idx; - bool voltage_found = false; - *std_voltage_hi_sidd = voltage_table->value * VOLTAGE_SCALE; - *std_voltage_lo_sidd = voltage_table->value * VOLTAGE_SCALE; - - if (adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries == NULL) - return -EINVAL; - - if (adev->pm.dpm.dyn_state.cac_leakage_table.entries) { - for (v_index = 0; (u32)v_index < adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.count; v_index++) { - if (voltage_table->value == - adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[v_index].v) { - voltage_found = true; - if ((u32)v_index < adev->pm.dpm.dyn_state.cac_leakage_table.count) - idx = v_index; - else - idx = adev->pm.dpm.dyn_state.cac_leakage_table.count - 1; - *std_voltage_lo_sidd = - adev->pm.dpm.dyn_state.cac_leakage_table.entries[idx].vddc * VOLTAGE_SCALE; - *std_voltage_hi_sidd = - adev->pm.dpm.dyn_state.cac_leakage_table.entries[idx].leakage * VOLTAGE_SCALE; - break; - } - } - - if (!voltage_found) { - for (v_index = 0; (u32)v_index < adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.count; v_index++) { - if (voltage_table->value <= - adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[v_index].v) { - voltage_found = true; - if ((u32)v_index < adev->pm.dpm.dyn_state.cac_leakage_table.count) - idx = v_index; - else - idx = adev->pm.dpm.dyn_state.cac_leakage_table.count - 1; - *std_voltage_lo_sidd = - adev->pm.dpm.dyn_state.cac_leakage_table.entries[idx].vddc * VOLTAGE_SCALE; - *std_voltage_hi_sidd = - adev->pm.dpm.dyn_state.cac_leakage_table.entries[idx].leakage * VOLTAGE_SCALE; - break; - } - } - } - } - - return 0; -} - -static void ci_populate_phase_value_based_on_sclk(struct amdgpu_device *adev, - const struct amdgpu_phase_shedding_limits_table *limits, - u32 sclk, - u32 *phase_shedding) -{ - unsigned int i; - - *phase_shedding = 1; - - for (i = 0; i < limits->count; i++) { - if (sclk < limits->entries[i].sclk) { - *phase_shedding = i; - break; - } - } -} - -static void ci_populate_phase_value_based_on_mclk(struct amdgpu_device *adev, - const struct amdgpu_phase_shedding_limits_table *limits, - u32 mclk, - u32 *phase_shedding) -{ - unsigned int i; - - *phase_shedding = 1; - - for (i = 0; i < limits->count; i++) { - if (mclk < limits->entries[i].mclk) { - *phase_shedding = i; - break; - } - } -} - -static int ci_init_arb_table_index(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 tmp; - int ret; - - ret = amdgpu_ci_read_smc_sram_dword(adev, pi->arb_table_start, - &tmp, pi->sram_end); - if (ret) - return ret; - - tmp &= 0x00FFFFFF; - tmp |= MC_CG_ARB_FREQ_F1 << 24; - - return amdgpu_ci_write_smc_sram_dword(adev, pi->arb_table_start, - tmp, pi->sram_end); -} - -static int ci_get_dependency_volt_by_clk(struct amdgpu_device *adev, - struct amdgpu_clock_voltage_dependency_table *allowed_clock_voltage_table, - u32 clock, u32 *voltage) -{ - u32 i = 0; - - if (allowed_clock_voltage_table->count == 0) - return -EINVAL; - - for (i = 0; i < allowed_clock_voltage_table->count; i++) { - if (allowed_clock_voltage_table->entries[i].clk >= clock) { - *voltage = allowed_clock_voltage_table->entries[i].v; - return 0; - } - } - - *voltage = allowed_clock_voltage_table->entries[i-1].v; - - return 0; -} - -static u8 ci_get_sleep_divider_id_from_clock(u32 sclk, u32 min_sclk_in_sr) -{ - u32 i; - u32 tmp; - u32 min = max(min_sclk_in_sr, (u32)CISLAND_MINIMUM_ENGINE_CLOCK); - - if (sclk < min) - return 0; - - for (i = CISLAND_MAX_DEEPSLEEP_DIVIDER_ID; ; i--) { - tmp = sclk >> i; - if (tmp >= min || i == 0) - break; - } - - return (u8)i; -} - -static int ci_initial_switch_from_arb_f0_to_f1(struct amdgpu_device *adev) -{ - return ci_copy_and_switch_arb_sets(adev, MC_CG_ARB_FREQ_F0, MC_CG_ARB_FREQ_F1); -} - -static int ci_reset_to_default(struct amdgpu_device *adev) -{ - return (amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_ResetToDefaults) == PPSMC_Result_OK) ? - 0 : -EINVAL; -} - -static int ci_force_switch_to_arb_f0(struct amdgpu_device *adev) -{ - u32 tmp; - - tmp = (RREG32_SMC(ixSMC_SCRATCH9) & 0x0000ff00) >> 8; - - if (tmp == MC_CG_ARB_FREQ_F0) - return 0; - - return ci_copy_and_switch_arb_sets(adev, tmp, MC_CG_ARB_FREQ_F0); -} - -static void ci_register_patching_mc_arb(struct amdgpu_device *adev, - const u32 engine_clock, - const u32 memory_clock, - u32 *dram_timimg2) -{ - bool patch; - u32 tmp, tmp2; - - tmp = RREG32(mmMC_SEQ_MISC0); - patch = ((tmp & 0x0000f00) == 0x300) ? true : false; - - if (patch && - ((adev->pdev->device == 0x67B0) || - (adev->pdev->device == 0x67B1))) { - if ((memory_clock > 100000) && (memory_clock <= 125000)) { - tmp2 = (((0x31 * engine_clock) / 125000) - 1) & 0xff; - *dram_timimg2 &= ~0x00ff0000; - *dram_timimg2 |= tmp2 << 16; - } else if ((memory_clock > 125000) && (memory_clock <= 137500)) { - tmp2 = (((0x36 * engine_clock) / 137500) - 1) & 0xff; - *dram_timimg2 &= ~0x00ff0000; - *dram_timimg2 |= tmp2 << 16; - } - } -} - -static int ci_populate_memory_timing_parameters(struct amdgpu_device *adev, - u32 sclk, - u32 mclk, - SMU7_Discrete_MCArbDramTimingTableEntry *arb_regs) -{ - u32 dram_timing; - u32 dram_timing2; - u32 burst_time; - - amdgpu_atombios_set_engine_dram_timings(adev, sclk, mclk); - - dram_timing = RREG32(mmMC_ARB_DRAM_TIMING); - dram_timing2 = RREG32(mmMC_ARB_DRAM_TIMING2); - burst_time = RREG32(mmMC_ARB_BURST_TIME) & MC_ARB_BURST_TIME__STATE0_MASK; - - ci_register_patching_mc_arb(adev, sclk, mclk, &dram_timing2); - - arb_regs->McArbDramTiming = cpu_to_be32(dram_timing); - arb_regs->McArbDramTiming2 = cpu_to_be32(dram_timing2); - arb_regs->McArbBurstTime = (u8)burst_time; - - return 0; -} - -static int ci_do_program_memory_timing_parameters(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - SMU7_Discrete_MCArbDramTimingTable arb_regs; - u32 i, j; - int ret = 0; - - memset(&arb_regs, 0, sizeof(SMU7_Discrete_MCArbDramTimingTable)); - - for (i = 0; i < pi->dpm_table.sclk_table.count; i++) { - for (j = 0; j < pi->dpm_table.mclk_table.count; j++) { - ret = ci_populate_memory_timing_parameters(adev, - pi->dpm_table.sclk_table.dpm_levels[i].value, - pi->dpm_table.mclk_table.dpm_levels[j].value, - &arb_regs.entries[i][j]); - if (ret) - break; - } - } - - if (ret == 0) - ret = amdgpu_ci_copy_bytes_to_smc(adev, - pi->arb_table_start, - (u8 *)&arb_regs, - sizeof(SMU7_Discrete_MCArbDramTimingTable), - pi->sram_end); - - return ret; -} - -static int ci_program_memory_timing_parameters(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (pi->need_update_smu7_dpm_table == 0) - return 0; - - return ci_do_program_memory_timing_parameters(adev); -} - -static void ci_populate_smc_initial_state(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_boot_state) -{ - struct ci_ps *boot_state = ci_get_ps(amdgpu_boot_state); - struct ci_power_info *pi = ci_get_pi(adev); - u32 level = 0; - - for (level = 0; level < adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.count; level++) { - if (adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[level].clk >= - boot_state->performance_levels[0].sclk) { - pi->smc_state_table.GraphicsBootLevel = level; - break; - } - } - - for (level = 0; level < adev->pm.dpm.dyn_state.vddc_dependency_on_mclk.count; level++) { - if (adev->pm.dpm.dyn_state.vddc_dependency_on_mclk.entries[level].clk >= - boot_state->performance_levels[0].mclk) { - pi->smc_state_table.MemoryBootLevel = level; - break; - } - } -} - -static u32 ci_get_dpm_level_enable_mask_value(struct ci_single_dpm_table *dpm_table) -{ - u32 i; - u32 mask_value = 0; - - for (i = dpm_table->count; i > 0; i--) { - mask_value = mask_value << 1; - if (dpm_table->dpm_levels[i-1].enabled) - mask_value |= 0x1; - else - mask_value &= 0xFFFFFFFE; - } - - return mask_value; -} - -static void ci_populate_smc_link_level(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_dpm_table *dpm_table = &pi->dpm_table; - u32 i; - - for (i = 0; i < dpm_table->pcie_speed_table.count; i++) { - table->LinkLevel[i].PcieGenSpeed = - (u8)dpm_table->pcie_speed_table.dpm_levels[i].value; - table->LinkLevel[i].PcieLaneCount = - amdgpu_encode_pci_lane_width(dpm_table->pcie_speed_table.dpm_levels[i].param1); - table->LinkLevel[i].EnabledForActivity = 1; - table->LinkLevel[i].DownT = cpu_to_be32(5); - table->LinkLevel[i].UpT = cpu_to_be32(30); - } - - pi->smc_state_table.LinkLevelCount = (u8)dpm_table->pcie_speed_table.count; - pi->dpm_level_enable_mask.pcie_dpm_enable_mask = - ci_get_dpm_level_enable_mask_value(&dpm_table->pcie_speed_table); -} - -static int ci_populate_smc_uvd_level(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - u32 count; - struct atom_clock_dividers dividers; - int ret = -EINVAL; - - table->UvdLevelCount = - adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.count; - - for (count = 0; count < table->UvdLevelCount; count++) { - table->UvdLevel[count].VclkFrequency = - adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries[count].vclk; - table->UvdLevel[count].DclkFrequency = - adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries[count].dclk; - table->UvdLevel[count].MinVddc = - adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries[count].v * VOLTAGE_SCALE; - table->UvdLevel[count].MinVddcPhases = 1; - - ret = amdgpu_atombios_get_clock_dividers(adev, - COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK, - table->UvdLevel[count].VclkFrequency, false, ÷rs); - if (ret) - return ret; - - table->UvdLevel[count].VclkDivider = (u8)dividers.post_divider; - - ret = amdgpu_atombios_get_clock_dividers(adev, - COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK, - table->UvdLevel[count].DclkFrequency, false, ÷rs); - if (ret) - return ret; - - table->UvdLevel[count].DclkDivider = (u8)dividers.post_divider; - - table->UvdLevel[count].VclkFrequency = cpu_to_be32(table->UvdLevel[count].VclkFrequency); - table->UvdLevel[count].DclkFrequency = cpu_to_be32(table->UvdLevel[count].DclkFrequency); - table->UvdLevel[count].MinVddc = cpu_to_be16(table->UvdLevel[count].MinVddc); - } - - return ret; -} - -static int ci_populate_smc_vce_level(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - u32 count; - struct atom_clock_dividers dividers; - int ret = -EINVAL; - - table->VceLevelCount = - adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.count; - - for (count = 0; count < table->VceLevelCount; count++) { - table->VceLevel[count].Frequency = - adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.entries[count].evclk; - table->VceLevel[count].MinVoltage = - (u16)adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.entries[count].v * VOLTAGE_SCALE; - table->VceLevel[count].MinPhases = 1; - - ret = amdgpu_atombios_get_clock_dividers(adev, - COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK, - table->VceLevel[count].Frequency, false, ÷rs); - if (ret) - return ret; - - table->VceLevel[count].Divider = (u8)dividers.post_divider; - - table->VceLevel[count].Frequency = cpu_to_be32(table->VceLevel[count].Frequency); - table->VceLevel[count].MinVoltage = cpu_to_be16(table->VceLevel[count].MinVoltage); - } - - return ret; - -} - -static int ci_populate_smc_acp_level(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - u32 count; - struct atom_clock_dividers dividers; - int ret = -EINVAL; - - table->AcpLevelCount = (u8) - (adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.count); - - for (count = 0; count < table->AcpLevelCount; count++) { - table->AcpLevel[count].Frequency = - adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.entries[count].clk; - table->AcpLevel[count].MinVoltage = - adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.entries[count].v; - table->AcpLevel[count].MinPhases = 1; - - ret = amdgpu_atombios_get_clock_dividers(adev, - COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK, - table->AcpLevel[count].Frequency, false, ÷rs); - if (ret) - return ret; - - table->AcpLevel[count].Divider = (u8)dividers.post_divider; - - table->AcpLevel[count].Frequency = cpu_to_be32(table->AcpLevel[count].Frequency); - table->AcpLevel[count].MinVoltage = cpu_to_be16(table->AcpLevel[count].MinVoltage); - } - - return ret; -} - -static int ci_populate_smc_samu_level(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - u32 count; - struct atom_clock_dividers dividers; - int ret = -EINVAL; - - table->SamuLevelCount = - adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.count; - - for (count = 0; count < table->SamuLevelCount; count++) { - table->SamuLevel[count].Frequency = - adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.entries[count].clk; - table->SamuLevel[count].MinVoltage = - adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.entries[count].v * VOLTAGE_SCALE; - table->SamuLevel[count].MinPhases = 1; - - ret = amdgpu_atombios_get_clock_dividers(adev, - COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK, - table->SamuLevel[count].Frequency, false, ÷rs); - if (ret) - return ret; - - table->SamuLevel[count].Divider = (u8)dividers.post_divider; - - table->SamuLevel[count].Frequency = cpu_to_be32(table->SamuLevel[count].Frequency); - table->SamuLevel[count].MinVoltage = cpu_to_be16(table->SamuLevel[count].MinVoltage); - } - - return ret; -} - -static int ci_calculate_mclk_params(struct amdgpu_device *adev, - u32 memory_clock, - SMU7_Discrete_MemoryLevel *mclk, - bool strobe_mode, - bool dll_state_on) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 dll_cntl = pi->clock_registers.dll_cntl; - u32 mclk_pwrmgt_cntl = pi->clock_registers.mclk_pwrmgt_cntl; - u32 mpll_ad_func_cntl = pi->clock_registers.mpll_ad_func_cntl; - u32 mpll_dq_func_cntl = pi->clock_registers.mpll_dq_func_cntl; - u32 mpll_func_cntl = pi->clock_registers.mpll_func_cntl; - u32 mpll_func_cntl_1 = pi->clock_registers.mpll_func_cntl_1; - u32 mpll_func_cntl_2 = pi->clock_registers.mpll_func_cntl_2; - u32 mpll_ss1 = pi->clock_registers.mpll_ss1; - u32 mpll_ss2 = pi->clock_registers.mpll_ss2; - struct atom_mpll_param mpll_param; - int ret; - - ret = amdgpu_atombios_get_memory_pll_dividers(adev, memory_clock, strobe_mode, &mpll_param); - if (ret) - return ret; - - mpll_func_cntl &= ~MPLL_FUNC_CNTL__BWCTRL_MASK; - mpll_func_cntl |= (mpll_param.bwcntl << MPLL_FUNC_CNTL__BWCTRL__SHIFT); - - mpll_func_cntl_1 &= ~(MPLL_FUNC_CNTL_1__CLKF_MASK | MPLL_FUNC_CNTL_1__CLKFRAC_MASK | - MPLL_FUNC_CNTL_1__VCO_MODE_MASK); - mpll_func_cntl_1 |= (mpll_param.clkf) << MPLL_FUNC_CNTL_1__CLKF__SHIFT | - (mpll_param.clkfrac << MPLL_FUNC_CNTL_1__CLKFRAC__SHIFT) | - (mpll_param.vco_mode << MPLL_FUNC_CNTL_1__VCO_MODE__SHIFT); - - mpll_ad_func_cntl &= ~MPLL_AD_FUNC_CNTL__YCLK_POST_DIV_MASK; - mpll_ad_func_cntl |= (mpll_param.post_div << MPLL_AD_FUNC_CNTL__YCLK_POST_DIV__SHIFT); - - if (adev->gmc.vram_type == AMDGPU_VRAM_TYPE_GDDR5) { - mpll_dq_func_cntl &= ~(MPLL_DQ_FUNC_CNTL__YCLK_SEL_MASK | - MPLL_AD_FUNC_CNTL__YCLK_POST_DIV_MASK); - mpll_dq_func_cntl |= (mpll_param.yclk_sel << MPLL_DQ_FUNC_CNTL__YCLK_SEL__SHIFT) | - (mpll_param.post_div << MPLL_AD_FUNC_CNTL__YCLK_POST_DIV__SHIFT); - } - - if (pi->caps_mclk_ss_support) { - struct amdgpu_atom_ss ss; - u32 freq_nom; - u32 tmp; - u32 reference_clock = adev->clock.mpll.reference_freq; - - if (mpll_param.qdr == 1) - freq_nom = memory_clock * 4 * (1 << mpll_param.post_div); - else - freq_nom = memory_clock * 2 * (1 << mpll_param.post_div); - - tmp = (freq_nom / reference_clock); - tmp = tmp * tmp; - if (amdgpu_atombios_get_asic_ss_info(adev, &ss, - ASIC_INTERNAL_MEMORY_SS, freq_nom)) { - u32 clks = reference_clock * 5 / ss.rate; - u32 clkv = (u32)((((131 * ss.percentage * ss.rate) / 100) * tmp) / freq_nom); - - mpll_ss1 &= ~MPLL_SS1__CLKV_MASK; - mpll_ss1 |= (clkv << MPLL_SS1__CLKV__SHIFT); - - mpll_ss2 &= ~MPLL_SS2__CLKS_MASK; - mpll_ss2 |= (clks << MPLL_SS2__CLKS__SHIFT); - } - } - - mclk_pwrmgt_cntl &= ~MCLK_PWRMGT_CNTL__DLL_SPEED_MASK; - mclk_pwrmgt_cntl |= (mpll_param.dll_speed << MCLK_PWRMGT_CNTL__DLL_SPEED__SHIFT); - - if (dll_state_on) - mclk_pwrmgt_cntl |= MCLK_PWRMGT_CNTL__MRDCK0_PDNB_MASK | - MCLK_PWRMGT_CNTL__MRDCK1_PDNB_MASK; - else - mclk_pwrmgt_cntl &= ~(MCLK_PWRMGT_CNTL__MRDCK0_PDNB_MASK | - MCLK_PWRMGT_CNTL__MRDCK1_PDNB_MASK); - - mclk->MclkFrequency = memory_clock; - mclk->MpllFuncCntl = mpll_func_cntl; - mclk->MpllFuncCntl_1 = mpll_func_cntl_1; - mclk->MpllFuncCntl_2 = mpll_func_cntl_2; - mclk->MpllAdFuncCntl = mpll_ad_func_cntl; - mclk->MpllDqFuncCntl = mpll_dq_func_cntl; - mclk->MclkPwrmgtCntl = mclk_pwrmgt_cntl; - mclk->DllCntl = dll_cntl; - mclk->MpllSs1 = mpll_ss1; - mclk->MpllSs2 = mpll_ss2; - - return 0; -} - -static int ci_populate_single_memory_level(struct amdgpu_device *adev, - u32 memory_clock, - SMU7_Discrete_MemoryLevel *memory_level) -{ - struct ci_power_info *pi = ci_get_pi(adev); - int ret; - bool dll_state_on; - - if (adev->pm.dpm.dyn_state.vddc_dependency_on_mclk.entries) { - ret = ci_get_dependency_volt_by_clk(adev, - &adev->pm.dpm.dyn_state.vddc_dependency_on_mclk, - memory_clock, &memory_level->MinVddc); - if (ret) - return ret; - } - - if (adev->pm.dpm.dyn_state.vddci_dependency_on_mclk.entries) { - ret = ci_get_dependency_volt_by_clk(adev, - &adev->pm.dpm.dyn_state.vddci_dependency_on_mclk, - memory_clock, &memory_level->MinVddci); - if (ret) - return ret; - } - - if (adev->pm.dpm.dyn_state.mvdd_dependency_on_mclk.entries) { - ret = ci_get_dependency_volt_by_clk(adev, - &adev->pm.dpm.dyn_state.mvdd_dependency_on_mclk, - memory_clock, &memory_level->MinMvdd); - if (ret) - return ret; - } - - memory_level->MinVddcPhases = 1; - - if (pi->vddc_phase_shed_control) - ci_populate_phase_value_based_on_mclk(adev, - &adev->pm.dpm.dyn_state.phase_shedding_limits_table, - memory_clock, - &memory_level->MinVddcPhases); - - memory_level->EnabledForActivity = 1; - memory_level->EnabledForThrottle = 1; - memory_level->UpH = 0; - memory_level->DownH = 100; - memory_level->VoltageDownH = 0; - memory_level->ActivityLevel = (u16)pi->mclk_activity_target; - - memory_level->StutterEnable = false; - memory_level->StrobeEnable = false; - memory_level->EdcReadEnable = false; - memory_level->EdcWriteEnable = false; - memory_level->RttEnable = false; - - memory_level->DisplayWatermark = PPSMC_DISPLAY_WATERMARK_LOW; - - if (pi->mclk_stutter_mode_threshold && - (memory_clock <= pi->mclk_stutter_mode_threshold) && - (!pi->uvd_enabled) && - (RREG32(mmDPG_PIPE_STUTTER_CONTROL) & DPG_PIPE_STUTTER_CONTROL__STUTTER_ENABLE_MASK) && - (adev->pm.dpm.new_active_crtc_count <= 2)) - memory_level->StutterEnable = true; - - if (pi->mclk_strobe_mode_threshold && - (memory_clock <= pi->mclk_strobe_mode_threshold)) - memory_level->StrobeEnable = 1; - - if (adev->gmc.vram_type == AMDGPU_VRAM_TYPE_GDDR5) { - memory_level->StrobeRatio = - ci_get_mclk_frequency_ratio(memory_clock, memory_level->StrobeEnable); - if (pi->mclk_edc_enable_threshold && - (memory_clock > pi->mclk_edc_enable_threshold)) - memory_level->EdcReadEnable = true; - - if (pi->mclk_edc_wr_enable_threshold && - (memory_clock > pi->mclk_edc_wr_enable_threshold)) - memory_level->EdcWriteEnable = true; - - if (memory_level->StrobeEnable) { - if (ci_get_mclk_frequency_ratio(memory_clock, true) >= - ((RREG32(mmMC_SEQ_MISC7) >> 16) & 0xf)) - dll_state_on = ((RREG32(mmMC_SEQ_MISC5) >> 1) & 0x1) ? true : false; - else - dll_state_on = ((RREG32(mmMC_SEQ_MISC6) >> 1) & 0x1) ? true : false; - } else { - dll_state_on = pi->dll_default_on; - } - } else { - memory_level->StrobeRatio = ci_get_ddr3_mclk_frequency_ratio(memory_clock); - dll_state_on = ((RREG32(mmMC_SEQ_MISC5) >> 1) & 0x1) ? true : false; - } - - ret = ci_calculate_mclk_params(adev, memory_clock, memory_level, memory_level->StrobeEnable, dll_state_on); - if (ret) - return ret; - - memory_level->MinVddc = cpu_to_be32(memory_level->MinVddc * VOLTAGE_SCALE); - memory_level->MinVddcPhases = cpu_to_be32(memory_level->MinVddcPhases); - memory_level->MinVddci = cpu_to_be32(memory_level->MinVddci * VOLTAGE_SCALE); - memory_level->MinMvdd = cpu_to_be32(memory_level->MinMvdd * VOLTAGE_SCALE); - - memory_level->MclkFrequency = cpu_to_be32(memory_level->MclkFrequency); - memory_level->ActivityLevel = cpu_to_be16(memory_level->ActivityLevel); - memory_level->MpllFuncCntl = cpu_to_be32(memory_level->MpllFuncCntl); - memory_level->MpllFuncCntl_1 = cpu_to_be32(memory_level->MpllFuncCntl_1); - memory_level->MpllFuncCntl_2 = cpu_to_be32(memory_level->MpllFuncCntl_2); - memory_level->MpllAdFuncCntl = cpu_to_be32(memory_level->MpllAdFuncCntl); - memory_level->MpllDqFuncCntl = cpu_to_be32(memory_level->MpllDqFuncCntl); - memory_level->MclkPwrmgtCntl = cpu_to_be32(memory_level->MclkPwrmgtCntl); - memory_level->DllCntl = cpu_to_be32(memory_level->DllCntl); - memory_level->MpllSs1 = cpu_to_be32(memory_level->MpllSs1); - memory_level->MpllSs2 = cpu_to_be32(memory_level->MpllSs2); - - return 0; -} - -static int ci_populate_smc_acpi_level(struct amdgpu_device *adev, - SMU7_Discrete_DpmTable *table) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct atom_clock_dividers dividers; - SMU7_Discrete_VoltageLevel voltage_level; - u32 spll_func_cntl = pi->clock_registers.cg_spll_func_cntl; - u32 spll_func_cntl_2 = pi->clock_registers.cg_spll_func_cntl_2; - u32 dll_cntl = pi->clock_registers.dll_cntl; - u32 mclk_pwrmgt_cntl = pi->clock_registers.mclk_pwrmgt_cntl; - int ret; - - table->ACPILevel.Flags &= ~PPSMC_SWSTATE_FLAG_DC; - - if (pi->acpi_vddc) - table->ACPILevel.MinVddc = cpu_to_be32(pi->acpi_vddc * VOLTAGE_SCALE); - else - table->ACPILevel.MinVddc = cpu_to_be32(pi->min_vddc_in_pp_table * VOLTAGE_SCALE); - - table->ACPILevel.MinVddcPhases = pi->vddc_phase_shed_control ? 0 : 1; - - table->ACPILevel.SclkFrequency = adev->clock.spll.reference_freq; - - ret = amdgpu_atombios_get_clock_dividers(adev, - COMPUTE_GPUCLK_INPUT_FLAG_SCLK, - table->ACPILevel.SclkFrequency, false, ÷rs); - if (ret) - return ret; - - table->ACPILevel.SclkDid = (u8)dividers.post_divider; - table->ACPILevel.DisplayWatermark = PPSMC_DISPLAY_WATERMARK_LOW; - table->ACPILevel.DeepSleepDivId = 0; - - spll_func_cntl &= ~CG_SPLL_FUNC_CNTL__SPLL_PWRON_MASK; - spll_func_cntl |= CG_SPLL_FUNC_CNTL__SPLL_RESET_MASK; - - spll_func_cntl_2 &= ~CG_SPLL_FUNC_CNTL_2__SCLK_MUX_SEL_MASK; - spll_func_cntl_2 |= (4 << CG_SPLL_FUNC_CNTL_2__SCLK_MUX_SEL__SHIFT); - - table->ACPILevel.CgSpllFuncCntl = spll_func_cntl; - table->ACPILevel.CgSpllFuncCntl2 = spll_func_cntl_2; - table->ACPILevel.CgSpllFuncCntl3 = pi->clock_registers.cg_spll_func_cntl_3; - table->ACPILevel.CgSpllFuncCntl4 = pi->clock_registers.cg_spll_func_cntl_4; - table->ACPILevel.SpllSpreadSpectrum = pi->clock_registers.cg_spll_spread_spectrum; - table->ACPILevel.SpllSpreadSpectrum2 = pi->clock_registers.cg_spll_spread_spectrum_2; - table->ACPILevel.CcPwrDynRm = 0; - table->ACPILevel.CcPwrDynRm1 = 0; - - table->ACPILevel.Flags = cpu_to_be32(table->ACPILevel.Flags); - table->ACPILevel.MinVddcPhases = cpu_to_be32(table->ACPILevel.MinVddcPhases); - table->ACPILevel.SclkFrequency = cpu_to_be32(table->ACPILevel.SclkFrequency); - table->ACPILevel.CgSpllFuncCntl = cpu_to_be32(table->ACPILevel.CgSpllFuncCntl); - table->ACPILevel.CgSpllFuncCntl2 = cpu_to_be32(table->ACPILevel.CgSpllFuncCntl2); - table->ACPILevel.CgSpllFuncCntl3 = cpu_to_be32(table->ACPILevel.CgSpllFuncCntl3); - table->ACPILevel.CgSpllFuncCntl4 = cpu_to_be32(table->ACPILevel.CgSpllFuncCntl4); - table->ACPILevel.SpllSpreadSpectrum = cpu_to_be32(table->ACPILevel.SpllSpreadSpectrum); - table->ACPILevel.SpllSpreadSpectrum2 = cpu_to_be32(table->ACPILevel.SpllSpreadSpectrum2); - table->ACPILevel.CcPwrDynRm = cpu_to_be32(table->ACPILevel.CcPwrDynRm); - table->ACPILevel.CcPwrDynRm1 = cpu_to_be32(table->ACPILevel.CcPwrDynRm1); - - table->MemoryACPILevel.MinVddc = table->ACPILevel.MinVddc; - table->MemoryACPILevel.MinVddcPhases = table->ACPILevel.MinVddcPhases; - - if (pi->vddci_control != CISLANDS_VOLTAGE_CONTROL_NONE) { - if (pi->acpi_vddci) - table->MemoryACPILevel.MinVddci = - cpu_to_be32(pi->acpi_vddci * VOLTAGE_SCALE); - else - table->MemoryACPILevel.MinVddci = - cpu_to_be32(pi->min_vddci_in_pp_table * VOLTAGE_SCALE); - } - - if (ci_populate_mvdd_value(adev, 0, &voltage_level)) - table->MemoryACPILevel.MinMvdd = 0; - else - table->MemoryACPILevel.MinMvdd = - cpu_to_be32(voltage_level.Voltage * VOLTAGE_SCALE); - - mclk_pwrmgt_cntl |= MCLK_PWRMGT_CNTL__MRDCK0_RESET_MASK | - MCLK_PWRMGT_CNTL__MRDCK1_RESET_MASK; - mclk_pwrmgt_cntl &= ~(MCLK_PWRMGT_CNTL__MRDCK0_PDNB_MASK | - MCLK_PWRMGT_CNTL__MRDCK1_PDNB_MASK); - - dll_cntl &= ~(DLL_CNTL__MRDCK0_BYPASS_MASK | DLL_CNTL__MRDCK1_BYPASS_MASK); - - table->MemoryACPILevel.DllCntl = cpu_to_be32(dll_cntl); - table->MemoryACPILevel.MclkPwrmgtCntl = cpu_to_be32(mclk_pwrmgt_cntl); - table->MemoryACPILevel.MpllAdFuncCntl = - cpu_to_be32(pi->clock_registers.mpll_ad_func_cntl); - table->MemoryACPILevel.MpllDqFuncCntl = - cpu_to_be32(pi->clock_registers.mpll_dq_func_cntl); - table->MemoryACPILevel.MpllFuncCntl = - cpu_to_be32(pi->clock_registers.mpll_func_cntl); - table->MemoryACPILevel.MpllFuncCntl_1 = - cpu_to_be32(pi->clock_registers.mpll_func_cntl_1); - table->MemoryACPILevel.MpllFuncCntl_2 = - cpu_to_be32(pi->clock_registers.mpll_func_cntl_2); - table->MemoryACPILevel.MpllSs1 = cpu_to_be32(pi->clock_registers.mpll_ss1); - table->MemoryACPILevel.MpllSs2 = cpu_to_be32(pi->clock_registers.mpll_ss2); - - table->MemoryACPILevel.EnabledForThrottle = 0; - table->MemoryACPILevel.EnabledForActivity = 0; - table->MemoryACPILevel.UpH = 0; - table->MemoryACPILevel.DownH = 100; - table->MemoryACPILevel.VoltageDownH = 0; - table->MemoryACPILevel.ActivityLevel = - cpu_to_be16((u16)pi->mclk_activity_target); - - table->MemoryACPILevel.StutterEnable = false; - table->MemoryACPILevel.StrobeEnable = false; - table->MemoryACPILevel.EdcReadEnable = false; - table->MemoryACPILevel.EdcWriteEnable = false; - table->MemoryACPILevel.RttEnable = false; - - return 0; -} - - -static int ci_enable_ulv(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ulv_parm *ulv = &pi->ulv; - - if (ulv->supported) { - if (enable) - return (amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_EnableULV) == PPSMC_Result_OK) ? - 0 : -EINVAL; - else - return (amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_DisableULV) == PPSMC_Result_OK) ? - 0 : -EINVAL; - } - - return 0; -} - -static int ci_populate_ulv_level(struct amdgpu_device *adev, - SMU7_Discrete_Ulv *state) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u16 ulv_voltage = adev->pm.dpm.backbias_response_time; - - state->CcPwrDynRm = 0; - state->CcPwrDynRm1 = 0; - - if (ulv_voltage == 0) { - pi->ulv.supported = false; - return 0; - } - - if (pi->voltage_control != CISLANDS_VOLTAGE_CONTROL_BY_SVID2) { - if (ulv_voltage > adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[0].v) - state->VddcOffset = 0; - else - state->VddcOffset = - adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[0].v - ulv_voltage; - } else { - if (ulv_voltage > adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[0].v) - state->VddcOffsetVid = 0; - else - state->VddcOffsetVid = (u8) - ((adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[0].v - ulv_voltage) * - VOLTAGE_VID_OFFSET_SCALE2 / VOLTAGE_VID_OFFSET_SCALE1); - } - state->VddcPhase = pi->vddc_phase_shed_control ? 0 : 1; - - state->CcPwrDynRm = cpu_to_be32(state->CcPwrDynRm); - state->CcPwrDynRm1 = cpu_to_be32(state->CcPwrDynRm1); - state->VddcOffset = cpu_to_be16(state->VddcOffset); - - return 0; -} - -static int ci_calculate_sclk_params(struct amdgpu_device *adev, - u32 engine_clock, - SMU7_Discrete_GraphicsLevel *sclk) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct atom_clock_dividers dividers; - u32 spll_func_cntl_3 = pi->clock_registers.cg_spll_func_cntl_3; - u32 spll_func_cntl_4 = pi->clock_registers.cg_spll_func_cntl_4; - u32 cg_spll_spread_spectrum = pi->clock_registers.cg_spll_spread_spectrum; - u32 cg_spll_spread_spectrum_2 = pi->clock_registers.cg_spll_spread_spectrum_2; - u32 reference_clock = adev->clock.spll.reference_freq; - u32 reference_divider; - u32 fbdiv; - int ret; - - ret = amdgpu_atombios_get_clock_dividers(adev, - COMPUTE_GPUCLK_INPUT_FLAG_SCLK, - engine_clock, false, ÷rs); - if (ret) - return ret; - - reference_divider = 1 + dividers.ref_div; - fbdiv = dividers.fb_div & 0x3FFFFFF; - - spll_func_cntl_3 &= ~CG_SPLL_FUNC_CNTL_3__SPLL_FB_DIV_MASK; - spll_func_cntl_3 |= (fbdiv << CG_SPLL_FUNC_CNTL_3__SPLL_FB_DIV__SHIFT); - spll_func_cntl_3 |= CG_SPLL_FUNC_CNTL_3__SPLL_DITHEN_MASK; - - if (pi->caps_sclk_ss_support) { - struct amdgpu_atom_ss ss; - u32 vco_freq = engine_clock * dividers.post_div; - - if (amdgpu_atombios_get_asic_ss_info(adev, &ss, - ASIC_INTERNAL_ENGINE_SS, vco_freq)) { - u32 clk_s = reference_clock * 5 / (reference_divider * ss.rate); - u32 clk_v = 4 * ss.percentage * fbdiv / (clk_s * 10000); - - cg_spll_spread_spectrum &= ~(CG_SPLL_SPREAD_SPECTRUM__CLKS_MASK | CG_SPLL_SPREAD_SPECTRUM__SSEN_MASK); - cg_spll_spread_spectrum |= (clk_s << CG_SPLL_SPREAD_SPECTRUM__CLKS__SHIFT); - cg_spll_spread_spectrum |= (1 << CG_SPLL_SPREAD_SPECTRUM__SSEN__SHIFT); - - cg_spll_spread_spectrum_2 &= ~CG_SPLL_SPREAD_SPECTRUM_2__CLKV_MASK; - cg_spll_spread_spectrum_2 |= (clk_v << CG_SPLL_SPREAD_SPECTRUM_2__CLKV__SHIFT); - } - } - - sclk->SclkFrequency = engine_clock; - sclk->CgSpllFuncCntl3 = spll_func_cntl_3; - sclk->CgSpllFuncCntl4 = spll_func_cntl_4; - sclk->SpllSpreadSpectrum = cg_spll_spread_spectrum; - sclk->SpllSpreadSpectrum2 = cg_spll_spread_spectrum_2; - sclk->SclkDid = (u8)dividers.post_divider; - - return 0; -} - -static int ci_populate_single_graphic_level(struct amdgpu_device *adev, - u32 engine_clock, - u16 sclk_activity_level_t, - SMU7_Discrete_GraphicsLevel *graphic_level) -{ - struct ci_power_info *pi = ci_get_pi(adev); - int ret; - - ret = ci_calculate_sclk_params(adev, engine_clock, graphic_level); - if (ret) - return ret; - - ret = ci_get_dependency_volt_by_clk(adev, - &adev->pm.dpm.dyn_state.vddc_dependency_on_sclk, - engine_clock, &graphic_level->MinVddc); - if (ret) - return ret; - - graphic_level->SclkFrequency = engine_clock; - - graphic_level->Flags = 0; - graphic_level->MinVddcPhases = 1; - - if (pi->vddc_phase_shed_control) - ci_populate_phase_value_based_on_sclk(adev, - &adev->pm.dpm.dyn_state.phase_shedding_limits_table, - engine_clock, - &graphic_level->MinVddcPhases); - - graphic_level->ActivityLevel = sclk_activity_level_t; - - graphic_level->CcPwrDynRm = 0; - graphic_level->CcPwrDynRm1 = 0; - graphic_level->EnabledForThrottle = 1; - graphic_level->UpH = 0; - graphic_level->DownH = 0; - graphic_level->VoltageDownH = 0; - graphic_level->PowerThrottle = 0; - - if (pi->caps_sclk_ds) - graphic_level->DeepSleepDivId = ci_get_sleep_divider_id_from_clock(engine_clock, - CISLAND_MINIMUM_ENGINE_CLOCK); - - graphic_level->DisplayWatermark = PPSMC_DISPLAY_WATERMARK_LOW; - - graphic_level->Flags = cpu_to_be32(graphic_level->Flags); - graphic_level->MinVddc = cpu_to_be32(graphic_level->MinVddc * VOLTAGE_SCALE); - graphic_level->MinVddcPhases = cpu_to_be32(graphic_level->MinVddcPhases); - graphic_level->SclkFrequency = cpu_to_be32(graphic_level->SclkFrequency); - graphic_level->ActivityLevel = cpu_to_be16(graphic_level->ActivityLevel); - graphic_level->CgSpllFuncCntl3 = cpu_to_be32(graphic_level->CgSpllFuncCntl3); - graphic_level->CgSpllFuncCntl4 = cpu_to_be32(graphic_level->CgSpllFuncCntl4); - graphic_level->SpllSpreadSpectrum = cpu_to_be32(graphic_level->SpllSpreadSpectrum); - graphic_level->SpllSpreadSpectrum2 = cpu_to_be32(graphic_level->SpllSpreadSpectrum2); - graphic_level->CcPwrDynRm = cpu_to_be32(graphic_level->CcPwrDynRm); - graphic_level->CcPwrDynRm1 = cpu_to_be32(graphic_level->CcPwrDynRm1); - - return 0; -} - -static int ci_populate_all_graphic_levels(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_dpm_table *dpm_table = &pi->dpm_table; - u32 level_array_address = pi->dpm_table_start + - offsetof(SMU7_Discrete_DpmTable, GraphicsLevel); - u32 level_array_size = sizeof(SMU7_Discrete_GraphicsLevel) * - SMU7_MAX_LEVELS_GRAPHICS; - SMU7_Discrete_GraphicsLevel *levels = pi->smc_state_table.GraphicsLevel; - u32 i, ret; - - memset(levels, 0, level_array_size); - - for (i = 0; i < dpm_table->sclk_table.count; i++) { - ret = ci_populate_single_graphic_level(adev, - dpm_table->sclk_table.dpm_levels[i].value, - (u16)pi->activity_target[i], - &pi->smc_state_table.GraphicsLevel[i]); - if (ret) - return ret; - if (i > 1) - pi->smc_state_table.GraphicsLevel[i].DeepSleepDivId = 0; - if (i == (dpm_table->sclk_table.count - 1)) - pi->smc_state_table.GraphicsLevel[i].DisplayWatermark = - PPSMC_DISPLAY_WATERMARK_HIGH; - } - pi->smc_state_table.GraphicsLevel[0].EnabledForActivity = 1; - - pi->smc_state_table.GraphicsDpmLevelCount = (u8)dpm_table->sclk_table.count; - pi->dpm_level_enable_mask.sclk_dpm_enable_mask = - ci_get_dpm_level_enable_mask_value(&dpm_table->sclk_table); - - ret = amdgpu_ci_copy_bytes_to_smc(adev, level_array_address, - (u8 *)levels, level_array_size, - pi->sram_end); - if (ret) - return ret; - - return 0; -} - -static int ci_populate_ulv_state(struct amdgpu_device *adev, - SMU7_Discrete_Ulv *ulv_level) -{ - return ci_populate_ulv_level(adev, ulv_level); -} - -static int ci_populate_all_memory_levels(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_dpm_table *dpm_table = &pi->dpm_table; - u32 level_array_address = pi->dpm_table_start + - offsetof(SMU7_Discrete_DpmTable, MemoryLevel); - u32 level_array_size = sizeof(SMU7_Discrete_MemoryLevel) * - SMU7_MAX_LEVELS_MEMORY; - SMU7_Discrete_MemoryLevel *levels = pi->smc_state_table.MemoryLevel; - u32 i, ret; - - memset(levels, 0, level_array_size); - - for (i = 0; i < dpm_table->mclk_table.count; i++) { - if (dpm_table->mclk_table.dpm_levels[i].value == 0) - return -EINVAL; - ret = ci_populate_single_memory_level(adev, - dpm_table->mclk_table.dpm_levels[i].value, - &pi->smc_state_table.MemoryLevel[i]); - if (ret) - return ret; - } - - if ((dpm_table->mclk_table.count >= 2) && - ((adev->pdev->device == 0x67B0) || (adev->pdev->device == 0x67B1))) { - pi->smc_state_table.MemoryLevel[1].MinVddc = - pi->smc_state_table.MemoryLevel[0].MinVddc; - pi->smc_state_table.MemoryLevel[1].MinVddcPhases = - pi->smc_state_table.MemoryLevel[0].MinVddcPhases; - } - - pi->smc_state_table.MemoryLevel[0].ActivityLevel = cpu_to_be16(0x1F); - - pi->smc_state_table.MemoryDpmLevelCount = (u8)dpm_table->mclk_table.count; - pi->dpm_level_enable_mask.mclk_dpm_enable_mask = - ci_get_dpm_level_enable_mask_value(&dpm_table->mclk_table); - - pi->smc_state_table.MemoryLevel[dpm_table->mclk_table.count - 1].DisplayWatermark = - PPSMC_DISPLAY_WATERMARK_HIGH; - - ret = amdgpu_ci_copy_bytes_to_smc(adev, level_array_address, - (u8 *)levels, level_array_size, - pi->sram_end); - if (ret) - return ret; - - return 0; -} - -static void ci_reset_single_dpm_table(struct amdgpu_device *adev, - struct ci_single_dpm_table* dpm_table, - u32 count) -{ - u32 i; - - dpm_table->count = count; - for (i = 0; i < MAX_REGULAR_DPM_NUMBER; i++) - dpm_table->dpm_levels[i].enabled = false; -} - -static void ci_setup_pcie_table_entry(struct ci_single_dpm_table* dpm_table, - u32 index, u32 pcie_gen, u32 pcie_lanes) -{ - dpm_table->dpm_levels[index].value = pcie_gen; - dpm_table->dpm_levels[index].param1 = pcie_lanes; - dpm_table->dpm_levels[index].enabled = true; -} - -static int ci_setup_default_pcie_tables(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (!pi->use_pcie_performance_levels && !pi->use_pcie_powersaving_levels) - return -EINVAL; - - if (pi->use_pcie_performance_levels && !pi->use_pcie_powersaving_levels) { - pi->pcie_gen_powersaving = pi->pcie_gen_performance; - pi->pcie_lane_powersaving = pi->pcie_lane_performance; - } else if (!pi->use_pcie_performance_levels && pi->use_pcie_powersaving_levels) { - pi->pcie_gen_performance = pi->pcie_gen_powersaving; - pi->pcie_lane_performance = pi->pcie_lane_powersaving; - } - - ci_reset_single_dpm_table(adev, - &pi->dpm_table.pcie_speed_table, - SMU7_MAX_LEVELS_LINK); - - if (adev->asic_type == CHIP_BONAIRE) - ci_setup_pcie_table_entry(&pi->dpm_table.pcie_speed_table, 0, - pi->pcie_gen_powersaving.min, - pi->pcie_lane_powersaving.max); - else - ci_setup_pcie_table_entry(&pi->dpm_table.pcie_speed_table, 0, - pi->pcie_gen_powersaving.min, - pi->pcie_lane_powersaving.min); - ci_setup_pcie_table_entry(&pi->dpm_table.pcie_speed_table, 1, - pi->pcie_gen_performance.min, - pi->pcie_lane_performance.min); - ci_setup_pcie_table_entry(&pi->dpm_table.pcie_speed_table, 2, - pi->pcie_gen_powersaving.min, - pi->pcie_lane_powersaving.max); - ci_setup_pcie_table_entry(&pi->dpm_table.pcie_speed_table, 3, - pi->pcie_gen_performance.min, - pi->pcie_lane_performance.max); - ci_setup_pcie_table_entry(&pi->dpm_table.pcie_speed_table, 4, - pi->pcie_gen_powersaving.max, - pi->pcie_lane_powersaving.max); - ci_setup_pcie_table_entry(&pi->dpm_table.pcie_speed_table, 5, - pi->pcie_gen_performance.max, - pi->pcie_lane_performance.max); - - pi->dpm_table.pcie_speed_table.count = 6; - - return 0; -} - -static int ci_setup_default_dpm_tables(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_clock_voltage_dependency_table *allowed_sclk_vddc_table = - &adev->pm.dpm.dyn_state.vddc_dependency_on_sclk; - struct amdgpu_clock_voltage_dependency_table *allowed_mclk_table = - &adev->pm.dpm.dyn_state.vddc_dependency_on_mclk; - struct amdgpu_cac_leakage_table *std_voltage_table = - &adev->pm.dpm.dyn_state.cac_leakage_table; - u32 i; - - if (allowed_sclk_vddc_table == NULL) - return -EINVAL; - if (allowed_sclk_vddc_table->count < 1) - return -EINVAL; - if (allowed_mclk_table == NULL) - return -EINVAL; - if (allowed_mclk_table->count < 1) - return -EINVAL; - - memset(&pi->dpm_table, 0, sizeof(struct ci_dpm_table)); - - ci_reset_single_dpm_table(adev, - &pi->dpm_table.sclk_table, - SMU7_MAX_LEVELS_GRAPHICS); - ci_reset_single_dpm_table(adev, - &pi->dpm_table.mclk_table, - SMU7_MAX_LEVELS_MEMORY); - ci_reset_single_dpm_table(adev, - &pi->dpm_table.vddc_table, - SMU7_MAX_LEVELS_VDDC); - ci_reset_single_dpm_table(adev, - &pi->dpm_table.vddci_table, - SMU7_MAX_LEVELS_VDDCI); - ci_reset_single_dpm_table(adev, - &pi->dpm_table.mvdd_table, - SMU7_MAX_LEVELS_MVDD); - - pi->dpm_table.sclk_table.count = 0; - for (i = 0; i < allowed_sclk_vddc_table->count; i++) { - if ((i == 0) || - (pi->dpm_table.sclk_table.dpm_levels[pi->dpm_table.sclk_table.count-1].value != - allowed_sclk_vddc_table->entries[i].clk)) { - pi->dpm_table.sclk_table.dpm_levels[pi->dpm_table.sclk_table.count].value = - allowed_sclk_vddc_table->entries[i].clk; - pi->dpm_table.sclk_table.dpm_levels[pi->dpm_table.sclk_table.count].enabled = - (i == 0) ? true : false; - pi->dpm_table.sclk_table.count++; - } - } - - pi->dpm_table.mclk_table.count = 0; - for (i = 0; i < allowed_mclk_table->count; i++) { - if ((i == 0) || - (pi->dpm_table.mclk_table.dpm_levels[pi->dpm_table.mclk_table.count-1].value != - allowed_mclk_table->entries[i].clk)) { - pi->dpm_table.mclk_table.dpm_levels[pi->dpm_table.mclk_table.count].value = - allowed_mclk_table->entries[i].clk; - pi->dpm_table.mclk_table.dpm_levels[pi->dpm_table.mclk_table.count].enabled = - (i == 0) ? true : false; - pi->dpm_table.mclk_table.count++; - } - } - - for (i = 0; i < allowed_sclk_vddc_table->count; i++) { - pi->dpm_table.vddc_table.dpm_levels[i].value = - allowed_sclk_vddc_table->entries[i].v; - pi->dpm_table.vddc_table.dpm_levels[i].param1 = - std_voltage_table->entries[i].leakage; - pi->dpm_table.vddc_table.dpm_levels[i].enabled = true; - } - pi->dpm_table.vddc_table.count = allowed_sclk_vddc_table->count; - - allowed_mclk_table = &adev->pm.dpm.dyn_state.vddci_dependency_on_mclk; - if (allowed_mclk_table) { - for (i = 0; i < allowed_mclk_table->count; i++) { - pi->dpm_table.vddci_table.dpm_levels[i].value = - allowed_mclk_table->entries[i].v; - pi->dpm_table.vddci_table.dpm_levels[i].enabled = true; - } - pi->dpm_table.vddci_table.count = allowed_mclk_table->count; - } - - allowed_mclk_table = &adev->pm.dpm.dyn_state.mvdd_dependency_on_mclk; - if (allowed_mclk_table) { - for (i = 0; i < allowed_mclk_table->count; i++) { - pi->dpm_table.mvdd_table.dpm_levels[i].value = - allowed_mclk_table->entries[i].v; - pi->dpm_table.mvdd_table.dpm_levels[i].enabled = true; - } - pi->dpm_table.mvdd_table.count = allowed_mclk_table->count; - } - - ci_setup_default_pcie_tables(adev); - - /* save a copy of the default DPM table */ - memcpy(&(pi->golden_dpm_table), &(pi->dpm_table), - sizeof(struct ci_dpm_table)); - - return 0; -} - -static int ci_find_boot_level(struct ci_single_dpm_table *table, - u32 value, u32 *boot_level) -{ - u32 i; - int ret = -EINVAL; - - for(i = 0; i < table->count; i++) { - if (value == table->dpm_levels[i].value) { - *boot_level = i; - ret = 0; - } - } - - return ret; -} - -static int ci_init_smc_table(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ulv_parm *ulv = &pi->ulv; - struct amdgpu_ps *amdgpu_boot_state = adev->pm.dpm.boot_ps; - SMU7_Discrete_DpmTable *table = &pi->smc_state_table; - int ret; - - ret = ci_setup_default_dpm_tables(adev); - if (ret) - return ret; - - if (pi->voltage_control != CISLANDS_VOLTAGE_CONTROL_NONE) - ci_populate_smc_voltage_tables(adev, table); - - ci_init_fps_limits(adev); - - if (adev->pm.dpm.platform_caps & ATOM_PP_PLATFORM_CAP_HARDWAREDC) - table->SystemFlags |= PPSMC_SYSTEMFLAG_GPIO_DC; - - if (adev->pm.dpm.platform_caps & ATOM_PP_PLATFORM_CAP_STEPVDDC) - table->SystemFlags |= PPSMC_SYSTEMFLAG_STEPVDDC; - - if (adev->gmc.vram_type == AMDGPU_VRAM_TYPE_GDDR5) - table->SystemFlags |= PPSMC_SYSTEMFLAG_GDDR5; - - if (ulv->supported) { - ret = ci_populate_ulv_state(adev, &pi->smc_state_table.Ulv); - if (ret) - return ret; - WREG32_SMC(ixCG_ULV_PARAMETER, ulv->cg_ulv_parameter); - } - - ret = ci_populate_all_graphic_levels(adev); - if (ret) - return ret; - - ret = ci_populate_all_memory_levels(adev); - if (ret) - return ret; - - ci_populate_smc_link_level(adev, table); - - ret = ci_populate_smc_acpi_level(adev, table); - if (ret) - return ret; - - ret = ci_populate_smc_vce_level(adev, table); - if (ret) - return ret; - - ret = ci_populate_smc_acp_level(adev, table); - if (ret) - return ret; - - ret = ci_populate_smc_samu_level(adev, table); - if (ret) - return ret; - - ret = ci_do_program_memory_timing_parameters(adev); - if (ret) - return ret; - - ret = ci_populate_smc_uvd_level(adev, table); - if (ret) - return ret; - - table->UvdBootLevel = 0; - table->VceBootLevel = 0; - table->AcpBootLevel = 0; - table->SamuBootLevel = 0; - table->GraphicsBootLevel = 0; - table->MemoryBootLevel = 0; - - ret = ci_find_boot_level(&pi->dpm_table.sclk_table, - pi->vbios_boot_state.sclk_bootup_value, - (u32 *)&pi->smc_state_table.GraphicsBootLevel); - - ret = ci_find_boot_level(&pi->dpm_table.mclk_table, - pi->vbios_boot_state.mclk_bootup_value, - (u32 *)&pi->smc_state_table.MemoryBootLevel); - - table->BootVddc = pi->vbios_boot_state.vddc_bootup_value; - table->BootVddci = pi->vbios_boot_state.vddci_bootup_value; - table->BootMVdd = pi->vbios_boot_state.mvdd_bootup_value; - - ci_populate_smc_initial_state(adev, amdgpu_boot_state); - - ret = ci_populate_bapm_parameters_in_dpm_table(adev); - if (ret) - return ret; - - table->UVDInterval = 1; - table->VCEInterval = 1; - table->ACPInterval = 1; - table->SAMUInterval = 1; - table->GraphicsVoltageChangeEnable = 1; - table->GraphicsThermThrottleEnable = 1; - table->GraphicsInterval = 1; - table->VoltageInterval = 1; - table->ThermalInterval = 1; - table->TemperatureLimitHigh = (u16)((pi->thermal_temp_setting.temperature_high * - CISLANDS_Q88_FORMAT_CONVERSION_UNIT) / 1000); - table->TemperatureLimitLow = (u16)((pi->thermal_temp_setting.temperature_low * - CISLANDS_Q88_FORMAT_CONVERSION_UNIT) / 1000); - table->MemoryVoltageChangeEnable = 1; - table->MemoryInterval = 1; - table->VoltageResponseTime = 0; - table->VddcVddciDelta = 4000; - table->PhaseResponseTime = 0; - table->MemoryThermThrottleEnable = 1; - table->PCIeBootLinkLevel = pi->dpm_table.pcie_speed_table.count - 1; - table->PCIeGenInterval = 1; - if (pi->voltage_control == CISLANDS_VOLTAGE_CONTROL_BY_SVID2) - table->SVI2Enable = 1; - else - table->SVI2Enable = 0; - - table->ThermGpio = 17; - table->SclkStepSize = 0x4000; - - table->SystemFlags = cpu_to_be32(table->SystemFlags); - table->SmioMaskVddcVid = cpu_to_be32(table->SmioMaskVddcVid); - table->SmioMaskVddcPhase = cpu_to_be32(table->SmioMaskVddcPhase); - table->SmioMaskVddciVid = cpu_to_be32(table->SmioMaskVddciVid); - table->SmioMaskMvddVid = cpu_to_be32(table->SmioMaskMvddVid); - table->SclkStepSize = cpu_to_be32(table->SclkStepSize); - table->TemperatureLimitHigh = cpu_to_be16(table->TemperatureLimitHigh); - table->TemperatureLimitLow = cpu_to_be16(table->TemperatureLimitLow); - table->VddcVddciDelta = cpu_to_be16(table->VddcVddciDelta); - table->VoltageResponseTime = cpu_to_be16(table->VoltageResponseTime); - table->PhaseResponseTime = cpu_to_be16(table->PhaseResponseTime); - table->BootVddc = cpu_to_be16(table->BootVddc * VOLTAGE_SCALE); - table->BootVddci = cpu_to_be16(table->BootVddci * VOLTAGE_SCALE); - table->BootMVdd = cpu_to_be16(table->BootMVdd * VOLTAGE_SCALE); - - ret = amdgpu_ci_copy_bytes_to_smc(adev, - pi->dpm_table_start + - offsetof(SMU7_Discrete_DpmTable, SystemFlags), - (u8 *)&table->SystemFlags, - sizeof(SMU7_Discrete_DpmTable) - 3 * sizeof(SMU7_PIDController), - pi->sram_end); - if (ret) - return ret; - - return 0; -} - -static void ci_trim_single_dpm_states(struct amdgpu_device *adev, - struct ci_single_dpm_table *dpm_table, - u32 low_limit, u32 high_limit) -{ - u32 i; - - for (i = 0; i < dpm_table->count; i++) { - if ((dpm_table->dpm_levels[i].value < low_limit) || - (dpm_table->dpm_levels[i].value > high_limit)) - dpm_table->dpm_levels[i].enabled = false; - else - dpm_table->dpm_levels[i].enabled = true; - } -} - -static void ci_trim_pcie_dpm_states(struct amdgpu_device *adev, - u32 speed_low, u32 lanes_low, - u32 speed_high, u32 lanes_high) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_single_dpm_table *pcie_table = &pi->dpm_table.pcie_speed_table; - u32 i, j; - - for (i = 0; i < pcie_table->count; i++) { - if ((pcie_table->dpm_levels[i].value < speed_low) || - (pcie_table->dpm_levels[i].param1 < lanes_low) || - (pcie_table->dpm_levels[i].value > speed_high) || - (pcie_table->dpm_levels[i].param1 > lanes_high)) - pcie_table->dpm_levels[i].enabled = false; - else - pcie_table->dpm_levels[i].enabled = true; - } - - for (i = 0; i < pcie_table->count; i++) { - if (pcie_table->dpm_levels[i].enabled) { - for (j = i + 1; j < pcie_table->count; j++) { - if (pcie_table->dpm_levels[j].enabled) { - if ((pcie_table->dpm_levels[i].value == pcie_table->dpm_levels[j].value) && - (pcie_table->dpm_levels[i].param1 == pcie_table->dpm_levels[j].param1)) - pcie_table->dpm_levels[j].enabled = false; - } - } - } - } -} - -static int ci_trim_dpm_states(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_state) -{ - struct ci_ps *state = ci_get_ps(amdgpu_state); - struct ci_power_info *pi = ci_get_pi(adev); - u32 high_limit_count; - - if (state->performance_level_count < 1) - return -EINVAL; - - if (state->performance_level_count == 1) - high_limit_count = 0; - else - high_limit_count = 1; - - ci_trim_single_dpm_states(adev, - &pi->dpm_table.sclk_table, - state->performance_levels[0].sclk, - state->performance_levels[high_limit_count].sclk); - - ci_trim_single_dpm_states(adev, - &pi->dpm_table.mclk_table, - state->performance_levels[0].mclk, - state->performance_levels[high_limit_count].mclk); - - ci_trim_pcie_dpm_states(adev, - state->performance_levels[0].pcie_gen, - state->performance_levels[0].pcie_lane, - state->performance_levels[high_limit_count].pcie_gen, - state->performance_levels[high_limit_count].pcie_lane); - - return 0; -} - -static int ci_apply_disp_minimum_voltage_request(struct amdgpu_device *adev) -{ - struct amdgpu_clock_voltage_dependency_table *disp_voltage_table = - &adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk; - struct amdgpu_clock_voltage_dependency_table *vddc_table = - &adev->pm.dpm.dyn_state.vddc_dependency_on_sclk; - u32 requested_voltage = 0; - u32 i; - - if (disp_voltage_table == NULL) - return -EINVAL; - if (!disp_voltage_table->count) - return -EINVAL; - - for (i = 0; i < disp_voltage_table->count; i++) { - if (adev->clock.current_dispclk == disp_voltage_table->entries[i].clk) - requested_voltage = disp_voltage_table->entries[i].v; - } - - for (i = 0; i < vddc_table->count; i++) { - if (requested_voltage <= vddc_table->entries[i].v) { - requested_voltage = vddc_table->entries[i].v; - return (amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_VddC_Request, - requested_voltage * VOLTAGE_SCALE) == PPSMC_Result_OK) ? - 0 : -EINVAL; - } - } - - return -EINVAL; -} - -static int ci_upload_dpm_level_enable_mask(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - PPSMC_Result result; - - ci_apply_disp_minimum_voltage_request(adev); - - if (!pi->sclk_dpm_key_disabled) { - if (pi->dpm_level_enable_mask.sclk_dpm_enable_mask) { - result = amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_SCLKDPM_SetEnabledMask, - pi->dpm_level_enable_mask.sclk_dpm_enable_mask); - if (result != PPSMC_Result_OK) - return -EINVAL; - } - } - - if (!pi->mclk_dpm_key_disabled) { - if (pi->dpm_level_enable_mask.mclk_dpm_enable_mask) { - result = amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_MCLKDPM_SetEnabledMask, - pi->dpm_level_enable_mask.mclk_dpm_enable_mask); - if (result != PPSMC_Result_OK) - return -EINVAL; - } - } - -#if 0 - if (!pi->pcie_dpm_key_disabled) { - if (pi->dpm_level_enable_mask.pcie_dpm_enable_mask) { - result = amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_PCIeDPM_SetEnabledMask, - pi->dpm_level_enable_mask.pcie_dpm_enable_mask); - if (result != PPSMC_Result_OK) - return -EINVAL; - } - } -#endif - - return 0; -} - -static void ci_find_dpm_states_clocks_in_dpm_table(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_state) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ps *state = ci_get_ps(amdgpu_state); - struct ci_single_dpm_table *sclk_table = &pi->dpm_table.sclk_table; - u32 sclk = state->performance_levels[state->performance_level_count-1].sclk; - struct ci_single_dpm_table *mclk_table = &pi->dpm_table.mclk_table; - u32 mclk = state->performance_levels[state->performance_level_count-1].mclk; - u32 i; - - pi->need_update_smu7_dpm_table = 0; - - for (i = 0; i < sclk_table->count; i++) { - if (sclk == sclk_table->dpm_levels[i].value) - break; - } - - if (i >= sclk_table->count) { - pi->need_update_smu7_dpm_table |= DPMTABLE_OD_UPDATE_SCLK; - } else { - /* XXX check display min clock requirements */ - if (CISLAND_MINIMUM_ENGINE_CLOCK != CISLAND_MINIMUM_ENGINE_CLOCK) - pi->need_update_smu7_dpm_table |= DPMTABLE_UPDATE_SCLK; - } - - for (i = 0; i < mclk_table->count; i++) { - if (mclk == mclk_table->dpm_levels[i].value) - break; - } - - if (i >= mclk_table->count) - pi->need_update_smu7_dpm_table |= DPMTABLE_OD_UPDATE_MCLK; - - if (adev->pm.dpm.current_active_crtc_count != - adev->pm.dpm.new_active_crtc_count) - pi->need_update_smu7_dpm_table |= DPMTABLE_UPDATE_MCLK; -} - -static int ci_populate_and_upload_sclk_mclk_dpm_levels(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_state) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ps *state = ci_get_ps(amdgpu_state); - u32 sclk = state->performance_levels[state->performance_level_count-1].sclk; - u32 mclk = state->performance_levels[state->performance_level_count-1].mclk; - struct ci_dpm_table *dpm_table = &pi->dpm_table; - int ret; - - if (!pi->need_update_smu7_dpm_table) - return 0; - - if (pi->need_update_smu7_dpm_table & DPMTABLE_OD_UPDATE_SCLK) - dpm_table->sclk_table.dpm_levels[dpm_table->sclk_table.count-1].value = sclk; - - if (pi->need_update_smu7_dpm_table & DPMTABLE_OD_UPDATE_MCLK) - dpm_table->mclk_table.dpm_levels[dpm_table->mclk_table.count-1].value = mclk; - - if (pi->need_update_smu7_dpm_table & (DPMTABLE_OD_UPDATE_SCLK | DPMTABLE_UPDATE_SCLK)) { - ret = ci_populate_all_graphic_levels(adev); - if (ret) - return ret; - } - - if (pi->need_update_smu7_dpm_table & (DPMTABLE_OD_UPDATE_MCLK | DPMTABLE_UPDATE_MCLK)) { - ret = ci_populate_all_memory_levels(adev); - if (ret) - return ret; - } - - return 0; -} - -static int ci_enable_uvd_dpm(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - const struct amdgpu_clock_and_voltage_limits *max_limits; - int i; - - if (adev->pm.ac_power) - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_ac; - else - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_dc; - - if (enable) { - pi->dpm_level_enable_mask.uvd_dpm_enable_mask = 0; - - for (i = adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.count - 1; i >= 0; i--) { - if (adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries[i].v <= max_limits->vddc) { - pi->dpm_level_enable_mask.uvd_dpm_enable_mask |= 1 << i; - - if (!pi->caps_uvd_dpm) - break; - } - } - - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_UVDDPM_SetEnabledMask, - pi->dpm_level_enable_mask.uvd_dpm_enable_mask); - - if (pi->last_mclk_dpm_enable_mask & 0x1) { - pi->uvd_enabled = true; - pi->dpm_level_enable_mask.mclk_dpm_enable_mask &= 0xFFFFFFFE; - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_MCLKDPM_SetEnabledMask, - pi->dpm_level_enable_mask.mclk_dpm_enable_mask); - } - } else { - if (pi->uvd_enabled) { - pi->uvd_enabled = false; - pi->dpm_level_enable_mask.mclk_dpm_enable_mask |= 1; - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_MCLKDPM_SetEnabledMask, - pi->dpm_level_enable_mask.mclk_dpm_enable_mask); - } - } - - return (amdgpu_ci_send_msg_to_smc(adev, enable ? - PPSMC_MSG_UVDDPM_Enable : PPSMC_MSG_UVDDPM_Disable) == PPSMC_Result_OK) ? - 0 : -EINVAL; -} - -static int ci_enable_vce_dpm(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - const struct amdgpu_clock_and_voltage_limits *max_limits; - int i; - - if (adev->pm.ac_power) - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_ac; - else - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_dc; - - if (enable) { - pi->dpm_level_enable_mask.vce_dpm_enable_mask = 0; - for (i = adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.count - 1; i >= 0; i--) { - if (adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.entries[i].v <= max_limits->vddc) { - pi->dpm_level_enable_mask.vce_dpm_enable_mask |= 1 << i; - - if (!pi->caps_vce_dpm) - break; - } - } - - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_VCEDPM_SetEnabledMask, - pi->dpm_level_enable_mask.vce_dpm_enable_mask); - } - - return (amdgpu_ci_send_msg_to_smc(adev, enable ? - PPSMC_MSG_VCEDPM_Enable : PPSMC_MSG_VCEDPM_Disable) == PPSMC_Result_OK) ? - 0 : -EINVAL; -} - -#if 0 -static int ci_enable_samu_dpm(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - const struct amdgpu_clock_and_voltage_limits *max_limits; - int i; - - if (adev->pm.ac_power) - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_ac; - else - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_dc; - - if (enable) { - pi->dpm_level_enable_mask.samu_dpm_enable_mask = 0; - for (i = adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.count - 1; i >= 0; i--) { - if (adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.entries[i].v <= max_limits->vddc) { - pi->dpm_level_enable_mask.samu_dpm_enable_mask |= 1 << i; - - if (!pi->caps_samu_dpm) - break; - } - } - - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_SAMUDPM_SetEnabledMask, - pi->dpm_level_enable_mask.samu_dpm_enable_mask); - } - return (amdgpu_ci_send_msg_to_smc(adev, enable ? - PPSMC_MSG_SAMUDPM_Enable : PPSMC_MSG_SAMUDPM_Disable) == PPSMC_Result_OK) ? - 0 : -EINVAL; -} - -static int ci_enable_acp_dpm(struct amdgpu_device *adev, bool enable) -{ - struct ci_power_info *pi = ci_get_pi(adev); - const struct amdgpu_clock_and_voltage_limits *max_limits; - int i; - - if (adev->pm.ac_power) - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_ac; - else - max_limits = &adev->pm.dpm.dyn_state.max_clock_voltage_on_dc; - - if (enable) { - pi->dpm_level_enable_mask.acp_dpm_enable_mask = 0; - for (i = adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.count - 1; i >= 0; i--) { - if (adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.entries[i].v <= max_limits->vddc) { - pi->dpm_level_enable_mask.acp_dpm_enable_mask |= 1 << i; - - if (!pi->caps_acp_dpm) - break; - } - } - - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_ACPDPM_SetEnabledMask, - pi->dpm_level_enable_mask.acp_dpm_enable_mask); - } - - return (amdgpu_ci_send_msg_to_smc(adev, enable ? - PPSMC_MSG_ACPDPM_Enable : PPSMC_MSG_ACPDPM_Disable) == PPSMC_Result_OK) ? - 0 : -EINVAL; -} -#endif - -static int ci_update_uvd_dpm(struct amdgpu_device *adev, bool gate) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 tmp; - int ret = 0; - - if (!gate) { - /* turn the clocks on when decoding */ - if (pi->caps_uvd_dpm || - (adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.count <= 0)) - pi->smc_state_table.UvdBootLevel = 0; - else - pi->smc_state_table.UvdBootLevel = - adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.count - 1; - - tmp = RREG32_SMC(ixDPM_TABLE_475); - tmp &= ~DPM_TABLE_475__UvdBootLevel_MASK; - tmp |= (pi->smc_state_table.UvdBootLevel << DPM_TABLE_475__UvdBootLevel__SHIFT); - WREG32_SMC(ixDPM_TABLE_475, tmp); - ret = ci_enable_uvd_dpm(adev, true); - } else { - ret = ci_enable_uvd_dpm(adev, false); - if (ret) - return ret; - } - - return ret; -} - -static u8 ci_get_vce_boot_level(struct amdgpu_device *adev) -{ - u8 i; - u32 min_evclk = 30000; /* ??? */ - struct amdgpu_vce_clock_voltage_dependency_table *table = - &adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table; - - for (i = 0; i < table->count; i++) { - if (table->entries[i].evclk >= min_evclk) - return i; - } - - return table->count - 1; -} - -static int ci_update_vce_dpm(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_new_state, - struct amdgpu_ps *amdgpu_current_state) -{ - struct ci_power_info *pi = ci_get_pi(adev); - int ret = 0; - u32 tmp; - - if (amdgpu_current_state->evclk != amdgpu_new_state->evclk) { - if (amdgpu_new_state->evclk) { - pi->smc_state_table.VceBootLevel = ci_get_vce_boot_level(adev); - tmp = RREG32_SMC(ixDPM_TABLE_475); - tmp &= ~DPM_TABLE_475__VceBootLevel_MASK; - tmp |= (pi->smc_state_table.VceBootLevel << DPM_TABLE_475__VceBootLevel__SHIFT); - WREG32_SMC(ixDPM_TABLE_475, tmp); - - ret = ci_enable_vce_dpm(adev, true); - } else { - ret = ci_enable_vce_dpm(adev, false); - if (ret) - return ret; - } - } - return ret; -} - -#if 0 -static int ci_update_samu_dpm(struct amdgpu_device *adev, bool gate) -{ - return ci_enable_samu_dpm(adev, gate); -} - -static int ci_update_acp_dpm(struct amdgpu_device *adev, bool gate) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 tmp; - - if (!gate) { - pi->smc_state_table.AcpBootLevel = 0; - - tmp = RREG32_SMC(ixDPM_TABLE_475); - tmp &= ~AcpBootLevel_MASK; - tmp |= AcpBootLevel(pi->smc_state_table.AcpBootLevel); - WREG32_SMC(ixDPM_TABLE_475, tmp); - } - - return ci_enable_acp_dpm(adev, !gate); -} -#endif - -static int ci_generate_dpm_level_enable_mask(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_state) -{ - struct ci_power_info *pi = ci_get_pi(adev); - int ret; - - ret = ci_trim_dpm_states(adev, amdgpu_state); - if (ret) - return ret; - - pi->dpm_level_enable_mask.sclk_dpm_enable_mask = - ci_get_dpm_level_enable_mask_value(&pi->dpm_table.sclk_table); - pi->dpm_level_enable_mask.mclk_dpm_enable_mask = - ci_get_dpm_level_enable_mask_value(&pi->dpm_table.mclk_table); - pi->last_mclk_dpm_enable_mask = - pi->dpm_level_enable_mask.mclk_dpm_enable_mask; - if (pi->uvd_enabled) { - if (pi->dpm_level_enable_mask.mclk_dpm_enable_mask & 1) - pi->dpm_level_enable_mask.mclk_dpm_enable_mask &= 0xFFFFFFFE; - } - pi->dpm_level_enable_mask.pcie_dpm_enable_mask = - ci_get_dpm_level_enable_mask_value(&pi->dpm_table.pcie_speed_table); - - return 0; -} - -static u32 ci_get_lowest_enabled_level(struct amdgpu_device *adev, - u32 level_mask) -{ - u32 level = 0; - - while ((level_mask & (1 << level)) == 0) - level++; - - return level; -} - - -static int ci_dpm_force_performance_level(void *handle, - enum amd_dpm_forced_level level) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - u32 tmp, levels, i; - int ret; - - if (level == AMD_DPM_FORCED_LEVEL_HIGH) { - if ((!pi->pcie_dpm_key_disabled) && - pi->dpm_level_enable_mask.pcie_dpm_enable_mask) { - levels = 0; - tmp = pi->dpm_level_enable_mask.pcie_dpm_enable_mask; - while (tmp >>= 1) - levels++; - if (levels) { - ret = ci_dpm_force_state_pcie(adev, level); - if (ret) - return ret; - for (i = 0; i < adev->usec_timeout; i++) { - tmp = (RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX_1) & - TARGET_AND_CURRENT_PROFILE_INDEX_1__CURR_PCIE_INDEX_MASK) >> - TARGET_AND_CURRENT_PROFILE_INDEX_1__CURR_PCIE_INDEX__SHIFT; - if (tmp == levels) - break; - udelay(1); - } - } - } - if ((!pi->sclk_dpm_key_disabled) && - pi->dpm_level_enable_mask.sclk_dpm_enable_mask) { - levels = 0; - tmp = pi->dpm_level_enable_mask.sclk_dpm_enable_mask; - while (tmp >>= 1) - levels++; - if (levels) { - ret = ci_dpm_force_state_sclk(adev, levels); - if (ret) - return ret; - for (i = 0; i < adev->usec_timeout; i++) { - tmp = (RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX) & - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_SCLK_INDEX_MASK) >> - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_SCLK_INDEX__SHIFT; - if (tmp == levels) - break; - udelay(1); - } - } - } - if ((!pi->mclk_dpm_key_disabled) && - pi->dpm_level_enable_mask.mclk_dpm_enable_mask) { - levels = 0; - tmp = pi->dpm_level_enable_mask.mclk_dpm_enable_mask; - while (tmp >>= 1) - levels++; - if (levels) { - ret = ci_dpm_force_state_mclk(adev, levels); - if (ret) - return ret; - for (i = 0; i < adev->usec_timeout; i++) { - tmp = (RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX) & - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_MCLK_INDEX_MASK) >> - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_MCLK_INDEX__SHIFT; - if (tmp == levels) - break; - udelay(1); - } - } - } - } else if (level == AMD_DPM_FORCED_LEVEL_LOW) { - if ((!pi->sclk_dpm_key_disabled) && - pi->dpm_level_enable_mask.sclk_dpm_enable_mask) { - levels = ci_get_lowest_enabled_level(adev, - pi->dpm_level_enable_mask.sclk_dpm_enable_mask); - ret = ci_dpm_force_state_sclk(adev, levels); - if (ret) - return ret; - for (i = 0; i < adev->usec_timeout; i++) { - tmp = (RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX) & - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_SCLK_INDEX_MASK) >> - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_SCLK_INDEX__SHIFT; - if (tmp == levels) - break; - udelay(1); - } - } - if ((!pi->mclk_dpm_key_disabled) && - pi->dpm_level_enable_mask.mclk_dpm_enable_mask) { - levels = ci_get_lowest_enabled_level(adev, - pi->dpm_level_enable_mask.mclk_dpm_enable_mask); - ret = ci_dpm_force_state_mclk(adev, levels); - if (ret) - return ret; - for (i = 0; i < adev->usec_timeout; i++) { - tmp = (RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX) & - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_MCLK_INDEX_MASK) >> - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_MCLK_INDEX__SHIFT; - if (tmp == levels) - break; - udelay(1); - } - } - if ((!pi->pcie_dpm_key_disabled) && - pi->dpm_level_enable_mask.pcie_dpm_enable_mask) { - levels = ci_get_lowest_enabled_level(adev, - pi->dpm_level_enable_mask.pcie_dpm_enable_mask); - ret = ci_dpm_force_state_pcie(adev, levels); - if (ret) - return ret; - for (i = 0; i < adev->usec_timeout; i++) { - tmp = (RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX_1) & - TARGET_AND_CURRENT_PROFILE_INDEX_1__CURR_PCIE_INDEX_MASK) >> - TARGET_AND_CURRENT_PROFILE_INDEX_1__CURR_PCIE_INDEX__SHIFT; - if (tmp == levels) - break; - udelay(1); - } - } - } else if (level == AMD_DPM_FORCED_LEVEL_AUTO) { - if (!pi->pcie_dpm_key_disabled) { - PPSMC_Result smc_result; - - smc_result = amdgpu_ci_send_msg_to_smc(adev, - PPSMC_MSG_PCIeDPM_UnForceLevel); - if (smc_result != PPSMC_Result_OK) - return -EINVAL; - } - ret = ci_upload_dpm_level_enable_mask(adev); - if (ret) - return ret; - } - - adev->pm.dpm.forced_level = level; - - return 0; -} - -static int ci_set_mc_special_registers(struct amdgpu_device *adev, - struct ci_mc_reg_table *table) -{ - u8 i, j, k; - u32 temp_reg; - - for (i = 0, j = table->last; i < table->last; i++) { - if (j >= SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE) - return -EINVAL; - switch(table->mc_reg_address[i].s1) { - case mmMC_SEQ_MISC1: - temp_reg = RREG32(mmMC_PMG_CMD_EMRS); - table->mc_reg_address[j].s1 = mmMC_PMG_CMD_EMRS; - table->mc_reg_address[j].s0 = mmMC_SEQ_PMG_CMD_EMRS_LP; - for (k = 0; k < table->num_entries; k++) { - table->mc_reg_table_entry[k].mc_data[j] = - ((temp_reg & 0xffff0000)) | ((table->mc_reg_table_entry[k].mc_data[i] & 0xffff0000) >> 16); - } - j++; - - if (j >= SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE) - return -EINVAL; - temp_reg = RREG32(mmMC_PMG_CMD_MRS); - table->mc_reg_address[j].s1 = mmMC_PMG_CMD_MRS; - table->mc_reg_address[j].s0 = mmMC_SEQ_PMG_CMD_MRS_LP; - for (k = 0; k < table->num_entries; k++) { - table->mc_reg_table_entry[k].mc_data[j] = - (temp_reg & 0xffff0000) | (table->mc_reg_table_entry[k].mc_data[i] & 0x0000ffff); - if (adev->gmc.vram_type != AMDGPU_VRAM_TYPE_GDDR5) - table->mc_reg_table_entry[k].mc_data[j] |= 0x100; - } - j++; - - if (adev->gmc.vram_type != AMDGPU_VRAM_TYPE_GDDR5) { - if (j >= SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE) - return -EINVAL; - table->mc_reg_address[j].s1 = mmMC_PMG_AUTO_CMD; - table->mc_reg_address[j].s0 = mmMC_PMG_AUTO_CMD; - for (k = 0; k < table->num_entries; k++) { - table->mc_reg_table_entry[k].mc_data[j] = - (table->mc_reg_table_entry[k].mc_data[i] & 0xffff0000) >> 16; - } - j++; - } - break; - case mmMC_SEQ_RESERVE_M: - temp_reg = RREG32(mmMC_PMG_CMD_MRS1); - table->mc_reg_address[j].s1 = mmMC_PMG_CMD_MRS1; - table->mc_reg_address[j].s0 = mmMC_SEQ_PMG_CMD_MRS1_LP; - for (k = 0; k < table->num_entries; k++) { - table->mc_reg_table_entry[k].mc_data[j] = - (temp_reg & 0xffff0000) | (table->mc_reg_table_entry[k].mc_data[i] & 0x0000ffff); - } - j++; - break; - default: - break; - } - - } - - table->last = j; - - return 0; -} - -static bool ci_check_s0_mc_reg_index(u16 in_reg, u16 *out_reg) -{ - bool result = true; - - switch(in_reg) { - case mmMC_SEQ_RAS_TIMING: - *out_reg = mmMC_SEQ_RAS_TIMING_LP; - break; - case mmMC_SEQ_DLL_STBY: - *out_reg = mmMC_SEQ_DLL_STBY_LP; - break; - case mmMC_SEQ_G5PDX_CMD0: - *out_reg = mmMC_SEQ_G5PDX_CMD0_LP; - break; - case mmMC_SEQ_G5PDX_CMD1: - *out_reg = mmMC_SEQ_G5PDX_CMD1_LP; - break; - case mmMC_SEQ_G5PDX_CTRL: - *out_reg = mmMC_SEQ_G5PDX_CTRL_LP; - break; - case mmMC_SEQ_CAS_TIMING: - *out_reg = mmMC_SEQ_CAS_TIMING_LP; - break; - case mmMC_SEQ_MISC_TIMING: - *out_reg = mmMC_SEQ_MISC_TIMING_LP; - break; - case mmMC_SEQ_MISC_TIMING2: - *out_reg = mmMC_SEQ_MISC_TIMING2_LP; - break; - case mmMC_SEQ_PMG_DVS_CMD: - *out_reg = mmMC_SEQ_PMG_DVS_CMD_LP; - break; - case mmMC_SEQ_PMG_DVS_CTL: - *out_reg = mmMC_SEQ_PMG_DVS_CTL_LP; - break; - case mmMC_SEQ_RD_CTL_D0: - *out_reg = mmMC_SEQ_RD_CTL_D0_LP; - break; - case mmMC_SEQ_RD_CTL_D1: - *out_reg = mmMC_SEQ_RD_CTL_D1_LP; - break; - case mmMC_SEQ_WR_CTL_D0: - *out_reg = mmMC_SEQ_WR_CTL_D0_LP; - break; - case mmMC_SEQ_WR_CTL_D1: - *out_reg = mmMC_SEQ_WR_CTL_D1_LP; - break; - case mmMC_PMG_CMD_EMRS: - *out_reg = mmMC_SEQ_PMG_CMD_EMRS_LP; - break; - case mmMC_PMG_CMD_MRS: - *out_reg = mmMC_SEQ_PMG_CMD_MRS_LP; - break; - case mmMC_PMG_CMD_MRS1: - *out_reg = mmMC_SEQ_PMG_CMD_MRS1_LP; - break; - case mmMC_SEQ_PMG_TIMING: - *out_reg = mmMC_SEQ_PMG_TIMING_LP; - break; - case mmMC_PMG_CMD_MRS2: - *out_reg = mmMC_SEQ_PMG_CMD_MRS2_LP; - break; - case mmMC_SEQ_WR_CTL_2: - *out_reg = mmMC_SEQ_WR_CTL_2_LP; - break; - default: - result = false; - break; - } - - return result; -} - -static void ci_set_valid_flag(struct ci_mc_reg_table *table) -{ - u8 i, j; - - for (i = 0; i < table->last; i++) { - for (j = 1; j < table->num_entries; j++) { - if (table->mc_reg_table_entry[j-1].mc_data[i] != - table->mc_reg_table_entry[j].mc_data[i]) { - table->valid_flag |= 1 << i; - break; - } - } - } -} - -static void ci_set_s0_mc_reg_index(struct ci_mc_reg_table *table) -{ - u32 i; - u16 address; - - for (i = 0; i < table->last; i++) { - table->mc_reg_address[i].s0 = - ci_check_s0_mc_reg_index(table->mc_reg_address[i].s1, &address) ? - address : table->mc_reg_address[i].s1; - } -} - -static int ci_copy_vbios_mc_reg_table(const struct atom_mc_reg_table *table, - struct ci_mc_reg_table *ci_table) -{ - u8 i, j; - - if (table->last > SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE) - return -EINVAL; - if (table->num_entries > MAX_AC_TIMING_ENTRIES) - return -EINVAL; - - for (i = 0; i < table->last; i++) - ci_table->mc_reg_address[i].s1 = table->mc_reg_address[i].s1; - - ci_table->last = table->last; - - for (i = 0; i < table->num_entries; i++) { - ci_table->mc_reg_table_entry[i].mclk_max = - table->mc_reg_table_entry[i].mclk_max; - for (j = 0; j < table->last; j++) - ci_table->mc_reg_table_entry[i].mc_data[j] = - table->mc_reg_table_entry[i].mc_data[j]; - } - ci_table->num_entries = table->num_entries; - - return 0; -} - -static int ci_register_patching_mc_seq(struct amdgpu_device *adev, - struct ci_mc_reg_table *table) -{ - u8 i, k; - u32 tmp; - bool patch; - - tmp = RREG32(mmMC_SEQ_MISC0); - patch = ((tmp & 0x0000f00) == 0x300) ? true : false; - - if (patch && - ((adev->pdev->device == 0x67B0) || - (adev->pdev->device == 0x67B1))) { - for (i = 0; i < table->last; i++) { - if (table->last >= SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE) - return -EINVAL; - switch (table->mc_reg_address[i].s1) { - case mmMC_SEQ_MISC1: - for (k = 0; k < table->num_entries; k++) { - if ((table->mc_reg_table_entry[k].mclk_max == 125000) || - (table->mc_reg_table_entry[k].mclk_max == 137500)) - table->mc_reg_table_entry[k].mc_data[i] = - (table->mc_reg_table_entry[k].mc_data[i] & 0xFFFFFFF8) | - 0x00000007; - } - break; - case mmMC_SEQ_WR_CTL_D0: - for (k = 0; k < table->num_entries; k++) { - if ((table->mc_reg_table_entry[k].mclk_max == 125000) || - (table->mc_reg_table_entry[k].mclk_max == 137500)) - table->mc_reg_table_entry[k].mc_data[i] = - (table->mc_reg_table_entry[k].mc_data[i] & 0xFFFF0F00) | - 0x0000D0DD; - } - break; - case mmMC_SEQ_WR_CTL_D1: - for (k = 0; k < table->num_entries; k++) { - if ((table->mc_reg_table_entry[k].mclk_max == 125000) || - (table->mc_reg_table_entry[k].mclk_max == 137500)) - table->mc_reg_table_entry[k].mc_data[i] = - (table->mc_reg_table_entry[k].mc_data[i] & 0xFFFF0F00) | - 0x0000D0DD; - } - break; - case mmMC_SEQ_WR_CTL_2: - for (k = 0; k < table->num_entries; k++) { - if ((table->mc_reg_table_entry[k].mclk_max == 125000) || - (table->mc_reg_table_entry[k].mclk_max == 137500)) - table->mc_reg_table_entry[k].mc_data[i] = 0; - } - break; - case mmMC_SEQ_CAS_TIMING: - for (k = 0; k < table->num_entries; k++) { - if (table->mc_reg_table_entry[k].mclk_max == 125000) - table->mc_reg_table_entry[k].mc_data[i] = - (table->mc_reg_table_entry[k].mc_data[i] & 0xFFE0FE0F) | - 0x000C0140; - else if (table->mc_reg_table_entry[k].mclk_max == 137500) - table->mc_reg_table_entry[k].mc_data[i] = - (table->mc_reg_table_entry[k].mc_data[i] & 0xFFE0FE0F) | - 0x000C0150; - } - break; - case mmMC_SEQ_MISC_TIMING: - for (k = 0; k < table->num_entries; k++) { - if (table->mc_reg_table_entry[k].mclk_max == 125000) - table->mc_reg_table_entry[k].mc_data[i] = - (table->mc_reg_table_entry[k].mc_data[i] & 0xFFFFFFE0) | - 0x00000030; - else if (table->mc_reg_table_entry[k].mclk_max == 137500) - table->mc_reg_table_entry[k].mc_data[i] = - (table->mc_reg_table_entry[k].mc_data[i] & 0xFFFFFFE0) | - 0x00000035; - } - break; - default: - break; - } - } - - WREG32(mmMC_SEQ_IO_DEBUG_INDEX, 3); - tmp = RREG32(mmMC_SEQ_IO_DEBUG_DATA); - tmp = (tmp & 0xFFF8FFFF) | (1 << 16); - WREG32(mmMC_SEQ_IO_DEBUG_INDEX, 3); - WREG32(mmMC_SEQ_IO_DEBUG_DATA, tmp); - } - - return 0; -} - -static int ci_initialize_mc_reg_table(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct atom_mc_reg_table *table; - struct ci_mc_reg_table *ci_table = &pi->mc_reg_table; - u8 module_index = ci_get_memory_module_index(adev); - int ret; - - table = kzalloc(sizeof(struct atom_mc_reg_table), GFP_KERNEL); - if (!table) - return -ENOMEM; - - WREG32(mmMC_SEQ_RAS_TIMING_LP, RREG32(mmMC_SEQ_RAS_TIMING)); - WREG32(mmMC_SEQ_CAS_TIMING_LP, RREG32(mmMC_SEQ_CAS_TIMING)); - WREG32(mmMC_SEQ_DLL_STBY_LP, RREG32(mmMC_SEQ_DLL_STBY)); - WREG32(mmMC_SEQ_G5PDX_CMD0_LP, RREG32(mmMC_SEQ_G5PDX_CMD0)); - WREG32(mmMC_SEQ_G5PDX_CMD1_LP, RREG32(mmMC_SEQ_G5PDX_CMD1)); - WREG32(mmMC_SEQ_G5PDX_CTRL_LP, RREG32(mmMC_SEQ_G5PDX_CTRL)); - WREG32(mmMC_SEQ_PMG_DVS_CMD_LP, RREG32(mmMC_SEQ_PMG_DVS_CMD)); - WREG32(mmMC_SEQ_PMG_DVS_CTL_LP, RREG32(mmMC_SEQ_PMG_DVS_CTL)); - WREG32(mmMC_SEQ_MISC_TIMING_LP, RREG32(mmMC_SEQ_MISC_TIMING)); - WREG32(mmMC_SEQ_MISC_TIMING2_LP, RREG32(mmMC_SEQ_MISC_TIMING2)); - WREG32(mmMC_SEQ_PMG_CMD_EMRS_LP, RREG32(mmMC_PMG_CMD_EMRS)); - WREG32(mmMC_SEQ_PMG_CMD_MRS_LP, RREG32(mmMC_PMG_CMD_MRS)); - WREG32(mmMC_SEQ_PMG_CMD_MRS1_LP, RREG32(mmMC_PMG_CMD_MRS1)); - WREG32(mmMC_SEQ_WR_CTL_D0_LP, RREG32(mmMC_SEQ_WR_CTL_D0)); - WREG32(mmMC_SEQ_WR_CTL_D1_LP, RREG32(mmMC_SEQ_WR_CTL_D1)); - WREG32(mmMC_SEQ_RD_CTL_D0_LP, RREG32(mmMC_SEQ_RD_CTL_D0)); - WREG32(mmMC_SEQ_RD_CTL_D1_LP, RREG32(mmMC_SEQ_RD_CTL_D1)); - WREG32(mmMC_SEQ_PMG_TIMING_LP, RREG32(mmMC_SEQ_PMG_TIMING)); - WREG32(mmMC_SEQ_PMG_CMD_MRS2_LP, RREG32(mmMC_PMG_CMD_MRS2)); - WREG32(mmMC_SEQ_WR_CTL_2_LP, RREG32(mmMC_SEQ_WR_CTL_2)); - - ret = amdgpu_atombios_init_mc_reg_table(adev, module_index, table); - if (ret) - goto init_mc_done; - - ret = ci_copy_vbios_mc_reg_table(table, ci_table); - if (ret) - goto init_mc_done; - - ci_set_s0_mc_reg_index(ci_table); - - ret = ci_register_patching_mc_seq(adev, ci_table); - if (ret) - goto init_mc_done; - - ret = ci_set_mc_special_registers(adev, ci_table); - if (ret) - goto init_mc_done; - - ci_set_valid_flag(ci_table); - -init_mc_done: - kfree(table); - - return ret; -} - -static int ci_populate_mc_reg_addresses(struct amdgpu_device *adev, - SMU7_Discrete_MCRegisters *mc_reg_table) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 i, j; - - for (i = 0, j = 0; j < pi->mc_reg_table.last; j++) { - if (pi->mc_reg_table.valid_flag & (1 << j)) { - if (i >= SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE) - return -EINVAL; - mc_reg_table->address[i].s0 = cpu_to_be16(pi->mc_reg_table.mc_reg_address[j].s0); - mc_reg_table->address[i].s1 = cpu_to_be16(pi->mc_reg_table.mc_reg_address[j].s1); - i++; - } - } - - mc_reg_table->last = (u8)i; - - return 0; -} - -static void ci_convert_mc_registers(const struct ci_mc_reg_entry *entry, - SMU7_Discrete_MCRegisterSet *data, - u32 num_entries, u32 valid_flag) -{ - u32 i, j; - - for (i = 0, j = 0; j < num_entries; j++) { - if (valid_flag & (1 << j)) { - data->value[i] = cpu_to_be32(entry->mc_data[j]); - i++; - } - } -} - -static void ci_convert_mc_reg_table_entry_to_smc(struct amdgpu_device *adev, - const u32 memory_clock, - SMU7_Discrete_MCRegisterSet *mc_reg_table_data) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 i = 0; - - for(i = 0; i < pi->mc_reg_table.num_entries; i++) { - if (memory_clock <= pi->mc_reg_table.mc_reg_table_entry[i].mclk_max) - break; - } - - if ((i == pi->mc_reg_table.num_entries) && (i > 0)) - --i; - - ci_convert_mc_registers(&pi->mc_reg_table.mc_reg_table_entry[i], - mc_reg_table_data, pi->mc_reg_table.last, - pi->mc_reg_table.valid_flag); -} - -static void ci_convert_mc_reg_table_to_smc(struct amdgpu_device *adev, - SMU7_Discrete_MCRegisters *mc_reg_table) -{ - struct ci_power_info *pi = ci_get_pi(adev); - u32 i; - - for (i = 0; i < pi->dpm_table.mclk_table.count; i++) - ci_convert_mc_reg_table_entry_to_smc(adev, - pi->dpm_table.mclk_table.dpm_levels[i].value, - &mc_reg_table->data[i]); -} - -static int ci_populate_initial_mc_reg_table(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - int ret; - - memset(&pi->smc_mc_reg_table, 0, sizeof(SMU7_Discrete_MCRegisters)); - - ret = ci_populate_mc_reg_addresses(adev, &pi->smc_mc_reg_table); - if (ret) - return ret; - ci_convert_mc_reg_table_to_smc(adev, &pi->smc_mc_reg_table); - - return amdgpu_ci_copy_bytes_to_smc(adev, - pi->mc_reg_table_start, - (u8 *)&pi->smc_mc_reg_table, - sizeof(SMU7_Discrete_MCRegisters), - pi->sram_end); -} - -static int ci_update_and_upload_mc_reg_table(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - - if (!(pi->need_update_smu7_dpm_table & DPMTABLE_OD_UPDATE_MCLK)) - return 0; - - memset(&pi->smc_mc_reg_table, 0, sizeof(SMU7_Discrete_MCRegisters)); - - ci_convert_mc_reg_table_to_smc(adev, &pi->smc_mc_reg_table); - - return amdgpu_ci_copy_bytes_to_smc(adev, - pi->mc_reg_table_start + - offsetof(SMU7_Discrete_MCRegisters, data[0]), - (u8 *)&pi->smc_mc_reg_table.data[0], - sizeof(SMU7_Discrete_MCRegisterSet) * - pi->dpm_table.mclk_table.count, - pi->sram_end); -} - -static void ci_enable_voltage_control(struct amdgpu_device *adev) -{ - u32 tmp = RREG32_SMC(ixGENERAL_PWRMGT); - - tmp |= GENERAL_PWRMGT__VOLT_PWRMGT_EN_MASK; - WREG32_SMC(ixGENERAL_PWRMGT, tmp); -} - -static enum amdgpu_pcie_gen ci_get_maximum_link_speed(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_state) -{ - struct ci_ps *state = ci_get_ps(amdgpu_state); - int i; - u16 pcie_speed, max_speed = 0; - - for (i = 0; i < state->performance_level_count; i++) { - pcie_speed = state->performance_levels[i].pcie_gen; - if (max_speed < pcie_speed) - max_speed = pcie_speed; - } - - return max_speed; -} - -static u16 ci_get_current_pcie_speed(struct amdgpu_device *adev) -{ - u32 speed_cntl = 0; - - speed_cntl = RREG32_PCIE(ixPCIE_LC_SPEED_CNTL) & - PCIE_LC_SPEED_CNTL__LC_CURRENT_DATA_RATE_MASK; - speed_cntl >>= PCIE_LC_SPEED_CNTL__LC_CURRENT_DATA_RATE__SHIFT; - - return (u16)speed_cntl; -} - -static int ci_get_current_pcie_lane_number(struct amdgpu_device *adev) -{ - u32 link_width = 0; - - link_width = RREG32_PCIE(ixPCIE_LC_LINK_WIDTH_CNTL) & - PCIE_LC_LINK_WIDTH_CNTL__LC_LINK_WIDTH_RD_MASK; - link_width >>= PCIE_LC_LINK_WIDTH_CNTL__LC_LINK_WIDTH_RD__SHIFT; - - switch (link_width) { - case 1: - return 1; - case 2: - return 2; - case 3: - return 4; - case 4: - return 8; - case 0: - case 6: - default: - return 16; - } -} - -static void ci_request_link_speed_change_before_state_change(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_new_state, - struct amdgpu_ps *amdgpu_current_state) -{ - struct ci_power_info *pi = ci_get_pi(adev); - enum amdgpu_pcie_gen target_link_speed = - ci_get_maximum_link_speed(adev, amdgpu_new_state); - enum amdgpu_pcie_gen current_link_speed; - - if (pi->force_pcie_gen == AMDGPU_PCIE_GEN_INVALID) - current_link_speed = ci_get_maximum_link_speed(adev, amdgpu_current_state); - else - current_link_speed = pi->force_pcie_gen; - - pi->force_pcie_gen = AMDGPU_PCIE_GEN_INVALID; - pi->pspp_notify_required = false; - if (target_link_speed > current_link_speed) { - switch (target_link_speed) { -#ifdef CONFIG_ACPI - case AMDGPU_PCIE_GEN3: - if (amdgpu_acpi_pcie_performance_request(adev, PCIE_PERF_REQ_PECI_GEN3, false) == 0) - break; - pi->force_pcie_gen = AMDGPU_PCIE_GEN2; - if (current_link_speed == AMDGPU_PCIE_GEN2) - break; - case AMDGPU_PCIE_GEN2: - if (amdgpu_acpi_pcie_performance_request(adev, PCIE_PERF_REQ_PECI_GEN2, false) == 0) - break; -#endif - default: - pi->force_pcie_gen = ci_get_current_pcie_speed(adev); - break; - } - } else { - if (target_link_speed < current_link_speed) - pi->pspp_notify_required = true; - } -} - -static void ci_notify_link_speed_change_after_state_change(struct amdgpu_device *adev, - struct amdgpu_ps *amdgpu_new_state, - struct amdgpu_ps *amdgpu_current_state) -{ - struct ci_power_info *pi = ci_get_pi(adev); - enum amdgpu_pcie_gen target_link_speed = - ci_get_maximum_link_speed(adev, amdgpu_new_state); - u8 request; - - if (pi->pspp_notify_required) { - if (target_link_speed == AMDGPU_PCIE_GEN3) - request = PCIE_PERF_REQ_PECI_GEN3; - else if (target_link_speed == AMDGPU_PCIE_GEN2) - request = PCIE_PERF_REQ_PECI_GEN2; - else - request = PCIE_PERF_REQ_PECI_GEN1; - - if ((request == PCIE_PERF_REQ_PECI_GEN1) && - (ci_get_current_pcie_speed(adev) > 0)) - return; - -#ifdef CONFIG_ACPI - amdgpu_acpi_pcie_performance_request(adev, request, false); -#endif - } -} - -static int ci_set_private_data_variables_based_on_pptable(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_clock_voltage_dependency_table *allowed_sclk_vddc_table = - &adev->pm.dpm.dyn_state.vddc_dependency_on_sclk; - struct amdgpu_clock_voltage_dependency_table *allowed_mclk_vddc_table = - &adev->pm.dpm.dyn_state.vddc_dependency_on_mclk; - struct amdgpu_clock_voltage_dependency_table *allowed_mclk_vddci_table = - &adev->pm.dpm.dyn_state.vddci_dependency_on_mclk; - - if (allowed_sclk_vddc_table == NULL) - return -EINVAL; - if (allowed_sclk_vddc_table->count < 1) - return -EINVAL; - if (allowed_mclk_vddc_table == NULL) - return -EINVAL; - if (allowed_mclk_vddc_table->count < 1) - return -EINVAL; - if (allowed_mclk_vddci_table == NULL) - return -EINVAL; - if (allowed_mclk_vddci_table->count < 1) - return -EINVAL; - - pi->min_vddc_in_pp_table = allowed_sclk_vddc_table->entries[0].v; - pi->max_vddc_in_pp_table = - allowed_sclk_vddc_table->entries[allowed_sclk_vddc_table->count - 1].v; - - pi->min_vddci_in_pp_table = allowed_mclk_vddci_table->entries[0].v; - pi->max_vddci_in_pp_table = - allowed_mclk_vddci_table->entries[allowed_mclk_vddci_table->count - 1].v; - - adev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk = - allowed_sclk_vddc_table->entries[allowed_sclk_vddc_table->count - 1].clk; - adev->pm.dpm.dyn_state.max_clock_voltage_on_ac.mclk = - allowed_mclk_vddc_table->entries[allowed_sclk_vddc_table->count - 1].clk; - adev->pm.dpm.dyn_state.max_clock_voltage_on_ac.vddc = - allowed_sclk_vddc_table->entries[allowed_sclk_vddc_table->count - 1].v; - adev->pm.dpm.dyn_state.max_clock_voltage_on_ac.vddci = - allowed_mclk_vddci_table->entries[allowed_mclk_vddci_table->count - 1].v; - - return 0; -} - -static void ci_patch_with_vddc_leakage(struct amdgpu_device *adev, u16 *vddc) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_leakage_voltage *leakage_table = &pi->vddc_leakage; - u32 leakage_index; - - for (leakage_index = 0; leakage_index < leakage_table->count; leakage_index++) { - if (leakage_table->leakage_id[leakage_index] == *vddc) { - *vddc = leakage_table->actual_voltage[leakage_index]; - break; - } - } -} - -static void ci_patch_with_vddci_leakage(struct amdgpu_device *adev, u16 *vddci) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_leakage_voltage *leakage_table = &pi->vddci_leakage; - u32 leakage_index; - - for (leakage_index = 0; leakage_index < leakage_table->count; leakage_index++) { - if (leakage_table->leakage_id[leakage_index] == *vddci) { - *vddci = leakage_table->actual_voltage[leakage_index]; - break; - } - } -} - -static void ci_patch_clock_voltage_dependency_table_with_vddc_leakage(struct amdgpu_device *adev, - struct amdgpu_clock_voltage_dependency_table *table) -{ - u32 i; - - if (table) { - for (i = 0; i < table->count; i++) - ci_patch_with_vddc_leakage(adev, &table->entries[i].v); - } -} - -static void ci_patch_clock_voltage_dependency_table_with_vddci_leakage(struct amdgpu_device *adev, - struct amdgpu_clock_voltage_dependency_table *table) -{ - u32 i; - - if (table) { - for (i = 0; i < table->count; i++) - ci_patch_with_vddci_leakage(adev, &table->entries[i].v); - } -} - -static void ci_patch_vce_clock_voltage_dependency_table_with_vddc_leakage(struct amdgpu_device *adev, - struct amdgpu_vce_clock_voltage_dependency_table *table) -{ - u32 i; - - if (table) { - for (i = 0; i < table->count; i++) - ci_patch_with_vddc_leakage(adev, &table->entries[i].v); - } -} - -static void ci_patch_uvd_clock_voltage_dependency_table_with_vddc_leakage(struct amdgpu_device *adev, - struct amdgpu_uvd_clock_voltage_dependency_table *table) -{ - u32 i; - - if (table) { - for (i = 0; i < table->count; i++) - ci_patch_with_vddc_leakage(adev, &table->entries[i].v); - } -} - -static void ci_patch_vddc_phase_shed_limit_table_with_vddc_leakage(struct amdgpu_device *adev, - struct amdgpu_phase_shedding_limits_table *table) -{ - u32 i; - - if (table) { - for (i = 0; i < table->count; i++) - ci_patch_with_vddc_leakage(adev, &table->entries[i].voltage); - } -} - -static void ci_patch_clock_voltage_limits_with_vddc_leakage(struct amdgpu_device *adev, - struct amdgpu_clock_and_voltage_limits *table) -{ - if (table) { - ci_patch_with_vddc_leakage(adev, (u16 *)&table->vddc); - ci_patch_with_vddci_leakage(adev, (u16 *)&table->vddci); - } -} - -static void ci_patch_cac_leakage_table_with_vddc_leakage(struct amdgpu_device *adev, - struct amdgpu_cac_leakage_table *table) -{ - u32 i; - - if (table) { - for (i = 0; i < table->count; i++) - ci_patch_with_vddc_leakage(adev, &table->entries[i].vddc); - } -} - -static void ci_patch_dependency_tables_with_leakage(struct amdgpu_device *adev) -{ - - ci_patch_clock_voltage_dependency_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.vddc_dependency_on_sclk); - ci_patch_clock_voltage_dependency_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.vddc_dependency_on_mclk); - ci_patch_clock_voltage_dependency_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk); - ci_patch_clock_voltage_dependency_table_with_vddci_leakage(adev, - &adev->pm.dpm.dyn_state.vddci_dependency_on_mclk); - ci_patch_vce_clock_voltage_dependency_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table); - ci_patch_uvd_clock_voltage_dependency_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table); - ci_patch_clock_voltage_dependency_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table); - ci_patch_clock_voltage_dependency_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table); - ci_patch_vddc_phase_shed_limit_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.phase_shedding_limits_table); - ci_patch_clock_voltage_limits_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.max_clock_voltage_on_ac); - ci_patch_clock_voltage_limits_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.max_clock_voltage_on_dc); - ci_patch_cac_leakage_table_with_vddc_leakage(adev, - &adev->pm.dpm.dyn_state.cac_leakage_table); - -} - -static void ci_update_current_ps(struct amdgpu_device *adev, - struct amdgpu_ps *rps) -{ - struct ci_ps *new_ps = ci_get_ps(rps); - struct ci_power_info *pi = ci_get_pi(adev); - - pi->current_rps = *rps; - pi->current_ps = *new_ps; - pi->current_rps.ps_priv = &pi->current_ps; - adev->pm.dpm.current_ps = &pi->current_rps; -} - -static void ci_update_requested_ps(struct amdgpu_device *adev, - struct amdgpu_ps *rps) -{ - struct ci_ps *new_ps = ci_get_ps(rps); - struct ci_power_info *pi = ci_get_pi(adev); - - pi->requested_rps = *rps; - pi->requested_ps = *new_ps; - pi->requested_rps.ps_priv = &pi->requested_ps; - adev->pm.dpm.requested_ps = &pi->requested_rps; -} - -static int ci_dpm_pre_set_power_state(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_ps requested_ps = *adev->pm.dpm.requested_ps; - struct amdgpu_ps *new_ps = &requested_ps; - - ci_update_requested_ps(adev, new_ps); - - ci_apply_state_adjust_rules(adev, &pi->requested_rps); - - return 0; -} - -static void ci_dpm_post_set_power_state(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_ps *new_ps = &pi->requested_rps; - - ci_update_current_ps(adev, new_ps); -} - - -static void ci_dpm_setup_asic(struct amdgpu_device *adev) -{ - ci_read_clock_registers(adev); - ci_enable_acpi_power_management(adev); - ci_init_sclk_t(adev); -} - -static int ci_dpm_enable(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_ps *boot_ps = adev->pm.dpm.boot_ps; - int ret; - - if (pi->voltage_control != CISLANDS_VOLTAGE_CONTROL_NONE) { - ci_enable_voltage_control(adev); - ret = ci_construct_voltage_tables(adev); - if (ret) { - DRM_ERROR("ci_construct_voltage_tables failed\n"); - return ret; - } - } - if (pi->caps_dynamic_ac_timing) { - ret = ci_initialize_mc_reg_table(adev); - if (ret) - pi->caps_dynamic_ac_timing = false; - } - if (pi->dynamic_ss) - ci_enable_spread_spectrum(adev, true); - if (pi->thermal_protection) - ci_enable_thermal_protection(adev, true); - ci_program_sstp(adev); - ci_enable_display_gap(adev); - ci_program_vc(adev); - ret = ci_upload_firmware(adev); - if (ret) { - DRM_ERROR("ci_upload_firmware failed\n"); - return ret; - } - ret = ci_process_firmware_header(adev); - if (ret) { - DRM_ERROR("ci_process_firmware_header failed\n"); - return ret; - } - ret = ci_initial_switch_from_arb_f0_to_f1(adev); - if (ret) { - DRM_ERROR("ci_initial_switch_from_arb_f0_to_f1 failed\n"); - return ret; - } - ret = ci_init_smc_table(adev); - if (ret) { - DRM_ERROR("ci_init_smc_table failed\n"); - return ret; - } - ret = ci_init_arb_table_index(adev); - if (ret) { - DRM_ERROR("ci_init_arb_table_index failed\n"); - return ret; - } - if (pi->caps_dynamic_ac_timing) { - ret = ci_populate_initial_mc_reg_table(adev); - if (ret) { - DRM_ERROR("ci_populate_initial_mc_reg_table failed\n"); - return ret; - } - } - ret = ci_populate_pm_base(adev); - if (ret) { - DRM_ERROR("ci_populate_pm_base failed\n"); - return ret; - } - ci_dpm_start_smc(adev); - ci_enable_vr_hot_gpio_interrupt(adev); - ret = ci_notify_smc_display_change(adev, false); - if (ret) { - DRM_ERROR("ci_notify_smc_display_change failed\n"); - return ret; - } - ci_enable_sclk_control(adev, true); - ret = ci_enable_ulv(adev, true); - if (ret) { - DRM_ERROR("ci_enable_ulv failed\n"); - return ret; - } - ret = ci_enable_ds_master_switch(adev, true); - if (ret) { - DRM_ERROR("ci_enable_ds_master_switch failed\n"); - return ret; - } - ret = ci_start_dpm(adev); - if (ret) { - DRM_ERROR("ci_start_dpm failed\n"); - return ret; - } - ret = ci_enable_didt(adev, true); - if (ret) { - DRM_ERROR("ci_enable_didt failed\n"); - return ret; - } - ret = ci_enable_smc_cac(adev, true); - if (ret) { - DRM_ERROR("ci_enable_smc_cac failed\n"); - return ret; - } - ret = ci_enable_power_containment(adev, true); - if (ret) { - DRM_ERROR("ci_enable_power_containment failed\n"); - return ret; - } - - ret = ci_power_control_set_level(adev); - if (ret) { - DRM_ERROR("ci_power_control_set_level failed\n"); - return ret; - } - - ci_enable_auto_throttle_source(adev, AMDGPU_DPM_AUTO_THROTTLE_SRC_THERMAL, true); - - ret = ci_enable_thermal_based_sclk_dpm(adev, true); - if (ret) { - DRM_ERROR("ci_enable_thermal_based_sclk_dpm failed\n"); - return ret; - } - - ci_thermal_start_thermal_controller(adev); - - ci_update_current_ps(adev, boot_ps); - - return 0; -} - -static void ci_dpm_disable(struct amdgpu_device *adev) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_ps *boot_ps = adev->pm.dpm.boot_ps; - - amdgpu_irq_put(adev, &adev->pm.dpm.thermal.irq, - AMDGPU_THERMAL_IRQ_LOW_TO_HIGH); - amdgpu_irq_put(adev, &adev->pm.dpm.thermal.irq, - AMDGPU_THERMAL_IRQ_HIGH_TO_LOW); - - ci_dpm_powergate_uvd(adev, true); - - if (!amdgpu_ci_is_smc_running(adev)) - return; - - ci_thermal_stop_thermal_controller(adev); - - if (pi->thermal_protection) - ci_enable_thermal_protection(adev, false); - ci_enable_power_containment(adev, false); - ci_enable_smc_cac(adev, false); - ci_enable_didt(adev, false); - ci_enable_spread_spectrum(adev, false); - ci_enable_auto_throttle_source(adev, AMDGPU_DPM_AUTO_THROTTLE_SRC_THERMAL, false); - ci_stop_dpm(adev); - ci_enable_ds_master_switch(adev, false); - ci_enable_ulv(adev, false); - ci_clear_vc(adev); - ci_reset_to_default(adev); - ci_dpm_stop_smc(adev); - ci_force_switch_to_arb_f0(adev); - ci_enable_thermal_based_sclk_dpm(adev, false); - - ci_update_current_ps(adev, boot_ps); -} - -static int ci_dpm_set_power_state(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_ps *new_ps = &pi->requested_rps; - struct amdgpu_ps *old_ps = &pi->current_rps; - int ret; - - ci_find_dpm_states_clocks_in_dpm_table(adev, new_ps); - if (pi->pcie_performance_request) - ci_request_link_speed_change_before_state_change(adev, new_ps, old_ps); - ret = ci_freeze_sclk_mclk_dpm(adev); - if (ret) { - DRM_ERROR("ci_freeze_sclk_mclk_dpm failed\n"); - return ret; - } - ret = ci_populate_and_upload_sclk_mclk_dpm_levels(adev, new_ps); - if (ret) { - DRM_ERROR("ci_populate_and_upload_sclk_mclk_dpm_levels failed\n"); - return ret; - } - ret = ci_generate_dpm_level_enable_mask(adev, new_ps); - if (ret) { - DRM_ERROR("ci_generate_dpm_level_enable_mask failed\n"); - return ret; - } - - ret = ci_update_vce_dpm(adev, new_ps, old_ps); - if (ret) { - DRM_ERROR("ci_update_vce_dpm failed\n"); - return ret; - } - - ret = ci_update_sclk_t(adev); - if (ret) { - DRM_ERROR("ci_update_sclk_t failed\n"); - return ret; - } - if (pi->caps_dynamic_ac_timing) { - ret = ci_update_and_upload_mc_reg_table(adev); - if (ret) { - DRM_ERROR("ci_update_and_upload_mc_reg_table failed\n"); - return ret; - } - } - ret = ci_program_memory_timing_parameters(adev); - if (ret) { - DRM_ERROR("ci_program_memory_timing_parameters failed\n"); - return ret; - } - ret = ci_unfreeze_sclk_mclk_dpm(adev); - if (ret) { - DRM_ERROR("ci_unfreeze_sclk_mclk_dpm failed\n"); - return ret; - } - ret = ci_upload_dpm_level_enable_mask(adev); - if (ret) { - DRM_ERROR("ci_upload_dpm_level_enable_mask failed\n"); - return ret; - } - if (pi->pcie_performance_request) - ci_notify_link_speed_change_after_state_change(adev, new_ps, old_ps); - - return 0; -} - -#if 0 -static void ci_dpm_reset_asic(struct amdgpu_device *adev) -{ - ci_set_boot_state(adev); -} -#endif - -static void ci_dpm_display_configuration_changed(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - ci_program_display_gap(adev); -} - -union power_info { - struct _ATOM_POWERPLAY_INFO info; - struct _ATOM_POWERPLAY_INFO_V2 info_2; - struct _ATOM_POWERPLAY_INFO_V3 info_3; - struct _ATOM_PPLIB_POWERPLAYTABLE pplib; - struct _ATOM_PPLIB_POWERPLAYTABLE2 pplib2; - struct _ATOM_PPLIB_POWERPLAYTABLE3 pplib3; -}; - -union pplib_clock_info { - struct _ATOM_PPLIB_R600_CLOCK_INFO r600; - struct _ATOM_PPLIB_RS780_CLOCK_INFO rs780; - struct _ATOM_PPLIB_EVERGREEN_CLOCK_INFO evergreen; - struct _ATOM_PPLIB_SUMO_CLOCK_INFO sumo; - struct _ATOM_PPLIB_SI_CLOCK_INFO si; - struct _ATOM_PPLIB_CI_CLOCK_INFO ci; -}; - -union pplib_power_state { - struct _ATOM_PPLIB_STATE v1; - struct _ATOM_PPLIB_STATE_V2 v2; -}; - -static void ci_parse_pplib_non_clock_info(struct amdgpu_device *adev, - struct amdgpu_ps *rps, - struct _ATOM_PPLIB_NONCLOCK_INFO *non_clock_info, - u8 table_rev) -{ - rps->caps = le32_to_cpu(non_clock_info->ulCapsAndSettings); - rps->class = le16_to_cpu(non_clock_info->usClassification); - rps->class2 = le16_to_cpu(non_clock_info->usClassification2); - - if (ATOM_PPLIB_NONCLOCKINFO_VER1 < table_rev) { - rps->vclk = le32_to_cpu(non_clock_info->ulVCLK); - rps->dclk = le32_to_cpu(non_clock_info->ulDCLK); - } else { - rps->vclk = 0; - rps->dclk = 0; - } - - if (rps->class & ATOM_PPLIB_CLASSIFICATION_BOOT) - adev->pm.dpm.boot_ps = rps; - if (rps->class & ATOM_PPLIB_CLASSIFICATION_UVDSTATE) - adev->pm.dpm.uvd_ps = rps; -} - -static void ci_parse_pplib_clock_info(struct amdgpu_device *adev, - struct amdgpu_ps *rps, int index, - union pplib_clock_info *clock_info) -{ - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ps *ps = ci_get_ps(rps); - struct ci_pl *pl = &ps->performance_levels[index]; - - ps->performance_level_count = index + 1; - - pl->sclk = le16_to_cpu(clock_info->ci.usEngineClockLow); - pl->sclk |= clock_info->ci.ucEngineClockHigh << 16; - pl->mclk = le16_to_cpu(clock_info->ci.usMemoryClockLow); - pl->mclk |= clock_info->ci.ucMemoryClockHigh << 16; - - pl->pcie_gen = amdgpu_get_pcie_gen_support(adev, - pi->sys_pcie_mask, - pi->vbios_boot_state.pcie_gen_bootup_value, - clock_info->ci.ucPCIEGen); - pl->pcie_lane = amdgpu_get_pcie_lane_support(adev, - pi->vbios_boot_state.pcie_lane_bootup_value, - le16_to_cpu(clock_info->ci.usPCIELane)); - - if (rps->class & ATOM_PPLIB_CLASSIFICATION_ACPI) { - pi->acpi_pcie_gen = pl->pcie_gen; - } - - if (rps->class2 & ATOM_PPLIB_CLASSIFICATION2_ULV) { - pi->ulv.supported = true; - pi->ulv.pl = *pl; - pi->ulv.cg_ulv_parameter = CISLANDS_CGULVPARAMETER_DFLT; - } - - /* patch up boot state */ - if (rps->class & ATOM_PPLIB_CLASSIFICATION_BOOT) { - pl->mclk = pi->vbios_boot_state.mclk_bootup_value; - pl->sclk = pi->vbios_boot_state.sclk_bootup_value; - pl->pcie_gen = pi->vbios_boot_state.pcie_gen_bootup_value; - pl->pcie_lane = pi->vbios_boot_state.pcie_lane_bootup_value; - } - - switch (rps->class & ATOM_PPLIB_CLASSIFICATION_UI_MASK) { - case ATOM_PPLIB_CLASSIFICATION_UI_BATTERY: - pi->use_pcie_powersaving_levels = true; - if (pi->pcie_gen_powersaving.max < pl->pcie_gen) - pi->pcie_gen_powersaving.max = pl->pcie_gen; - if (pi->pcie_gen_powersaving.min > pl->pcie_gen) - pi->pcie_gen_powersaving.min = pl->pcie_gen; - if (pi->pcie_lane_powersaving.max < pl->pcie_lane) - pi->pcie_lane_powersaving.max = pl->pcie_lane; - if (pi->pcie_lane_powersaving.min > pl->pcie_lane) - pi->pcie_lane_powersaving.min = pl->pcie_lane; - break; - case ATOM_PPLIB_CLASSIFICATION_UI_PERFORMANCE: - pi->use_pcie_performance_levels = true; - if (pi->pcie_gen_performance.max < pl->pcie_gen) - pi->pcie_gen_performance.max = pl->pcie_gen; - if (pi->pcie_gen_performance.min > pl->pcie_gen) - pi->pcie_gen_performance.min = pl->pcie_gen; - if (pi->pcie_lane_performance.max < pl->pcie_lane) - pi->pcie_lane_performance.max = pl->pcie_lane; - if (pi->pcie_lane_performance.min > pl->pcie_lane) - pi->pcie_lane_performance.min = pl->pcie_lane; - break; - default: - break; - } -} - -static int ci_parse_power_table(struct amdgpu_device *adev) -{ - struct amdgpu_mode_info *mode_info = &adev->mode_info; - struct _ATOM_PPLIB_NONCLOCK_INFO *non_clock_info; - union pplib_power_state *power_state; - int i, j, k, non_clock_array_index, clock_array_index; - union pplib_clock_info *clock_info; - struct _StateArray *state_array; - struct _ClockInfoArray *clock_info_array; - struct _NonClockInfoArray *non_clock_info_array; - union power_info *power_info; - int index = GetIndexIntoMasterTable(DATA, PowerPlayInfo); - u16 data_offset; - u8 frev, crev; - u8 *power_state_offset; - struct ci_ps *ps; - - if (!amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL, - &frev, &crev, &data_offset)) - return -EINVAL; - power_info = (union power_info *)(mode_info->atom_context->bios + data_offset); - - amdgpu_add_thermal_controller(adev); - - state_array = (struct _StateArray *) - (mode_info->atom_context->bios + data_offset + - le16_to_cpu(power_info->pplib.usStateArrayOffset)); - clock_info_array = (struct _ClockInfoArray *) - (mode_info->atom_context->bios + data_offset + - le16_to_cpu(power_info->pplib.usClockInfoArrayOffset)); - non_clock_info_array = (struct _NonClockInfoArray *) - (mode_info->atom_context->bios + data_offset + - le16_to_cpu(power_info->pplib.usNonClockInfoArrayOffset)); - - adev->pm.dpm.ps = kcalloc(state_array->ucNumEntries, - sizeof(struct amdgpu_ps), - GFP_KERNEL); - if (!adev->pm.dpm.ps) - return -ENOMEM; - power_state_offset = (u8 *)state_array->states; - for (i = 0; i < state_array->ucNumEntries; i++) { - u8 *idx; - power_state = (union pplib_power_state *)power_state_offset; - non_clock_array_index = power_state->v2.nonClockInfoIndex; - non_clock_info = (struct _ATOM_PPLIB_NONCLOCK_INFO *) - &non_clock_info_array->nonClockInfo[non_clock_array_index]; - ps = kzalloc(sizeof(struct ci_ps), GFP_KERNEL); - if (ps == NULL) { - kfree(adev->pm.dpm.ps); - return -ENOMEM; - } - adev->pm.dpm.ps[i].ps_priv = ps; - ci_parse_pplib_non_clock_info(adev, &adev->pm.dpm.ps[i], - non_clock_info, - non_clock_info_array->ucEntrySize); - k = 0; - idx = (u8 *)&power_state->v2.clockInfoIndex[0]; - for (j = 0; j < power_state->v2.ucNumDPMLevels; j++) { - clock_array_index = idx[j]; - if (clock_array_index >= clock_info_array->ucNumEntries) - continue; - if (k >= CISLANDS_MAX_HARDWARE_POWERLEVELS) - break; - clock_info = (union pplib_clock_info *) - ((u8 *)&clock_info_array->clockInfo[0] + - (clock_array_index * clock_info_array->ucEntrySize)); - ci_parse_pplib_clock_info(adev, - &adev->pm.dpm.ps[i], k, - clock_info); - k++; - } - power_state_offset += 2 + power_state->v2.ucNumDPMLevels; - } - adev->pm.dpm.num_ps = state_array->ucNumEntries; - - /* fill in the vce power states */ - for (i = 0; i < adev->pm.dpm.num_of_vce_states; i++) { - u32 sclk, mclk; - clock_array_index = adev->pm.dpm.vce_states[i].clk_idx; - clock_info = (union pplib_clock_info *) - &clock_info_array->clockInfo[clock_array_index * clock_info_array->ucEntrySize]; - sclk = le16_to_cpu(clock_info->ci.usEngineClockLow); - sclk |= clock_info->ci.ucEngineClockHigh << 16; - mclk = le16_to_cpu(clock_info->ci.usMemoryClockLow); - mclk |= clock_info->ci.ucMemoryClockHigh << 16; - adev->pm.dpm.vce_states[i].sclk = sclk; - adev->pm.dpm.vce_states[i].mclk = mclk; - } - - return 0; -} - -static int ci_get_vbios_boot_values(struct amdgpu_device *adev, - struct ci_vbios_boot_state *boot_state) -{ - struct amdgpu_mode_info *mode_info = &adev->mode_info; - int index = GetIndexIntoMasterTable(DATA, FirmwareInfo); - ATOM_FIRMWARE_INFO_V2_2 *firmware_info; - u8 frev, crev; - u16 data_offset; - - if (amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL, - &frev, &crev, &data_offset)) { - firmware_info = - (ATOM_FIRMWARE_INFO_V2_2 *)(mode_info->atom_context->bios + - data_offset); - boot_state->mvdd_bootup_value = le16_to_cpu(firmware_info->usBootUpMVDDCVoltage); - boot_state->vddc_bootup_value = le16_to_cpu(firmware_info->usBootUpVDDCVoltage); - boot_state->vddci_bootup_value = le16_to_cpu(firmware_info->usBootUpVDDCIVoltage); - boot_state->pcie_gen_bootup_value = ci_get_current_pcie_speed(adev); - boot_state->pcie_lane_bootup_value = ci_get_current_pcie_lane_number(adev); - boot_state->sclk_bootup_value = le32_to_cpu(firmware_info->ulDefaultEngineClock); - boot_state->mclk_bootup_value = le32_to_cpu(firmware_info->ulDefaultMemoryClock); - - return 0; - } - return -EINVAL; -} - -static void ci_dpm_fini(struct amdgpu_device *adev) -{ - int i; - - for (i = 0; i < adev->pm.dpm.num_ps; i++) { - kfree(adev->pm.dpm.ps[i].ps_priv); - } - kfree(adev->pm.dpm.ps); - kfree(adev->pm.dpm.priv); - kfree(adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries); - amdgpu_free_extended_power_table(adev); -} - -/** - * ci_dpm_init_microcode - load ucode images from disk - * - * @adev: amdgpu_device pointer - * - * Use the firmware interface to load the ucode images into - * the driver (not loaded into hw). - * Returns 0 on success, error on failure. - */ -static int ci_dpm_init_microcode(struct amdgpu_device *adev) -{ - const char *chip_name; - char fw_name[30]; - int err; - - DRM_DEBUG("\n"); - - switch (adev->asic_type) { - case CHIP_BONAIRE: - if ((adev->pdev->revision == 0x80) || - (adev->pdev->revision == 0x81) || - (adev->pdev->device == 0x665f)) - chip_name = "bonaire_k"; - else - chip_name = "bonaire"; - break; - case CHIP_HAWAII: - if (adev->pdev->revision == 0x80) - chip_name = "hawaii_k"; - else - chip_name = "hawaii"; - break; - case CHIP_KAVERI: - case CHIP_KABINI: - case CHIP_MULLINS: - default: BUG(); - } - - snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_smc.bin", chip_name); - err = request_firmware(&adev->pm.fw, fw_name, adev->dev); - if (err) - goto out; - err = amdgpu_ucode_validate(adev->pm.fw); - -out: - if (err) { - pr_err("cik_smc: Failed to load firmware \"%s\"\n", fw_name); - release_firmware(adev->pm.fw); - adev->pm.fw = NULL; - } - return err; -} - -static int ci_dpm_init(struct amdgpu_device *adev) -{ - int index = GetIndexIntoMasterTable(DATA, ASIC_InternalSS_Info); - SMU7_Discrete_DpmTable *dpm_table; - struct amdgpu_gpio_rec gpio; - u16 data_offset, size; - u8 frev, crev; - struct ci_power_info *pi; - int ret; - - pi = kzalloc(sizeof(struct ci_power_info), GFP_KERNEL); - if (pi == NULL) - return -ENOMEM; - adev->pm.dpm.priv = pi; - - pi->sys_pcie_mask = - adev->pm.pcie_gen_mask & CAIL_PCIE_LINK_SPEED_SUPPORT_MASK; - - pi->force_pcie_gen = AMDGPU_PCIE_GEN_INVALID; - - pi->pcie_gen_performance.max = AMDGPU_PCIE_GEN1; - pi->pcie_gen_performance.min = AMDGPU_PCIE_GEN3; - pi->pcie_gen_powersaving.max = AMDGPU_PCIE_GEN1; - pi->pcie_gen_powersaving.min = AMDGPU_PCIE_GEN3; - - pi->pcie_lane_performance.max = 0; - pi->pcie_lane_performance.min = 16; - pi->pcie_lane_powersaving.max = 0; - pi->pcie_lane_powersaving.min = 16; - - ret = ci_get_vbios_boot_values(adev, &pi->vbios_boot_state); - if (ret) { - ci_dpm_fini(adev); - return ret; - } - - ret = amdgpu_get_platform_caps(adev); - if (ret) { - ci_dpm_fini(adev); - return ret; - } - - ret = amdgpu_parse_extended_power_table(adev); - if (ret) { - ci_dpm_fini(adev); - return ret; - } - - ret = ci_parse_power_table(adev); - if (ret) { - ci_dpm_fini(adev); - return ret; - } - - pi->dll_default_on = false; - pi->sram_end = SMC_RAM_END; - - pi->activity_target[0] = CISLAND_TARGETACTIVITY_DFLT; - pi->activity_target[1] = CISLAND_TARGETACTIVITY_DFLT; - pi->activity_target[2] = CISLAND_TARGETACTIVITY_DFLT; - pi->activity_target[3] = CISLAND_TARGETACTIVITY_DFLT; - pi->activity_target[4] = CISLAND_TARGETACTIVITY_DFLT; - pi->activity_target[5] = CISLAND_TARGETACTIVITY_DFLT; - pi->activity_target[6] = CISLAND_TARGETACTIVITY_DFLT; - pi->activity_target[7] = CISLAND_TARGETACTIVITY_DFLT; - - pi->mclk_activity_target = CISLAND_MCLK_TARGETACTIVITY_DFLT; - - pi->sclk_dpm_key_disabled = 0; - pi->mclk_dpm_key_disabled = 0; - pi->pcie_dpm_key_disabled = 0; - pi->thermal_sclk_dpm_enabled = 0; - - if (adev->powerplay.pp_feature & PP_SCLK_DEEP_SLEEP_MASK) - pi->caps_sclk_ds = true; - else - pi->caps_sclk_ds = false; - - pi->mclk_strobe_mode_threshold = 40000; - pi->mclk_stutter_mode_threshold = 40000; - pi->mclk_edc_enable_threshold = 40000; - pi->mclk_edc_wr_enable_threshold = 40000; - - ci_initialize_powertune_defaults(adev); - - pi->caps_fps = false; - - pi->caps_sclk_throttle_low_notification = false; - - pi->caps_uvd_dpm = true; - pi->caps_vce_dpm = true; - - ci_get_leakage_voltages(adev); - ci_patch_dependency_tables_with_leakage(adev); - ci_set_private_data_variables_based_on_pptable(adev); - - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries = - kcalloc(4, - sizeof(struct amdgpu_clock_voltage_dependency_entry), - GFP_KERNEL); - if (!adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries) { - ci_dpm_fini(adev); - return -ENOMEM; - } - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.count = 4; - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries[0].clk = 0; - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries[0].v = 0; - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries[1].clk = 36000; - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries[1].v = 720; - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries[2].clk = 54000; - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries[2].v = 810; - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries[3].clk = 72000; - adev->pm.dpm.dyn_state.vddc_dependency_on_dispclk.entries[3].v = 900; - - adev->pm.dpm.dyn_state.mclk_sclk_ratio = 4; - adev->pm.dpm.dyn_state.sclk_mclk_delta = 15000; - adev->pm.dpm.dyn_state.vddc_vddci_delta = 200; - - adev->pm.dpm.dyn_state.valid_sclk_values.count = 0; - adev->pm.dpm.dyn_state.valid_sclk_values.values = NULL; - adev->pm.dpm.dyn_state.valid_mclk_values.count = 0; - adev->pm.dpm.dyn_state.valid_mclk_values.values = NULL; - - if (adev->asic_type == CHIP_HAWAII) { - pi->thermal_temp_setting.temperature_low = 94500; - pi->thermal_temp_setting.temperature_high = 95000; - pi->thermal_temp_setting.temperature_shutdown = 104000; - } else { - pi->thermal_temp_setting.temperature_low = 99500; - pi->thermal_temp_setting.temperature_high = 100000; - pi->thermal_temp_setting.temperature_shutdown = 104000; - } - - pi->uvd_enabled = false; - - dpm_table = &pi->smc_state_table; - - gpio = amdgpu_atombios_lookup_gpio(adev, VDDC_VRHOT_GPIO_PINID); - if (gpio.valid) { - dpm_table->VRHotGpio = gpio.shift; - adev->pm.dpm.platform_caps |= ATOM_PP_PLATFORM_CAP_REGULATOR_HOT; - } else { - dpm_table->VRHotGpio = CISLANDS_UNUSED_GPIO_PIN; - adev->pm.dpm.platform_caps &= ~ATOM_PP_PLATFORM_CAP_REGULATOR_HOT; - } - - gpio = amdgpu_atombios_lookup_gpio(adev, PP_AC_DC_SWITCH_GPIO_PINID); - if (gpio.valid) { - dpm_table->AcDcGpio = gpio.shift; - adev->pm.dpm.platform_caps |= ATOM_PP_PLATFORM_CAP_HARDWAREDC; - } else { - dpm_table->AcDcGpio = CISLANDS_UNUSED_GPIO_PIN; - adev->pm.dpm.platform_caps &= ~ATOM_PP_PLATFORM_CAP_HARDWAREDC; - } - - gpio = amdgpu_atombios_lookup_gpio(adev, VDDC_PCC_GPIO_PINID); - if (gpio.valid) { - u32 tmp = RREG32_SMC(ixCNB_PWRMGT_CNTL); - - switch (gpio.shift) { - case 0: - tmp &= ~CNB_PWRMGT_CNTL__GNB_SLOW_MODE_MASK; - tmp |= 1 << CNB_PWRMGT_CNTL__GNB_SLOW_MODE__SHIFT; - break; - case 1: - tmp &= ~CNB_PWRMGT_CNTL__GNB_SLOW_MODE_MASK; - tmp |= 2 << CNB_PWRMGT_CNTL__GNB_SLOW_MODE__SHIFT; - break; - case 2: - tmp |= CNB_PWRMGT_CNTL__GNB_SLOW_MASK; - break; - case 3: - tmp |= CNB_PWRMGT_CNTL__FORCE_NB_PS1_MASK; - break; - case 4: - tmp |= CNB_PWRMGT_CNTL__DPM_ENABLED_MASK; - break; - default: - DRM_INFO("Invalid PCC GPIO: %u!\n", gpio.shift); - break; - } - WREG32_SMC(ixCNB_PWRMGT_CNTL, tmp); - } - - pi->voltage_control = CISLANDS_VOLTAGE_CONTROL_NONE; - pi->vddci_control = CISLANDS_VOLTAGE_CONTROL_NONE; - pi->mvdd_control = CISLANDS_VOLTAGE_CONTROL_NONE; - if (amdgpu_atombios_is_voltage_gpio(adev, VOLTAGE_TYPE_VDDC, VOLTAGE_OBJ_GPIO_LUT)) - pi->voltage_control = CISLANDS_VOLTAGE_CONTROL_BY_GPIO; - else if (amdgpu_atombios_is_voltage_gpio(adev, VOLTAGE_TYPE_VDDC, VOLTAGE_OBJ_SVID2)) - pi->voltage_control = CISLANDS_VOLTAGE_CONTROL_BY_SVID2; - - if (adev->pm.dpm.platform_caps & ATOM_PP_PLATFORM_CAP_VDDCI_CONTROL) { - if (amdgpu_atombios_is_voltage_gpio(adev, VOLTAGE_TYPE_VDDCI, VOLTAGE_OBJ_GPIO_LUT)) - pi->vddci_control = CISLANDS_VOLTAGE_CONTROL_BY_GPIO; - else if (amdgpu_atombios_is_voltage_gpio(adev, VOLTAGE_TYPE_VDDCI, VOLTAGE_OBJ_SVID2)) - pi->vddci_control = CISLANDS_VOLTAGE_CONTROL_BY_SVID2; - else - adev->pm.dpm.platform_caps &= ~ATOM_PP_PLATFORM_CAP_VDDCI_CONTROL; - } - - if (adev->pm.dpm.platform_caps & ATOM_PP_PLATFORM_CAP_MVDDCONTROL) { - if (amdgpu_atombios_is_voltage_gpio(adev, VOLTAGE_TYPE_MVDDC, VOLTAGE_OBJ_GPIO_LUT)) - pi->mvdd_control = CISLANDS_VOLTAGE_CONTROL_BY_GPIO; - else if (amdgpu_atombios_is_voltage_gpio(adev, VOLTAGE_TYPE_MVDDC, VOLTAGE_OBJ_SVID2)) - pi->mvdd_control = CISLANDS_VOLTAGE_CONTROL_BY_SVID2; - else - adev->pm.dpm.platform_caps &= ~ATOM_PP_PLATFORM_CAP_MVDDCONTROL; - } - - pi->vddc_phase_shed_control = true; - -#if defined(CONFIG_ACPI) - pi->pcie_performance_request = - amdgpu_acpi_is_pcie_performance_request_supported(adev); -#else - pi->pcie_performance_request = false; -#endif - - if (amdgpu_atom_parse_data_header(adev->mode_info.atom_context, index, &size, - &frev, &crev, &data_offset)) { - pi->caps_sclk_ss_support = true; - pi->caps_mclk_ss_support = true; - pi->dynamic_ss = true; - } else { - pi->caps_sclk_ss_support = false; - pi->caps_mclk_ss_support = false; - pi->dynamic_ss = true; - } - - if (adev->pm.int_thermal_type != THERMAL_TYPE_NONE) - pi->thermal_protection = true; - else - pi->thermal_protection = false; - - pi->caps_dynamic_ac_timing = true; - - pi->uvd_power_gated = true; - - /* make sure dc limits are valid */ - if ((adev->pm.dpm.dyn_state.max_clock_voltage_on_dc.sclk == 0) || - (adev->pm.dpm.dyn_state.max_clock_voltage_on_dc.mclk == 0)) - adev->pm.dpm.dyn_state.max_clock_voltage_on_dc = - adev->pm.dpm.dyn_state.max_clock_voltage_on_ac; - - pi->fan_ctrl_is_in_default_mode = true; - - return 0; -} - -static void -ci_dpm_debugfs_print_current_performance_level(void *handle, - struct seq_file *m) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct amdgpu_ps *rps = &pi->current_rps; - u32 sclk = ci_get_average_sclk_freq(adev); - u32 mclk = ci_get_average_mclk_freq(adev); - u32 activity_percent = 50; - int ret; - - ret = ci_read_smc_soft_register(adev, offsetof(SMU7_SoftRegisters, AverageGraphicsA), - &activity_percent); - - if (ret == 0) { - activity_percent += 0x80; - activity_percent >>= 8; - activity_percent = activity_percent > 100 ? 100 : activity_percent; - } - - seq_printf(m, "uvd %sabled\n", pi->uvd_power_gated ? "dis" : "en"); - seq_printf(m, "vce %sabled\n", rps->vce_active ? "en" : "dis"); - seq_printf(m, "power level avg sclk: %u mclk: %u\n", - sclk, mclk); - seq_printf(m, "GPU load: %u %%\n", activity_percent); -} - -static void ci_dpm_print_power_state(void *handle, void *current_ps) -{ - struct amdgpu_ps *rps = (struct amdgpu_ps *)current_ps; - struct ci_ps *ps = ci_get_ps(rps); - struct ci_pl *pl; - int i; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - amdgpu_dpm_print_class_info(rps->class, rps->class2); - amdgpu_dpm_print_cap_info(rps->caps); - printk("\tuvd vclk: %d dclk: %d\n", rps->vclk, rps->dclk); - for (i = 0; i < ps->performance_level_count; i++) { - pl = &ps->performance_levels[i]; - printk("\t\tpower level %d sclk: %u mclk: %u pcie gen: %u pcie lanes: %u\n", - i, pl->sclk, pl->mclk, pl->pcie_gen + 1, pl->pcie_lane); - } - amdgpu_dpm_print_ps_status(adev, rps); -} - -static inline bool ci_are_power_levels_equal(const struct ci_pl *ci_cpl1, - const struct ci_pl *ci_cpl2) -{ - return ((ci_cpl1->mclk == ci_cpl2->mclk) && - (ci_cpl1->sclk == ci_cpl2->sclk) && - (ci_cpl1->pcie_gen == ci_cpl2->pcie_gen) && - (ci_cpl1->pcie_lane == ci_cpl2->pcie_lane)); -} - -static int ci_check_state_equal(void *handle, - void *current_ps, - void *request_ps, - bool *equal) -{ - struct ci_ps *ci_cps; - struct ci_ps *ci_rps; - int i; - struct amdgpu_ps *cps = (struct amdgpu_ps *)current_ps; - struct amdgpu_ps *rps = (struct amdgpu_ps *)request_ps; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - if (adev == NULL || cps == NULL || rps == NULL || equal == NULL) - return -EINVAL; - - ci_cps = ci_get_ps((struct amdgpu_ps *)cps); - ci_rps = ci_get_ps((struct amdgpu_ps *)rps); - - if (ci_cps == NULL) { - *equal = false; - return 0; - } - - if (ci_cps->performance_level_count != ci_rps->performance_level_count) { - - *equal = false; - return 0; - } - - for (i = 0; i < ci_cps->performance_level_count; i++) { - if (!ci_are_power_levels_equal(&(ci_cps->performance_levels[i]), - &(ci_rps->performance_levels[i]))) { - *equal = false; - return 0; - } - } - - /* If all performance levels are the same try to use the UVD clocks to break the tie.*/ - *equal = ((cps->vclk == rps->vclk) && (cps->dclk == rps->dclk)); - *equal &= ((cps->evclk == rps->evclk) && (cps->ecclk == rps->ecclk)); - - return 0; -} - -static u32 ci_dpm_get_sclk(void *handle, bool low) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ps *requested_state = ci_get_ps(&pi->requested_rps); - - if (low) - return requested_state->performance_levels[0].sclk; - else - return requested_state->performance_levels[requested_state->performance_level_count - 1].sclk; -} - -static u32 ci_dpm_get_mclk(void *handle, bool low) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ps *requested_state = ci_get_ps(&pi->requested_rps); - - if (low) - return requested_state->performance_levels[0].mclk; - else - return requested_state->performance_levels[requested_state->performance_level_count - 1].mclk; -} - -/* get temperature in millidegrees */ -static int ci_dpm_get_temp(void *handle) -{ - u32 temp; - int actual_temp = 0; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - temp = (RREG32_SMC(ixCG_MULT_THERMAL_STATUS) & CG_MULT_THERMAL_STATUS__CTF_TEMP_MASK) >> - CG_MULT_THERMAL_STATUS__CTF_TEMP__SHIFT; - - if (temp & 0x200) - actual_temp = 255; - else - actual_temp = temp & 0x1ff; - - actual_temp = actual_temp * 1000; - - return actual_temp; -} - -static int ci_set_temperature_range(struct amdgpu_device *adev) -{ - int ret; - - ret = ci_thermal_enable_alert(adev, false); - if (ret) - return ret; - ret = ci_thermal_set_temperature_range(adev, CISLANDS_TEMP_RANGE_MIN, - CISLANDS_TEMP_RANGE_MAX); - if (ret) - return ret; - ret = ci_thermal_enable_alert(adev, true); - if (ret) - return ret; - return ret; -} - -static int ci_dpm_early_init(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - adev->powerplay.pp_funcs = &ci_dpm_funcs; - adev->powerplay.pp_handle = adev; - ci_dpm_set_irq_funcs(adev); - - return 0; -} - -static int ci_dpm_late_init(void *handle) -{ - int ret; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - if (!adev->pm.dpm_enabled) - return 0; - - /* init the sysfs and debugfs files late */ - ret = amdgpu_pm_sysfs_init(adev); - if (ret) - return ret; - - ret = ci_set_temperature_range(adev); - if (ret) - return ret; - - return 0; -} - -static int ci_dpm_sw_init(void *handle) -{ - int ret; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - ret = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 230, - &adev->pm.dpm.thermal.irq); - if (ret) - return ret; - - ret = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 231, - &adev->pm.dpm.thermal.irq); - if (ret) - return ret; - - /* default to balanced state */ - adev->pm.dpm.state = POWER_STATE_TYPE_BALANCED; - adev->pm.dpm.user_state = POWER_STATE_TYPE_BALANCED; - adev->pm.dpm.forced_level = AMD_DPM_FORCED_LEVEL_AUTO; - adev->pm.default_sclk = adev->clock.default_sclk; - adev->pm.default_mclk = adev->clock.default_mclk; - adev->pm.current_sclk = adev->clock.default_sclk; - adev->pm.current_mclk = adev->clock.default_mclk; - adev->pm.int_thermal_type = THERMAL_TYPE_NONE; - - ret = ci_dpm_init_microcode(adev); - if (ret) - return ret; - - if (amdgpu_dpm == 0) - return 0; - - INIT_WORK(&adev->pm.dpm.thermal.work, amdgpu_dpm_thermal_work_handler); - mutex_lock(&adev->pm.mutex); - ret = ci_dpm_init(adev); - if (ret) - goto dpm_failed; - adev->pm.dpm.current_ps = adev->pm.dpm.requested_ps = adev->pm.dpm.boot_ps; - if (amdgpu_dpm == 1) - amdgpu_pm_print_power_states(adev); - mutex_unlock(&adev->pm.mutex); - DRM_INFO("amdgpu: dpm initialized\n"); - - return 0; - -dpm_failed: - ci_dpm_fini(adev); - mutex_unlock(&adev->pm.mutex); - DRM_ERROR("amdgpu: dpm initialization failed\n"); - return ret; -} - -static int ci_dpm_sw_fini(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - flush_work(&adev->pm.dpm.thermal.work); - - mutex_lock(&adev->pm.mutex); - ci_dpm_fini(adev); - mutex_unlock(&adev->pm.mutex); - - release_firmware(adev->pm.fw); - adev->pm.fw = NULL; - - return 0; -} - -static int ci_dpm_hw_init(void *handle) -{ - int ret; - - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - if (!amdgpu_dpm) { - ret = ci_upload_firmware(adev); - if (ret) { - DRM_ERROR("ci_upload_firmware failed\n"); - return ret; - } - ci_dpm_start_smc(adev); - return 0; - } - - mutex_lock(&adev->pm.mutex); - ci_dpm_setup_asic(adev); - ret = ci_dpm_enable(adev); - if (ret) - adev->pm.dpm_enabled = false; - else - adev->pm.dpm_enabled = true; - mutex_unlock(&adev->pm.mutex); - - return ret; -} - -static int ci_dpm_hw_fini(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - if (adev->pm.dpm_enabled) { - mutex_lock(&adev->pm.mutex); - ci_dpm_disable(adev); - mutex_unlock(&adev->pm.mutex); - } else { - ci_dpm_stop_smc(adev); - } - - return 0; -} - -static int ci_dpm_suspend(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - if (adev->pm.dpm_enabled) { - mutex_lock(&adev->pm.mutex); - amdgpu_irq_put(adev, &adev->pm.dpm.thermal.irq, - AMDGPU_THERMAL_IRQ_LOW_TO_HIGH); - amdgpu_irq_put(adev, &adev->pm.dpm.thermal.irq, - AMDGPU_THERMAL_IRQ_HIGH_TO_LOW); - adev->pm.dpm.last_user_state = adev->pm.dpm.user_state; - adev->pm.dpm.last_state = adev->pm.dpm.state; - adev->pm.dpm.user_state = POWER_STATE_TYPE_INTERNAL_BOOT; - adev->pm.dpm.state = POWER_STATE_TYPE_INTERNAL_BOOT; - mutex_unlock(&adev->pm.mutex); - amdgpu_pm_compute_clocks(adev); - - } - - return 0; -} - -static int ci_dpm_resume(void *handle) -{ - int ret; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - if (adev->pm.dpm_enabled) { - /* asic init will reset to the boot state */ - mutex_lock(&adev->pm.mutex); - ci_dpm_setup_asic(adev); - ret = ci_dpm_enable(adev); - if (ret) - adev->pm.dpm_enabled = false; - else - adev->pm.dpm_enabled = true; - adev->pm.dpm.user_state = adev->pm.dpm.last_user_state; - adev->pm.dpm.state = adev->pm.dpm.last_state; - mutex_unlock(&adev->pm.mutex); - if (adev->pm.dpm_enabled) - amdgpu_pm_compute_clocks(adev); - } - return 0; -} - -static bool ci_dpm_is_idle(void *handle) -{ - /* XXX */ - return true; -} - -static int ci_dpm_wait_for_idle(void *handle) -{ - /* XXX */ - return 0; -} - -static int ci_dpm_soft_reset(void *handle) -{ - return 0; -} - -static int ci_dpm_set_interrupt_state(struct amdgpu_device *adev, - struct amdgpu_irq_src *source, - unsigned type, - enum amdgpu_interrupt_state state) -{ - u32 cg_thermal_int; - - switch (type) { - case AMDGPU_THERMAL_IRQ_LOW_TO_HIGH: - switch (state) { - case AMDGPU_IRQ_STATE_DISABLE: - cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; - WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); - break; - case AMDGPU_IRQ_STATE_ENABLE: - cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; - WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); - break; - default: - break; - } - break; - - case AMDGPU_THERMAL_IRQ_HIGH_TO_LOW: - switch (state) { - case AMDGPU_IRQ_STATE_DISABLE: - cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; - WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); - break; - case AMDGPU_IRQ_STATE_ENABLE: - cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; - WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); - break; - default: - break; - } - break; - - default: - break; - } - return 0; -} - -static int ci_dpm_process_interrupt(struct amdgpu_device *adev, - struct amdgpu_irq_src *source, - struct amdgpu_iv_entry *entry) -{ - bool queue_thermal = false; - - if (entry == NULL) - return -EINVAL; - - switch (entry->src_id) { - case 230: /* thermal low to high */ - DRM_DEBUG("IH: thermal low to high\n"); - adev->pm.dpm.thermal.high_to_low = false; - queue_thermal = true; - break; - case 231: /* thermal high to low */ - DRM_DEBUG("IH: thermal high to low\n"); - adev->pm.dpm.thermal.high_to_low = true; - queue_thermal = true; - break; - default: - break; - } - - if (queue_thermal) - schedule_work(&adev->pm.dpm.thermal.work); - - return 0; -} - -static int ci_dpm_set_clockgating_state(void *handle, - enum amd_clockgating_state state) -{ - return 0; -} - -static int ci_dpm_set_powergating_state(void *handle, - enum amd_powergating_state state) -{ - return 0; -} - -static int ci_dpm_print_clock_levels(void *handle, - enum pp_clock_type type, char *buf) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_single_dpm_table *sclk_table = &pi->dpm_table.sclk_table; - struct ci_single_dpm_table *mclk_table = &pi->dpm_table.mclk_table; - struct ci_single_dpm_table *pcie_table = &pi->dpm_table.pcie_speed_table; - - int i, now, size = 0; - uint32_t clock, pcie_speed; - - switch (type) { - case PP_SCLK: - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_API_GetSclkFrequency); - clock = RREG32(mmSMC_MSG_ARG_0); - - for (i = 0; i < sclk_table->count; i++) { - if (clock > sclk_table->dpm_levels[i].value) - continue; - break; - } - now = i; - - for (i = 0; i < sclk_table->count; i++) - size += sprintf(buf + size, "%d: %uMhz %s\n", - i, sclk_table->dpm_levels[i].value / 100, - (i == now) ? "*" : ""); - break; - case PP_MCLK: - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_API_GetMclkFrequency); - clock = RREG32(mmSMC_MSG_ARG_0); - - for (i = 0; i < mclk_table->count; i++) { - if (clock > mclk_table->dpm_levels[i].value) - continue; - break; - } - now = i; - - for (i = 0; i < mclk_table->count; i++) - size += sprintf(buf + size, "%d: %uMhz %s\n", - i, mclk_table->dpm_levels[i].value / 100, - (i == now) ? "*" : ""); - break; - case PP_PCIE: - pcie_speed = ci_get_current_pcie_speed(adev); - for (i = 0; i < pcie_table->count; i++) { - if (pcie_speed != pcie_table->dpm_levels[i].value) - continue; - break; - } - now = i; - - for (i = 0; i < pcie_table->count; i++) - size += sprintf(buf + size, "%d: %s %s\n", i, - (pcie_table->dpm_levels[i].value == 0) ? "2.5GT/s, x1" : - (pcie_table->dpm_levels[i].value == 1) ? "5.0GT/s, x16" : - (pcie_table->dpm_levels[i].value == 2) ? "8.0GT/s, x16" : "", - (i == now) ? "*" : ""); - break; - default: - break; - } - - return size; -} - -static int ci_dpm_force_clock_level(void *handle, - enum pp_clock_type type, uint32_t mask) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - - if (adev->pm.dpm.forced_level != AMD_DPM_FORCED_LEVEL_MANUAL) - return -EINVAL; - - if (mask == 0) - return -EINVAL; - - switch (type) { - case PP_SCLK: - if (!pi->sclk_dpm_key_disabled) - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_SCLKDPM_SetEnabledMask, - pi->dpm_level_enable_mask.sclk_dpm_enable_mask & mask); - break; - - case PP_MCLK: - if (!pi->mclk_dpm_key_disabled) - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_MCLKDPM_SetEnabledMask, - pi->dpm_level_enable_mask.mclk_dpm_enable_mask & mask); - break; - - case PP_PCIE: - { - uint32_t tmp = mask & pi->dpm_level_enable_mask.pcie_dpm_enable_mask; - - if (!pi->pcie_dpm_key_disabled) { - if (fls(tmp) != ffs(tmp)) - amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_PCIeDPM_UnForceLevel); - else - amdgpu_ci_send_msg_to_smc_with_parameter(adev, - PPSMC_MSG_PCIeDPM_ForceLevel, - fls(tmp) - 1); - } - break; - } - default: - break; - } - - return 0; -} - -static int ci_dpm_get_sclk_od(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_single_dpm_table *sclk_table = &(pi->dpm_table.sclk_table); - struct ci_single_dpm_table *golden_sclk_table = - &(pi->golden_dpm_table.sclk_table); - int value; - - value = (sclk_table->dpm_levels[sclk_table->count - 1].value - - golden_sclk_table->dpm_levels[golden_sclk_table->count - 1].value) * - 100 / - golden_sclk_table->dpm_levels[golden_sclk_table->count - 1].value; - - return value; -} - -static int ci_dpm_set_sclk_od(void *handle, uint32_t value) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ps *ps = ci_get_ps(adev->pm.dpm.requested_ps); - struct ci_single_dpm_table *golden_sclk_table = - &(pi->golden_dpm_table.sclk_table); - - if (value > 20) - value = 20; - - ps->performance_levels[ps->performance_level_count - 1].sclk = - golden_sclk_table->dpm_levels[golden_sclk_table->count - 1].value * - value / 100 + - golden_sclk_table->dpm_levels[golden_sclk_table->count - 1].value; - - return 0; -} - -static int ci_dpm_get_mclk_od(void *handle) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_single_dpm_table *mclk_table = &(pi->dpm_table.mclk_table); - struct ci_single_dpm_table *golden_mclk_table = - &(pi->golden_dpm_table.mclk_table); - int value; - - value = (mclk_table->dpm_levels[mclk_table->count - 1].value - - golden_mclk_table->dpm_levels[golden_mclk_table->count - 1].value) * - 100 / - golden_mclk_table->dpm_levels[golden_mclk_table->count - 1].value; - - return value; -} - -static int ci_dpm_set_mclk_od(void *handle, uint32_t value) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct ci_power_info *pi = ci_get_pi(adev); - struct ci_ps *ps = ci_get_ps(adev->pm.dpm.requested_ps); - struct ci_single_dpm_table *golden_mclk_table = - &(pi->golden_dpm_table.mclk_table); - - if (value > 20) - value = 20; - - ps->performance_levels[ps->performance_level_count - 1].mclk = - golden_mclk_table->dpm_levels[golden_mclk_table->count - 1].value * - value / 100 + - golden_mclk_table->dpm_levels[golden_mclk_table->count - 1].value; - - return 0; -} - -static int ci_dpm_read_sensor(void *handle, int idx, - void *value, int *size) -{ - u32 activity_percent = 50; - int ret; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - /* size must be at least 4 bytes for all sensors */ - if (*size < 4) - return -EINVAL; - - switch (idx) { - case AMDGPU_PP_SENSOR_GFX_SCLK: - *((uint32_t *)value) = ci_get_average_sclk_freq(adev); - *size = 4; - return 0; - case AMDGPU_PP_SENSOR_GFX_MCLK: - *((uint32_t *)value) = ci_get_average_mclk_freq(adev); - *size = 4; - return 0; - case AMDGPU_PP_SENSOR_GPU_TEMP: - *((uint32_t *)value) = ci_dpm_get_temp(adev); - *size = 4; - return 0; - case AMDGPU_PP_SENSOR_GPU_LOAD: - ret = ci_read_smc_soft_register(adev, - offsetof(SMU7_SoftRegisters, - AverageGraphicsA), - &activity_percent); - if (ret == 0) { - activity_percent += 0x80; - activity_percent >>= 8; - activity_percent = - activity_percent > 100 ? 100 : activity_percent; - } - *((uint32_t *)value) = activity_percent; - *size = 4; - return 0; - default: - return -EINVAL; - } -} - -static int ci_set_powergating_by_smu(void *handle, - uint32_t block_type, bool gate) -{ - switch (block_type) { - case AMD_IP_BLOCK_TYPE_UVD: - ci_dpm_powergate_uvd(handle, gate); - break; - default: - break; - } - return 0; -} - -static const struct amd_ip_funcs ci_dpm_ip_funcs = { - .name = "ci_dpm", - .early_init = ci_dpm_early_init, - .late_init = ci_dpm_late_init, - .sw_init = ci_dpm_sw_init, - .sw_fini = ci_dpm_sw_fini, - .hw_init = ci_dpm_hw_init, - .hw_fini = ci_dpm_hw_fini, - .suspend = ci_dpm_suspend, - .resume = ci_dpm_resume, - .is_idle = ci_dpm_is_idle, - .wait_for_idle = ci_dpm_wait_for_idle, - .soft_reset = ci_dpm_soft_reset, - .set_clockgating_state = ci_dpm_set_clockgating_state, - .set_powergating_state = ci_dpm_set_powergating_state, -}; - -const struct amdgpu_ip_block_version ci_smu_ip_block = -{ - .type = AMD_IP_BLOCK_TYPE_SMC, - .major = 7, - .minor = 0, - .rev = 0, - .funcs = &ci_dpm_ip_funcs, -}; - -static const struct amd_pm_funcs ci_dpm_funcs = { - .pre_set_power_state = &ci_dpm_pre_set_power_state, - .set_power_state = &ci_dpm_set_power_state, - .post_set_power_state = &ci_dpm_post_set_power_state, - .display_configuration_changed = &ci_dpm_display_configuration_changed, - .get_sclk = &ci_dpm_get_sclk, - .get_mclk = &ci_dpm_get_mclk, - .print_power_state = &ci_dpm_print_power_state, - .debugfs_print_current_performance_level = &ci_dpm_debugfs_print_current_performance_level, - .force_performance_level = &ci_dpm_force_performance_level, - .vblank_too_short = &ci_dpm_vblank_too_short, - .set_powergating_by_smu = &ci_set_powergating_by_smu, - .set_fan_control_mode = &ci_dpm_set_fan_control_mode, - .get_fan_control_mode = &ci_dpm_get_fan_control_mode, - .set_fan_speed_percent = &ci_dpm_set_fan_speed_percent, - .get_fan_speed_percent = &ci_dpm_get_fan_speed_percent, - .print_clock_levels = ci_dpm_print_clock_levels, - .force_clock_level = ci_dpm_force_clock_level, - .get_sclk_od = ci_dpm_get_sclk_od, - .set_sclk_od = ci_dpm_set_sclk_od, - .get_mclk_od = ci_dpm_get_mclk_od, - .set_mclk_od = ci_dpm_set_mclk_od, - .check_state_equal = ci_check_state_equal, - .get_vce_clock_state = amdgpu_get_vce_clock_state, - .read_sensor = ci_dpm_read_sensor, -}; - -static const struct amdgpu_irq_src_funcs ci_dpm_irq_funcs = { - .set = ci_dpm_set_interrupt_state, - .process = ci_dpm_process_interrupt, -}; - -static void ci_dpm_set_irq_funcs(struct amdgpu_device *adev) -{ - adev->pm.dpm.thermal.irq.num_types = AMDGPU_THERMAL_IRQ_LAST; - adev->pm.dpm.thermal.irq.funcs = &ci_dpm_irq_funcs; -} diff --git a/drivers/gpu/drm/amd/amdgpu/ci_dpm.h b/drivers/gpu/drm/amd/amdgpu/ci_dpm.h deleted file mode 100644 index 91be2996ae7c..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/ci_dpm.h +++ /dev/null @@ -1,349 +0,0 @@ -/* - * Copyright 2013 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ -#ifndef __CI_DPM_H__ -#define __CI_DPM_H__ - -#include "amdgpu_atombios.h" -#include "ppsmc.h" - -#define SMU__NUM_SCLK_DPM_STATE 8 -#define SMU__NUM_MCLK_DPM_LEVELS 6 -#define SMU__NUM_LCLK_DPM_LEVELS 8 -#define SMU__NUM_PCIE_DPM_LEVELS 8 -#include "smu7_discrete.h" - -#define CISLANDS_MAX_HARDWARE_POWERLEVELS 2 - -#define CISLANDS_UNUSED_GPIO_PIN 0x7F - -struct ci_pl { - u32 mclk; - u32 sclk; - enum amdgpu_pcie_gen pcie_gen; - u16 pcie_lane; -}; - -struct ci_ps { - u16 performance_level_count; - bool dc_compatible; - u32 sclk_t; - struct ci_pl performance_levels[CISLANDS_MAX_HARDWARE_POWERLEVELS]; -}; - -struct ci_dpm_level { - bool enabled; - u32 value; - u32 param1; -}; - -#define CISLAND_MAX_DEEPSLEEP_DIVIDER_ID 5 -#define MAX_REGULAR_DPM_NUMBER 8 -#define CISLAND_MINIMUM_ENGINE_CLOCK 800 - -struct ci_single_dpm_table { - u32 count; - struct ci_dpm_level dpm_levels[MAX_REGULAR_DPM_NUMBER]; -}; - -struct ci_dpm_table { - struct ci_single_dpm_table sclk_table; - struct ci_single_dpm_table mclk_table; - struct ci_single_dpm_table pcie_speed_table; - struct ci_single_dpm_table vddc_table; - struct ci_single_dpm_table vddci_table; - struct ci_single_dpm_table mvdd_table; -}; - -struct ci_mc_reg_entry { - u32 mclk_max; - u32 mc_data[SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE]; -}; - -struct ci_mc_reg_table { - u8 last; - u8 num_entries; - u16 valid_flag; - struct ci_mc_reg_entry mc_reg_table_entry[MAX_AC_TIMING_ENTRIES]; - SMU7_Discrete_MCRegisterAddress mc_reg_address[SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE]; -}; - -struct ci_ulv_parm -{ - bool supported; - u32 cg_ulv_parameter; - u32 volt_change_delay; - struct ci_pl pl; -}; - -#define CISLANDS_MAX_LEAKAGE_COUNT 8 - -struct ci_leakage_voltage { - u16 count; - u16 leakage_id[CISLANDS_MAX_LEAKAGE_COUNT]; - u16 actual_voltage[CISLANDS_MAX_LEAKAGE_COUNT]; -}; - -struct ci_dpm_level_enable_mask { - u32 uvd_dpm_enable_mask; - u32 vce_dpm_enable_mask; - u32 acp_dpm_enable_mask; - u32 samu_dpm_enable_mask; - u32 sclk_dpm_enable_mask; - u32 mclk_dpm_enable_mask; - u32 pcie_dpm_enable_mask; -}; - -struct ci_vbios_boot_state -{ - u16 mvdd_bootup_value; - u16 vddc_bootup_value; - u16 vddci_bootup_value; - u32 sclk_bootup_value; - u32 mclk_bootup_value; - u16 pcie_gen_bootup_value; - u16 pcie_lane_bootup_value; -}; - -struct ci_clock_registers { - u32 cg_spll_func_cntl; - u32 cg_spll_func_cntl_2; - u32 cg_spll_func_cntl_3; - u32 cg_spll_func_cntl_4; - u32 cg_spll_spread_spectrum; - u32 cg_spll_spread_spectrum_2; - u32 dll_cntl; - u32 mclk_pwrmgt_cntl; - u32 mpll_ad_func_cntl; - u32 mpll_dq_func_cntl; - u32 mpll_func_cntl; - u32 mpll_func_cntl_1; - u32 mpll_func_cntl_2; - u32 mpll_ss1; - u32 mpll_ss2; -}; - -struct ci_thermal_temperature_setting { - s32 temperature_low; - s32 temperature_high; - s32 temperature_shutdown; -}; - -struct ci_pcie_perf_range { - u16 max; - u16 min; -}; - -enum ci_pt_config_reg_type { - CISLANDS_CONFIGREG_MMR = 0, - CISLANDS_CONFIGREG_SMC_IND, - CISLANDS_CONFIGREG_DIDT_IND, - CISLANDS_CONFIGREG_CACHE, - CISLANDS_CONFIGREG_MAX -}; - -#define POWERCONTAINMENT_FEATURE_BAPM 0x00000001 -#define POWERCONTAINMENT_FEATURE_TDCLimit 0x00000002 -#define POWERCONTAINMENT_FEATURE_PkgPwrLimit 0x00000004 - -struct ci_pt_config_reg { - u32 offset; - u32 mask; - u32 shift; - u32 value; - enum ci_pt_config_reg_type type; -}; - -struct ci_pt_defaults { - u8 svi_load_line_en; - u8 svi_load_line_vddc; - u8 tdc_vddc_throttle_release_limit_perc; - u8 tdc_mawt; - u8 tdc_waterfall_ctl; - u8 dte_ambient_temp_base; - u32 display_cac; - u32 bapm_temp_gradient; - u16 bapmti_r[SMU7_DTE_ITERATIONS * SMU7_DTE_SOURCES * SMU7_DTE_SINKS]; - u16 bapmti_rc[SMU7_DTE_ITERATIONS * SMU7_DTE_SOURCES * SMU7_DTE_SINKS]; -}; - -#define DPMTABLE_OD_UPDATE_SCLK 0x00000001 -#define DPMTABLE_OD_UPDATE_MCLK 0x00000002 -#define DPMTABLE_UPDATE_SCLK 0x00000004 -#define DPMTABLE_UPDATE_MCLK 0x00000008 - -struct ci_power_info { - struct ci_dpm_table dpm_table; - struct ci_dpm_table golden_dpm_table; - u32 voltage_control; - u32 mvdd_control; - u32 vddci_control; - u32 active_auto_throttle_sources; - struct ci_clock_registers clock_registers; - u16 acpi_vddc; - u16 acpi_vddci; - enum amdgpu_pcie_gen force_pcie_gen; - enum amdgpu_pcie_gen acpi_pcie_gen; - struct ci_leakage_voltage vddc_leakage; - struct ci_leakage_voltage vddci_leakage; - u16 max_vddc_in_pp_table; - u16 min_vddc_in_pp_table; - u16 max_vddci_in_pp_table; - u16 min_vddci_in_pp_table; - u32 mclk_strobe_mode_threshold; - u32 mclk_stutter_mode_threshold; - u32 mclk_edc_enable_threshold; - u32 mclk_edc_wr_enable_threshold; - struct ci_vbios_boot_state vbios_boot_state; - /* smc offsets */ - u32 sram_end; - u32 dpm_table_start; - u32 soft_regs_start; - u32 mc_reg_table_start; - u32 fan_table_start; - u32 arb_table_start; - /* smc tables */ - SMU7_Discrete_DpmTable smc_state_table; - SMU7_Discrete_MCRegisters smc_mc_reg_table; - SMU7_Discrete_PmFuses smc_powertune_table; - /* other stuff */ - struct ci_mc_reg_table mc_reg_table; - struct atom_voltage_table vddc_voltage_table; - struct atom_voltage_table vddci_voltage_table; - struct atom_voltage_table mvdd_voltage_table; - struct ci_ulv_parm ulv; - u32 power_containment_features; - const struct ci_pt_defaults *powertune_defaults; - u32 dte_tj_offset; - bool vddc_phase_shed_control; - struct ci_thermal_temperature_setting thermal_temp_setting; - struct ci_dpm_level_enable_mask dpm_level_enable_mask; - u32 need_update_smu7_dpm_table; - u32 sclk_dpm_key_disabled; - u32 mclk_dpm_key_disabled; - u32 pcie_dpm_key_disabled; - u32 thermal_sclk_dpm_enabled; - struct ci_pcie_perf_range pcie_gen_performance; - struct ci_pcie_perf_range pcie_lane_performance; - struct ci_pcie_perf_range pcie_gen_powersaving; - struct ci_pcie_perf_range pcie_lane_powersaving; - u32 activity_target[SMU7_MAX_LEVELS_GRAPHICS]; - u32 mclk_activity_target; - u32 low_sclk_interrupt_t; - u32 last_mclk_dpm_enable_mask; - u32 sys_pcie_mask; - /* caps */ - bool caps_power_containment; - bool caps_cac; - bool caps_sq_ramping; - bool caps_db_ramping; - bool caps_td_ramping; - bool caps_tcp_ramping; - bool caps_fps; - bool caps_sclk_ds; - bool caps_sclk_ss_support; - bool caps_mclk_ss_support; - bool caps_uvd_dpm; - bool caps_vce_dpm; - bool caps_samu_dpm; - bool caps_acp_dpm; - bool caps_automatic_dc_transition; - bool caps_sclk_throttle_low_notification; - bool caps_dynamic_ac_timing; - bool caps_od_fuzzy_fan_control_support; - /* flags */ - bool thermal_protection; - bool pcie_performance_request; - bool dynamic_ss; - bool dll_default_on; - bool cac_enabled; - bool uvd_enabled; - bool battery_state; - bool pspp_notify_required; - bool enable_bapm_feature; - bool enable_tdc_limit_feature; - bool enable_pkg_pwr_tracking_feature; - bool use_pcie_performance_levels; - bool use_pcie_powersaving_levels; - bool uvd_power_gated; - /* driver states */ - struct amdgpu_ps current_rps; - struct ci_ps current_ps; - struct amdgpu_ps requested_rps; - struct ci_ps requested_ps; - /* fan control */ - bool fan_ctrl_is_in_default_mode; - bool fan_is_controlled_by_smc; - u32 t_min; - u32 fan_ctrl_default_mode; -}; - -#define CISLANDS_VOLTAGE_CONTROL_NONE 0x0 -#define CISLANDS_VOLTAGE_CONTROL_BY_GPIO 0x1 -#define CISLANDS_VOLTAGE_CONTROL_BY_SVID2 0x2 - -#define CISLANDS_Q88_FORMAT_CONVERSION_UNIT 256 - -#define CISLANDS_VRC_DFLT0 0x3FFFC000 -#define CISLANDS_VRC_DFLT1 0x000400 -#define CISLANDS_VRC_DFLT2 0xC00080 -#define CISLANDS_VRC_DFLT3 0xC00200 -#define CISLANDS_VRC_DFLT4 0xC01680 -#define CISLANDS_VRC_DFLT5 0xC00033 -#define CISLANDS_VRC_DFLT6 0xC00033 -#define CISLANDS_VRC_DFLT7 0x3FFFC000 - -#define CISLANDS_CGULVPARAMETER_DFLT 0x00040035 -#define CISLAND_TARGETACTIVITY_DFLT 30 -#define CISLAND_MCLK_TARGETACTIVITY_DFLT 10 - -#define PCIE_PERF_REQ_REMOVE_REGISTRY 0 -#define PCIE_PERF_REQ_FORCE_LOWPOWER 1 -#define PCIE_PERF_REQ_PECI_GEN1 2 -#define PCIE_PERF_REQ_PECI_GEN2 3 -#define PCIE_PERF_REQ_PECI_GEN3 4 - -#define CISLANDS_SSTU_DFLT 0 -#define CISLANDS_SST_DFLT 0x00C8 - -/* XXX are these ok? */ -#define CISLANDS_TEMP_RANGE_MIN (90 * 1000) -#define CISLANDS_TEMP_RANGE_MAX (120 * 1000) - -int amdgpu_ci_copy_bytes_to_smc(struct amdgpu_device *adev, - u32 smc_start_address, - const u8 *src, u32 byte_count, u32 limit); -void amdgpu_ci_start_smc(struct amdgpu_device *adev); -void amdgpu_ci_reset_smc(struct amdgpu_device *adev); -int amdgpu_ci_program_jump_on_start(struct amdgpu_device *adev); -void amdgpu_ci_stop_smc_clock(struct amdgpu_device *adev); -void amdgpu_ci_start_smc_clock(struct amdgpu_device *adev); -bool amdgpu_ci_is_smc_running(struct amdgpu_device *adev); -PPSMC_Result amdgpu_ci_send_msg_to_smc(struct amdgpu_device *adev, PPSMC_Msg msg); -PPSMC_Result amdgpu_ci_wait_for_smc_inactive(struct amdgpu_device *adev); -int amdgpu_ci_load_smc_ucode(struct amdgpu_device *adev, u32 limit); -int amdgpu_ci_read_smc_sram_dword(struct amdgpu_device *adev, - u32 smc_address, u32 *value, u32 limit); -int amdgpu_ci_write_smc_sram_dword(struct amdgpu_device *adev, - u32 smc_address, u32 value, u32 limit); - -#endif diff --git a/drivers/gpu/drm/amd/amdgpu/ci_smc.c b/drivers/gpu/drm/amd/amdgpu/ci_smc.c deleted file mode 100644 index b8ba51e045b5..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/ci_smc.c +++ /dev/null @@ -1,279 +0,0 @@ -/* - * Copyright 2011 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - * Authors: Alex Deucher - */ - -#include <linux/firmware.h> -#include <drm/drmP.h> -#include "amdgpu.h" -#include "cikd.h" -#include "ppsmc.h" -#include "amdgpu_ucode.h" -#include "ci_dpm.h" - -#include "smu/smu_7_0_1_d.h" -#include "smu/smu_7_0_1_sh_mask.h" - -static int ci_set_smc_sram_address(struct amdgpu_device *adev, - u32 smc_address, u32 limit) -{ - if (smc_address & 3) - return -EINVAL; - if ((smc_address + 3) > limit) - return -EINVAL; - - WREG32(mmSMC_IND_INDEX_0, smc_address); - WREG32_P(mmSMC_IND_ACCESS_CNTL, 0, ~SMC_IND_ACCESS_CNTL__AUTO_INCREMENT_IND_0_MASK); - - return 0; -} - -int amdgpu_ci_copy_bytes_to_smc(struct amdgpu_device *adev, - u32 smc_start_address, - const u8 *src, u32 byte_count, u32 limit) -{ - unsigned long flags; - u32 data, original_data; - u32 addr; - u32 extra_shift; - int ret = 0; - - if (smc_start_address & 3) - return -EINVAL; - if ((smc_start_address + byte_count) > limit) - return -EINVAL; - - addr = smc_start_address; - - spin_lock_irqsave(&adev->smc_idx_lock, flags); - while (byte_count >= 4) { - /* SMC address space is BE */ - data = (src[0] << 24) | (src[1] << 16) | (src[2] << 8) | src[3]; - - ret = ci_set_smc_sram_address(adev, addr, limit); - if (ret) - goto done; - - WREG32(mmSMC_IND_DATA_0, data); - - src += 4; - byte_count -= 4; - addr += 4; - } - - /* RMW for the final bytes */ - if (byte_count > 0) { - data = 0; - - ret = ci_set_smc_sram_address(adev, addr, limit); - if (ret) - goto done; - - original_data = RREG32(mmSMC_IND_DATA_0); - - extra_shift = 8 * (4 - byte_count); - - while (byte_count > 0) { - data = (data << 8) + *src++; - byte_count--; - } - - data <<= extra_shift; - - data |= (original_data & ~((~0UL) << extra_shift)); - - ret = ci_set_smc_sram_address(adev, addr, limit); - if (ret) - goto done; - - WREG32(mmSMC_IND_DATA_0, data); - } - -done: - spin_unlock_irqrestore(&adev->smc_idx_lock, flags); - - return ret; -} - -void amdgpu_ci_start_smc(struct amdgpu_device *adev) -{ - u32 tmp = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL); - - tmp &= ~SMC_SYSCON_RESET_CNTL__rst_reg_MASK; - WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, tmp); -} - -void amdgpu_ci_reset_smc(struct amdgpu_device *adev) -{ - u32 tmp = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL); - - tmp |= SMC_SYSCON_RESET_CNTL__rst_reg_MASK; - WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, tmp); -} - -int amdgpu_ci_program_jump_on_start(struct amdgpu_device *adev) -{ - static u8 data[] = { 0xE0, 0x00, 0x80, 0x40 }; - - return amdgpu_ci_copy_bytes_to_smc(adev, 0x0, data, 4, sizeof(data)+1); -} - -void amdgpu_ci_stop_smc_clock(struct amdgpu_device *adev) -{ - u32 tmp = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0); - - tmp |= SMC_SYSCON_CLOCK_CNTL_0__ck_disable_MASK; - - WREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0, tmp); -} - -void amdgpu_ci_start_smc_clock(struct amdgpu_device *adev) -{ - u32 tmp = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0); - - tmp &= ~SMC_SYSCON_CLOCK_CNTL_0__ck_disable_MASK; - - WREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0, tmp); -} - -bool amdgpu_ci_is_smc_running(struct amdgpu_device *adev) -{ - u32 clk = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0); - u32 pc_c = RREG32_SMC(ixSMC_PC_C); - - if (!(clk & SMC_SYSCON_CLOCK_CNTL_0__ck_disable_MASK) && (0x20100 <= pc_c)) - return true; - - return false; -} - -PPSMC_Result amdgpu_ci_send_msg_to_smc(struct amdgpu_device *adev, PPSMC_Msg msg) -{ - u32 tmp; - int i; - - if (!amdgpu_ci_is_smc_running(adev)) - return PPSMC_Result_Failed; - - WREG32(mmSMC_MESSAGE_0, msg); - - for (i = 0; i < adev->usec_timeout; i++) { - tmp = RREG32(mmSMC_RESP_0); - if (tmp != 0) - break; - udelay(1); - } - tmp = RREG32(mmSMC_RESP_0); - - return (PPSMC_Result)tmp; -} - -PPSMC_Result amdgpu_ci_wait_for_smc_inactive(struct amdgpu_device *adev) -{ - u32 tmp; - int i; - - if (!amdgpu_ci_is_smc_running(adev)) - return PPSMC_Result_OK; - - for (i = 0; i < adev->usec_timeout; i++) { - tmp = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0); - if ((tmp & SMC_SYSCON_CLOCK_CNTL_0__cken_MASK) == 0) - break; - udelay(1); - } - - return PPSMC_Result_OK; -} - -int amdgpu_ci_load_smc_ucode(struct amdgpu_device *adev, u32 limit) -{ - const struct smc_firmware_header_v1_0 *hdr; - unsigned long flags; - u32 ucode_start_address; - u32 ucode_size; - const u8 *src; - u32 data; - - if (!adev->pm.fw) - return -EINVAL; - - hdr = (const struct smc_firmware_header_v1_0 *)adev->pm.fw->data; - amdgpu_ucode_print_smc_hdr(&hdr->header); - - adev->pm.fw_version = le32_to_cpu(hdr->header.ucode_version); - ucode_start_address = le32_to_cpu(hdr->ucode_start_addr); - ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes); - src = (const u8 *) - (adev->pm.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - - if (ucode_size & 3) - return -EINVAL; - - spin_lock_irqsave(&adev->smc_idx_lock, flags); - WREG32(mmSMC_IND_INDEX_0, ucode_start_address); - WREG32_P(mmSMC_IND_ACCESS_CNTL, SMC_IND_ACCESS_CNTL__AUTO_INCREMENT_IND_0_MASK, - ~SMC_IND_ACCESS_CNTL__AUTO_INCREMENT_IND_0_MASK); - while (ucode_size >= 4) { - /* SMC address space is BE */ - data = (src[0] << 24) | (src[1] << 16) | (src[2] << 8) | src[3]; - - WREG32(mmSMC_IND_DATA_0, data); - - src += 4; - ucode_size -= 4; - } - WREG32_P(mmSMC_IND_ACCESS_CNTL, 0, ~SMC_IND_ACCESS_CNTL__AUTO_INCREMENT_IND_0_MASK); - spin_unlock_irqrestore(&adev->smc_idx_lock, flags); - - return 0; -} - -int amdgpu_ci_read_smc_sram_dword(struct amdgpu_device *adev, - u32 smc_address, u32 *value, u32 limit) -{ - unsigned long flags; - int ret; - - spin_lock_irqsave(&adev->smc_idx_lock, flags); - ret = ci_set_smc_sram_address(adev, smc_address, limit); - if (ret == 0) - *value = RREG32(mmSMC_IND_DATA_0); - spin_unlock_irqrestore(&adev->smc_idx_lock, flags); - - return ret; -} - -int amdgpu_ci_write_smc_sram_dword(struct amdgpu_device *adev, - u32 smc_address, u32 value, u32 limit) -{ - unsigned long flags; - int ret; - - spin_lock_irqsave(&adev->smc_idx_lock, flags); - ret = ci_set_smc_sram_address(adev, smc_address, limit); - if (ret == 0) - WREG32(mmSMC_IND_DATA_0, value); - spin_unlock_irqrestore(&adev->smc_idx_lock, flags); - - return ret; -} diff --git a/drivers/gpu/drm/amd/amdgpu/cik.c b/drivers/gpu/drm/amd/amdgpu/cik.c index f41f5f57e9f3..07c1f239e9c3 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik.c +++ b/drivers/gpu/drm/amd/amdgpu/cik.c @@ -1741,6 +1741,69 @@ static bool cik_need_full_reset(struct amdgpu_device *adev) return true; } +static void cik_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0, + uint64_t *count1) +{ + uint32_t perfctr = 0; + uint64_t cnt0_of, cnt1_of; + int tmp; + + /* This reports 0 on APUs, so return to avoid writing/reading registers + * that may or may not be different from their GPU counterparts + */ + if (adev->flags & AMD_IS_APU) + return; + + /* Set the 2 events that we wish to watch, defined above */ + /* Reg 40 is # received msgs, Reg 104 is # of posted requests sent */ + perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40); + perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104); + + /* Write to enable desired perf counters */ + WREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK, perfctr); + /* Zero out and enable the perf counters + * Write 0x5: + * Bit 0 = Start all counters(1) + * Bit 2 = Global counter reset enable(1) + */ + WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000005); + + msleep(1000); + + /* Load the shadow and disable the perf counters + * Write 0x2: + * Bit 0 = Stop counters(0) + * Bit 1 = Load the shadow counters(1) + */ + WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000002); + + /* Read register values to get any >32bit overflow */ + tmp = RREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK); + cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER); + cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER); + + /* Get the values and add the overflow */ + *count0 = RREG32_PCIE(ixPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32); + *count1 = RREG32_PCIE(ixPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32); +} + +static bool cik_need_reset_on_init(struct amdgpu_device *adev) +{ + u32 clock_cntl, pc; + + if (adev->flags & AMD_IS_APU) + return false; + + /* check if the SMC is already running */ + clock_cntl = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0); + pc = RREG32_SMC(ixSMC_PC_C); + if ((0 == REG_GET_FIELD(clock_cntl, SMC_SYSCON_CLOCK_CNTL_0, ck_disable)) && + (0x20100 <= pc)) + return true; + + return false; +} + static const struct amdgpu_asic_funcs cik_asic_funcs = { .read_disabled_bios = &cik_read_disabled_bios, @@ -1755,6 +1818,9 @@ static const struct amdgpu_asic_funcs cik_asic_funcs = .flush_hdp = &cik_flush_hdp, .invalidate_hdp = &cik_invalidate_hdp, .need_full_reset = &cik_need_full_reset, + .init_doorbell_index = &legacy_doorbell_index_init, + .get_pcie_usage = &cik_get_pcie_usage, + .need_reset_on_init = &cik_need_reset_on_init, }; static int cik_common_early_init(void *handle) @@ -2004,10 +2070,7 @@ int cik_set_ip_blocks(struct amdgpu_device *adev) amdgpu_device_ip_block_add(adev, &cik_ih_ip_block); amdgpu_device_ip_block_add(adev, &gfx_v7_2_ip_block); amdgpu_device_ip_block_add(adev, &cik_sdma_ip_block); - if (amdgpu_dpm == -1) - amdgpu_device_ip_block_add(adev, &pp_smu_ip_block); - else - amdgpu_device_ip_block_add(adev, &ci_smu_ip_block); + amdgpu_device_ip_block_add(adev, &pp_smu_ip_block); if (adev->enable_virtual_display) amdgpu_device_ip_block_add(adev, &dce_virtual_ip_block); #if defined(CONFIG_DRM_AMD_DC) @@ -2025,10 +2088,7 @@ int cik_set_ip_blocks(struct amdgpu_device *adev) amdgpu_device_ip_block_add(adev, &cik_ih_ip_block); amdgpu_device_ip_block_add(adev, &gfx_v7_3_ip_block); amdgpu_device_ip_block_add(adev, &cik_sdma_ip_block); - if (amdgpu_dpm == -1) - amdgpu_device_ip_block_add(adev, &pp_smu_ip_block); - else - amdgpu_device_ip_block_add(adev, &ci_smu_ip_block); + amdgpu_device_ip_block_add(adev, &pp_smu_ip_block); if (adev->enable_virtual_display) amdgpu_device_ip_block_add(adev, &dce_virtual_ip_block); #if defined(CONFIG_DRM_AMD_DC) diff --git a/drivers/gpu/drm/amd/amdgpu/cik.h b/drivers/gpu/drm/amd/amdgpu/cik.h index e49c6f15a0a0..54c625a2e570 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik.h +++ b/drivers/gpu/drm/amd/amdgpu/cik.h @@ -30,4 +30,5 @@ void cik_srbm_select(struct amdgpu_device *adev, u32 me, u32 pipe, u32 queue, u32 vmid); int cik_set_ip_blocks(struct amdgpu_device *adev); +void legacy_doorbell_index_init(struct amdgpu_device *adev); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/cik_dpm.h b/drivers/gpu/drm/amd/amdgpu/cik_dpm.h index 2a086610f74d..2fcc4b60153c 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_dpm.h +++ b/drivers/gpu/drm/amd/amdgpu/cik_dpm.h @@ -24,7 +24,6 @@ #ifndef __CIK_DPM_H__ #define __CIK_DPM_H__ -extern const struct amdgpu_ip_block_version ci_smu_ip_block; extern const struct amdgpu_ip_block_version kv_smu_ip_block; #endif diff --git a/drivers/gpu/drm/amd/amdgpu/cik_ih.c b/drivers/gpu/drm/amd/amdgpu/cik_ih.c index b5775c6a857b..721c757156e8 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_ih.c @@ -103,9 +103,9 @@ static void cik_ih_disable_interrupts(struct amdgpu_device *adev) */ static int cik_ih_irq_init(struct amdgpu_device *adev) { + struct amdgpu_ih_ring *ih = &adev->irq.ih; int rb_bufsz; u32 interrupt_cntl, ih_cntl, ih_rb_cntl; - u64 wptr_off; /* disable irqs */ cik_ih_disable_interrupts(adev); @@ -131,9 +131,8 @@ static int cik_ih_irq_init(struct amdgpu_device *adev) ih_rb_cntl |= IH_RB_CNTL__WPTR_WRITEBACK_ENABLE_MASK; /* set the writeback address whether it's enabled or not */ - wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4); - WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off)); - WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF); + WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr)); + WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF); WREG32(mmIH_RB_CNTL, ih_rb_cntl); @@ -183,11 +182,12 @@ static void cik_ih_irq_disable(struct amdgpu_device *adev) * Used by cik_irq_process(). * Returns the value of the wptr. */ -static u32 cik_ih_get_wptr(struct amdgpu_device *adev) +static u32 cik_ih_get_wptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { u32 wptr, tmp; - wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]); + wptr = le32_to_cpu(*ih->wptr_cpu); if (wptr & IH_RB_WPTR__RB_OVERFLOW_MASK) { wptr &= ~IH_RB_WPTR__RB_OVERFLOW_MASK; @@ -196,13 +196,13 @@ static u32 cik_ih_get_wptr(struct amdgpu_device *adev) * this should allow us to catchup. */ dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", - wptr, adev->irq.ih.rptr, (wptr + 16) & adev->irq.ih.ptr_mask); - adev->irq.ih.rptr = (wptr + 16) & adev->irq.ih.ptr_mask; + wptr, ih->rptr, (wptr + 16) & ih->ptr_mask); + ih->rptr = (wptr + 16) & ih->ptr_mask; tmp = RREG32(mmIH_RB_CNTL); tmp |= IH_RB_CNTL__WPTR_OVERFLOW_CLEAR_MASK; WREG32(mmIH_RB_CNTL, tmp); } - return (wptr & adev->irq.ih.ptr_mask); + return (wptr & ih->ptr_mask); } /* CIK IV Ring @@ -228,34 +228,6 @@ static u32 cik_ih_get_wptr(struct amdgpu_device *adev) * [127:96] - reserved */ -/** - * cik_ih_prescreen_iv - prescreen an interrupt vector - * - * @adev: amdgpu_device pointer - * - * Returns true if the interrupt vector should be further processed. - */ -static bool cik_ih_prescreen_iv(struct amdgpu_device *adev) -{ - u32 ring_index = adev->irq.ih.rptr >> 2; - u16 pasid; - - switch (le32_to_cpu(adev->irq.ih.ring[ring_index]) & 0xff) { - case 146: - case 147: - pasid = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]) >> 16; - if (!pasid || amdgpu_vm_pasid_fault_credit(adev, pasid)) - return true; - break; - default: - /* Not a VM fault */ - return true; - } - - adev->irq.ih.rptr += 16; - return false; -} - /** * cik_ih_decode_iv - decode an interrupt vector * @@ -265,16 +237,17 @@ static bool cik_ih_prescreen_iv(struct amdgpu_device *adev) * position and also advance the position. */ static void cik_ih_decode_iv(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih, struct amdgpu_iv_entry *entry) { /* wptr/rptr are in bytes! */ - u32 ring_index = adev->irq.ih.rptr >> 2; + u32 ring_index = ih->rptr >> 2; uint32_t dw[4]; - dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); - dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]); - dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]); - dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); + dw[0] = le32_to_cpu(ih->ring[ring_index + 0]); + dw[1] = le32_to_cpu(ih->ring[ring_index + 1]); + dw[2] = le32_to_cpu(ih->ring[ring_index + 2]); + dw[3] = le32_to_cpu(ih->ring[ring_index + 3]); entry->client_id = AMDGPU_IRQ_CLIENTID_LEGACY; entry->src_id = dw[0] & 0xff; @@ -284,7 +257,7 @@ static void cik_ih_decode_iv(struct amdgpu_device *adev, entry->pasid = (dw[2] >> 16) & 0xffff; /* wptr/rptr are in bytes! */ - adev->irq.ih.rptr += 16; + ih->rptr += 16; } /** @@ -294,9 +267,10 @@ static void cik_ih_decode_iv(struct amdgpu_device *adev, * * Set the IH ring buffer rptr. */ -static void cik_ih_set_rptr(struct amdgpu_device *adev) +static void cik_ih_set_rptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { - WREG32(mmIH_RB_RPTR, adev->irq.ih.rptr); + WREG32(mmIH_RB_RPTR, ih->rptr); } static int cik_ih_early_init(void *handle) @@ -461,7 +435,6 @@ static const struct amd_ip_funcs cik_ih_ip_funcs = { static const struct amdgpu_ih_funcs cik_ih_funcs = { .get_wptr = cik_ih_get_wptr, - .prescreen_iv = cik_ih_prescreen_iv, .decode_iv = cik_ih_decode_iv, .set_rptr = cik_ih_set_rptr }; diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c index b918c8886b75..189599b694e8 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c @@ -198,7 +198,7 @@ static void cik_sdma_ring_set_wptr(struct amdgpu_ring *ring) static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) { - struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring); int i; for (i = 0; i < count; i++) @@ -218,9 +218,11 @@ static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (CIK). */ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 extra_bits = vmid & 0xf; /* IB packet must end on a 8 DW boundary */ @@ -316,8 +318,8 @@ static void cik_sdma_gfx_stop(struct amdgpu_device *adev) WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); WREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i], 0); } - sdma0->ready = false; - sdma1->ready = false; + sdma0->sched.ready = false; + sdma1->sched.ready = false; } /** @@ -494,18 +496,16 @@ static int cik_sdma_gfx_resume(struct amdgpu_device *adev) /* enable DMA IBs */ WREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i], ib_cntl); - ring->ready = true; + ring->sched.ready = true; } cik_sdma_enable(adev, true); for (i = 0; i < adev->sdma.num_instances; i++) { ring = &adev->sdma.instance[i].ring; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) return r; - } if (adev->mman.buffer_funcs_ring == ring) amdgpu_ttm_set_buffer_funcs_status(adev, true); @@ -618,21 +618,17 @@ static int cik_sdma_ring_test_ring(struct amdgpu_ring *ring) u64 gpu_addr; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); r = amdgpu_ring_alloc(ring, 5); - if (r) { - DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); - amdgpu_device_wb_free(adev, index); - return r; - } + if (r) + goto error_free_wb; + amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0)); amdgpu_ring_write(ring, lower_32_bits(gpu_addr)); amdgpu_ring_write(ring, upper_32_bits(gpu_addr)); @@ -647,15 +643,11 @@ static int cik_sdma_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } - amdgpu_device_wb_free(adev, index); + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; +error_free_wb: + amdgpu_device_wb_free(adev, index); return r; } @@ -678,20 +670,16 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 256, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err0; - } ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0); @@ -706,21 +694,16 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out\n"); r = -ETIMEDOUT; goto err1; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto err1; } tmp = le32_to_cpu(adev->wb.wb[index]); - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + if (tmp == 0xDEADBEEF) r = 0; - } else { - DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp); + else r = -EINVAL; - } err1: amdgpu_ib_free(adev, &ib, NULL); @@ -822,7 +805,7 @@ static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib, uint64_t pe, */ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring); u32 pad_count; int i; @@ -1214,8 +1197,11 @@ static int cik_sdma_process_illegal_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { + u8 instance_id; + DRM_ERROR("Illegal instruction in SDMA command stream\n"); - schedule_work(&adev->reset_work); + instance_id = (entry->ring_id & 0x3) >> 0; + drm_sched_fault(&adev->sdma.instance[instance_id].ring.sched); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/cz_ih.c b/drivers/gpu/drm/amd/amdgpu/cz_ih.c index df5ac4d85a00..61024b9c7a4b 100644 --- a/drivers/gpu/drm/amd/amdgpu/cz_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/cz_ih.c @@ -103,9 +103,9 @@ static void cz_ih_disable_interrupts(struct amdgpu_device *adev) */ static int cz_ih_irq_init(struct amdgpu_device *adev) { - int rb_bufsz; + struct amdgpu_ih_ring *ih = &adev->irq.ih; u32 interrupt_cntl, ih_cntl, ih_rb_cntl; - u64 wptr_off; + int rb_bufsz; /* disable irqs */ cz_ih_disable_interrupts(adev); @@ -133,9 +133,8 @@ static int cz_ih_irq_init(struct amdgpu_device *adev) ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1); /* set the writeback address whether it's enabled or not */ - wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4); - WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off)); - WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF); + WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr)); + WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF); WREG32(mmIH_RB_CNTL, ih_rb_cntl); @@ -185,11 +184,12 @@ static void cz_ih_irq_disable(struct amdgpu_device *adev) * Used by cz_irq_process(VI). * Returns the value of the wptr. */ -static u32 cz_ih_get_wptr(struct amdgpu_device *adev) +static u32 cz_ih_get_wptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { u32 wptr, tmp; - wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]); + wptr = le32_to_cpu(*ih->wptr_cpu); if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) { wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0); @@ -198,41 +198,13 @@ static u32 cz_ih_get_wptr(struct amdgpu_device *adev) * this should allow us to catchup. */ dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", - wptr, adev->irq.ih.rptr, (wptr + 16) & adev->irq.ih.ptr_mask); - adev->irq.ih.rptr = (wptr + 16) & adev->irq.ih.ptr_mask; + wptr, ih->rptr, (wptr + 16) & ih->ptr_mask); + ih->rptr = (wptr + 16) & ih->ptr_mask; tmp = RREG32(mmIH_RB_CNTL); tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); WREG32(mmIH_RB_CNTL, tmp); } - return (wptr & adev->irq.ih.ptr_mask); -} - -/** - * cz_ih_prescreen_iv - prescreen an interrupt vector - * - * @adev: amdgpu_device pointer - * - * Returns true if the interrupt vector should be further processed. - */ -static bool cz_ih_prescreen_iv(struct amdgpu_device *adev) -{ - u32 ring_index = adev->irq.ih.rptr >> 2; - u16 pasid; - - switch (le32_to_cpu(adev->irq.ih.ring[ring_index]) & 0xff) { - case 146: - case 147: - pasid = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]) >> 16; - if (!pasid || amdgpu_vm_pasid_fault_credit(adev, pasid)) - return true; - break; - default: - /* Not a VM fault */ - return true; - } - - adev->irq.ih.rptr += 16; - return false; + return (wptr & ih->ptr_mask); } /** @@ -244,16 +216,17 @@ static bool cz_ih_prescreen_iv(struct amdgpu_device *adev) * position and also advance the position. */ static void cz_ih_decode_iv(struct amdgpu_device *adev, - struct amdgpu_iv_entry *entry) + struct amdgpu_ih_ring *ih, + struct amdgpu_iv_entry *entry) { /* wptr/rptr are in bytes! */ - u32 ring_index = adev->irq.ih.rptr >> 2; + u32 ring_index = ih->rptr >> 2; uint32_t dw[4]; - dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); - dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]); - dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]); - dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); + dw[0] = le32_to_cpu(ih->ring[ring_index + 0]); + dw[1] = le32_to_cpu(ih->ring[ring_index + 1]); + dw[2] = le32_to_cpu(ih->ring[ring_index + 2]); + dw[3] = le32_to_cpu(ih->ring[ring_index + 3]); entry->client_id = AMDGPU_IRQ_CLIENTID_LEGACY; entry->src_id = dw[0] & 0xff; @@ -263,7 +236,7 @@ static void cz_ih_decode_iv(struct amdgpu_device *adev, entry->pasid = (dw[2] >> 16) & 0xffff; /* wptr/rptr are in bytes! */ - adev->irq.ih.rptr += 16; + ih->rptr += 16; } /** @@ -273,9 +246,10 @@ static void cz_ih_decode_iv(struct amdgpu_device *adev, * * Set the IH ring buffer rptr. */ -static void cz_ih_set_rptr(struct amdgpu_device *adev) +static void cz_ih_set_rptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { - WREG32(mmIH_RB_RPTR, adev->irq.ih.rptr); + WREG32(mmIH_RB_RPTR, ih->rptr); } static int cz_ih_early_init(void *handle) @@ -442,7 +416,6 @@ static const struct amd_ip_funcs cz_ih_ip_funcs = { static const struct amdgpu_ih_funcs cz_ih_funcs = { .get_wptr = cz_ih_get_wptr, - .prescreen_iv = cz_ih_prescreen_iv, .decode_iv = cz_ih_decode_iv, .set_rptr = cz_ih_set_rptr }; diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c index 4cfecdce29a3..1f0426d2fc2a 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c @@ -1682,7 +1682,7 @@ static void dce_v10_0_afmt_setmode(struct drm_encoder *encoder, dce_v10_0_audio_write_sad_regs(encoder); dce_v10_0_audio_write_latency_fields(encoder, mode); - err = drm_hdmi_avi_infoframe_from_display_mode(&frame, mode, false); + err = drm_hdmi_avi_infoframe_from_display_mode(&frame, connector, mode); if (err < 0) { DRM_ERROR("failed to setup AVI infoframe: %zd\n", err); return; diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c index 7c868916d90f..2280b971d758 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c @@ -1724,7 +1724,7 @@ static void dce_v11_0_afmt_setmode(struct drm_encoder *encoder, dce_v11_0_audio_write_sad_regs(encoder); dce_v11_0_audio_write_latency_fields(encoder, mode); - err = drm_hdmi_avi_infoframe_from_display_mode(&frame, mode, false); + err = drm_hdmi_avi_infoframe_from_display_mode(&frame, connector, mode); if (err < 0) { DRM_ERROR("failed to setup AVI infoframe: %zd\n", err); return; diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c index 17eaaba36017..bea32f076b91 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c @@ -1423,6 +1423,7 @@ static void dce_v6_0_audio_set_avi_infoframe(struct drm_encoder *encoder, struct amdgpu_device *adev = dev->dev_private; struct amdgpu_encoder *amdgpu_encoder = to_amdgpu_encoder(encoder); struct amdgpu_encoder_atom_dig *dig = amdgpu_encoder->enc_priv; + struct drm_connector *connector = amdgpu_get_connector_for_encoder(encoder); struct hdmi_avi_infoframe frame; u8 buffer[HDMI_INFOFRAME_HEADER_SIZE + HDMI_AVI_INFOFRAME_SIZE]; uint8_t *payload = buffer + 3; @@ -1430,7 +1431,7 @@ static void dce_v6_0_audio_set_avi_infoframe(struct drm_encoder *encoder, ssize_t err; u32 tmp; - err = drm_hdmi_avi_infoframe_from_display_mode(&frame, mode, false); + err = drm_hdmi_avi_infoframe_from_display_mode(&frame, connector, mode); if (err < 0) { DRM_ERROR("failed to setup AVI infoframe: %zd\n", err); return; @@ -2979,7 +2980,7 @@ static int dce_v6_0_pageflip_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { - unsigned long flags; + unsigned long flags; unsigned crtc_id; struct amdgpu_crtc *amdgpu_crtc; struct amdgpu_flip_work *works; diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c index 8c0576978d36..13da915991dd 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c @@ -1616,7 +1616,7 @@ static void dce_v8_0_afmt_setmode(struct drm_encoder *encoder, dce_v8_0_audio_write_sad_regs(encoder); dce_v8_0_audio_write_latency_fields(encoder, mode); - err = drm_hdmi_avi_infoframe_from_display_mode(&frame, mode, false); + err = drm_hdmi_avi_infoframe_from_display_mode(&frame, connector, mode); if (err < 0) { DRM_ERROR("failed to setup AVI infoframe: %zd\n", err); return; diff --git a/drivers/gpu/drm/amd/amdgpu/dce_virtual.c b/drivers/gpu/drm/amd/amdgpu/dce_virtual.c index fdace004544d..e4cc1d48eaab 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_virtual.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_virtual.c @@ -167,19 +167,6 @@ static void dce_virtual_crtc_disable(struct drm_crtc *crtc) struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); dce_virtual_crtc_dpms(crtc, DRM_MODE_DPMS_OFF); - if (crtc->primary->fb) { - int r; - struct amdgpu_bo *abo; - - abo = gem_to_amdgpu_bo(crtc->primary->fb->obj[0]); - r = amdgpu_bo_reserve(abo, true); - if (unlikely(r)) - DRM_ERROR("failed to reserve abo before unpin\n"); - else { - amdgpu_bo_unpin(abo); - amdgpu_bo_unreserve(abo); - } - } amdgpu_crtc->pll_id = ATOM_PPLL_INVALID; amdgpu_crtc->encoder = NULL; @@ -692,7 +679,9 @@ static int dce_virtual_pageflip(struct amdgpu_device *adev, spin_unlock_irqrestore(&adev->ddev->event_lock, flags); drm_crtc_vblank_put(&amdgpu_crtc->base); - schedule_work(&works->unpin_work); + amdgpu_bo_unref(&works->old_abo); + kfree(works->shared); + kfree(works); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c index d76eb27945dc..305276c7e4bf 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c @@ -1775,18 +1775,15 @@ static int gfx_v6_0_ring_test_ring(struct amdgpu_ring *ring) int r; r = amdgpu_gfx_scratch_get(adev, &scratch); - if (r) { - DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r); + if (r) return r; - } + WREG32(scratch, 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r); - amdgpu_gfx_scratch_free(adev, scratch); - return r; - } + if (r) + goto error_free_scratch; + amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); amdgpu_ring_write(ring, (scratch - PACKET3_SET_CONFIG_REG_START)); amdgpu_ring_write(ring, 0xDEADBEEF); @@ -1798,13 +1795,11 @@ static int gfx_v6_0_ring_test_ring(struct amdgpu_ring *ring) break; DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n", - ring->idx, scratch, tmp); - r = -EINVAL; - } + + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + +error_free_scratch: amdgpu_gfx_scratch_free(adev, scratch); return r; } @@ -1845,13 +1840,15 @@ static void gfx_v6_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, } static void gfx_v6_0_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; /* insert SWITCH_BUFFER packet before first IB in the ring frame */ - if (ctx_switch) { + if (flags & AMDGPU_HAVE_CTX_SWITCH) { amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); amdgpu_ring_write(ring, 0); } @@ -1892,17 +1889,15 @@ static int gfx_v6_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_gfx_scratch_get(adev, &scratch); - if (r) { - DRM_ERROR("amdgpu: failed to get scratch reg (%ld).\n", r); + if (r) return r; - } + WREG32(scratch, 0xCAFEDEAD); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 256, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err1; - } + ib.ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1); ib.ptr[1] = ((scratch - PACKET3_SET_CONFIG_REG_START)); ib.ptr[2] = 0xDEADBEEF; @@ -1914,22 +1909,16 @@ static int gfx_v6_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out\n"); r = -ETIMEDOUT; goto err2; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto err2; } tmp = RREG32(scratch); - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + if (tmp == 0xDEADBEEF) r = 0; - } else { - DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n", - scratch, tmp); + else r = -EINVAL; - } err2: amdgpu_ib_free(adev, &ib, NULL); @@ -1950,9 +1939,9 @@ static void gfx_v6_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) CP_ME_CNTL__CE_HALT_MASK)); WREG32(mmSCRATCH_UMSK, 0); for (i = 0; i < adev->gfx.num_gfx_rings; i++) - adev->gfx.gfx_ring[i].ready = false; + adev->gfx.gfx_ring[i].sched.ready = false; for (i = 0; i < adev->gfx.num_compute_rings; i++) - adev->gfx.compute_ring[i].ready = false; + adev->gfx.compute_ring[i].sched.ready = false; } udelay(50); } @@ -2124,12 +2113,9 @@ static int gfx_v6_0_cp_gfx_resume(struct amdgpu_device *adev) /* start the rings */ gfx_v6_0_cp_gfx_start(adev); - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) return r; - } return 0; } @@ -2227,14 +2213,11 @@ static int gfx_v6_0_cp_compute_resume(struct amdgpu_device *adev) WREG32(mmCP_RB2_CNTL, tmp); WREG32(mmCP_RB2_BASE, ring->gpu_addr >> 8); - adev->gfx.compute_ring[0].ready = false; - adev->gfx.compute_ring[1].ready = false; for (i = 0; i < 2; i++) { - r = amdgpu_ring_test_ring(&adev->gfx.compute_ring[i]); + r = amdgpu_ring_test_helper(&adev->gfx.compute_ring[i]); if (r) return r; - adev->gfx.compute_ring[i].ready = true; } return 0; @@ -2368,18 +2351,11 @@ static void gfx_v6_0_ring_emit_wreg(struct amdgpu_ring *ring, amdgpu_ring_write(ring, val); } -static void gfx_v6_0_rlc_fini(struct amdgpu_device *adev) -{ - amdgpu_bo_free_kernel(&adev->gfx.rlc.save_restore_obj, NULL, NULL); - amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, NULL, NULL); - amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, NULL, NULL); -} - static int gfx_v6_0_rlc_init(struct amdgpu_device *adev) { const u32 *src_ptr; volatile u32 *dst_ptr; - u32 dws, i; + u32 dws; u64 reg_list_mc_addr; const struct cs_section_def *cs_data; int r; @@ -2394,26 +2370,10 @@ static int gfx_v6_0_rlc_init(struct amdgpu_device *adev) cs_data = adev->gfx.rlc.cs_data; if (src_ptr) { - /* save restore block */ - r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.save_restore_obj, - &adev->gfx.rlc.save_restore_gpu_addr, - (void **)&adev->gfx.rlc.sr_ptr); - if (r) { - dev_warn(adev->dev, "(%d) create RLC sr bo failed\n", - r); - gfx_v6_0_rlc_fini(adev); + /* init save restore block */ + r = amdgpu_gfx_rlc_init_sr(adev, dws); + if (r) return r; - } - - /* write the sr buffer */ - dst_ptr = adev->gfx.rlc.sr_ptr; - for (i = 0; i < adev->gfx.rlc.reg_list_size; i++) - dst_ptr[i] = cpu_to_le32(src_ptr[i]); - - amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj); } if (cs_data) { @@ -2428,7 +2388,7 @@ static int gfx_v6_0_rlc_init(struct amdgpu_device *adev) (void **)&adev->gfx.rlc.cs_ptr); if (r) { dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r); - gfx_v6_0_rlc_fini(adev); + amdgpu_gfx_rlc_fini(adev); return r; } @@ -2549,8 +2509,8 @@ static int gfx_v6_0_rlc_resume(struct amdgpu_device *adev) if (!adev->gfx.rlc_fw) return -EINVAL; - gfx_v6_0_rlc_stop(adev); - gfx_v6_0_rlc_reset(adev); + adev->gfx.rlc.funcs->stop(adev); + adev->gfx.rlc.funcs->reset(adev); gfx_v6_0_init_pg(adev); gfx_v6_0_init_cg(adev); @@ -2578,7 +2538,7 @@ static int gfx_v6_0_rlc_resume(struct amdgpu_device *adev) WREG32(mmRLC_UCODE_ADDR, 0); gfx_v6_0_enable_lbpw(adev, gfx_v6_0_lbpw_supported(adev)); - gfx_v6_0_rlc_start(adev); + adev->gfx.rlc.funcs->start(adev); return 0; } @@ -3075,6 +3035,14 @@ static const struct amdgpu_gfx_funcs gfx_v6_0_gfx_funcs = { .select_me_pipe_q = &gfx_v6_0_select_me_pipe_q }; +static const struct amdgpu_rlc_funcs gfx_v6_0_rlc_funcs = { + .init = gfx_v6_0_rlc_init, + .resume = gfx_v6_0_rlc_resume, + .stop = gfx_v6_0_rlc_stop, + .reset = gfx_v6_0_rlc_reset, + .start = gfx_v6_0_rlc_start +}; + static int gfx_v6_0_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; @@ -3082,6 +3050,7 @@ static int gfx_v6_0_early_init(void *handle) adev->gfx.num_gfx_rings = GFX6_NUM_GFX_RINGS; adev->gfx.num_compute_rings = GFX6_NUM_COMPUTE_RINGS; adev->gfx.funcs = &gfx_v6_0_gfx_funcs; + adev->gfx.rlc.funcs = &gfx_v6_0_rlc_funcs; gfx_v6_0_set_ring_funcs(adev); gfx_v6_0_set_irq_funcs(adev); @@ -3114,7 +3083,7 @@ static int gfx_v6_0_sw_init(void *handle) return r; } - r = gfx_v6_0_rlc_init(adev); + r = adev->gfx.rlc.funcs->init(adev); if (r) { DRM_ERROR("Failed to init rlc BOs!\n"); return r; @@ -3165,7 +3134,7 @@ static int gfx_v6_0_sw_fini(void *handle) for (i = 0; i < adev->gfx.num_compute_rings; i++) amdgpu_ring_fini(&adev->gfx.compute_ring[i]); - gfx_v6_0_rlc_fini(adev); + amdgpu_gfx_rlc_fini(adev); return 0; } @@ -3177,7 +3146,7 @@ static int gfx_v6_0_hw_init(void *handle) gfx_v6_0_constants_init(adev); - r = gfx_v6_0_rlc_resume(adev); + r = adev->gfx.rlc.funcs->resume(adev); if (r) return r; @@ -3195,7 +3164,7 @@ static int gfx_v6_0_hw_fini(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; gfx_v6_0_cp_enable(adev, false); - gfx_v6_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); gfx_v6_0_fini_pg(adev); return 0; @@ -3393,12 +3362,31 @@ static int gfx_v6_0_eop_irq(struct amdgpu_device *adev, return 0; } +static void gfx_v6_0_fault(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry) +{ + struct amdgpu_ring *ring; + + switch (entry->ring_id) { + case 0: + ring = &adev->gfx.gfx_ring[0]; + break; + case 1: + case 2: + ring = &adev->gfx.compute_ring[entry->ring_id - 1]; + break; + default: + return; + } + drm_sched_fault(&ring->sched); +} + static int gfx_v6_0_priv_reg_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { DRM_ERROR("Illegal register access in command stream\n"); - schedule_work(&adev->reset_work); + gfx_v6_0_fault(adev, entry); return 0; } @@ -3407,7 +3395,7 @@ static int gfx_v6_0_priv_inst_irq(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry) { DRM_ERROR("Illegal instruction in command stream\n"); - schedule_work(&adev->reset_work); + gfx_v6_0_fault(adev, entry); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index 0e72bc09939a..a59e0fdf5a97 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -882,7 +882,6 @@ static const u32 kalindi_rlc_save_restore_register_list[] = static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev); static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer); -static void gfx_v7_0_init_cp_pg_table(struct amdgpu_device *adev); static void gfx_v7_0_init_pg(struct amdgpu_device *adev); static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev); @@ -2064,17 +2063,14 @@ static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring) int r; r = amdgpu_gfx_scratch_get(adev, &scratch); - if (r) { - DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r); + if (r) return r; - } + WREG32(scratch, 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r); - amdgpu_gfx_scratch_free(adev, scratch); - return r; - } + if (r) + goto error_free_scratch; + amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START)); amdgpu_ring_write(ring, 0xDEADBEEF); @@ -2086,13 +2082,10 @@ static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring) break; DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n", - ring->idx, scratch, tmp); - r = -EINVAL; - } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + +error_free_scratch: amdgpu_gfx_scratch_free(adev, scratch); return r; } @@ -2233,13 +2226,15 @@ static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring, * on the gfx ring for execution by the GPU. */ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; /* insert SWITCH_BUFFER packet before first IB in the ring frame */ - if (ctx_switch) { + if (flags & AMDGPU_HAVE_CTX_SWITCH) { amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); amdgpu_ring_write(ring, 0); } @@ -2262,11 +2257,29 @@ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, } static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); + /* Currently, there is a high possibility to get wave ID mismatch + * between ME and GDS, leading to a hw deadlock, because ME generates + * different wave IDs than the GDS expects. This situation happens + * randomly when at least 5 compute pipes use GDS ordered append. + * The wave IDs generated by ME are also wrong after suspend/resume. + * Those are probably bugs somewhere else in the kernel driver. + * + * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and + * GDS to 0 for this ring (me/pipe). + */ + if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) { + amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); + amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START); + amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id); + } + amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); amdgpu_ring_write(ring, #ifdef __BIG_ENDIAN @@ -2316,17 +2329,15 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_gfx_scratch_get(adev, &scratch); - if (r) { - DRM_ERROR("amdgpu: failed to get scratch reg (%ld).\n", r); + if (r) return r; - } + WREG32(scratch, 0xCAFEDEAD); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 256, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err1; - } + ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START)); ib.ptr[2] = 0xDEADBEEF; @@ -2338,22 +2349,16 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out\n"); r = -ETIMEDOUT; goto err2; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto err2; } tmp = RREG32(scratch); - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + if (tmp == 0xDEADBEEF) r = 0; - } else { - DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n", - scratch, tmp); + else r = -EINVAL; - } err2: amdgpu_ib_free(adev, &ib, NULL); @@ -2403,7 +2408,7 @@ static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) } else { WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK)); for (i = 0; i < adev->gfx.num_gfx_rings; i++) - adev->gfx.gfx_ring[i].ready = false; + adev->gfx.gfx_ring[i].sched.ready = false; } udelay(50); } @@ -2613,12 +2618,9 @@ static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev) /* start the ring */ gfx_v7_0_cp_gfx_start(adev); - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) return r; - } return 0; } @@ -2675,7 +2677,7 @@ static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) } else { WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK)); for (i = 0; i < adev->gfx.num_compute_rings; i++) - adev->gfx.compute_ring[i].ready = false; + adev->gfx.compute_ring[i].sched.ready = false; } udelay(50); } @@ -2781,7 +2783,7 @@ static int gfx_v7_0_mec_init(struct amdgpu_device *adev) * GFX7_MEC_HPD_SIZE * 2; r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_GTT, + AMDGPU_GEM_DOMAIN_VRAM, &adev->gfx.mec.hpd_eop_obj, &adev->gfx.mec.hpd_eop_gpu_addr, (void **)&hpd); @@ -3106,10 +3108,7 @@ static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev) for (i = 0; i < adev->gfx.num_compute_rings; i++) { ring = &adev->gfx.compute_ring[i]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) - ring->ready = false; + amdgpu_ring_test_helper(ring); } return 0; @@ -3268,18 +3267,10 @@ static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring, * The RLC is a multi-purpose microengine that handles a * variety of functions. */ -static void gfx_v7_0_rlc_fini(struct amdgpu_device *adev) -{ - amdgpu_bo_free_kernel(&adev->gfx.rlc.save_restore_obj, NULL, NULL); - amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, NULL, NULL); - amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, NULL, NULL); -} - static int gfx_v7_0_rlc_init(struct amdgpu_device *adev) { const u32 *src_ptr; - volatile u32 *dst_ptr; - u32 dws, i; + u32 dws; const struct cs_section_def *cs_data; int r; @@ -3306,66 +3297,23 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev) cs_data = adev->gfx.rlc.cs_data; if (src_ptr) { - /* save restore block */ - r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.save_restore_obj, - &adev->gfx.rlc.save_restore_gpu_addr, - (void **)&adev->gfx.rlc.sr_ptr); - if (r) { - dev_warn(adev->dev, "(%d) create, pin or map of RLC sr bo failed\n", r); - gfx_v7_0_rlc_fini(adev); + /* init save restore block */ + r = amdgpu_gfx_rlc_init_sr(adev, dws); + if (r) return r; - } - - /* write the sr buffer */ - dst_ptr = adev->gfx.rlc.sr_ptr; - for (i = 0; i < adev->gfx.rlc.reg_list_size; i++) - dst_ptr[i] = cpu_to_le32(src_ptr[i]); - amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj); } if (cs_data) { - /* clear state block */ - adev->gfx.rlc.clear_state_size = dws = gfx_v7_0_get_csb_size(adev); - - r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.clear_state_obj, - &adev->gfx.rlc.clear_state_gpu_addr, - (void **)&adev->gfx.rlc.cs_ptr); - if (r) { - dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r); - gfx_v7_0_rlc_fini(adev); + /* init clear state block */ + r = amdgpu_gfx_rlc_init_csb(adev); + if (r) return r; - } - - /* set up the cs buffer */ - dst_ptr = adev->gfx.rlc.cs_ptr; - gfx_v7_0_get_csb_buffer(adev, dst_ptr); - amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); } if (adev->gfx.rlc.cp_table_size) { - - r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size, - PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.cp_table_obj, - &adev->gfx.rlc.cp_table_gpu_addr, - (void **)&adev->gfx.rlc.cp_table_ptr); - if (r) { - dev_warn(adev->dev, "(%d) create RLC cp table bo failed\n", r); - gfx_v7_0_rlc_fini(adev); + r = amdgpu_gfx_rlc_init_cpt(adev); + if (r) return r; - } - - gfx_v7_0_init_cp_pg_table(adev); - - amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj); - } return 0; @@ -3446,7 +3394,12 @@ static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev) return orig; } -static void gfx_v7_0_enter_rlc_safe_mode(struct amdgpu_device *adev) +static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev) +{ + return true; +} + +static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev) { u32 tmp, i, mask; @@ -3468,7 +3421,7 @@ static void gfx_v7_0_enter_rlc_safe_mode(struct amdgpu_device *adev) } } -static void gfx_v7_0_exit_rlc_safe_mode(struct amdgpu_device *adev) +static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev) { u32 tmp; @@ -3545,13 +3498,13 @@ static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev) adev->gfx.rlc_feature_version = le32_to_cpu( hdr->ucode_feature_version); - gfx_v7_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); /* disable CG */ tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc; WREG32(mmRLC_CGCG_CGLS_CTRL, tmp); - gfx_v7_0_rlc_reset(adev); + adev->gfx.rlc.funcs->reset(adev); gfx_v7_0_init_pg(adev); @@ -3582,7 +3535,7 @@ static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev) if (adev->asic_type == CHIP_BONAIRE) WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0); - gfx_v7_0_rlc_start(adev); + adev->gfx.rlc.funcs->start(adev); return 0; } @@ -3784,72 +3737,12 @@ static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable) WREG32(mmRLC_PG_CNTL, data); } -static void gfx_v7_0_init_cp_pg_table(struct amdgpu_device *adev) +static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev) { - const __le32 *fw_data; - volatile u32 *dst_ptr; - int me, i, max_me = 4; - u32 bo_offset = 0; - u32 table_offset, table_size; - if (adev->asic_type == CHIP_KAVERI) - max_me = 5; - - if (adev->gfx.rlc.cp_table_ptr == NULL) - return; - - /* write the cp table buffer */ - dst_ptr = adev->gfx.rlc.cp_table_ptr; - for (me = 0; me < max_me; me++) { - if (me == 0) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; - fw_data = (const __le32 *) - (adev->gfx.ce_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 1) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; - fw_data = (const __le32 *) - (adev->gfx.pfp_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 2) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; - fw_data = (const __le32 *) - (adev->gfx.me_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 3) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; - fw_data = (const __le32 *) - (adev->gfx.mec_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; - fw_data = (const __le32 *) - (adev->gfx.mec2_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } - - for (i = 0; i < table_size; i ++) { - dst_ptr[bo_offset + i] = - cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); - } - - bo_offset += table_size; - } + return 5; + else + return 4; } static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev, @@ -4288,8 +4181,17 @@ static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = { }; static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = { - .enter_safe_mode = gfx_v7_0_enter_rlc_safe_mode, - .exit_safe_mode = gfx_v7_0_exit_rlc_safe_mode + .is_rlc_enabled = gfx_v7_0_is_rlc_enabled, + .set_safe_mode = gfx_v7_0_set_safe_mode, + .unset_safe_mode = gfx_v7_0_unset_safe_mode, + .init = gfx_v7_0_rlc_init, + .get_csb_size = gfx_v7_0_get_csb_size, + .get_csb_buffer = gfx_v7_0_get_csb_buffer, + .get_cp_table_num = gfx_v7_0_cp_pg_table_num, + .resume = gfx_v7_0_rlc_resume, + .stop = gfx_v7_0_rlc_stop, + .reset = gfx_v7_0_rlc_reset, + .start = gfx_v7_0_rlc_start }; static int gfx_v7_0_early_init(void *handle) @@ -4477,7 +4379,7 @@ static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id, ring->ring_obj = NULL; ring->use_doorbell = true; - ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id; + ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id; sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue); irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP @@ -4540,7 +4442,7 @@ static int gfx_v7_0_sw_init(void *handle) return r; } - r = gfx_v7_0_rlc_init(adev); + r = adev->gfx.rlc.funcs->init(adev); if (r) { DRM_ERROR("Failed to init rlc BOs!\n"); return r; @@ -4604,7 +4506,7 @@ static int gfx_v7_0_sw_fini(void *handle) amdgpu_ring_fini(&adev->gfx.compute_ring[i]); gfx_v7_0_cp_compute_fini(adev); - gfx_v7_0_rlc_fini(adev); + amdgpu_gfx_rlc_fini(adev); gfx_v7_0_mec_fini(adev); amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, &adev->gfx.rlc.clear_state_gpu_addr, @@ -4627,7 +4529,7 @@ static int gfx_v7_0_hw_init(void *handle) gfx_v7_0_constants_init(adev); /* init rlc */ - r = gfx_v7_0_rlc_resume(adev); + r = adev->gfx.rlc.funcs->resume(adev); if (r) return r; @@ -4645,7 +4547,7 @@ static int gfx_v7_0_hw_fini(void *handle) amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); gfx_v7_0_cp_enable(adev, false); - gfx_v7_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); gfx_v7_0_fini_pg(adev); return 0; @@ -4730,7 +4632,7 @@ static int gfx_v7_0_soft_reset(void *handle) gfx_v7_0_update_cg(adev, false); /* stop the rlc */ - gfx_v7_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); /* Disable GFX parsing/prefetching */ WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK); @@ -4959,12 +4861,36 @@ static int gfx_v7_0_eop_irq(struct amdgpu_device *adev, return 0; } +static void gfx_v7_0_fault(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry) +{ + struct amdgpu_ring *ring; + u8 me_id, pipe_id; + int i; + + me_id = (entry->ring_id & 0x0c) >> 2; + pipe_id = (entry->ring_id & 0x03) >> 0; + switch (me_id) { + case 0: + drm_sched_fault(&adev->gfx.gfx_ring[0].sched); + break; + case 1: + case 2: + for (i = 0; i < adev->gfx.num_compute_rings; i++) { + ring = &adev->gfx.compute_ring[i]; + if ((ring->me == me_id) && (ring->pipe == pipe_id)) + drm_sched_fault(&ring->sched); + } + break; + } +} + static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { DRM_ERROR("Illegal register access in command stream\n"); - schedule_work(&adev->reset_work); + gfx_v7_0_fault(adev, entry); return 0; } @@ -4974,7 +4900,7 @@ static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev, { DRM_ERROR("Illegal instruction in command stream\n"); // XXX soft reset the gfx block only - schedule_work(&adev->reset_work); + gfx_v7_0_fault(adev, entry); return 0; } @@ -5090,7 +5016,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = { 7 + /* gfx_v7_0_ring_emit_pipeline_sync */ CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */ 7 + 7 + 7, /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */ - .emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_compute */ + .emit_ib_size = 7, /* gfx_v7_0_ring_emit_ib_compute */ .emit_ib = gfx_v7_0_ring_emit_ib_compute, .emit_fence = gfx_v7_0_ring_emit_fence_compute, .emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync, @@ -5147,6 +5073,7 @@ static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev) adev->gds.mem.total_size = RREG32(mmGDS_VMID0_SIZE); adev->gds.gws.total_size = 64; adev->gds.oa.total_size = 16; + adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID); if (adev->gds.mem.total_size == 64 * 1024) { adev->gds.mem.gfx_partition_size = 4096; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 617b0c8908a3..b8e50a34bdb3 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -44,7 +44,6 @@ #include "gca/gfx_8_0_d.h" #include "gca/gfx_8_0_enum.h" #include "gca/gfx_8_0_sh_mask.h" -#include "gca/gfx_8_0_enum.h" #include "dce/dce_10_0_d.h" #include "dce/dce_10_0_sh_mask.h" @@ -54,7 +53,7 @@ #include "ivsrcid/ivsrcid_vislands30.h" #define GFX8_NUM_GFX_RINGS 1 -#define GFX8_MEC_HPD_SIZE 2048 +#define GFX8_MEC_HPD_SIZE 4096 #define TOPAZ_GB_ADDR_CONFIG_GOLDEN 0x22010001 #define CARRIZO_GB_ADDR_CONFIG_GOLDEN 0x22010001 @@ -839,18 +838,14 @@ static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring) int r; r = amdgpu_gfx_scratch_get(adev, &scratch); - if (r) { - DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r); + if (r) return r; - } + WREG32(scratch, 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", - ring->idx, r); - amdgpu_gfx_scratch_free(adev, scratch); - return r; - } + if (r) + goto error_free_scratch; + amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START)); amdgpu_ring_write(ring, 0xDEADBEEF); @@ -862,14 +857,11 @@ static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring) break; DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n", - ring->idx, scratch, tmp); - r = -EINVAL; - } + + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + +error_free_scratch: amdgpu_gfx_scratch_free(adev, scratch); return r; } @@ -886,19 +878,16 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 16, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err1; - } + ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3); ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM; ib.ptr[2] = lower_32_bits(gpu_addr); @@ -912,22 +901,17 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out.\n"); r = -ETIMEDOUT; goto err2; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto err2; } tmp = adev->wb.wb[index]; - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + if (tmp == 0xDEADBEEF) r = 0; - } else { - DRM_ERROR("ib test on ring %d failed\n", ring->idx); + else r = -EINVAL; - } err2: amdgpu_ib_free(adev, &ib, NULL); @@ -1298,81 +1282,16 @@ static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev, buffer[count++] = cpu_to_le32(0); } -static void cz_init_cp_jump_table(struct amdgpu_device *adev) +static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev) { - const __le32 *fw_data; - volatile u32 *dst_ptr; - int me, i, max_me = 4; - u32 bo_offset = 0; - u32 table_offset, table_size; - if (adev->asic_type == CHIP_CARRIZO) - max_me = 5; - - /* write the cp table buffer */ - dst_ptr = adev->gfx.rlc.cp_table_ptr; - for (me = 0; me < max_me; me++) { - if (me == 0) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; - fw_data = (const __le32 *) - (adev->gfx.ce_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 1) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; - fw_data = (const __le32 *) - (adev->gfx.pfp_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 2) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; - fw_data = (const __le32 *) - (adev->gfx.me_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 3) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; - fw_data = (const __le32 *) - (adev->gfx.mec_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 4) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; - fw_data = (const __le32 *) - (adev->gfx.mec2_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } - - for (i = 0; i < table_size; i ++) { - dst_ptr[bo_offset + i] = - cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); - } - - bo_offset += table_size; - } -} - -static void gfx_v8_0_rlc_fini(struct amdgpu_device *adev) -{ - amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, NULL, NULL); - amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, NULL, NULL); + return 5; + else + return 4; } static int gfx_v8_0_rlc_init(struct amdgpu_device *adev) { - volatile u32 *dst_ptr; - u32 dws; const struct cs_section_def *cs_data; int r; @@ -1381,44 +1300,18 @@ static int gfx_v8_0_rlc_init(struct amdgpu_device *adev) cs_data = adev->gfx.rlc.cs_data; if (cs_data) { - /* clear state block */ - adev->gfx.rlc.clear_state_size = dws = gfx_v8_0_get_csb_size(adev); - - r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.clear_state_obj, - &adev->gfx.rlc.clear_state_gpu_addr, - (void **)&adev->gfx.rlc.cs_ptr); - if (r) { - dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r); - gfx_v8_0_rlc_fini(adev); + /* init clear state block */ + r = amdgpu_gfx_rlc_init_csb(adev); + if (r) return r; - } - - /* set up the cs buffer */ - dst_ptr = adev->gfx.rlc.cs_ptr; - gfx_v8_0_get_csb_buffer(adev, dst_ptr); - amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); } if ((adev->asic_type == CHIP_CARRIZO) || (adev->asic_type == CHIP_STONEY)) { adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */ - r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size, - PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.cp_table_obj, - &adev->gfx.rlc.cp_table_gpu_addr, - (void **)&adev->gfx.rlc.cp_table_ptr); - if (r) { - dev_warn(adev->dev, "(%d) create RLC cp table bo failed\n", r); + r = amdgpu_gfx_rlc_init_cpt(adev); + if (r) return r; - } - - cz_init_cp_jump_table(adev); - - amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj); } return 0; @@ -1443,7 +1336,7 @@ static int gfx_v8_0_mec_init(struct amdgpu_device *adev) mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE; r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_GTT, + AMDGPU_GEM_DOMAIN_VRAM, &adev->gfx.mec.hpd_eop_obj, &adev->gfx.mec.hpd_eop_gpu_addr, (void **)&hpd); @@ -1629,7 +1522,7 @@ static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev) return 0; /* bail if the compute ring is not ready */ - if (!ring->ready) + if (!ring->sched.ready) return 0; tmp = RREG32(mmGB_EDC_MODE); @@ -1997,7 +1890,7 @@ static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id, ring->ring_obj = NULL; ring->use_doorbell = true; - ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id; + ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id; ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (ring_id * GFX8_MEC_HPD_SIZE); sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue); @@ -2088,7 +1981,7 @@ static int gfx_v8_0_sw_init(void *handle) return r; } - r = gfx_v8_0_rlc_init(adev); + r = adev->gfx.rlc.funcs->init(adev); if (r) { DRM_ERROR("Failed to init rlc BOs!\n"); return r; @@ -2108,7 +2001,7 @@ static int gfx_v8_0_sw_init(void *handle) /* no gfx doorbells on iceland */ if (adev->asic_type != CHIP_TOPAZ) { ring->use_doorbell = true; - ring->doorbell_index = AMDGPU_DOORBELL_GFX_RING0; + ring->doorbell_index = adev->doorbell_index.gfx_ring0; } r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, @@ -2181,7 +2074,7 @@ static int gfx_v8_0_sw_fini(void *handle) amdgpu_gfx_kiq_fini(adev); gfx_v8_0_mec_fini(adev); - gfx_v8_0_rlc_fini(adev); + amdgpu_gfx_rlc_fini(adev); amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, &adev->gfx.rlc.clear_state_gpu_addr, (void **)&adev->gfx.rlc.cs_ptr); @@ -4175,10 +4068,15 @@ static void gfx_v8_0_rlc_start(struct amdgpu_device *adev) static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev) { - gfx_v8_0_rlc_stop(adev); - gfx_v8_0_rlc_reset(adev); + if (amdgpu_sriov_vf(adev)) { + gfx_v8_0_init_csb(adev); + return 0; + } + + adev->gfx.rlc.funcs->stop(adev); + adev->gfx.rlc.funcs->reset(adev); gfx_v8_0_init_pg(adev); - gfx_v8_0_rlc_start(adev); + adev->gfx.rlc.funcs->start(adev); return 0; } @@ -4197,7 +4095,7 @@ static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1); tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1); for (i = 0; i < adev->gfx.num_gfx_rings; i++) - adev->gfx.gfx_ring[i].ready = false; + adev->gfx.gfx_ring[i].sched.ready = false; } WREG32(mmCP_ME_CNTL, tmp); udelay(50); @@ -4322,7 +4220,7 @@ static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER, DOORBELL_RANGE_LOWER, - AMDGPU_DOORBELL_GFX_RING0); + adev->doorbell_index.gfx_ring0); WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp); WREG32(mmCP_RB_DOORBELL_RANGE_UPPER, @@ -4335,7 +4233,6 @@ static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev) u32 tmp; u32 rb_bufsz; u64 rb_addr, rptr_addr, wptr_gpu_addr; - int r; /* Set the write pointer delay */ WREG32(mmCP_RB_WPTR_DELAY, 0); @@ -4379,12 +4276,9 @@ static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev) /* start the ring */ amdgpu_ring_clear_ring(ring); gfx_v8_0_cp_gfx_start(adev); - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) - ring->ready = false; + ring->sched.ready = true; - return r; + return 0; } static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) @@ -4396,8 +4290,8 @@ static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) } else { WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK)); for (i = 0; i < adev->gfx.num_compute_rings; i++) - adev->gfx.compute_ring[i].ready = false; - adev->gfx.kiq.ring.ready = false; + adev->gfx.compute_ring[i].sched.ready = false; + adev->gfx.kiq.ring.sched.ready = false; } udelay(50); } @@ -4473,12 +4367,9 @@ static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev) amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr)); } - r = amdgpu_ring_test_ring(kiq_ring); - if (r) { - DRM_ERROR("KCQ enable failed\n"); - kiq_ring->ready = false; - } - return r; + amdgpu_ring_commit(kiq_ring); + + return 0; } static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req) @@ -4755,8 +4646,8 @@ static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring) static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev) { if (adev->asic_type > CHIP_TONGA) { - WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, AMDGPU_DOORBELL_KIQ << 2); - WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, AMDGPU_DOORBELL_MEC_RING7 << 2); + WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2); + WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2); } /* enable doorbells */ WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1); @@ -4781,7 +4672,7 @@ static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev) amdgpu_bo_kunmap(ring->mqd_obj); ring->mqd_ptr = NULL; amdgpu_bo_unreserve(ring->mqd_obj); - ring->ready = true; + ring->sched.ready = true; return 0; } @@ -4815,19 +4706,32 @@ static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev) if (r) goto done; - /* Test KCQs - reversing the order of rings seems to fix ring test failure - * after GPU reset - */ - for (i = adev->gfx.num_compute_rings - 1; i >= 0; i--) { +done: + return r; +} + +static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev) +{ + int r, i; + struct amdgpu_ring *ring; + + /* collect all the ring_tests here, gfx, kiq, compute */ + ring = &adev->gfx.gfx_ring[0]; + r = amdgpu_ring_test_helper(ring); + if (r) + return r; + + ring = &adev->gfx.kiq.ring; + r = amdgpu_ring_test_helper(ring); + if (r) + return r; + + for (i = 0; i < adev->gfx.num_compute_rings; i++) { ring = &adev->gfx.compute_ring[i]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) - ring->ready = false; + amdgpu_ring_test_helper(ring); } -done: - return r; + return 0; } static int gfx_v8_0_cp_resume(struct amdgpu_device *adev) @@ -4848,6 +4752,11 @@ static int gfx_v8_0_cp_resume(struct amdgpu_device *adev) r = gfx_v8_0_kcq_resume(adev); if (r) return r; + + r = gfx_v8_0_cp_test_all_rings(adev); + if (r) + return r; + gfx_v8_0_enable_gui_idle_interrupt(adev, true); return 0; @@ -4867,7 +4776,7 @@ static int gfx_v8_0_hw_init(void *handle) gfx_v8_0_init_golden_registers(adev); gfx_v8_0_constants_init(adev); - r = gfx_v8_0_rlc_resume(adev); + r = adev->gfx.rlc.funcs->resume(adev); if (r) return r; @@ -4899,7 +4808,7 @@ static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev) amdgpu_ring_write(kiq_ring, 0); amdgpu_ring_write(kiq_ring, 0); } - r = amdgpu_ring_test_ring(kiq_ring); + r = amdgpu_ring_test_helper(kiq_ring); if (r) DRM_ERROR("KCQ disable failed\n"); @@ -4973,16 +4882,16 @@ static int gfx_v8_0_hw_fini(void *handle) pr_debug("For SRIOV client, shouldn't do anything.\n"); return 0; } - adev->gfx.rlc.funcs->enter_safe_mode(adev); + amdgpu_gfx_rlc_enter_safe_mode(adev); if (!gfx_v8_0_wait_for_idle(adev)) gfx_v8_0_cp_enable(adev, false); else pr_err("cp is busy, skip halt cp\n"); if (!gfx_v8_0_wait_for_rlc_idle(adev)) - gfx_v8_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); else pr_err("rlc is busy, skip halt rlc\n"); - adev->gfx.rlc.funcs->exit_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); return 0; } @@ -5061,17 +4970,16 @@ static bool gfx_v8_0_check_soft_reset(void *handle) static int gfx_v8_0_pre_soft_reset(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; - u32 grbm_soft_reset = 0, srbm_soft_reset = 0; + u32 grbm_soft_reset = 0; if ((!adev->gfx.grbm_soft_reset) && (!adev->gfx.srbm_soft_reset)) return 0; grbm_soft_reset = adev->gfx.grbm_soft_reset; - srbm_soft_reset = adev->gfx.srbm_soft_reset; /* stop the rlc */ - gfx_v8_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) || REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX)) @@ -5165,14 +5073,13 @@ static int gfx_v8_0_soft_reset(void *handle) static int gfx_v8_0_post_soft_reset(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; - u32 grbm_soft_reset = 0, srbm_soft_reset = 0; + u32 grbm_soft_reset = 0; if ((!adev->gfx.grbm_soft_reset) && (!adev->gfx.srbm_soft_reset)) return 0; grbm_soft_reset = adev->gfx.grbm_soft_reset; - srbm_soft_reset = adev->gfx.srbm_soft_reset; if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) || REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) || @@ -5197,7 +5104,9 @@ static int gfx_v8_0_post_soft_reset(void *handle) REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX)) gfx_v8_0_cp_gfx_resume(adev); - gfx_v8_0_rlc_start(adev); + gfx_v8_0_cp_test_all_rings(adev); + + adev->gfx.rlc.funcs->start(adev); return 0; } @@ -5445,7 +5354,7 @@ static int gfx_v8_0_set_powergating_state(void *handle, AMD_PG_SUPPORT_RLC_SMU_HS | AMD_PG_SUPPORT_CP | AMD_PG_SUPPORT_GFX_DMG)) - adev->gfx.rlc.funcs->enter_safe_mode(adev); + amdgpu_gfx_rlc_enter_safe_mode(adev); switch (adev->asic_type) { case CHIP_CARRIZO: case CHIP_STONEY: @@ -5499,7 +5408,7 @@ static int gfx_v8_0_set_powergating_state(void *handle, AMD_PG_SUPPORT_RLC_SMU_HS | AMD_PG_SUPPORT_CP | AMD_PG_SUPPORT_GFX_DMG)) - adev->gfx.rlc.funcs->exit_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); return 0; } @@ -5593,57 +5502,53 @@ static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev, #define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001 #define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e -static void iceland_enter_rlc_safe_mode(struct amdgpu_device *adev) +static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev) { - u32 data; - unsigned i; + uint32_t rlc_setting; - data = RREG32(mmRLC_CNTL); - if (!(data & RLC_CNTL__RLC_ENABLE_F32_MASK)) - return; + rlc_setting = RREG32(mmRLC_CNTL); + if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) + return false; - if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG)) { - data |= RLC_SAFE_MODE__CMD_MASK; - data &= ~RLC_SAFE_MODE__MESSAGE_MASK; - data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); - WREG32(mmRLC_SAFE_MODE, data); + return true; +} - for (i = 0; i < adev->usec_timeout; i++) { - if ((RREG32(mmRLC_GPM_STAT) & - (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | - RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) == - (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | - RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) - break; - udelay(1); - } +static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev) +{ + uint32_t data; + unsigned i; + data = RREG32(mmRLC_CNTL); + data |= RLC_SAFE_MODE__CMD_MASK; + data &= ~RLC_SAFE_MODE__MESSAGE_MASK; + data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); + WREG32(mmRLC_SAFE_MODE, data); - for (i = 0; i < adev->usec_timeout; i++) { - if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) - break; - udelay(1); - } - adev->gfx.rlc.in_safe_mode = true; + /* wait for RLC_SAFE_MODE */ + for (i = 0; i < adev->usec_timeout; i++) { + if ((RREG32(mmRLC_GPM_STAT) & + (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | + RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) == + (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | + RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) + break; + udelay(1); + } + for (i = 0; i < adev->usec_timeout; i++) { + if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) + break; + udelay(1); } } -static void iceland_exit_rlc_safe_mode(struct amdgpu_device *adev) +static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev) { - u32 data = 0; + uint32_t data; unsigned i; data = RREG32(mmRLC_CNTL); - if (!(data & RLC_CNTL__RLC_ENABLE_F32_MASK)) - return; - - if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG)) { - if (adev->gfx.rlc.in_safe_mode) { - data |= RLC_SAFE_MODE__CMD_MASK; - data &= ~RLC_SAFE_MODE__MESSAGE_MASK; - WREG32(mmRLC_SAFE_MODE, data); - adev->gfx.rlc.in_safe_mode = false; - } - } + data |= RLC_SAFE_MODE__CMD_MASK; + data &= ~RLC_SAFE_MODE__MESSAGE_MASK; + WREG32(mmRLC_SAFE_MODE, data); for (i = 0; i < adev->usec_timeout; i++) { if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) @@ -5653,8 +5558,17 @@ static void iceland_exit_rlc_safe_mode(struct amdgpu_device *adev) } static const struct amdgpu_rlc_funcs iceland_rlc_funcs = { - .enter_safe_mode = iceland_enter_rlc_safe_mode, - .exit_safe_mode = iceland_exit_rlc_safe_mode + .is_rlc_enabled = gfx_v8_0_is_rlc_enabled, + .set_safe_mode = gfx_v8_0_set_safe_mode, + .unset_safe_mode = gfx_v8_0_unset_safe_mode, + .init = gfx_v8_0_rlc_init, + .get_csb_size = gfx_v8_0_get_csb_size, + .get_csb_buffer = gfx_v8_0_get_csb_buffer, + .get_cp_table_num = gfx_v8_0_cp_jump_table_num, + .resume = gfx_v8_0_rlc_resume, + .stop = gfx_v8_0_rlc_stop, + .reset = gfx_v8_0_rlc_reset, + .start = gfx_v8_0_rlc_start }; static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev, @@ -5662,7 +5576,7 @@ static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev { uint32_t temp, data; - adev->gfx.rlc.funcs->enter_safe_mode(adev); + amdgpu_gfx_rlc_enter_safe_mode(adev); /* It is disabled by HW by default */ if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { @@ -5758,7 +5672,7 @@ static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev gfx_v8_0_wait_for_rlc_serdes(adev); } - adev->gfx.rlc.funcs->exit_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); } static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev, @@ -5768,7 +5682,7 @@ static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL); - adev->gfx.rlc.funcs->enter_safe_mode(adev); + amdgpu_gfx_rlc_enter_safe_mode(adev); if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { temp1 = data1 = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); @@ -5851,7 +5765,7 @@ static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev gfx_v8_0_wait_for_rlc_serdes(adev); - adev->gfx.rlc.funcs->exit_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); } static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev, bool enable) @@ -6131,9 +6045,11 @@ static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring) } static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; if (ib->flags & AMDGPU_IB_FLAG_CE) @@ -6161,11 +6077,29 @@ static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, } static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); + /* Currently, there is a high possibility to get wave ID mismatch + * between ME and GDS, leading to a hw deadlock, because ME generates + * different wave IDs than the GDS expects. This situation happens + * randomly when at least 5 compute pipes use GDS ordered append. + * The wave IDs generated by ME are also wrong after suspend/resume. + * Those are probably bugs somewhere else in the kernel driver. + * + * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and + * GDS to 0 for this ring (me/pipe). + */ + if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) { + amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); + amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START); + amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id); + } + amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); amdgpu_ring_write(ring, #ifdef __BIG_ENDIAN @@ -6738,12 +6672,39 @@ static int gfx_v8_0_eop_irq(struct amdgpu_device *adev, return 0; } +static void gfx_v8_0_fault(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry) +{ + u8 me_id, pipe_id, queue_id; + struct amdgpu_ring *ring; + int i; + + me_id = (entry->ring_id & 0x0c) >> 2; + pipe_id = (entry->ring_id & 0x03) >> 0; + queue_id = (entry->ring_id & 0x70) >> 4; + + switch (me_id) { + case 0: + drm_sched_fault(&adev->gfx.gfx_ring[0].sched); + break; + case 1: + case 2: + for (i = 0; i < adev->gfx.num_compute_rings; i++) { + ring = &adev->gfx.compute_ring[i]; + if (ring->me == me_id && ring->pipe == pipe_id && + ring->queue == queue_id) + drm_sched_fault(&ring->sched); + } + break; + } +} + static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { DRM_ERROR("Illegal register access in command stream\n"); - schedule_work(&adev->reset_work); + gfx_v8_0_fault(adev, entry); return 0; } @@ -6752,7 +6713,7 @@ static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry) { DRM_ERROR("Illegal instruction in command stream\n"); - schedule_work(&adev->reset_work); + gfx_v8_0_fault(adev, entry); return 0; } @@ -6945,7 +6906,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = { 7 + /* gfx_v8_0_ring_emit_pipeline_sync */ VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */ 7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */ - .emit_ib_size = 4, /* gfx_v8_0_ring_emit_ib_compute */ + .emit_ib_size = 7, /* gfx_v8_0_ring_emit_ib_compute */ .emit_ib = gfx_v8_0_ring_emit_ib_compute, .emit_fence = gfx_v8_0_ring_emit_fence_compute, .emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync, @@ -6975,11 +6936,9 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = { 7 + /* gfx_v8_0_ring_emit_pipeline_sync */ 17 + /* gfx_v8_0_ring_emit_vm_flush */ 7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */ - .emit_ib_size = 4, /* gfx_v8_0_ring_emit_ib_compute */ - .emit_ib = gfx_v8_0_ring_emit_ib_compute, + .emit_ib_size = 7, /* gfx_v8_0_ring_emit_ib_compute */ .emit_fence = gfx_v8_0_ring_emit_fence_kiq, .test_ring = gfx_v8_0_ring_test_ring, - .test_ib = gfx_v8_0_ring_test_ib, .insert_nop = amdgpu_ring_insert_nop, .pad_ib = amdgpu_ring_generic_pad_ib, .emit_rreg = gfx_v8_0_ring_emit_rreg, @@ -7053,6 +7012,7 @@ static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev) adev->gds.mem.total_size = RREG32(mmGDS_VMID0_SIZE); adev->gds.gws.total_size = 64; adev->gds.oa.total_size = 16; + adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID); if (adev->gds.mem.total_size == 64 * 1024) { adev->gds.mem.gfx_partition_size = 4096; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index 21363b2b2ee5..a11db2b1a63f 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -41,7 +41,7 @@ #include "ivsrcid/gfx/irqsrcs_gfx_9_0.h" #define GFX9_NUM_GFX_RINGS 1 -#define GFX9_MEC_HPD_SIZE 2048 +#define GFX9_MEC_HPD_SIZE 4096 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L #define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L @@ -86,6 +86,7 @@ MODULE_FIRMWARE("amdgpu/picasso_me.bin"); MODULE_FIRMWARE("amdgpu/picasso_mec.bin"); MODULE_FIRMWARE("amdgpu/picasso_mec2.bin"); MODULE_FIRMWARE("amdgpu/picasso_rlc.bin"); +MODULE_FIRMWARE("amdgpu/picasso_rlc_am4.bin"); MODULE_FIRMWARE("amdgpu/raven2_ce.bin"); MODULE_FIRMWARE("amdgpu/raven2_pfp.bin"); @@ -112,7 +113,10 @@ static const struct soc15_reg_golden golden_settings_gc_9_0[] = SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68), SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197), SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000), - SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff) + SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff), + SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000000, 0x00000800), + SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000000, 0x00000800), + SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00000000, 0x00008000) }; static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] = @@ -134,10 +138,7 @@ static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] = SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000), SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107), SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800), - SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080), - SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000000, 0x00000800), - SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000000, 0x00000800), - SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00000000, 0x00008000) + SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080) }; static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] = @@ -219,6 +220,7 @@ static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] = static const struct soc15_reg_golden golden_settings_gc_9_x_common[] = { + SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff), SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000), SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382) }; @@ -396,18 +398,14 @@ static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring) int r; r = amdgpu_gfx_scratch_get(adev, &scratch); - if (r) { - DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r); + if (r) return r; - } + WREG32(scratch, 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", - ring->idx, r); - amdgpu_gfx_scratch_free(adev, scratch); - return r; - } + if (r) + goto error_free_scratch; + amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START)); amdgpu_ring_write(ring, 0xDEADBEEF); @@ -419,14 +417,11 @@ static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring) break; DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n", - ring->idx, scratch, tmp); - r = -EINVAL; - } + + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + +error_free_scratch: amdgpu_gfx_scratch_free(adev, scratch); return r; } @@ -443,19 +438,16 @@ static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 16, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err1; - } + ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3); ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM; ib.ptr[2] = lower_32_bits(gpu_addr); @@ -469,22 +461,17 @@ static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out.\n"); - r = -ETIMEDOUT; - goto err2; + r = -ETIMEDOUT; + goto err2; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); - goto err2; + goto err2; } tmp = adev->wb.wb[index]; - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); - r = 0; - } else { - DRM_ERROR("ib test on ring %d failed\n", ring->idx); - r = -EINVAL; - } + if (tmp == 0xDEADBEEF) + r = 0; + else + r = -EINVAL; err2: amdgpu_ib_free(adev, &ib, NULL); @@ -660,7 +647,20 @@ static int gfx_v9_0_init_microcode(struct amdgpu_device *adev) adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); - snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name); + /* + * For Picasso && AM4 SOCKET board, we use picasso_rlc_am4.bin + * instead of picasso_rlc.bin. + * Judgment method: + * PCO AM4: revision >= 0xC8 && revision <= 0xCF + * or revision >= 0xD8 && revision <= 0xDF + * otherwise is PCO FP5 + */ + if (!strcmp(chip_name, "picasso") && + (((adev->pdev->revision >= 0xC8) && (adev->pdev->revision <= 0xCF)) || + ((adev->pdev->revision >= 0xD8) && (adev->pdev->revision <= 0xDF)))) + snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc_am4.bin", chip_name); + else + snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name); err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev); if (err) goto out; @@ -1065,85 +1065,13 @@ static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable) WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0); } -static void rv_init_cp_jump_table(struct amdgpu_device *adev) -{ - const __le32 *fw_data; - volatile u32 *dst_ptr; - int me, i, max_me = 5; - u32 bo_offset = 0; - u32 table_offset, table_size; - - /* write the cp table buffer */ - dst_ptr = adev->gfx.rlc.cp_table_ptr; - for (me = 0; me < max_me; me++) { - if (me == 0) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; - fw_data = (const __le32 *) - (adev->gfx.ce_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 1) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; - fw_data = (const __le32 *) - (adev->gfx.pfp_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 2) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; - fw_data = (const __le32 *) - (adev->gfx.me_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 3) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; - fw_data = (const __le32 *) - (adev->gfx.mec_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 4) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; - fw_data = (const __le32 *) - (adev->gfx.mec2_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } - - for (i = 0; i < table_size; i ++) { - dst_ptr[bo_offset + i] = - cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); - } - - bo_offset += table_size; - } -} - -static void gfx_v9_0_rlc_fini(struct amdgpu_device *adev) +static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev) { - /* clear state block */ - amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, - &adev->gfx.rlc.clear_state_gpu_addr, - (void **)&adev->gfx.rlc.cs_ptr); - - /* jump table block */ - amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, - &adev->gfx.rlc.cp_table_gpu_addr, - (void **)&adev->gfx.rlc.cp_table_ptr); + return 5; } static int gfx_v9_0_rlc_init(struct amdgpu_device *adev) { - volatile u32 *dst_ptr; - u32 dws; const struct cs_section_def *cs_data; int r; @@ -1152,45 +1080,18 @@ static int gfx_v9_0_rlc_init(struct amdgpu_device *adev) cs_data = adev->gfx.rlc.cs_data; if (cs_data) { - /* clear state block */ - adev->gfx.rlc.clear_state_size = dws = gfx_v9_0_get_csb_size(adev); - r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.clear_state_obj, - &adev->gfx.rlc.clear_state_gpu_addr, - (void **)&adev->gfx.rlc.cs_ptr); - if (r) { - dev_err(adev->dev, "(%d) failed to create rlc csb bo\n", - r); - gfx_v9_0_rlc_fini(adev); + /* init clear state block */ + r = amdgpu_gfx_rlc_init_csb(adev); + if (r) return r; - } - /* set up the cs buffer */ - dst_ptr = adev->gfx.rlc.cs_ptr; - gfx_v9_0_get_csb_buffer(adev, dst_ptr); - amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); - amdgpu_bo_unpin(adev->gfx.rlc.clear_state_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); } if (adev->asic_type == CHIP_RAVEN) { /* TODO: double check the cp_table_size for RV */ adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */ - r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size, - PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.cp_table_obj, - &adev->gfx.rlc.cp_table_gpu_addr, - (void **)&adev->gfx.rlc.cp_table_ptr); - if (r) { - dev_err(adev->dev, - "(%d) failed to create cp table bo\n", r); - gfx_v9_0_rlc_fini(adev); + r = amdgpu_gfx_rlc_init_cpt(adev); + if (r) return r; - } - - rv_init_cp_jump_table(adev); - amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj); } switch (adev->asic_type) { @@ -1264,7 +1165,7 @@ static int gfx_v9_0_mec_init(struct amdgpu_device *adev) mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE; r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_GTT, + AMDGPU_GEM_DOMAIN_VRAM, &adev->gfx.mec.hpd_eop_obj, &adev->gfx.mec.hpd_eop_gpu_addr, (void **)&hpd); @@ -1635,8 +1536,8 @@ static int gfx_v9_0_ngg_en(struct amdgpu_device *adev) /* Clear GDS reserved memory */ r = amdgpu_ring_alloc(ring, 17); if (r) { - DRM_ERROR("amdgpu: NGG failed to lock ring %d (%d).\n", - ring->idx, r); + DRM_ERROR("amdgpu: NGG failed to lock ring %s (%d).\n", + ring->name, r); return r; } @@ -1680,7 +1581,7 @@ static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id, ring->ring_obj = NULL; ring->use_doorbell = true; - ring->doorbell_index = (AMDGPU_DOORBELL_MEC_RING0 + ring_id) << 1; + ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1; ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (ring_id * GFX9_MEC_HPD_SIZE); sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue); @@ -1748,7 +1649,7 @@ static int gfx_v9_0_sw_init(void *handle) return r; } - r = gfx_v9_0_rlc_init(adev); + r = adev->gfx.rlc.funcs->init(adev); if (r) { DRM_ERROR("Failed to init rlc BOs!\n"); return r; @@ -1769,7 +1670,7 @@ static int gfx_v9_0_sw_init(void *handle) else sprintf(ring->name, "gfx_%d", i); ring->use_doorbell = true; - ring->doorbell_index = AMDGPU_DOORBELL64_GFX_RING0 << 1; + ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1; r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_EOP); if (r) @@ -2499,13 +2400,11 @@ static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev) return 0; } - gfx_v9_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); /* disable CG */ WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0); - gfx_v9_0_rlc_reset(adev); - gfx_v9_0_init_pg(adev); if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { @@ -2515,15 +2414,24 @@ static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev) return r; } - if (adev->asic_type == CHIP_RAVEN || - adev->asic_type == CHIP_VEGA20) { - if (amdgpu_lbpw != 0) + switch (adev->asic_type) { + case CHIP_RAVEN: + if (amdgpu_lbpw == 0) + gfx_v9_0_enable_lbpw(adev, false); + else + gfx_v9_0_enable_lbpw(adev, true); + break; + case CHIP_VEGA20: + if (amdgpu_lbpw > 0) gfx_v9_0_enable_lbpw(adev, true); else gfx_v9_0_enable_lbpw(adev, false); + break; + default: + break; } - gfx_v9_0_rlc_start(adev); + adev->gfx.rlc.funcs->start(adev); return 0; } @@ -2538,7 +2446,7 @@ static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1); if (!enable) { for (i = 0; i < adev->gfx.num_gfx_rings; i++) - adev->gfx.gfx_ring[i].ready = false; + adev->gfx.gfx_ring[i].sched.ready = false; } WREG32_SOC15(GC, 0, mmCP_ME_CNTL, tmp); udelay(50); @@ -2728,7 +2636,7 @@ static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev) /* start the ring */ gfx_v9_0_cp_gfx_start(adev); - ring->ready = true; + ring->sched.ready = true; return 0; } @@ -2743,8 +2651,8 @@ static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) WREG32_SOC15(GC, 0, mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK)); for (i = 0; i < adev->gfx.num_compute_rings; i++) - adev->gfx.compute_ring[i].ready = false; - adev->gfx.kiq.ring.ready = false; + adev->gfx.compute_ring[i].sched.ready = false; + adev->gfx.kiq.ring.sched.ready = false; } udelay(50); } @@ -2867,11 +2775,9 @@ static int gfx_v9_0_kiq_kcq_enable(struct amdgpu_device *adev) amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr)); } - r = amdgpu_ring_test_ring(kiq_ring); - if (r) { + r = amdgpu_ring_test_helper(kiq_ring); + if (r) DRM_ERROR("KCQ enable failed\n"); - kiq_ring->ready = false; - } return r; } @@ -3089,9 +2995,9 @@ static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring) /* enable the doorbell if requested */ if (ring->use_doorbell) { WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER, - (AMDGPU_DOORBELL64_KIQ *2) << 2); + (adev->doorbell_index.kiq * 2) << 2); WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER, - (AMDGPU_DOORBELL64_USERQUEUE_END * 2) << 2); + (adev->doorbell_index.userqueue_end * 2) << 2); } WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, @@ -3250,7 +3156,7 @@ static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev) amdgpu_bo_kunmap(ring->mqd_obj); ring->mqd_ptr = NULL; amdgpu_bo_unreserve(ring->mqd_obj); - ring->ready = true; + ring->sched.ready = true; return 0; } @@ -3315,19 +3221,13 @@ static int gfx_v9_0_cp_resume(struct amdgpu_device *adev) return r; ring = &adev->gfx.gfx_ring[0]; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) return r; - } for (i = 0; i < adev->gfx.num_compute_rings; i++) { ring = &adev->gfx.compute_ring[i]; - - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) - ring->ready = false; + amdgpu_ring_test_helper(ring); } gfx_v9_0_enable_gui_idle_interrupt(adev, true); @@ -3354,7 +3254,7 @@ static int gfx_v9_0_hw_init(void *handle) if (r) return r; - r = gfx_v9_0_rlc_resume(adev); + r = adev->gfx.rlc.funcs->resume(adev); if (r) return r; @@ -3392,7 +3292,7 @@ static int gfx_v9_0_kcq_disable(struct amdgpu_device *adev) amdgpu_ring_write(kiq_ring, 0); amdgpu_ring_write(kiq_ring, 0); } - r = amdgpu_ring_test_ring(kiq_ring); + r = amdgpu_ring_test_helper(kiq_ring); if (r) DRM_ERROR("KCQ disable failed\n"); @@ -3434,7 +3334,7 @@ static int gfx_v9_0_hw_fini(void *handle) } gfx_v9_0_cp_enable(adev, false); - gfx_v9_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); gfx_v9_0_csb_vram_unpin(adev); @@ -3509,7 +3409,7 @@ static int gfx_v9_0_soft_reset(void *handle) if (grbm_soft_reset) { /* stop the rlc */ - gfx_v9_0_rlc_stop(adev); + adev->gfx.rlc.funcs->stop(adev); /* Disable GFX parsing/prefetching */ gfx_v9_0_cp_gfx_enable(adev, false); @@ -3608,64 +3508,47 @@ static int gfx_v9_0_late_init(void *handle) return 0; } -static void gfx_v9_0_enter_rlc_safe_mode(struct amdgpu_device *adev) +static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev) { - uint32_t rlc_setting, data; - unsigned i; - - if (adev->gfx.rlc.in_safe_mode) - return; + uint32_t rlc_setting; /* if RLC is not enabled, do nothing */ rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL); if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) - return; - - if (adev->cg_flags & - (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG | - AMD_CG_SUPPORT_GFX_3D_CGCG)) { - data = RLC_SAFE_MODE__CMD_MASK; - data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); - WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); + return false; - /* wait for RLC_SAFE_MODE */ - for (i = 0; i < adev->usec_timeout; i++) { - if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) - break; - udelay(1); - } - adev->gfx.rlc.in_safe_mode = true; - } + return true; } -static void gfx_v9_0_exit_rlc_safe_mode(struct amdgpu_device *adev) +static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev) { - uint32_t rlc_setting, data; + uint32_t data; + unsigned i; - if (!adev->gfx.rlc.in_safe_mode) - return; + data = RLC_SAFE_MODE__CMD_MASK; + data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); + WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); - /* if RLC is not enabled, do nothing */ - rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL); - if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) - return; - - if (adev->cg_flags & - (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG)) { - /* - * Try to exit safe mode only if it is already in safe - * mode. - */ - data = RLC_SAFE_MODE__CMD_MASK; - WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); - adev->gfx.rlc.in_safe_mode = false; + /* wait for RLC_SAFE_MODE */ + for (i = 0; i < adev->usec_timeout; i++) { + if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) + break; + udelay(1); } } +static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev) +{ + uint32_t data; + + data = RLC_SAFE_MODE__CMD_MASK; + WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); +} + static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev, bool enable) { - gfx_v9_0_enter_rlc_safe_mode(adev); + amdgpu_gfx_rlc_enter_safe_mode(adev); if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) { gfx_v9_0_enable_gfx_cg_power_gating(adev, true); @@ -3676,7 +3559,7 @@ static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev, gfx_v9_0_enable_gfx_pipeline_powergating(adev, false); } - gfx_v9_0_exit_rlc_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); } static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev, @@ -3703,6 +3586,8 @@ static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev { uint32_t data, def; + amdgpu_gfx_rlc_enter_safe_mode(adev); + /* It is disabled by HW by default */ if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { /* 1 - RLC_CGTT_MGCG_OVERRIDE */ @@ -3767,6 +3652,8 @@ static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data); } } + + amdgpu_gfx_rlc_exit_safe_mode(adev); } static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev, @@ -3774,7 +3661,7 @@ static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev, { uint32_t data, def; - adev->gfx.rlc.funcs->enter_safe_mode(adev); + amdgpu_gfx_rlc_enter_safe_mode(adev); /* Enable 3D CGCG/CGLS */ if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)) { @@ -3814,7 +3701,7 @@ static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev, WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data); } - adev->gfx.rlc.funcs->exit_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); } static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev, @@ -3822,7 +3709,7 @@ static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev { uint32_t def, data; - adev->gfx.rlc.funcs->enter_safe_mode(adev); + amdgpu_gfx_rlc_enter_safe_mode(adev); if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); @@ -3862,7 +3749,7 @@ static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data); } - adev->gfx.rlc.funcs->exit_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); } static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev, @@ -3891,8 +3778,17 @@ static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev, } static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = { - .enter_safe_mode = gfx_v9_0_enter_rlc_safe_mode, - .exit_safe_mode = gfx_v9_0_exit_rlc_safe_mode + .is_rlc_enabled = gfx_v9_0_is_rlc_enabled, + .set_safe_mode = gfx_v9_0_set_safe_mode, + .unset_safe_mode = gfx_v9_0_unset_safe_mode, + .init = gfx_v9_0_rlc_init, + .get_csb_size = gfx_v9_0_get_csb_size, + .get_csb_buffer = gfx_v9_0_get_csb_buffer, + .get_cp_table_num = gfx_v9_0_cp_jump_table_num, + .resume = gfx_v9_0_rlc_resume, + .stop = gfx_v9_0_rlc_stop, + .reset = gfx_v9_0_rlc_reset, + .start = gfx_v9_0_rlc_start }; static int gfx_v9_0_set_powergating_state(void *handle, @@ -4073,9 +3969,11 @@ static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) } static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; if (ib->flags & AMDGPU_IB_FLAG_CE) @@ -4104,20 +4002,38 @@ static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, } static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) -{ - u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) +{ + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); + + /* Currently, there is a high possibility to get wave ID mismatch + * between ME and GDS, leading to a hw deadlock, because ME generates + * different wave IDs than the GDS expects. This situation happens + * randomly when at least 5 compute pipes use GDS ordered append. + * The wave IDs generated by ME are also wrong after suspend/resume. + * Those are probably bugs somewhere else in the kernel driver. + * + * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and + * GDS to 0 for this ring (me/pipe). + */ + if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) { + amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); + amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID); + amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id); + } - amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); + amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); BUG_ON(ib->gpu_addr & 0x3); /* Dword align */ - amdgpu_ring_write(ring, + amdgpu_ring_write(ring, #ifdef __BIG_ENDIAN - (2 << 0) | + (2 << 0) | #endif - lower_32_bits(ib->gpu_addr)); - amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); - amdgpu_ring_write(ring, control); + lower_32_bits(ib->gpu_addr)); + amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); + amdgpu_ring_write(ring, control); } static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, @@ -4696,12 +4612,39 @@ static int gfx_v9_0_eop_irq(struct amdgpu_device *adev, return 0; } +static void gfx_v9_0_fault(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry) +{ + u8 me_id, pipe_id, queue_id; + struct amdgpu_ring *ring; + int i; + + me_id = (entry->ring_id & 0x0c) >> 2; + pipe_id = (entry->ring_id & 0x03) >> 0; + queue_id = (entry->ring_id & 0x70) >> 4; + + switch (me_id) { + case 0: + drm_sched_fault(&adev->gfx.gfx_ring[0].sched); + break; + case 1: + case 2: + for (i = 0; i < adev->gfx.num_compute_rings; i++) { + ring = &adev->gfx.compute_ring[i]; + if (ring->me == me_id && ring->pipe == pipe_id && + ring->queue == queue_id) + drm_sched_fault(&ring->sched); + } + break; + } +} + static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { DRM_ERROR("Illegal register access in command stream\n"); - schedule_work(&adev->reset_work); + gfx_v9_0_fault(adev, entry); return 0; } @@ -4710,7 +4653,7 @@ static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry) { DRM_ERROR("Illegal instruction in command stream\n"); - schedule_work(&adev->reset_work); + gfx_v9_0_fault(adev, entry); return 0; } @@ -4801,7 +4744,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = { SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 2 + /* gfx_v9_0_ring_emit_vm_flush */ 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */ - .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_compute */ + .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */ .emit_ib = gfx_v9_0_ring_emit_ib_compute, .emit_fence = gfx_v9_0_ring_emit_fence, .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync, @@ -4836,11 +4779,9 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = { SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 2 + /* gfx_v9_0_ring_emit_vm_flush */ 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */ - .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_compute */ - .emit_ib = gfx_v9_0_ring_emit_ib_compute, + .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */ .emit_fence = gfx_v9_0_ring_emit_fence_kiq, .test_ring = gfx_v9_0_ring_test_ring, - .test_ib = gfx_v9_0_ring_test_ib, .insert_nop = amdgpu_ring_insert_nop, .pad_ib = amdgpu_ring_generic_pad_ib, .emit_rreg = gfx_v9_0_ring_emit_rreg, @@ -4920,6 +4861,26 @@ static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev) break; } + switch (adev->asic_type) { + case CHIP_VEGA10: + case CHIP_VEGA20: + adev->gds.gds_compute_max_wave_id = 0x7ff; + break; + case CHIP_VEGA12: + adev->gds.gds_compute_max_wave_id = 0x27f; + break; + case CHIP_RAVEN: + if (adev->rev_id >= 0x8) + adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */ + else + adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */ + break; + default: + /* this really depends on the chip */ + adev->gds.gds_compute_max_wave_id = 0x7ff; + break; + } + adev->gds.gws.total_size = 64; adev->gds.oa.total_size = 16; diff --git a/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c b/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c index bfa317ad20a9..f5edddf3b29d 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c @@ -35,20 +35,25 @@ u64 gfxhub_v1_0_get_mc_fb_offset(struct amdgpu_device *adev) return (u64)RREG32_SOC15(GC, 0, mmMC_VM_FB_OFFSET) << 24; } -static void gfxhub_v1_0_init_gart_pt_regs(struct amdgpu_device *adev) +void gfxhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid, + uint64_t page_table_base) { - uint64_t value = amdgpu_gmc_pd_addr(adev->gart.bo); + /* two registers distance between mmVM_CONTEXT0_* to mmVM_CONTEXT1_* */ + int offset = mmVM_CONTEXT1_PAGE_TABLE_BASE_ADDR_LO32 + - mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32; - WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32, - lower_32_bits(value)); + WREG32_SOC15_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32, + offset * vmid, lower_32_bits(page_table_base)); - WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32, - upper_32_bits(value)); + WREG32_SOC15_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32, + offset * vmid, upper_32_bits(page_table_base)); } static void gfxhub_v1_0_init_gart_aperture_regs(struct amdgpu_device *adev) { - gfxhub_v1_0_init_gart_pt_regs(adev); + uint64_t pt_base = amdgpu_gmc_pd_addr(adev->gart.bo); + + gfxhub_v1_0_setup_vm_pt_regs(adev, 0, pt_base); WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32, (u32)(adev->gmc.gart_start >> 12)); diff --git a/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.h b/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.h index 206e29cad753..92d3a70cd9b1 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.h +++ b/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.h @@ -30,5 +30,7 @@ void gfxhub_v1_0_set_fault_enable_default(struct amdgpu_device *adev, bool value); void gfxhub_v1_0_init(struct amdgpu_device *adev); u64 gfxhub_v1_0_get_mc_fb_offset(struct amdgpu_device *adev); +void gfxhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid, + uint64_t page_table_base); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c index 73ad02aea2b2..9fc3296592fe 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c @@ -359,7 +359,8 @@ static int gmc_v6_0_mc_init(struct amdgpu_device *adev) return 0; } -static void gmc_v6_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid) +static void gmc_v6_0_flush_gpu_tlb(struct amdgpu_device *adev, + uint32_t vmid, uint32_t flush_type) { WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid); } @@ -581,7 +582,7 @@ static int gmc_v6_0_gart_enable(struct amdgpu_device *adev) else gmc_v6_0_set_fault_enable_default(adev, true); - gmc_v6_0_flush_gpu_tlb(adev, 0); + gmc_v6_0_flush_gpu_tlb(adev, 0, 0); dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n", (unsigned)(adev->gmc.gart_size >> 20), (unsigned long long)table_addr); diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index 910c4ce19cb3..761dcfb2fec0 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -430,7 +430,8 @@ static int gmc_v7_0_mc_init(struct amdgpu_device *adev) * * Flush the TLB for the requested page table (CIK). */ -static void gmc_v7_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid) +static void gmc_v7_0_flush_gpu_tlb(struct amdgpu_device *adev, + uint32_t vmid, uint32_t flush_type) { /* bits 0-15 are the VM contexts0-15 */ WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid); @@ -698,7 +699,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) WREG32(mmCHUB_CONTROL, tmp); } - gmc_v7_0_flush_gpu_tlb(adev, 0); + gmc_v7_0_flush_gpu_tlb(adev, 0, 0); DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", (unsigned)(adev->gmc.gart_size >> 20), (unsigned long long)table_addr); diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c index 747c068379dc..34440672f938 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c @@ -633,7 +633,7 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev) * Flush the TLB for the requested page table (CIK). */ static void gmc_v8_0_flush_gpu_tlb(struct amdgpu_device *adev, - uint32_t vmid) + uint32_t vmid, uint32_t flush_type) { /* bits 0-15 are the VM contexts0-15 */ WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid); @@ -942,7 +942,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) else gmc_v8_0_set_fault_enable_default(adev, true); - gmc_v8_0_flush_gpu_tlb(adev, 0); + gmc_v8_0_flush_gpu_tlb(adev, 0, 0); DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", (unsigned)(adev->gmc.gart_size >> 20), (unsigned long long)table_addr); @@ -1471,8 +1471,9 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev, gmc_v8_0_set_fault_enable_default(adev, false); if (printk_ratelimit()) { - struct amdgpu_task_info task_info = { 0 }; + struct amdgpu_task_info task_info; + memset(&task_info, 0, sizeof(struct amdgpu_task_info)); amdgpu_vm_get_task_info(adev, entry->pasid, &task_info); dev_err(adev->dev, "GPU fault detected: %d 0x%08x for process %s pid %d thread %s pid %d\n", diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c index f35d7a554ad5..2fe8397241ea 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c @@ -244,30 +244,93 @@ static int gmc_v9_0_vm_fault_interrupt_state(struct amdgpu_device *adev, return 0; } +/** + * vega10_ih_prescreen_iv - prescreen an interrupt vector + * + * @adev: amdgpu_device pointer + * + * Returns true if the interrupt vector should be further processed. + */ +static bool gmc_v9_0_prescreen_iv(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry, + uint64_t addr) +{ + struct amdgpu_vm *vm; + u64 key; + int r; + + /* No PASID, can't identify faulting process */ + if (!entry->pasid) + return true; + + /* Not a retry fault */ + if (!(entry->src_data[1] & 0x80)) + return true; + + /* Track retry faults in per-VM fault FIFO. */ + spin_lock(&adev->vm_manager.pasid_lock); + vm = idr_find(&adev->vm_manager.pasid_idr, entry->pasid); + if (!vm) { + /* VM not found, process it normally */ + spin_unlock(&adev->vm_manager.pasid_lock); + return true; + } + + key = AMDGPU_VM_FAULT(entry->pasid, addr); + r = amdgpu_vm_add_fault(vm->fault_hash, key); + + /* Hash table is full or the fault is already being processed, + * ignore further page faults + */ + if (r != 0) { + spin_unlock(&adev->vm_manager.pasid_lock); + return false; + } + /* No locking required with single writer and single reader */ + r = kfifo_put(&vm->faults, key); + if (!r) { + /* FIFO is full. Ignore it until there is space */ + amdgpu_vm_clear_fault(vm->fault_hash, key); + spin_unlock(&adev->vm_manager.pasid_lock); + return false; + } + + spin_unlock(&adev->vm_manager.pasid_lock); + /* It's the first fault for this address, process it normally */ + return true; +} + static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { struct amdgpu_vmhub *hub = &adev->vmhub[entry->vmid_src]; + bool retry_fault = !!(entry->src_data[1] & 0x80); uint32_t status = 0; u64 addr; addr = (u64)entry->src_data[0] << 12; addr |= ((u64)entry->src_data[1] & 0xf) << 44; + if (!gmc_v9_0_prescreen_iv(adev, entry, addr)) + return 1; /* This also prevents sending it to KFD */ + if (!amdgpu_sriov_vf(adev)) { status = RREG32(hub->vm_l2_pro_fault_status); WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1); } if (printk_ratelimit()) { - struct amdgpu_task_info task_info = { 0 }; + struct amdgpu_task_info task_info; + memset(&task_info, 0, sizeof(struct amdgpu_task_info)); amdgpu_vm_get_task_info(adev, entry->pasid, &task_info); dev_err(adev->dev, - "[%s] VMC page fault (src_id:%u ring:%u vmid:%u pasid:%u, for process %s pid %d thread %s pid %d)\n", + "[%s] %s page fault (src_id:%u ring:%u vmid:%u " + "pasid:%u, for process %s pid %d thread %s pid %d)\n", entry->vmid_src ? "mmhub" : "gfxhub", + retry_fault ? "retry" : "no-retry", entry->src_id, entry->ring_id, entry->vmid, entry->pasid, task_info.process_name, task_info.tgid, task_info.task_name, task_info.pid); @@ -293,14 +356,14 @@ static void gmc_v9_0_set_irq_funcs(struct amdgpu_device *adev) adev->gmc.vm_fault.funcs = &gmc_v9_0_irq_funcs; } -static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid) +static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid, + uint32_t flush_type) { u32 req = 0; - /* invalidate using legacy mode on vmid*/ req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, PER_VMID_INVALIDATE_REQ, 1 << vmid); - req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 0); + req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, flush_type); req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PTES, 1); req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE0, 1); req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE1, 1); @@ -312,48 +375,6 @@ static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid) return req; } -static signed long amdgpu_kiq_reg_write_reg_wait(struct amdgpu_device *adev, - uint32_t reg0, uint32_t reg1, - uint32_t ref, uint32_t mask) -{ - signed long r, cnt = 0; - unsigned long flags; - uint32_t seq; - struct amdgpu_kiq *kiq = &adev->gfx.kiq; - struct amdgpu_ring *ring = &kiq->ring; - - spin_lock_irqsave(&kiq->ring_lock, flags); - - amdgpu_ring_alloc(ring, 32); - amdgpu_ring_emit_reg_write_reg_wait(ring, reg0, reg1, - ref, mask); - amdgpu_fence_emit_polling(ring, &seq); - amdgpu_ring_commit(ring); - spin_unlock_irqrestore(&kiq->ring_lock, flags); - - r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); - - /* don't wait anymore for IRQ context */ - if (r < 1 && in_interrupt()) - goto failed_kiq; - - might_sleep(); - - while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) { - msleep(MAX_KIQ_REG_BAILOUT_INTERVAL); - r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); - } - - if (cnt > MAX_KIQ_REG_TRY) - goto failed_kiq; - - return 0; - -failed_kiq: - pr_err("failed to invalidate tlb with kiq\n"); - return r; -} - /* * GART * VMID 0 is the physical GPU addresses as used by the kernel. @@ -362,64 +383,50 @@ failed_kiq: */ /** - * gmc_v9_0_flush_gpu_tlb - gart tlb flush callback + * gmc_v9_0_flush_gpu_tlb - tlb flush with certain type * * @adev: amdgpu_device pointer * @vmid: vm instance to flush + * @flush_type: the flush type * - * Flush the TLB for the requested page table. + * Flush the TLB for the requested page table using certain type. */ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, - uint32_t vmid) + uint32_t vmid, uint32_t flush_type) { - /* Use register 17 for GART */ const unsigned eng = 17; unsigned i, j; - int r; for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) { struct amdgpu_vmhub *hub = &adev->vmhub[i]; - u32 tmp = gmc_v9_0_get_invalidate_req(vmid); + u32 tmp = gmc_v9_0_get_invalidate_req(vmid, flush_type); - if (adev->gfx.kiq.ring.ready && + /* This is necessary for a HW workaround under SRIOV as well + * as GFXOFF under bare metal + */ + if (adev->gfx.kiq.ring.sched.ready && (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev)) && !adev->in_gpu_reset) { - r = amdgpu_kiq_reg_write_reg_wait(adev, hub->vm_inv_eng0_req + eng, - hub->vm_inv_eng0_ack + eng, tmp, 1 << vmid); - if (!r) - continue; - } + uint32_t req = hub->vm_inv_eng0_req + eng; + uint32_t ack = hub->vm_inv_eng0_ack + eng; - spin_lock(&adev->gmc.invalidate_lock); - - WREG32_NO_KIQ(hub->vm_inv_eng0_req + eng, tmp); - - /* Busy wait for ACK.*/ - for (j = 0; j < 100; j++) { - tmp = RREG32_NO_KIQ(hub->vm_inv_eng0_ack + eng); - tmp &= 1 << vmid; - if (tmp) - break; - cpu_relax(); - } - if (j < 100) { - spin_unlock(&adev->gmc.invalidate_lock); + amdgpu_virt_kiq_reg_write_reg_wait(adev, req, ack, tmp, + 1 << vmid); continue; } - /* Wait for ACK with a delay.*/ + spin_lock(&adev->gmc.invalidate_lock); + WREG32_NO_KIQ(hub->vm_inv_eng0_req + eng, tmp); for (j = 0; j < adev->usec_timeout; j++) { tmp = RREG32_NO_KIQ(hub->vm_inv_eng0_ack + eng); - tmp &= 1 << vmid; - if (tmp) + if (tmp & (1 << vmid)) break; udelay(1); } - if (j < adev->usec_timeout) { - spin_unlock(&adev->gmc.invalidate_lock); - continue; - } spin_unlock(&adev->gmc.invalidate_lock); + if (j < adev->usec_timeout) + continue; + DRM_ERROR("Timeout waiting for VM flush ACK!\n"); } } @@ -429,7 +436,7 @@ static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring, { struct amdgpu_device *adev = ring->adev; struct amdgpu_vmhub *hub = &adev->vmhub[ring->funcs->vmhub]; - uint32_t req = gmc_v9_0_get_invalidate_req(vmid); + uint32_t req = gmc_v9_0_get_invalidate_req(vmid, 0); unsigned eng = ring->vm_inv_eng; amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_lo32 + (2 * vmid), @@ -715,38 +722,46 @@ static bool gmc_v9_0_keep_stolen_memory(struct amdgpu_device *adev) } } +static int gmc_v9_0_allocate_vm_inv_eng(struct amdgpu_device *adev) +{ + struct amdgpu_ring *ring; + unsigned vm_inv_engs[AMDGPU_MAX_VMHUBS] = + {GFXHUB_FREE_VM_INV_ENGS_BITMAP, MMHUB_FREE_VM_INV_ENGS_BITMAP}; + unsigned i; + unsigned vmhub, inv_eng; + + for (i = 0; i < adev->num_rings; ++i) { + ring = adev->rings[i]; + vmhub = ring->funcs->vmhub; + + inv_eng = ffs(vm_inv_engs[vmhub]); + if (!inv_eng) { + dev_err(adev->dev, "no VM inv eng for ring %s\n", + ring->name); + return -EINVAL; + } + + ring->vm_inv_eng = inv_eng - 1; + vm_inv_engs[vmhub] &= ~(1 << ring->vm_inv_eng); + + dev_info(adev->dev, "ring %s uses VM inv eng %u on hub %u\n", + ring->name, ring->vm_inv_eng, ring->funcs->vmhub); + } + + return 0; +} + static int gmc_v9_0_late_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; - /* - * The latest engine allocation on gfx9 is: - * Engine 0, 1: idle - * Engine 2, 3: firmware - * Engine 4~13: amdgpu ring, subject to change when ring number changes - * Engine 14~15: idle - * Engine 16: kfd tlb invalidation - * Engine 17: Gart flushes - */ - unsigned vm_inv_eng[AMDGPU_MAX_VMHUBS] = { 4, 4 }; - unsigned i; int r; if (!gmc_v9_0_keep_stolen_memory(adev)) amdgpu_bo_late_init(adev); - for(i = 0; i < adev->num_rings; ++i) { - struct amdgpu_ring *ring = adev->rings[i]; - unsigned vmhub = ring->funcs->vmhub; - - ring->vm_inv_eng = vm_inv_eng[vmhub]++; - dev_info(adev->dev, "ring %u(%s) uses VM inv eng %u on hub %u\n", - ring->idx, ring->name, ring->vm_inv_eng, - ring->funcs->vmhub); - } - - /* Engine 16 is used for KFD and 17 for GART flushes */ - for(i = 0; i < AMDGPU_MAX_VMHUBS; ++i) - BUG_ON(vm_inv_eng[i] > 16); + r = gmc_v9_0_allocate_vm_inv_eng(adev); + if (r) + return r; if (adev->asic_type == CHIP_VEGA10 && !amdgpu_sriov_vf(adev)) { r = gmc_v9_0_ecc_available(adev); @@ -950,7 +965,11 @@ static int gmc_v9_0_sw_init(void *handle) * vm size is 256TB (48bit), maximum size of Vega10, * block size 512 (9bit) */ - amdgpu_vm_adjust_size(adev, 256 * 1024, 9, 3, 48); + /* sriov restrict max_pfn below AMDGPU_GMC_HOLE */ + if (amdgpu_sriov_vf(adev)) + amdgpu_vm_adjust_size(adev, 256 * 1024, 9, 3, 47); + else + amdgpu_vm_adjust_size(adev, 256 * 1024, 9, 3, 48); break; default: break; @@ -959,6 +978,9 @@ static int gmc_v9_0_sw_init(void *handle) /* This interrupt is VMC page fault.*/ r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VMC, VMC_1_0__SRCID__VM_FAULT, &adev->gmc.vm_fault); + if (r) + return r; + r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_UTCL2, UTCL2_1_0__SRCID__FAULT, &adev->gmc.vm_fault); @@ -991,7 +1013,7 @@ static int gmc_v9_0_sw_init(void *handle) } adev->need_swiotlb = drm_get_max_iomem() > ((u64)1 << dma_bits); - if (adev->asic_type == CHIP_VEGA20) { + if (adev->gmc.xgmi.supported) { r = gfxhub_v1_1_get_xgmi_info(adev); if (r) return r; @@ -1122,7 +1144,7 @@ static int gmc_v9_0_gart_enable(struct amdgpu_device *adev) gfxhub_v1_0_set_fault_enable_default(adev, value); mmhub_v1_0_set_fault_enable_default(adev, value); - gmc_v9_0_flush_gpu_tlb(adev, 0); + gmc_v9_0_flush_gpu_tlb(adev, 0, 0); DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", (unsigned)(adev->gmc.gart_size >> 20), diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h index b030ca5ea107..5c8deac65580 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h @@ -24,6 +24,16 @@ #ifndef __GMC_V9_0_H__ #define __GMC_V9_0_H__ + /* + * The latest engine allocation on gfx9 is: + * Engine 2, 3: firmware + * Engine 0, 1, 4~16: amdgpu ring, + * subject to change when ring number changes + * Engine 17: Gart flushes + */ +#define GFXHUB_FREE_VM_INV_ENGS_BITMAP 0x1FFF3 +#define MMHUB_FREE_VM_INV_ENGS_BITMAP 0x1FFF3 + extern const struct amd_ip_funcs gmc_v9_0_ip_funcs; extern const struct amdgpu_ip_block_version gmc_v9_0_ip_block; diff --git a/drivers/gpu/drm/amd/amdgpu/iceland_ih.c b/drivers/gpu/drm/amd/amdgpu/iceland_ih.c index cf0fc61aebe6..b1626e1d2f5d 100644 --- a/drivers/gpu/drm/amd/amdgpu/iceland_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/iceland_ih.c @@ -103,9 +103,9 @@ static void iceland_ih_disable_interrupts(struct amdgpu_device *adev) */ static int iceland_ih_irq_init(struct amdgpu_device *adev) { + struct amdgpu_ih_ring *ih = &adev->irq.ih; int rb_bufsz; u32 interrupt_cntl, ih_cntl, ih_rb_cntl; - u64 wptr_off; /* disable irqs */ iceland_ih_disable_interrupts(adev); @@ -133,9 +133,8 @@ static int iceland_ih_irq_init(struct amdgpu_device *adev) ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1); /* set the writeback address whether it's enabled or not */ - wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4); - WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off)); - WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF); + WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr)); + WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF); WREG32(mmIH_RB_CNTL, ih_rb_cntl); @@ -185,11 +184,12 @@ static void iceland_ih_irq_disable(struct amdgpu_device *adev) * Used by cz_irq_process(VI). * Returns the value of the wptr. */ -static u32 iceland_ih_get_wptr(struct amdgpu_device *adev) +static u32 iceland_ih_get_wptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { u32 wptr, tmp; - wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]); + wptr = le32_to_cpu(*ih->wptr_cpu); if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) { wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0); @@ -198,41 +198,13 @@ static u32 iceland_ih_get_wptr(struct amdgpu_device *adev) * this should allow us to catchup. */ dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", - wptr, adev->irq.ih.rptr, (wptr + 16) & adev->irq.ih.ptr_mask); - adev->irq.ih.rptr = (wptr + 16) & adev->irq.ih.ptr_mask; + wptr, ih->rptr, (wptr + 16) & ih->ptr_mask); + ih->rptr = (wptr + 16) & ih->ptr_mask; tmp = RREG32(mmIH_RB_CNTL); tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); WREG32(mmIH_RB_CNTL, tmp); } - return (wptr & adev->irq.ih.ptr_mask); -} - -/** - * iceland_ih_prescreen_iv - prescreen an interrupt vector - * - * @adev: amdgpu_device pointer - * - * Returns true if the interrupt vector should be further processed. - */ -static bool iceland_ih_prescreen_iv(struct amdgpu_device *adev) -{ - u32 ring_index = adev->irq.ih.rptr >> 2; - u16 pasid; - - switch (le32_to_cpu(adev->irq.ih.ring[ring_index]) & 0xff) { - case 146: - case 147: - pasid = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]) >> 16; - if (!pasid || amdgpu_vm_pasid_fault_credit(adev, pasid)) - return true; - break; - default: - /* Not a VM fault */ - return true; - } - - adev->irq.ih.rptr += 16; - return false; + return (wptr & ih->ptr_mask); } /** @@ -244,16 +216,17 @@ static bool iceland_ih_prescreen_iv(struct amdgpu_device *adev) * position and also advance the position. */ static void iceland_ih_decode_iv(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih, struct amdgpu_iv_entry *entry) { /* wptr/rptr are in bytes! */ - u32 ring_index = adev->irq.ih.rptr >> 2; + u32 ring_index = ih->rptr >> 2; uint32_t dw[4]; - dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); - dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]); - dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]); - dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); + dw[0] = le32_to_cpu(ih->ring[ring_index + 0]); + dw[1] = le32_to_cpu(ih->ring[ring_index + 1]); + dw[2] = le32_to_cpu(ih->ring[ring_index + 2]); + dw[3] = le32_to_cpu(ih->ring[ring_index + 3]); entry->client_id = AMDGPU_IRQ_CLIENTID_LEGACY; entry->src_id = dw[0] & 0xff; @@ -263,7 +236,7 @@ static void iceland_ih_decode_iv(struct amdgpu_device *adev, entry->pasid = (dw[2] >> 16) & 0xffff; /* wptr/rptr are in bytes! */ - adev->irq.ih.rptr += 16; + ih->rptr += 16; } /** @@ -273,9 +246,10 @@ static void iceland_ih_decode_iv(struct amdgpu_device *adev, * * Set the IH ring buffer rptr. */ -static void iceland_ih_set_rptr(struct amdgpu_device *adev) +static void iceland_ih_set_rptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { - WREG32(mmIH_RB_RPTR, adev->irq.ih.rptr); + WREG32(mmIH_RB_RPTR, ih->rptr); } static int iceland_ih_early_init(void *handle) @@ -440,7 +414,6 @@ static const struct amd_ip_funcs iceland_ih_ip_funcs = { static const struct amdgpu_ih_funcs iceland_ih_funcs = { .get_wptr = iceland_ih_get_wptr, - .prescreen_iv = iceland_ih_prescreen_iv, .decode_iv = iceland_ih_decode_iv, .set_rptr = iceland_ih_set_rptr }; diff --git a/drivers/gpu/drm/amd/amdgpu/kv_dpm.c b/drivers/gpu/drm/amd/amdgpu/kv_dpm.c index d0e478f43443..0c9a2c03504e 100644 --- a/drivers/gpu/drm/amd/amdgpu/kv_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/kv_dpm.c @@ -508,19 +508,19 @@ static int kv_enable_didt(struct amdgpu_device *adev, bool enable) pi->caps_db_ramping || pi->caps_td_ramping || pi->caps_tcp_ramping) { - adev->gfx.rlc.funcs->enter_safe_mode(adev); + amdgpu_gfx_rlc_enter_safe_mode(adev); if (enable) { ret = kv_program_pt_config_registers(adev, didt_config_kv); if (ret) { - adev->gfx.rlc.funcs->exit_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); return ret; } } kv_do_enable_didt(adev, enable); - adev->gfx.rlc.funcs->exit_safe_mode(adev); + amdgpu_gfx_rlc_exit_safe_mode(adev); } return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c index a0db67adc34c..1696644ec022 100644 --- a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c @@ -52,20 +52,25 @@ u64 mmhub_v1_0_get_fb_location(struct amdgpu_device *adev) return base; } -static void mmhub_v1_0_init_gart_pt_regs(struct amdgpu_device *adev) +void mmhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid, + uint64_t page_table_base) { - uint64_t value = amdgpu_gmc_pd_addr(adev->gart.bo); + /* two registers distance between mmVM_CONTEXT0_* to mmVM_CONTEXT1_* */ + int offset = mmVM_CONTEXT1_PAGE_TABLE_BASE_ADDR_LO32 + - mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32; - WREG32_SOC15(MMHUB, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32, - lower_32_bits(value)); + WREG32_SOC15_OFFSET(MMHUB, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32, + offset * vmid, lower_32_bits(page_table_base)); - WREG32_SOC15(MMHUB, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32, - upper_32_bits(value)); + WREG32_SOC15_OFFSET(MMHUB, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32, + offset * vmid, upper_32_bits(page_table_base)); } static void mmhub_v1_0_init_gart_aperture_regs(struct amdgpu_device *adev) { - mmhub_v1_0_init_gart_pt_regs(adev); + uint64_t pt_base = amdgpu_gmc_pd_addr(adev->gart.bo); + + mmhub_v1_0_setup_vm_pt_regs(adev, 0, pt_base); WREG32_SOC15(MMHUB, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32, (u32)(adev->gmc.gart_start >> 12)); @@ -177,6 +182,7 @@ static void mmhub_v1_0_init_cache_regs(struct amdgpu_device *adev) tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, 6); } + WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL3, tmp); tmp = mmVM_L2_CNTL4_DEFAULT; tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 0); diff --git a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.h b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.h index bef3d0c0c117..0de0fdf98c00 100644 --- a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.h +++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.h @@ -34,5 +34,7 @@ int mmhub_v1_0_set_clockgating(struct amdgpu_device *adev, void mmhub_v1_0_get_clockgating(struct amdgpu_device *adev, u32 *flags); void mmhub_v1_0_update_power_gating(struct amdgpu_device *adev, bool enable); +void mmhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid, + uint64_t page_table_base); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c index 8cbb4655896a..73851ebb3833 100644 --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c @@ -174,7 +174,7 @@ static int xgpu_ai_send_access_requests(struct amdgpu_device *adev, return r; } /* Retrieve checksum from mailbox2 */ - if (req == IDH_REQ_GPU_INIT_ACCESS) { + if (req == IDH_REQ_GPU_INIT_ACCESS || req == IDH_REQ_GPU_RESET_ACCESS) { adev->virt.fw_reserve.checksum_key = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_MSGBUF_RCV_DW2)); @@ -266,7 +266,8 @@ flr_done: } /* Trigger recovery for world switch failure if no TDR */ - if (amdgpu_device_should_recover_gpu(adev)) + if (amdgpu_device_should_recover_gpu(adev) + && amdgpu_lockup_timeout == MAX_SCHEDULE_TIMEOUT) amdgpu_device_gpu_recover(adev, NULL); } diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c index 64e875d528dd..6a0fcd67662a 100644 --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c @@ -37,7 +37,6 @@ #include "gmc/gmc_8_2_sh_mask.h" #include "oss/oss_3_0_d.h" #include "oss/oss_3_0_sh_mask.h" -#include "gca/gfx_8_0_sh_mask.h" #include "dce/dce_10_0_d.h" #include "dce/dce_10_0_sh_mask.h" #include "smu/smu_7_1_3_d.h" diff --git a/drivers/gpu/drm/amd/amdgpu/nbio_v6_1.c b/drivers/gpu/drm/amd/amdgpu/nbio_v6_1.c index 6f9c54978cc1..cc967dbfd631 100644 --- a/drivers/gpu/drm/amd/amdgpu/nbio_v6_1.c +++ b/drivers/gpu/drm/amd/amdgpu/nbio_v6_1.c @@ -27,12 +27,9 @@ #include "nbio/nbio_6_1_default.h" #include "nbio/nbio_6_1_offset.h" #include "nbio/nbio_6_1_sh_mask.h" +#include "nbio/nbio_6_1_smn.h" #include "vega10_enum.h" -#define smnCPM_CONTROL 0x11180460 -#define smnPCIE_CNTL2 0x11180070 -#define smnPCIE_CONFIG_CNTL 0x11180044 - static u32 nbio_v6_1_get_rev_id(struct amdgpu_device *adev) { u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0); @@ -71,7 +68,7 @@ static u32 nbio_v6_1_get_memsize(struct amdgpu_device *adev) } static void nbio_v6_1_sdma_doorbell_range(struct amdgpu_device *adev, int instance, - bool use_doorbell, int doorbell_index) + bool use_doorbell, int doorbell_index, int doorbell_size) { u32 reg = instance == 0 ? SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA0_DOORBELL_RANGE) : SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA1_DOORBELL_RANGE); @@ -80,7 +77,7 @@ static void nbio_v6_1_sdma_doorbell_range(struct amdgpu_device *adev, int instan if (use_doorbell) { doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, OFFSET, doorbell_index); - doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, 2); + doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, doorbell_size); } else doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, 0); @@ -270,6 +267,12 @@ static void nbio_v6_1_init_registers(struct amdgpu_device *adev) if (def != data) WREG32_PCIE(smnPCIE_CONFIG_CNTL, data); + + def = data = RREG32_PCIE(smnPCIE_CI_CNTL); + data = REG_SET_FIELD(data, PCIE_CI_CNTL, CI_SLV_ORDERING_DIS, 1); + + if (def != data) + WREG32_PCIE(smnPCIE_CI_CNTL, data); } const struct amdgpu_nbio_funcs nbio_v6_1_funcs = { diff --git a/drivers/gpu/drm/amd/amdgpu/nbio_v7_0.c b/drivers/gpu/drm/amd/amdgpu/nbio_v7_0.c index df34dc79d444..1cdb98ad2db3 100644 --- a/drivers/gpu/drm/amd/amdgpu/nbio_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/nbio_v7_0.c @@ -27,13 +27,11 @@ #include "nbio/nbio_7_0_default.h" #include "nbio/nbio_7_0_offset.h" #include "nbio/nbio_7_0_sh_mask.h" +#include "nbio/nbio_7_0_smn.h" #include "vega10_enum.h" #define smnNBIF_MGCG_CTRL_LCLK 0x1013a05c -#define smnCPM_CONTROL 0x11180460 -#define smnPCIE_CNTL2 0x11180070 - static u32 nbio_v7_0_get_rev_id(struct amdgpu_device *adev) { u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0); @@ -69,7 +67,7 @@ static u32 nbio_v7_0_get_memsize(struct amdgpu_device *adev) } static void nbio_v7_0_sdma_doorbell_range(struct amdgpu_device *adev, int instance, - bool use_doorbell, int doorbell_index) + bool use_doorbell, int doorbell_index, int doorbell_size) { u32 reg = instance == 0 ? SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA0_DOORBELL_RANGE) : SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA1_DOORBELL_RANGE); @@ -78,7 +76,7 @@ static void nbio_v7_0_sdma_doorbell_range(struct amdgpu_device *adev, int instan if (use_doorbell) { doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, OFFSET, doorbell_index); - doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, 2); + doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, doorbell_size); } else doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, 0); diff --git a/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c b/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c index f8cee95d61cc..c69d51598cfe 100644 --- a/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c +++ b/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c @@ -26,15 +26,13 @@ #include "nbio/nbio_7_4_offset.h" #include "nbio/nbio_7_4_sh_mask.h" +#include "nbio/nbio_7_4_0_smn.h" #define smnNBIF_MGCG_CTRL_LCLK 0x1013a21c -#define smnCPM_CONTROL 0x11180460 -#define smnPCIE_CNTL2 0x11180070 - static u32 nbio_v7_4_get_rev_id(struct amdgpu_device *adev) { - u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0); + u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0); tmp &= RCC_DEV0_EPF0_STRAP0__STRAP_ATI_REV_ID_DEV0_F0_MASK; tmp >>= RCC_DEV0_EPF0_STRAP0__STRAP_ATI_REV_ID_DEV0_F0__SHIFT; @@ -67,7 +65,7 @@ static u32 nbio_v7_4_get_memsize(struct amdgpu_device *adev) } static void nbio_v7_4_sdma_doorbell_range(struct amdgpu_device *adev, int instance, - bool use_doorbell, int doorbell_index) + bool use_doorbell, int doorbell_index, int doorbell_size) { u32 reg = instance == 0 ? SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA0_DOORBELL_RANGE) : SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA1_DOORBELL_RANGE); @@ -76,7 +74,7 @@ static void nbio_v7_4_sdma_doorbell_range(struct amdgpu_device *adev, int instan if (use_doorbell) { doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, OFFSET, doorbell_index); - doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, 2); + doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, doorbell_size); } else doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, 0); @@ -92,7 +90,20 @@ static void nbio_v7_4_enable_doorbell_aperture(struct amdgpu_device *adev, static void nbio_v7_4_enable_doorbell_selfring_aperture(struct amdgpu_device *adev, bool enable) { + u32 tmp = 0; + + if (enable) { + tmp = REG_SET_FIELD(tmp, DOORBELL_SELFRING_GPA_APER_CNTL, DOORBELL_SELFRING_GPA_APER_EN, 1) | + REG_SET_FIELD(tmp, DOORBELL_SELFRING_GPA_APER_CNTL, DOORBELL_SELFRING_GPA_APER_MODE, 1) | + REG_SET_FIELD(tmp, DOORBELL_SELFRING_GPA_APER_CNTL, DOORBELL_SELFRING_GPA_APER_SIZE, 0); + + WREG32_SOC15(NBIO, 0, mmDOORBELL_SELFRING_GPA_APER_BASE_LOW, + lower_32_bits(adev->doorbell.base)); + WREG32_SOC15(NBIO, 0, mmDOORBELL_SELFRING_GPA_APER_BASE_HIGH, + upper_32_bits(adev->doorbell.base)); + } + WREG32_SOC15(NBIO, 0, mmDOORBELL_SELFRING_GPA_APER_CNTL, tmp); } static void nbio_v7_4_ih_doorbell_range(struct amdgpu_device *adev, @@ -222,7 +233,13 @@ static void nbio_v7_4_detect_hw_virt(struct amdgpu_device *adev) static void nbio_v7_4_init_registers(struct amdgpu_device *adev) { + uint32_t def, data; + + def = data = RREG32_PCIE(smnPCIE_CI_CNTL); + data = REG_SET_FIELD(data, PCIE_CI_CNTL, CI_SLV_ORDERING_DIS, 1); + if (def != data) + WREG32_PCIE(smnPCIE_CI_CNTL, data); } const struct amdgpu_nbio_funcs nbio_v7_4_funcs = { diff --git a/drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h b/drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h index 882bd83a28c4..f3a7d207af07 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h +++ b/drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h @@ -43,6 +43,8 @@ enum psp_gfx_crtl_cmd_id GFX_CTRL_CMD_ID_ENABLE_INT = 0x00050000, /* enable PSP-to-Gfx interrupt */ GFX_CTRL_CMD_ID_DISABLE_INT = 0x00060000, /* disable PSP-to-Gfx interrupt */ GFX_CTRL_CMD_ID_MODE1_RST = 0x00070000, /* trigger the Mode 1 reset */ + GFX_CTRL_CMD_ID_CONSUME_CMD = 0x000A0000, /* send interrupt to psp for updating write pointer of vf */ + GFX_CTRL_CMD_ID_DESTROY_GPCOM_RING = 0x000C0000, /* destroy GPCOM ring */ GFX_CTRL_CMD_ID_MAX = 0x000F0000, /* max command ID */ }; @@ -89,7 +91,8 @@ enum psp_gfx_cmd_id GFX_CMD_ID_LOAD_IP_FW = 0x00000006, /* load HW IP FW */ GFX_CMD_ID_DESTROY_TMR = 0x00000007, /* destroy TMR region */ GFX_CMD_ID_SAVE_RESTORE = 0x00000008, /* save/restore HW IP FW */ - + GFX_CMD_ID_SETUP_VMR = 0x00000009, /* setup VMR region */ + GFX_CMD_ID_DESTROY_VMR = 0x0000000A, /* destroy VMR region */ }; @@ -188,7 +191,7 @@ enum psp_gfx_fw_type GFX_FW_TYPE_MMSCH = 19, GFX_FW_TYPE_RLC_RESTORE_LIST_GPM_MEM = 20, GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_MEM = 21, - GFX_FW_TYPE_RLC_RESTORE_LIST_CNTL = 22, + GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_CNTL = 22, GFX_FW_TYPE_UVD1 = 23, GFX_FW_TYPE_MAX = 24 }; diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v10_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v10_0.c index 295c2205485a..77c2bc344dfc 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v10_0.c @@ -38,75 +38,6 @@ MODULE_FIRMWARE("amdgpu/raven_asd.bin"); MODULE_FIRMWARE("amdgpu/picasso_asd.bin"); MODULE_FIRMWARE("amdgpu/raven2_asd.bin"); -static int -psp_v10_0_get_fw_type(struct amdgpu_firmware_info *ucode, enum psp_gfx_fw_type *type) -{ - switch(ucode->ucode_id) { - case AMDGPU_UCODE_ID_SDMA0: - *type = GFX_FW_TYPE_SDMA0; - break; - case AMDGPU_UCODE_ID_SDMA1: - *type = GFX_FW_TYPE_SDMA1; - break; - case AMDGPU_UCODE_ID_CP_CE: - *type = GFX_FW_TYPE_CP_CE; - break; - case AMDGPU_UCODE_ID_CP_PFP: - *type = GFX_FW_TYPE_CP_PFP; - break; - case AMDGPU_UCODE_ID_CP_ME: - *type = GFX_FW_TYPE_CP_ME; - break; - case AMDGPU_UCODE_ID_CP_MEC1: - *type = GFX_FW_TYPE_CP_MEC; - break; - case AMDGPU_UCODE_ID_CP_MEC1_JT: - *type = GFX_FW_TYPE_CP_MEC_ME1; - break; - case AMDGPU_UCODE_ID_CP_MEC2: - *type = GFX_FW_TYPE_CP_MEC; - break; - case AMDGPU_UCODE_ID_CP_MEC2_JT: - *type = GFX_FW_TYPE_CP_MEC_ME2; - break; - case AMDGPU_UCODE_ID_RLC_G: - *type = GFX_FW_TYPE_RLC_G; - break; - case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL: - *type = GFX_FW_TYPE_RLC_RESTORE_LIST_CNTL; - break; - case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM: - *type = GFX_FW_TYPE_RLC_RESTORE_LIST_GPM_MEM; - break; - case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM: - *type = GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_MEM; - break; - case AMDGPU_UCODE_ID_SMC: - *type = GFX_FW_TYPE_SMU; - break; - case AMDGPU_UCODE_ID_UVD: - *type = GFX_FW_TYPE_UVD; - break; - case AMDGPU_UCODE_ID_VCE: - *type = GFX_FW_TYPE_VCE; - break; - case AMDGPU_UCODE_ID_VCN: - *type = GFX_FW_TYPE_VCN; - break; - case AMDGPU_UCODE_ID_DMCU_ERAM: - *type = GFX_FW_TYPE_DMCU_ERAM; - break; - case AMDGPU_UCODE_ID_DMCU_INTV: - *type = GFX_FW_TYPE_DMCU_ISR; - break; - case AMDGPU_UCODE_ID_MAXIMUM: - default: - return -EINVAL; - } - - return 0; -} - static int psp_v10_0_init_microcode(struct psp_context *psp) { struct amdgpu_device *adev = psp->adev; @@ -158,26 +89,6 @@ out: return err; } -static int psp_v10_0_prep_cmd_buf(struct amdgpu_firmware_info *ucode, - struct psp_gfx_cmd_resp *cmd) -{ - int ret; - uint64_t fw_mem_mc_addr = ucode->mc_addr; - - memset(cmd, 0, sizeof(struct psp_gfx_cmd_resp)); - - cmd->cmd_id = GFX_CMD_ID_LOAD_IP_FW; - cmd->cmd.cmd_load_ip_fw.fw_phy_addr_lo = lower_32_bits(fw_mem_mc_addr); - cmd->cmd.cmd_load_ip_fw.fw_phy_addr_hi = upper_32_bits(fw_mem_mc_addr); - cmd->cmd.cmd_load_ip_fw.fw_size = ucode->ucode_size; - - ret = psp_v10_0_get_fw_type(ucode, &cmd->cmd.cmd_load_ip_fw.fw_type); - if (ret) - DRM_ERROR("Unknown firmware type\n"); - - return ret; -} - static int psp_v10_0_ring_init(struct psp_context *psp, enum psp_ring_type ring_type) { @@ -240,12 +151,9 @@ static int psp_v10_0_ring_stop(struct psp_context *psp, enum psp_ring_type ring_type) { int ret = 0; - struct psp_ring *ring; unsigned int psp_ring_reg = 0; struct amdgpu_device *adev = psp->adev; - ring = &psp->km_ring; - /* Write the ring destroy command to C2PMSG_64 */ psp_ring_reg = 3 << 16; WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_64, psp_ring_reg); @@ -457,7 +365,6 @@ static int psp_v10_0_mode1_reset(struct psp_context *psp) static const struct psp_funcs psp_v10_0_funcs = { .init_microcode = psp_v10_0_init_microcode, - .prep_cmd_buf = psp_v10_0_prep_cmd_buf, .ring_init = psp_v10_0_ring_init, .ring_create = psp_v10_0_ring_create, .ring_stop = psp_v10_0_ring_stop, diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c index 3f3fac2d50cd..860b70d80d3c 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c @@ -34,71 +34,21 @@ #include "nbio/nbio_7_4_offset.h" MODULE_FIRMWARE("amdgpu/vega20_sos.bin"); +MODULE_FIRMWARE("amdgpu/vega20_asd.bin"); +MODULE_FIRMWARE("amdgpu/vega20_ta.bin"); /* address block */ #define smnMP1_FIRMWARE_FLAGS 0x3010024 -static int -psp_v11_0_get_fw_type(struct amdgpu_firmware_info *ucode, enum psp_gfx_fw_type *type) -{ - switch (ucode->ucode_id) { - case AMDGPU_UCODE_ID_SDMA0: - *type = GFX_FW_TYPE_SDMA0; - break; - case AMDGPU_UCODE_ID_SDMA1: - *type = GFX_FW_TYPE_SDMA1; - break; - case AMDGPU_UCODE_ID_CP_CE: - *type = GFX_FW_TYPE_CP_CE; - break; - case AMDGPU_UCODE_ID_CP_PFP: - *type = GFX_FW_TYPE_CP_PFP; - break; - case AMDGPU_UCODE_ID_CP_ME: - *type = GFX_FW_TYPE_CP_ME; - break; - case AMDGPU_UCODE_ID_CP_MEC1: - *type = GFX_FW_TYPE_CP_MEC; - break; - case AMDGPU_UCODE_ID_CP_MEC1_JT: - *type = GFX_FW_TYPE_CP_MEC_ME1; - break; - case AMDGPU_UCODE_ID_CP_MEC2: - *type = GFX_FW_TYPE_CP_MEC; - break; - case AMDGPU_UCODE_ID_CP_MEC2_JT: - *type = GFX_FW_TYPE_CP_MEC_ME2; - break; - case AMDGPU_UCODE_ID_RLC_G: - *type = GFX_FW_TYPE_RLC_G; - break; - case AMDGPU_UCODE_ID_SMC: - *type = GFX_FW_TYPE_SMU; - break; - case AMDGPU_UCODE_ID_UVD: - *type = GFX_FW_TYPE_UVD; - break; - case AMDGPU_UCODE_ID_VCE: - *type = GFX_FW_TYPE_VCE; - break; - case AMDGPU_UCODE_ID_UVD1: - *type = GFX_FW_TYPE_UVD1; - break; - case AMDGPU_UCODE_ID_MAXIMUM: - default: - return -EINVAL; - } - - return 0; -} - static int psp_v11_0_init_microcode(struct psp_context *psp) { struct amdgpu_device *adev = psp->adev; const char *chip_name; char fw_name[30]; int err = 0; - const struct psp_firmware_header_v1_0 *hdr; + const struct psp_firmware_header_v1_0 *sos_hdr; + const struct psp_firmware_header_v1_0 *asd_hdr; + const struct ta_firmware_header_v1_0 *ta_hdr; DRM_DEBUG("\n"); @@ -119,26 +69,66 @@ static int psp_v11_0_init_microcode(struct psp_context *psp) if (err) goto out; - hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.sos_fw->data; - adev->psp.sos_fw_version = le32_to_cpu(hdr->header.ucode_version); - adev->psp.sos_feature_version = le32_to_cpu(hdr->ucode_feature_version); - adev->psp.sos_bin_size = le32_to_cpu(hdr->sos_size_bytes); - adev->psp.sys_bin_size = le32_to_cpu(hdr->header.ucode_size_bytes) - - le32_to_cpu(hdr->sos_size_bytes); - adev->psp.sys_start_addr = (uint8_t *)hdr + - le32_to_cpu(hdr->header.ucode_array_offset_bytes); + sos_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.sos_fw->data; + adev->psp.sos_fw_version = le32_to_cpu(sos_hdr->header.ucode_version); + adev->psp.sos_feature_version = le32_to_cpu(sos_hdr->ucode_feature_version); + adev->psp.sos_bin_size = le32_to_cpu(sos_hdr->sos_size_bytes); + adev->psp.sys_bin_size = le32_to_cpu(sos_hdr->header.ucode_size_bytes) - + le32_to_cpu(sos_hdr->sos_size_bytes); + adev->psp.sys_start_addr = (uint8_t *)sos_hdr + + le32_to_cpu(sos_hdr->header.ucode_array_offset_bytes); adev->psp.sos_start_addr = (uint8_t *)adev->psp.sys_start_addr + - le32_to_cpu(hdr->sos_offset_bytes); - return 0; -out: + le32_to_cpu(sos_hdr->sos_offset_bytes); + + snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_asd.bin", chip_name); + err = request_firmware(&adev->psp.asd_fw, fw_name, adev->dev); + if (err) + goto out1; + + err = amdgpu_ucode_validate(adev->psp.asd_fw); + if (err) + goto out1; + + asd_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.asd_fw->data; + adev->psp.asd_fw_version = le32_to_cpu(asd_hdr->header.ucode_version); + adev->psp.asd_feature_version = le32_to_cpu(asd_hdr->ucode_feature_version); + adev->psp.asd_ucode_size = le32_to_cpu(asd_hdr->header.ucode_size_bytes); + adev->psp.asd_start_addr = (uint8_t *)asd_hdr + + le32_to_cpu(asd_hdr->header.ucode_array_offset_bytes); + + snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ta.bin", chip_name); + err = request_firmware(&adev->psp.ta_fw, fw_name, adev->dev); if (err) { - dev_err(adev->dev, - "psp v11.0: Failed to load firmware \"%s\"\n", - fw_name); - release_firmware(adev->psp.sos_fw); - adev->psp.sos_fw = NULL; + release_firmware(adev->psp.ta_fw); + adev->psp.ta_fw = NULL; + dev_info(adev->dev, + "psp v11.0: Failed to load firmware \"%s\"\n", fw_name); + } else { + err = amdgpu_ucode_validate(adev->psp.ta_fw); + if (err) + goto out2; + + ta_hdr = (const struct ta_firmware_header_v1_0 *)adev->psp.ta_fw->data; + adev->psp.ta_xgmi_ucode_version = le32_to_cpu(ta_hdr->ta_xgmi_ucode_version); + adev->psp.ta_xgmi_ucode_size = le32_to_cpu(ta_hdr->ta_xgmi_size_bytes); + adev->psp.ta_xgmi_start_addr = (uint8_t *)ta_hdr + + le32_to_cpu(ta_hdr->header.ucode_array_offset_bytes); } + return 0; + +out2: + release_firmware(adev->psp.ta_fw); + adev->psp.ta_fw = NULL; +out1: + release_firmware(adev->psp.asd_fw); + adev->psp.asd_fw = NULL; +out: + dev_err(adev->dev, + "psp v11.0: Failed to load firmware \"%s\"\n", fw_name); + release_firmware(adev->psp.sos_fw); + adev->psp.sos_fw = NULL; + return err; } @@ -153,8 +143,11 @@ static int psp_v11_0_bootloader_load_sysdrv(struct psp_context *psp) * are already been loaded. */ sol_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_81); - if (sol_reg) + if (sol_reg) { + psp->sos_fw_version = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_58); + printk("sos fw version = 0x%x.\n", psp->sos_fw_version); return 0; + } /* Wait for bootloader to signify that is ready having bit 31 of C2PMSG_35 set to 1 */ ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_35), @@ -167,7 +160,7 @@ static int psp_v11_0_bootloader_load_sysdrv(struct psp_context *psp) /* Copy PSP System Driver binary to memory */ memcpy(psp->fw_pri_buf, psp->sys_start_addr, psp->sys_bin_size); - /* Provide the sys driver to bootrom */ + /* Provide the sys driver to bootloader */ WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_36, (uint32_t)(psp->fw_pri_mc_addr >> 20)); psp_gfxdrv_command_reg = 1 << 16; @@ -208,7 +201,7 @@ static int psp_v11_0_bootloader_load_sos(struct psp_context *psp) /* Copy Secure OS binary to PSP memory */ memcpy(psp->fw_pri_buf, psp->sos_start_addr, psp->sos_bin_size); - /* Provide the PSP secure OS to bootrom */ + /* Provide the PSP secure OS to bootloader */ WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_36, (uint32_t)(psp->fw_pri_mc_addr >> 20)); psp_gfxdrv_command_reg = 2 << 16; @@ -224,26 +217,6 @@ static int psp_v11_0_bootloader_load_sos(struct psp_context *psp) return ret; } -static int psp_v11_0_prep_cmd_buf(struct amdgpu_firmware_info *ucode, - struct psp_gfx_cmd_resp *cmd) -{ - int ret; - uint64_t fw_mem_mc_addr = ucode->mc_addr; - - memset(cmd, 0, sizeof(struct psp_gfx_cmd_resp)); - - cmd->cmd_id = GFX_CMD_ID_LOAD_IP_FW; - cmd->cmd.cmd_load_ip_fw.fw_phy_addr_lo = lower_32_bits(fw_mem_mc_addr); - cmd->cmd.cmd_load_ip_fw.fw_phy_addr_hi = upper_32_bits(fw_mem_mc_addr); - cmd->cmd.cmd_load_ip_fw.fw_size = ucode->ucode_size; - - ret = psp_v11_0_get_fw_type(ucode, &cmd->cmd.cmd_load_ip_fw.fw_type); - if (ret) - DRM_ERROR("Unknown firmware type\n"); - - return ret; -} - static int psp_v11_0_ring_init(struct psp_context *psp, enum psp_ring_type ring_type) { @@ -270,6 +243,13 @@ static int psp_v11_0_ring_init(struct psp_context *psp, return 0; } +static bool psp_v11_0_support_vmr_ring(struct psp_context *psp) +{ + if (amdgpu_sriov_vf(psp->adev) && psp->sos_fw_version > 0x80045) + return true; + return false; +} + static int psp_v11_0_ring_create(struct psp_context *psp, enum psp_ring_type ring_type) { @@ -278,26 +258,47 @@ static int psp_v11_0_ring_create(struct psp_context *psp, struct psp_ring *ring = &psp->km_ring; struct amdgpu_device *adev = psp->adev; - /* Write low address of the ring to C2PMSG_69 */ - psp_ring_reg = lower_32_bits(ring->ring_mem_mc_addr); - WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_69, psp_ring_reg); - /* Write high address of the ring to C2PMSG_70 */ - psp_ring_reg = upper_32_bits(ring->ring_mem_mc_addr); - WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_70, psp_ring_reg); - /* Write size of ring to C2PMSG_71 */ - psp_ring_reg = ring->ring_size; - WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_71, psp_ring_reg); - /* Write the ring initialization command to C2PMSG_64 */ - psp_ring_reg = ring_type; - psp_ring_reg = psp_ring_reg << 16; - WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_64, psp_ring_reg); - - /* there might be handshake issue with hardware which needs delay */ - mdelay(20); - - /* Wait for response flag (bit 31) in C2PMSG_64 */ - ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_64), - 0x80000000, 0x8000FFFF, false); + if (psp_v11_0_support_vmr_ring(psp)) { + /* Write low address of the ring to C2PMSG_102 */ + psp_ring_reg = lower_32_bits(ring->ring_mem_mc_addr); + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_102, psp_ring_reg); + /* Write high address of the ring to C2PMSG_103 */ + psp_ring_reg = upper_32_bits(ring->ring_mem_mc_addr); + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_103, psp_ring_reg); + + /* Write the ring initialization command to C2PMSG_101 */ + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_101, + GFX_CTRL_CMD_ID_INIT_GPCOM_RING); + + /* there might be handshake issue with hardware which needs delay */ + mdelay(20); + + /* Wait for response flag (bit 31) in C2PMSG_101 */ + ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_101), + 0x80000000, 0x8000FFFF, false); + + } else { + /* Write low address of the ring to C2PMSG_69 */ + psp_ring_reg = lower_32_bits(ring->ring_mem_mc_addr); + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_69, psp_ring_reg); + /* Write high address of the ring to C2PMSG_70 */ + psp_ring_reg = upper_32_bits(ring->ring_mem_mc_addr); + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_70, psp_ring_reg); + /* Write size of ring to C2PMSG_71 */ + psp_ring_reg = ring->ring_size; + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_71, psp_ring_reg); + /* Write the ring initialization command to C2PMSG_64 */ + psp_ring_reg = ring_type; + psp_ring_reg = psp_ring_reg << 16; + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_64, psp_ring_reg); + + /* there might be handshake issue with hardware which needs delay */ + mdelay(20); + + /* Wait for response flag (bit 31) in C2PMSG_64 */ + ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_64), + 0x80000000, 0x8000FFFF, false); + } return ret; } @@ -308,15 +309,24 @@ static int psp_v11_0_ring_stop(struct psp_context *psp, int ret = 0; struct amdgpu_device *adev = psp->adev; - /* Write the ring destroy command to C2PMSG_64 */ - WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_64, GFX_CTRL_CMD_ID_DESTROY_RINGS); + /* Write the ring destroy command*/ + if (psp_v11_0_support_vmr_ring(psp)) + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_101, + GFX_CTRL_CMD_ID_DESTROY_GPCOM_RING); + else + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_64, + GFX_CTRL_CMD_ID_DESTROY_RINGS); /* there might be handshake issue with hardware which needs delay */ mdelay(20); - /* Wait for response flag (bit 31) in C2PMSG_64 */ - ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_64), - 0x80000000, 0x80000000, false); + /* Wait for response flag (bit 31) */ + if (psp_v11_0_support_vmr_ring(psp)) + ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_101), + 0x80000000, 0x80000000, false); + else + ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_64), + 0x80000000, 0x80000000, false); return ret; } @@ -355,7 +365,10 @@ static int psp_v11_0_cmd_submit(struct psp_context *psp, uint32_t rb_frame_size_dw = sizeof(struct psp_gfx_rb_frame) / 4; /* KM (GPCOM) prepare write pointer */ - psp_write_ptr_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_67); + if (psp_v11_0_support_vmr_ring(psp)) + psp_write_ptr_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_102); + else + psp_write_ptr_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_67); /* Update KM RB frame pointer to new frame */ /* write_frame ptr increments by size of rb_frame in bytes */ @@ -384,7 +397,11 @@ static int psp_v11_0_cmd_submit(struct psp_context *psp, /* Update the write Pointer in DWORDs */ psp_write_ptr_reg = (psp_write_ptr_reg + rb_frame_size_dw) % ring_size_dw; - WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_67, psp_write_ptr_reg); + if (psp_v11_0_support_vmr_ring(psp)) { + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_102, psp_write_ptr_reg); + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_101, GFX_CTRL_CMD_ID_CONSUME_CMD); + } else + WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_67, psp_write_ptr_reg); return 0; } @@ -529,7 +546,7 @@ static int psp_v11_0_mode1_reset(struct psp_context *psp) /*send the mode 1 reset command*/ WREG32(offset, GFX_CTRL_CMD_ID_MODE1_RST); - mdelay(1000); + msleep(500); offset = SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_33); @@ -552,31 +569,120 @@ static int psp_v11_0_mode1_reset(struct psp_context *psp) static int psp_v11_0_xgmi_get_topology_info(struct psp_context *psp, int number_devices, struct psp_xgmi_topology_info *topology) { + struct ta_xgmi_shared_memory *xgmi_cmd; + struct ta_xgmi_cmd_get_topology_info_input *topology_info_input; + struct ta_xgmi_cmd_get_topology_info_output *topology_info_output; + int i; + int ret; + + if (!topology || topology->num_nodes > TA_XGMI__MAX_CONNECTED_NODES) + return -EINVAL; + + xgmi_cmd = (struct ta_xgmi_shared_memory*)psp->xgmi_context.xgmi_shared_buf; + memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory)); + + /* Fill in the shared memory with topology information as input */ + topology_info_input = &xgmi_cmd->xgmi_in_message.get_topology_info; + xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_GET_TOPOLOGY_INFO; + topology_info_input->num_nodes = number_devices; + + for (i = 0; i < topology_info_input->num_nodes; i++) { + topology_info_input->nodes[i].node_id = topology->nodes[i].node_id; + topology_info_input->nodes[i].num_hops = topology->nodes[i].num_hops; + topology_info_input->nodes[i].is_sharing_enabled = topology->nodes[i].is_sharing_enabled; + topology_info_input->nodes[i].sdma_engine = topology->nodes[i].sdma_engine; + } + + /* Invoke xgmi ta to get the topology information */ + ret = psp_xgmi_invoke(psp, TA_COMMAND_XGMI__GET_GET_TOPOLOGY_INFO); + if (ret) + return ret; + + /* Read the output topology information from the shared memory */ + topology_info_output = &xgmi_cmd->xgmi_out_message.get_topology_info; + topology->num_nodes = xgmi_cmd->xgmi_out_message.get_topology_info.num_nodes; + for (i = 0; i < topology->num_nodes; i++) { + topology->nodes[i].node_id = topology_info_output->nodes[i].node_id; + topology->nodes[i].num_hops = topology_info_output->nodes[i].num_hops; + topology->nodes[i].is_sharing_enabled = topology_info_output->nodes[i].is_sharing_enabled; + topology->nodes[i].sdma_engine = topology_info_output->nodes[i].sdma_engine; + } + return 0; } static int psp_v11_0_xgmi_set_topology_info(struct psp_context *psp, int number_devices, struct psp_xgmi_topology_info *topology) { + struct ta_xgmi_shared_memory *xgmi_cmd; + struct ta_xgmi_cmd_get_topology_info_input *topology_info_input; + int i; + + if (!topology || topology->num_nodes > TA_XGMI__MAX_CONNECTED_NODES) + return -EINVAL; + + xgmi_cmd = (struct ta_xgmi_shared_memory*)psp->xgmi_context.xgmi_shared_buf; + memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory)); + + topology_info_input = &xgmi_cmd->xgmi_in_message.get_topology_info; + xgmi_cmd->cmd_id = TA_COMMAND_XGMI__SET_TOPOLOGY_INFO; + topology_info_input->num_nodes = number_devices; + + for (i = 0; i < topology_info_input->num_nodes; i++) { + topology_info_input->nodes[i].node_id = topology->nodes[i].node_id; + topology_info_input->nodes[i].num_hops = topology->nodes[i].num_hops; + topology_info_input->nodes[i].is_sharing_enabled = topology->nodes[i].is_sharing_enabled; + topology_info_input->nodes[i].sdma_engine = topology->nodes[i].sdma_engine; + } + + /* Invoke xgmi ta to set topology information */ + return psp_xgmi_invoke(psp, TA_COMMAND_XGMI__SET_TOPOLOGY_INFO); +} + +static int psp_v11_0_xgmi_get_hive_id(struct psp_context *psp, uint64_t *hive_id) +{ + struct ta_xgmi_shared_memory *xgmi_cmd; + int ret; + + xgmi_cmd = (struct ta_xgmi_shared_memory*)psp->xgmi_context.xgmi_shared_buf; + memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory)); + + xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_HIVE_ID; + + /* Invoke xgmi ta to get hive id */ + ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id); + if (ret) + return ret; + + *hive_id = xgmi_cmd->xgmi_out_message.get_hive_id.hive_id; + return 0; } -static u64 psp_v11_0_xgmi_get_hive_id(struct psp_context *psp) +static int psp_v11_0_xgmi_get_node_id(struct psp_context *psp, uint64_t *node_id) { - u64 hive_id = 0; + struct ta_xgmi_shared_memory *xgmi_cmd; + int ret; - /* Remove me when we can get correct hive_id through PSP */ - if (psp->adev->gmc.xgmi.num_physical_nodes) - hive_id = 0x123456789abcdef; + xgmi_cmd = (struct ta_xgmi_shared_memory*)psp->xgmi_context.xgmi_shared_buf; + memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory)); - return hive_id; + xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_NODE_ID; + + /* Invoke xgmi ta to get the node id */ + ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id); + if (ret) + return ret; + + *node_id = xgmi_cmd->xgmi_out_message.get_node_id.node_id; + + return 0; } static const struct psp_funcs psp_v11_0_funcs = { .init_microcode = psp_v11_0_init_microcode, .bootloader_load_sysdrv = psp_v11_0_bootloader_load_sysdrv, .bootloader_load_sos = psp_v11_0_bootloader_load_sos, - .prep_cmd_buf = psp_v11_0_prep_cmd_buf, .ring_init = psp_v11_0_ring_init, .ring_create = psp_v11_0_ring_create, .ring_stop = psp_v11_0_ring_stop, @@ -587,6 +693,8 @@ static const struct psp_funcs psp_v11_0_funcs = { .xgmi_get_topology_info = psp_v11_0_xgmi_get_topology_info, .xgmi_set_topology_info = psp_v11_0_xgmi_set_topology_info, .xgmi_get_hive_id = psp_v11_0_xgmi_get_hive_id, + .xgmi_get_node_id = psp_v11_0_xgmi_get_node_id, + .support_vmr_ring = psp_v11_0_support_vmr_ring, }; void psp_v11_0_set_psp_funcs(struct psp_context *psp) diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v3_1.c b/drivers/gpu/drm/amd/amdgpu/psp_v3_1.c index e1ebf770c303..0487e3a4e9e7 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v3_1.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v3_1.c @@ -47,57 +47,6 @@ MODULE_FIRMWARE("amdgpu/vega12_asd.bin"); static uint32_t sos_old_versions[] = {1517616, 1510592, 1448594, 1446554}; -static int -psp_v3_1_get_fw_type(struct amdgpu_firmware_info *ucode, enum psp_gfx_fw_type *type) -{ - switch(ucode->ucode_id) { - case AMDGPU_UCODE_ID_SDMA0: - *type = GFX_FW_TYPE_SDMA0; - break; - case AMDGPU_UCODE_ID_SDMA1: - *type = GFX_FW_TYPE_SDMA1; - break; - case AMDGPU_UCODE_ID_CP_CE: - *type = GFX_FW_TYPE_CP_CE; - break; - case AMDGPU_UCODE_ID_CP_PFP: - *type = GFX_FW_TYPE_CP_PFP; - break; - case AMDGPU_UCODE_ID_CP_ME: - *type = GFX_FW_TYPE_CP_ME; - break; - case AMDGPU_UCODE_ID_CP_MEC1: - *type = GFX_FW_TYPE_CP_MEC; - break; - case AMDGPU_UCODE_ID_CP_MEC1_JT: - *type = GFX_FW_TYPE_CP_MEC_ME1; - break; - case AMDGPU_UCODE_ID_CP_MEC2: - *type = GFX_FW_TYPE_CP_MEC; - break; - case AMDGPU_UCODE_ID_CP_MEC2_JT: - *type = GFX_FW_TYPE_CP_MEC_ME2; - break; - case AMDGPU_UCODE_ID_RLC_G: - *type = GFX_FW_TYPE_RLC_G; - break; - case AMDGPU_UCODE_ID_SMC: - *type = GFX_FW_TYPE_SMU; - break; - case AMDGPU_UCODE_ID_UVD: - *type = GFX_FW_TYPE_UVD; - break; - case AMDGPU_UCODE_ID_VCE: - *type = GFX_FW_TYPE_VCE; - break; - case AMDGPU_UCODE_ID_MAXIMUM: - default: - return -EINVAL; - } - - return 0; -} - static int psp_v3_1_init_microcode(struct psp_context *psp) { struct amdgpu_device *adev = psp->adev; @@ -194,7 +143,7 @@ static int psp_v3_1_bootloader_load_sysdrv(struct psp_context *psp) /* Copy PSP System Driver binary to memory */ memcpy(psp->fw_pri_buf, psp->sys_start_addr, psp->sys_bin_size); - /* Provide the sys driver to bootrom */ + /* Provide the sys driver to bootloader */ WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_36, (uint32_t)(psp->fw_pri_mc_addr >> 20)); psp_gfxdrv_command_reg = 1 << 16; @@ -240,8 +189,11 @@ static int psp_v3_1_bootloader_load_sos(struct psp_context *psp) * are already been loaded. */ sol_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_81); - if (sol_reg) + if (sol_reg) { + psp->sos_fw_version = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_58); + printk("sos fw version = 0x%x.\n", psp->sos_fw_version); return 0; + } /* Wait for bootloader to signify that is ready having bit 31 of C2PMSG_35 set to 1 */ ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_35), @@ -254,7 +206,7 @@ static int psp_v3_1_bootloader_load_sos(struct psp_context *psp) /* Copy Secure OS binary to PSP memory */ memcpy(psp->fw_pri_buf, psp->sos_start_addr, psp->sos_bin_size); - /* Provide the PSP secure OS to bootrom */ + /* Provide the PSP secure OS to bootloader */ WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_36, (uint32_t)(psp->fw_pri_mc_addr >> 20)); psp_gfxdrv_command_reg = 2 << 16; @@ -274,26 +226,6 @@ static int psp_v3_1_bootloader_load_sos(struct psp_context *psp) return ret; } -static int psp_v3_1_prep_cmd_buf(struct amdgpu_firmware_info *ucode, - struct psp_gfx_cmd_resp *cmd) -{ - int ret; - uint64_t fw_mem_mc_addr = ucode->mc_addr; - - memset(cmd, 0, sizeof(struct psp_gfx_cmd_resp)); - - cmd->cmd_id = GFX_CMD_ID_LOAD_IP_FW; - cmd->cmd.cmd_load_ip_fw.fw_phy_addr_lo = lower_32_bits(fw_mem_mc_addr); - cmd->cmd.cmd_load_ip_fw.fw_phy_addr_hi = upper_32_bits(fw_mem_mc_addr); - cmd->cmd.cmd_load_ip_fw.fw_size = ucode->ucode_size; - - ret = psp_v3_1_get_fw_type(ucode, &cmd->cmd.cmd_load_ip_fw.fw_type); - if (ret) - DRM_ERROR("Unknown firmware type\n"); - - return ret; -} - static int psp_v3_1_ring_init(struct psp_context *psp, enum psp_ring_type ring_type) { @@ -356,12 +288,9 @@ static int psp_v3_1_ring_stop(struct psp_context *psp, enum psp_ring_type ring_type) { int ret = 0; - struct psp_ring *ring; unsigned int psp_ring_reg = 0; struct amdgpu_device *adev = psp->adev; - ring = &psp->km_ring; - /* Write the ring destroy command to C2PMSG_64 */ psp_ring_reg = 3 << 16; WREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_64, psp_ring_reg); @@ -571,9 +500,7 @@ static bool psp_v3_1_smu_reload_quirk(struct psp_context *psp) struct amdgpu_device *adev = psp->adev; uint32_t reg; - reg = smnMP1_FIRMWARE_FLAGS | 0x03b00000; - WREG32_SOC15(NBIO, 0, mmPCIE_INDEX2, reg); - reg = RREG32_SOC15(NBIO, 0, mmPCIE_DATA2); + reg = RREG32_PCIE(smnMP1_FIRMWARE_FLAGS | 0x03b00000); return (reg & MP1_FIRMWARE_FLAGS__INTERRUPTS_ENABLED_MASK) ? true : false; } @@ -593,9 +520,9 @@ static int psp_v3_1_mode1_reset(struct psp_context *psp) } /*send the mode 1 reset command*/ - WREG32(offset, 0x70000); + WREG32(offset, GFX_CTRL_CMD_ID_MODE1_RST); - mdelay(1000); + msleep(500); offset = SOC15_REG_OFFSET(MP0, 0, mmMP0_SMN_C2PMSG_33); @@ -615,7 +542,6 @@ static const struct psp_funcs psp_v3_1_funcs = { .init_microcode = psp_v3_1_init_microcode, .bootloader_load_sysdrv = psp_v3_1_bootloader_load_sysdrv, .bootloader_load_sos = psp_v3_1_bootloader_load_sos, - .prep_cmd_buf = psp_v3_1_prep_cmd_buf, .ring_init = psp_v3_1_ring_init, .ring_create = psp_v3_1_ring_create, .ring_stop = psp_v3_1_ring_stop, diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index 2d4770e173dd..cca3552b36ed 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c @@ -225,7 +225,7 @@ static void sdma_v2_4_ring_set_wptr(struct amdgpu_ring *ring) static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) { - struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring); int i; for (i = 0; i < count; i++) @@ -245,9 +245,12 @@ static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (VI). */ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + /* IB packet must end on a 8 DW boundary */ sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8); @@ -349,8 +352,8 @@ static void sdma_v2_4_gfx_stop(struct amdgpu_device *adev) ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 0); WREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i], ib_cntl); } - sdma0->ready = false; - sdma1->ready = false; + sdma0->sched.ready = false; + sdma1->sched.ready = false; } /** @@ -471,17 +474,15 @@ static int sdma_v2_4_gfx_resume(struct amdgpu_device *adev) /* enable DMA IBs */ WREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i], ib_cntl); - ring->ready = true; + ring->sched.ready = true; } sdma_v2_4_enable(adev, true); for (i = 0; i < adev->sdma.num_instances; i++) { ring = &adev->sdma.instance[i].ring; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) return r; - } if (adev->mman.buffer_funcs_ring == ring) amdgpu_ttm_set_buffer_funcs_status(adev, true); @@ -550,21 +551,16 @@ static int sdma_v2_4_ring_test_ring(struct amdgpu_ring *ring) u64 gpu_addr; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); r = amdgpu_ring_alloc(ring, 5); - if (r) { - DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); - amdgpu_device_wb_free(adev, index); - return r; - } + if (r) + goto error_free_wb; amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR)); @@ -581,15 +577,11 @@ static int sdma_v2_4_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } - amdgpu_device_wb_free(adev, index); + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; +error_free_wb: + amdgpu_device_wb_free(adev, index); return r; } @@ -612,20 +604,16 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 256, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err0; - } ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR); @@ -644,21 +632,16 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out\n"); r = -ETIMEDOUT; goto err1; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto err1; } tmp = le32_to_cpu(adev->wb.wb[index]); - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + if (tmp == 0xDEADBEEF) r = 0; - } else { - DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp); + else r = -EINVAL; - } err1: amdgpu_ib_free(adev, &ib, NULL); @@ -760,7 +743,7 @@ static void sdma_v2_4_vm_set_pte_pde(struct amdgpu_ib *ib, uint64_t pe, */ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring); u32 pad_count; int i; @@ -1105,8 +1088,14 @@ static int sdma_v2_4_process_illegal_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { + u8 instance_id, queue_id; + DRM_ERROR("Illegal instruction in SDMA command stream\n"); - schedule_work(&adev->reset_work); + instance_id = (entry->ring_id & 0x3) >> 0; + queue_id = (entry->ring_id & 0xc) >> 2; + + if (instance_id <= 1 && queue_id == 0) + drm_sched_fault(&adev->sdma.instance[instance_id].ring.sched); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c index 6fb3edaba0ec..0ce8331baeb2 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c @@ -399,7 +399,7 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring) static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) { - struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring); int i; for (i = 0; i < count; i++) @@ -419,9 +419,12 @@ static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (VI). */ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + /* IB packet must end on a 8 DW boundary */ sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8); @@ -523,8 +526,8 @@ static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev) ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 0); WREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i], ib_cntl); } - sdma0->ready = false; - sdma1->ready = false; + sdma0->sched.ready = false; + sdma1->sched.ready = false; } /** @@ -739,7 +742,7 @@ static int sdma_v3_0_gfx_resume(struct amdgpu_device *adev) /* enable DMA IBs */ WREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i], ib_cntl); - ring->ready = true; + ring->sched.ready = true; } /* unhalt the MEs */ @@ -749,11 +752,9 @@ static int sdma_v3_0_gfx_resume(struct amdgpu_device *adev) for (i = 0; i < adev->sdma.num_instances; i++) { ring = &adev->sdma.instance[i].ring; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) return r; - } if (adev->mman.buffer_funcs_ring == ring) amdgpu_ttm_set_buffer_funcs_status(adev, true); @@ -822,21 +823,16 @@ static int sdma_v3_0_ring_test_ring(struct amdgpu_ring *ring) u64 gpu_addr; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); r = amdgpu_ring_alloc(ring, 5); - if (r) { - DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); - amdgpu_device_wb_free(adev, index); - return r; - } + if (r) + goto error_free_wb; amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR)); @@ -853,15 +849,11 @@ static int sdma_v3_0_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } - amdgpu_device_wb_free(adev, index); + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; +error_free_wb: + amdgpu_device_wb_free(adev, index); return r; } @@ -884,20 +876,16 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 256, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err0; - } ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR); @@ -916,21 +904,16 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out\n"); r = -ETIMEDOUT; goto err1; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto err1; } tmp = le32_to_cpu(adev->wb.wb[index]); - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + if (tmp == 0xDEADBEEF) r = 0; - } else { - DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp); + else r = -EINVAL; - } err1: amdgpu_ib_free(adev, &ib, NULL); dma_fence_put(f); @@ -1031,7 +1014,7 @@ static void sdma_v3_0_vm_set_pte_pde(struct amdgpu_ib *ib, uint64_t pe, */ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring); u32 pad_count; int i; @@ -1162,8 +1145,7 @@ static int sdma_v3_0_sw_init(void *handle) ring->ring_obj = NULL; if (!amdgpu_sriov_vf(adev)) { ring->use_doorbell = true; - ring->doorbell_index = (i == 0) ? - AMDGPU_DOORBELL_sDMA_ENGINE0 : AMDGPU_DOORBELL_sDMA_ENGINE1; + ring->doorbell_index = adev->doorbell_index.sdma_engine[i]; } else { ring->use_pollmem = true; } @@ -1440,8 +1422,14 @@ static int sdma_v3_0_process_illegal_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { + u8 instance_id, queue_id; + DRM_ERROR("Illegal instruction in SDMA command stream\n"); - schedule_work(&adev->reset_work); + instance_id = (entry->ring_id & 0x3) >> 0; + queue_id = (entry->ring_id & 0xc) >> 2; + + if (instance_id <= 1 && queue_id == 0) + drm_sched_fault(&adev->sdma.instance[instance_id].ring.sched); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c index 7a8c9172d30a..c816e55d43a9 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c @@ -54,6 +54,11 @@ MODULE_FIRMWARE("amdgpu/raven2_sdma.bin"); #define SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK 0x000000F8L #define SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK 0xFC000000L +#define WREG32_SDMA(instance, offset, value) \ + WREG32(sdma_v4_0_get_reg_offset(adev, (instance), (offset)), value) +#define RREG32_SDMA(instance, offset) \ + RREG32(sdma_v4_0_get_reg_offset(adev, (instance), (offset))) + static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev); static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev); static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev); @@ -73,7 +78,6 @@ static const struct soc15_reg_golden golden_settings_sdma_4[] = { SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000), SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0), SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x00000000), - SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831f07), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_IB_CNTL, 0x800f0100, 0x00000100), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000), @@ -91,6 +95,7 @@ static const struct soc15_reg_golden golden_settings_sdma_4[] = { static const struct soc15_reg_golden golden_settings_sdma_vg10[] = { SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104002), SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002), + SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104002), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002) }; @@ -98,6 +103,7 @@ static const struct soc15_reg_golden golden_settings_sdma_vg10[] = { static const struct soc15_reg_golden golden_settings_sdma_vg12[] = { SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104001), SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104001), + SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104001), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104001) }; @@ -122,7 +128,7 @@ static const struct soc15_reg_golden golden_settings_sdma0_4_2_init[] = { static const struct soc15_reg_golden golden_settings_sdma0_4_2[] = { - SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07), + SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831f07), SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100), SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0000773f, 0x00004002), SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002), @@ -152,7 +158,7 @@ static const struct soc15_reg_golden golden_settings_sdma0_4_2[] = }; static const struct soc15_reg_golden golden_settings_sdma1_4_2[] = { - SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07), + SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831f07), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0000773f, 0x00004002), SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002), @@ -367,16 +373,11 @@ static uint64_t sdma_v4_0_ring_get_wptr(struct amdgpu_ring *ring) wptr = READ_ONCE(*((u64 *)&adev->wb.wb[ring->wptr_offs])); DRM_DEBUG("wptr/doorbell before shift == 0x%016llx\n", wptr); } else { - u32 lowbit, highbit; - - lowbit = RREG32(sdma_v4_0_get_reg_offset(adev, ring->me, mmSDMA0_GFX_RB_WPTR)) >> 2; - highbit = RREG32(sdma_v4_0_get_reg_offset(adev, ring->me, mmSDMA0_GFX_RB_WPTR_HI)) >> 2; - - DRM_DEBUG("wptr [%i]high== 0x%08x low==0x%08x\n", - ring->me, highbit, lowbit); - wptr = highbit; + wptr = RREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR_HI); wptr = wptr << 32; - wptr |= lowbit; + wptr |= RREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR); + DRM_DEBUG("wptr before shift [%i] wptr == 0x%016llx\n", + ring->me, wptr); } return wptr >> 2; @@ -417,14 +418,67 @@ static void sdma_v4_0_ring_set_wptr(struct amdgpu_ring *ring) lower_32_bits(ring->wptr << 2), ring->me, upper_32_bits(ring->wptr << 2)); - WREG32(sdma_v4_0_get_reg_offset(adev, ring->me, mmSDMA0_GFX_RB_WPTR), lower_32_bits(ring->wptr << 2)); - WREG32(sdma_v4_0_get_reg_offset(adev, ring->me, mmSDMA0_GFX_RB_WPTR_HI), upper_32_bits(ring->wptr << 2)); + WREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR, + lower_32_bits(ring->wptr << 2)); + WREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR_HI, + upper_32_bits(ring->wptr << 2)); + } +} + +/** + * sdma_v4_0_page_ring_get_wptr - get the current write pointer + * + * @ring: amdgpu ring pointer + * + * Get the current wptr from the hardware (VEGA10+). + */ +static uint64_t sdma_v4_0_page_ring_get_wptr(struct amdgpu_ring *ring) +{ + struct amdgpu_device *adev = ring->adev; + u64 wptr; + + if (ring->use_doorbell) { + /* XXX check if swapping is necessary on BE */ + wptr = READ_ONCE(*((u64 *)&adev->wb.wb[ring->wptr_offs])); + } else { + wptr = RREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR_HI); + wptr = wptr << 32; + wptr |= RREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR); + } + + return wptr >> 2; +} + +/** + * sdma_v4_0_ring_set_wptr - commit the write pointer + * + * @ring: amdgpu ring pointer + * + * Write the wptr back to the hardware (VEGA10+). + */ +static void sdma_v4_0_page_ring_set_wptr(struct amdgpu_ring *ring) +{ + struct amdgpu_device *adev = ring->adev; + + if (ring->use_doorbell) { + u64 *wb = (u64 *)&adev->wb.wb[ring->wptr_offs]; + + /* XXX check if swapping is necessary on BE */ + WRITE_ONCE(*wb, (ring->wptr << 2)); + WDOORBELL64(ring->doorbell_index, ring->wptr << 2); + } else { + uint64_t wptr = ring->wptr << 2; + + WREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR, + lower_32_bits(wptr)); + WREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR_HI, + upper_32_bits(wptr)); } } static void sdma_v4_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) { - struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring); int i; for (i = 0; i < count; i++) @@ -444,9 +498,12 @@ static void sdma_v4_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (VEGA10). */ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + /* IB packet must end on a 8 DW boundary */ sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8); @@ -568,16 +625,16 @@ static void sdma_v4_0_gfx_stop(struct amdgpu_device *adev) amdgpu_ttm_set_buffer_funcs_status(adev, false); for (i = 0; i < adev->sdma.num_instances; i++) { - rb_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL)); + rb_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL); rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL), rb_cntl); - ib_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_IB_CNTL)); + WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl); + ib_cntl = RREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL); ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 0); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_IB_CNTL), ib_cntl); + WREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL, ib_cntl); } - sdma0->ready = false; - sdma1->ready = false; + sdma0->sched.ready = false; + sdma1->sched.ready = false; } /** @@ -593,6 +650,39 @@ static void sdma_v4_0_rlc_stop(struct amdgpu_device *adev) } /** + * sdma_v4_0_page_stop - stop the page async dma engines + * + * @adev: amdgpu_device pointer + * + * Stop the page async dma ring buffers (VEGA10). + */ +static void sdma_v4_0_page_stop(struct amdgpu_device *adev) +{ + struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].page; + struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].page; + u32 rb_cntl, ib_cntl; + int i; + + if ((adev->mman.buffer_funcs_ring == sdma0) || + (adev->mman.buffer_funcs_ring == sdma1)) + amdgpu_ttm_set_buffer_funcs_status(adev, false); + + for (i = 0; i < adev->sdma.num_instances; i++) { + rb_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL); + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL, + RB_ENABLE, 0); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl); + ib_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL); + ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, + IB_ENABLE, 0); + WREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL, ib_cntl); + } + + sdma0->sched.ready = false; + sdma1->sched.ready = false; +} + +/** * sdma_v_0_ctx_switch_enable - stop the async dma engines context switch * * @adev: amdgpu_device pointer @@ -630,18 +720,15 @@ static void sdma_v4_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable) } for (i = 0; i < adev->sdma.num_instances; i++) { - f32_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_CNTL)); + f32_cntl = RREG32_SDMA(i, mmSDMA0_CNTL); f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL, AUTO_CTXSW_ENABLE, enable ? 1 : 0); if (enable && amdgpu_sdma_phase_quantum) { - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_PHASE0_QUANTUM), - phase_quantum); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_PHASE1_QUANTUM), - phase_quantum); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_PHASE2_QUANTUM), - phase_quantum); + WREG32_SDMA(i, mmSDMA0_PHASE0_QUANTUM, phase_quantum); + WREG32_SDMA(i, mmSDMA0_PHASE1_QUANTUM, phase_quantum); + WREG32_SDMA(i, mmSDMA0_PHASE2_QUANTUM, phase_quantum); } - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_CNTL), f32_cntl); + WREG32_SDMA(i, mmSDMA0_CNTL, f32_cntl); } } @@ -662,156 +749,213 @@ static void sdma_v4_0_enable(struct amdgpu_device *adev, bool enable) if (enable == false) { sdma_v4_0_gfx_stop(adev); sdma_v4_0_rlc_stop(adev); + if (adev->sdma.has_page_queue) + sdma_v4_0_page_stop(adev); } for (i = 0; i < adev->sdma.num_instances; i++) { - f32_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_F32_CNTL)); + f32_cntl = RREG32_SDMA(i, mmSDMA0_F32_CNTL); f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, enable ? 0 : 1); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_F32_CNTL), f32_cntl); + WREG32_SDMA(i, mmSDMA0_F32_CNTL, f32_cntl); } } /** + * sdma_v4_0_rb_cntl - get parameters for rb_cntl + */ +static uint32_t sdma_v4_0_rb_cntl(struct amdgpu_ring *ring, uint32_t rb_cntl) +{ + /* Set ring buffer size in dwords */ + uint32_t rb_bufsz = order_base_2(ring->ring_size / 4); + + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SIZE, rb_bufsz); +#ifdef __BIG_ENDIAN + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SWAP_ENABLE, 1); + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, + RPTR_WRITEBACK_SWAP_ENABLE, 1); +#endif + return rb_cntl; +} + +/** * sdma_v4_0_gfx_resume - setup and start the async dma engines * * @adev: amdgpu_device pointer + * @i: instance to resume * * Set up the gfx DMA ring buffers and enable them (VEGA10). * Returns 0 for success, error for failure. */ -static int sdma_v4_0_gfx_resume(struct amdgpu_device *adev) +static void sdma_v4_0_gfx_resume(struct amdgpu_device *adev, unsigned int i) { - struct amdgpu_ring *ring; + struct amdgpu_ring *ring = &adev->sdma.instance[i].ring; u32 rb_cntl, ib_cntl, wptr_poll_cntl; - u32 rb_bufsz; u32 wb_offset; u32 doorbell; u32 doorbell_offset; - u32 temp; u64 wptr_gpu_addr; - int i, r; - for (i = 0; i < adev->sdma.num_instances; i++) { - ring = &adev->sdma.instance[i].ring; - wb_offset = (ring->rptr_offs * 4); + wb_offset = (ring->rptr_offs * 4); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL), 0); + rb_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL); + rb_cntl = sdma_v4_0_rb_cntl(ring, rb_cntl); + WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl); - /* Set ring buffer size in dwords */ - rb_bufsz = order_base_2(ring->ring_size / 4); - rb_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL)); - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SIZE, rb_bufsz); -#ifdef __BIG_ENDIAN - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SWAP_ENABLE, 1); - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, - RPTR_WRITEBACK_SWAP_ENABLE, 1); -#endif - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL), rb_cntl); + /* Initialize the ring buffer's read and write pointers */ + WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR, 0); + WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_HI, 0); + WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR, 0); + WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_HI, 0); - /* Initialize the ring buffer's read and write pointers */ - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_RPTR), 0); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_RPTR_HI), 0); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR), 0); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_HI), 0); + /* set the wb address whether it's enabled or not */ + WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_ADDR_HI, + upper_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF); + WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_ADDR_LO, + lower_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC); - /* set the wb address whether it's enabled or not */ - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_RPTR_ADDR_HI), - upper_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_RPTR_ADDR_LO), - lower_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC); + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, + RPTR_WRITEBACK_ENABLE, 1); - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RPTR_WRITEBACK_ENABLE, 1); + WREG32_SDMA(i, mmSDMA0_GFX_RB_BASE, ring->gpu_addr >> 8); + WREG32_SDMA(i, mmSDMA0_GFX_RB_BASE_HI, ring->gpu_addr >> 40); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_BASE), ring->gpu_addr >> 8); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_BASE_HI), ring->gpu_addr >> 40); + ring->wptr = 0; - ring->wptr = 0; + /* before programing wptr to a less value, need set minor_ptr_update first */ + WREG32_SDMA(i, mmSDMA0_GFX_MINOR_PTR_UPDATE, 1); - /* before programing wptr to a less value, need set minor_ptr_update first */ - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_MINOR_PTR_UPDATE), 1); + doorbell = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL); + doorbell_offset = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET); - if (!amdgpu_sriov_vf(adev)) { /* only bare-metal use register write for wptr */ - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR), lower_32_bits(ring->wptr) << 2); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_HI), upper_32_bits(ring->wptr) << 2); - } - - doorbell = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL)); - doorbell_offset = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL_OFFSET)); - - if (ring->use_doorbell) { - doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE, 1); - doorbell_offset = REG_SET_FIELD(doorbell_offset, SDMA0_GFX_DOORBELL_OFFSET, + doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE, + ring->use_doorbell); + doorbell_offset = REG_SET_FIELD(doorbell_offset, + SDMA0_GFX_DOORBELL_OFFSET, OFFSET, ring->doorbell_index); - } else { - doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE, 0); - } - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL), doorbell); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL_OFFSET), doorbell_offset); - adev->nbio_funcs->sdma_doorbell_range(adev, i, ring->use_doorbell, - ring->doorbell_index); + WREG32_SDMA(i, mmSDMA0_GFX_DOORBELL, doorbell); + WREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET, doorbell_offset); + + sdma_v4_0_ring_set_wptr(ring); + + /* set minor_ptr_update to 0 after wptr programed */ + WREG32_SDMA(i, mmSDMA0_GFX_MINOR_PTR_UPDATE, 0); + + /* setup the wptr shadow polling */ + wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4); + WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_LO, + lower_32_bits(wptr_gpu_addr)); + WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_HI, + upper_32_bits(wptr_gpu_addr)); + wptr_poll_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL); + wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl, + SDMA0_GFX_RB_WPTR_POLL_CNTL, + F32_POLL_ENABLE, amdgpu_sriov_vf(adev)); + WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, wptr_poll_cntl); + + /* enable DMA RB */ + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 1); + WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl); + + ib_cntl = RREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL); + ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 1); +#ifdef __BIG_ENDIAN + ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_SWAP_ENABLE, 1); +#endif + /* enable DMA IBs */ + WREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL, ib_cntl); - if (amdgpu_sriov_vf(adev)) - sdma_v4_0_ring_set_wptr(ring); + ring->sched.ready = true; +} - /* set minor_ptr_update to 0 after wptr programed */ - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_MINOR_PTR_UPDATE), 0); +/** + * sdma_v4_0_page_resume - setup and start the async dma engines + * + * @adev: amdgpu_device pointer + * @i: instance to resume + * + * Set up the page DMA ring buffers and enable them (VEGA10). + * Returns 0 for success, error for failure. + */ +static void sdma_v4_0_page_resume(struct amdgpu_device *adev, unsigned int i) +{ + struct amdgpu_ring *ring = &adev->sdma.instance[i].page; + u32 rb_cntl, ib_cntl, wptr_poll_cntl; + u32 wb_offset; + u32 doorbell; + u32 doorbell_offset; + u64 wptr_gpu_addr; - /* set utc l1 enable flag always to 1 */ - temp = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_CNTL)); - temp = REG_SET_FIELD(temp, SDMA0_CNTL, UTC_L1_ENABLE, 1); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_CNTL), temp); + wb_offset = (ring->rptr_offs * 4); - if (!amdgpu_sriov_vf(adev)) { - /* unhalt engine */ - temp = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_F32_CNTL)); - temp = REG_SET_FIELD(temp, SDMA0_F32_CNTL, HALT, 0); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_F32_CNTL), temp); - } + rb_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL); + rb_cntl = sdma_v4_0_rb_cntl(ring, rb_cntl); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl); - /* setup the wptr shadow polling */ - wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_LO), - lower_32_bits(wptr_gpu_addr)); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_HI), - upper_32_bits(wptr_gpu_addr)); - wptr_poll_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL)); - if (amdgpu_sriov_vf(adev)) - wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl, SDMA0_GFX_RB_WPTR_POLL_CNTL, F32_POLL_ENABLE, 1); - else - wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl, SDMA0_GFX_RB_WPTR_POLL_CNTL, F32_POLL_ENABLE, 0); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL), wptr_poll_cntl); + /* Initialize the ring buffer's read and write pointers */ + WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR, 0); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_HI, 0); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR, 0); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_HI, 0); - /* enable DMA RB */ - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 1); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL), rb_cntl); + /* set the wb address whether it's enabled or not */ + WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_ADDR_HI, + upper_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_ADDR_LO, + lower_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC); - ib_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_IB_CNTL)); - ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 1); -#ifdef __BIG_ENDIAN - ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_SWAP_ENABLE, 1); -#endif - /* enable DMA IBs */ - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_IB_CNTL), ib_cntl); + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL, + RPTR_WRITEBACK_ENABLE, 1); - ring->ready = true; + WREG32_SDMA(i, mmSDMA0_PAGE_RB_BASE, ring->gpu_addr >> 8); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_BASE_HI, ring->gpu_addr >> 40); - if (amdgpu_sriov_vf(adev)) { /* bare-metal sequence doesn't need below to lines */ - sdma_v4_0_ctx_switch_enable(adev, true); - sdma_v4_0_enable(adev, true); - } + ring->wptr = 0; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; - return r; - } + /* before programing wptr to a less value, need set minor_ptr_update first */ + WREG32_SDMA(i, mmSDMA0_PAGE_MINOR_PTR_UPDATE, 1); - if (adev->mman.buffer_funcs_ring == ring) - amdgpu_ttm_set_buffer_funcs_status(adev, true); + doorbell = RREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL); + doorbell_offset = RREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL_OFFSET); - } + doorbell = REG_SET_FIELD(doorbell, SDMA0_PAGE_DOORBELL, ENABLE, + ring->use_doorbell); + doorbell_offset = REG_SET_FIELD(doorbell_offset, + SDMA0_PAGE_DOORBELL_OFFSET, + OFFSET, ring->doorbell_index); + WREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL, doorbell); + WREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL_OFFSET, doorbell_offset); + + /* paging queue doorbell range is setup at sdma_v4_0_gfx_resume */ + sdma_v4_0_page_ring_set_wptr(ring); + + /* set minor_ptr_update to 0 after wptr programed */ + WREG32_SDMA(i, mmSDMA0_PAGE_MINOR_PTR_UPDATE, 0); + + /* setup the wptr shadow polling */ + wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_ADDR_LO, + lower_32_bits(wptr_gpu_addr)); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_ADDR_HI, + upper_32_bits(wptr_gpu_addr)); + wptr_poll_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL); + wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl, + SDMA0_PAGE_RB_WPTR_POLL_CNTL, + F32_POLL_ENABLE, amdgpu_sriov_vf(adev)); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, wptr_poll_cntl); + + /* enable DMA RB */ + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL, RB_ENABLE, 1); + WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl); + + ib_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL); + ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, IB_ENABLE, 1); +#ifdef __BIG_ENDIAN + ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, IB_SWAP_ENABLE, 1); +#endif + /* enable DMA IBs */ + WREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL, ib_cntl); - return 0; + ring->sched.ready = true; } static void @@ -922,12 +1066,14 @@ static int sdma_v4_0_load_microcode(struct amdgpu_device *adev) (adev->sdma.instance[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_UCODE_ADDR), 0); + WREG32_SDMA(i, mmSDMA0_UCODE_ADDR, 0); for (j = 0; j < fw_size; j++) - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_UCODE_DATA), le32_to_cpup(fw_data++)); + WREG32_SDMA(i, mmSDMA0_UCODE_DATA, + le32_to_cpup(fw_data++)); - WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_UCODE_ADDR), adev->sdma.instance[i].fw_version); + WREG32_SDMA(i, mmSDMA0_UCODE_ADDR, + adev->sdma.instance[i].fw_version); } return 0; @@ -943,33 +1089,78 @@ static int sdma_v4_0_load_microcode(struct amdgpu_device *adev) */ static int sdma_v4_0_start(struct amdgpu_device *adev) { - int r = 0; + struct amdgpu_ring *ring; + int i, r; if (amdgpu_sriov_vf(adev)) { sdma_v4_0_ctx_switch_enable(adev, false); sdma_v4_0_enable(adev, false); + } else { - /* set RB registers */ - r = sdma_v4_0_gfx_resume(adev); - return r; + if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { + r = sdma_v4_0_load_microcode(adev); + if (r) + return r; + } + + /* unhalt the MEs */ + sdma_v4_0_enable(adev, true); + /* enable sdma ring preemption */ + sdma_v4_0_ctx_switch_enable(adev, true); + } + + /* start the gfx rings and rlc compute queues */ + for (i = 0; i < adev->sdma.num_instances; i++) { + uint32_t temp; + + WREG32_SDMA(i, mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL, 0); + sdma_v4_0_gfx_resume(adev, i); + if (adev->sdma.has_page_queue) + sdma_v4_0_page_resume(adev, i); + + /* set utc l1 enable flag always to 1 */ + temp = RREG32_SDMA(i, mmSDMA0_CNTL); + temp = REG_SET_FIELD(temp, SDMA0_CNTL, UTC_L1_ENABLE, 1); + WREG32_SDMA(i, mmSDMA0_CNTL, temp); + + if (!amdgpu_sriov_vf(adev)) { + /* unhalt engine */ + temp = RREG32_SDMA(i, mmSDMA0_F32_CNTL); + temp = REG_SET_FIELD(temp, SDMA0_F32_CNTL, HALT, 0); + WREG32_SDMA(i, mmSDMA0_F32_CNTL, temp); + } } - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { - r = sdma_v4_0_load_microcode(adev); + if (amdgpu_sriov_vf(adev)) { + sdma_v4_0_ctx_switch_enable(adev, true); + sdma_v4_0_enable(adev, true); + } else { + r = sdma_v4_0_rlc_resume(adev); if (r) return r; } - /* unhalt the MEs */ - sdma_v4_0_enable(adev, true); - /* enable sdma ring preemption */ - sdma_v4_0_ctx_switch_enable(adev, true); + for (i = 0; i < adev->sdma.num_instances; i++) { + ring = &adev->sdma.instance[i].ring; - /* start the gfx rings and rlc compute queues */ - r = sdma_v4_0_gfx_resume(adev); - if (r) - return r; - r = sdma_v4_0_rlc_resume(adev); + r = amdgpu_ring_test_helper(ring); + if (r) + return r; + + if (adev->sdma.has_page_queue) { + struct amdgpu_ring *page = &adev->sdma.instance[i].page; + + r = amdgpu_ring_test_helper(page); + if (r) + return r; + + if (adev->mman.buffer_funcs_ring == page) + amdgpu_ttm_set_buffer_funcs_status(adev, true); + } + + if (adev->mman.buffer_funcs_ring == ring) + amdgpu_ttm_set_buffer_funcs_status(adev, true); + } return r; } @@ -993,21 +1184,16 @@ static int sdma_v4_0_ring_test_ring(struct amdgpu_ring *ring) u64 gpu_addr; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); r = amdgpu_ring_alloc(ring, 5); - if (r) { - DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); - amdgpu_device_wb_free(adev, index); - return r; - } + if (r) + goto error_free_wb; amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR)); @@ -1024,15 +1210,11 @@ static int sdma_v4_0_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } - amdgpu_device_wb_free(adev, index); + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; +error_free_wb: + amdgpu_device_wb_free(adev, index); return r; } @@ -1055,20 +1237,16 @@ static int sdma_v4_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) u64 gpu_addr; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 256, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err0; - } ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR); @@ -1087,21 +1265,17 @@ static int sdma_v4_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out\n"); r = -ETIMEDOUT; goto err1; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto err1; } tmp = le32_to_cpu(adev->wb.wb[index]); - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + if (tmp == 0xDEADBEEF) r = 0; - } else { - DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp); + else r = -EINVAL; - } + err1: amdgpu_ib_free(adev, &ib, NULL); dma_fence_put(f); @@ -1206,7 +1380,7 @@ static void sdma_v4_0_vm_set_pte_pde(struct amdgpu_ib *ib, */ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring); u32 pad_count; int i; @@ -1272,15 +1446,45 @@ static void sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg, sdma_v4_0_wait_reg_mem(ring, 0, 0, reg, 0, val, mask, 10); } +static bool sdma_v4_0_fw_support_paging_queue(struct amdgpu_device *adev) +{ + uint fw_version = adev->sdma.instance[0].fw_version; + + switch (adev->asic_type) { + case CHIP_VEGA10: + return fw_version >= 430; + case CHIP_VEGA12: + /*return fw_version >= 31;*/ + return false; + case CHIP_VEGA20: + return fw_version >= 123; + default: + return false; + } +} + static int sdma_v4_0_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int r; if (adev->asic_type == CHIP_RAVEN) adev->sdma.num_instances = 1; else adev->sdma.num_instances = 2; + r = sdma_v4_0_init_microcode(adev); + if (r) { + DRM_ERROR("Failed to load sdma firmware!\n"); + return r; + } + + /* TODO: Page queue breaks driver reload under SRIOV */ + if ((adev->asic_type == CHIP_VEGA10) && amdgpu_sriov_vf((adev))) + adev->sdma.has_page_queue = false; + else if (sdma_v4_0_fw_support_paging_queue(adev)) + adev->sdma.has_page_queue = true; + sdma_v4_0_set_ring_funcs(adev); sdma_v4_0_set_buffer_funcs(adev); sdma_v4_0_set_vm_pte_funcs(adev); @@ -1289,7 +1493,6 @@ static int sdma_v4_0_early_init(void *handle) return 0; } - static int sdma_v4_0_sw_init(void *handle) { struct amdgpu_ring *ring; @@ -1308,12 +1511,6 @@ static int sdma_v4_0_sw_init(void *handle) if (r) return r; - r = sdma_v4_0_init_microcode(adev); - if (r) { - DRM_ERROR("Failed to load sdma firmware!\n"); - return r; - } - for (i = 0; i < adev->sdma.num_instances; i++) { ring = &adev->sdma.instance[i].ring; ring->ring_obj = NULL; @@ -1322,15 +1519,8 @@ static int sdma_v4_0_sw_init(void *handle) DRM_INFO("use_doorbell being set to: [%s]\n", ring->use_doorbell?"true":"false"); - if (adev->asic_type == CHIP_VEGA10) - ring->doorbell_index = (i == 0) ? - (AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE0 << 1) //get DWORD offset - : (AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE1 << 1); // get DWORD offset - else - ring->doorbell_index = (i == 0) ? - (AMDGPU_DOORBELL64_sDMA_ENGINE0 << 1) //get DWORD offset - : (AMDGPU_DOORBELL64_sDMA_ENGINE1 << 1); // get DWORD offset - + /* doorbell size is 2 dwords, get DWORD offset */ + ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1; sprintf(ring->name, "sdma%d", i); r = amdgpu_ring_init(adev, ring, 1024, @@ -1340,6 +1530,27 @@ static int sdma_v4_0_sw_init(void *handle) AMDGPU_SDMA_IRQ_TRAP1); if (r) return r; + + if (adev->sdma.has_page_queue) { + ring = &adev->sdma.instance[i].page; + ring->ring_obj = NULL; + ring->use_doorbell = true; + + /* paging queue use same doorbell index/routing as gfx queue + * with 0x400 (4096 dwords) offset on second doorbell page + */ + ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1; + ring->doorbell_index += 0x400; + + sprintf(ring->name, "page%d", i); + r = amdgpu_ring_init(adev, ring, 1024, + &adev->sdma.trap_irq, + (i == 0) ? + AMDGPU_SDMA_IRQ_TRAP0 : + AMDGPU_SDMA_IRQ_TRAP1); + if (r) + return r; + } } return r; @@ -1350,8 +1561,11 @@ static int sdma_v4_0_sw_fini(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; int i; - for (i = 0; i < adev->sdma.num_instances; i++) + for (i = 0; i < adev->sdma.num_instances; i++) { amdgpu_ring_fini(&adev->sdma.instance[i].ring); + if (adev->sdma.has_page_queue) + amdgpu_ring_fini(&adev->sdma.instance[i].page); + } for (i = 0; i < adev->sdma.num_instances; i++) { release_firmware(adev->sdma.instance[i].fw); @@ -1414,7 +1628,7 @@ static bool sdma_v4_0_is_idle(void *handle) u32 i; for (i = 0; i < adev->sdma.num_instances; i++) { - u32 tmp = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_STATUS_REG)); + u32 tmp = RREG32_SDMA(i, mmSDMA0_STATUS_REG); if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK)) return false; @@ -1430,8 +1644,8 @@ static int sdma_v4_0_wait_for_idle(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; for (i = 0; i < adev->usec_timeout; i++) { - sdma0 = RREG32(sdma_v4_0_get_reg_offset(adev, 0, mmSDMA0_STATUS_REG)); - sdma1 = RREG32(sdma_v4_0_get_reg_offset(adev, 1, mmSDMA0_STATUS_REG)); + sdma0 = RREG32_SDMA(0, mmSDMA0_STATUS_REG); + sdma1 = RREG32_SDMA(1, mmSDMA0_STATUS_REG); if (sdma0 & sdma1 & SDMA0_STATUS_REG__IDLE_MASK) return 0; @@ -1452,16 +1666,13 @@ static int sdma_v4_0_set_trap_irq_state(struct amdgpu_device *adev, unsigned type, enum amdgpu_interrupt_state state) { + unsigned int instance = (type == AMDGPU_SDMA_IRQ_TRAP0) ? 0 : 1; u32 sdma_cntl; - u32 reg_offset = (type == AMDGPU_SDMA_IRQ_TRAP0) ? - sdma_v4_0_get_reg_offset(adev, 0, mmSDMA0_CNTL) : - sdma_v4_0_get_reg_offset(adev, 1, mmSDMA0_CNTL); - - sdma_cntl = RREG32(reg_offset); + sdma_cntl = RREG32_SDMA(instance, mmSDMA0_CNTL); sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE, state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); - WREG32(reg_offset, sdma_cntl); + WREG32_SDMA(instance, mmSDMA0_CNTL, sdma_cntl); return 0; } @@ -1470,39 +1681,34 @@ static int sdma_v4_0_process_trap_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { + uint32_t instance; + DRM_DEBUG("IH: SDMA trap\n"); switch (entry->client_id) { case SOC15_IH_CLIENTID_SDMA0: - switch (entry->ring_id) { - case 0: - amdgpu_fence_process(&adev->sdma.instance[0].ring); - break; - case 1: - /* XXX compute */ - break; - case 2: - /* XXX compute */ - break; - case 3: - /* XXX page queue*/ - break; - } + instance = 0; break; case SOC15_IH_CLIENTID_SDMA1: - switch (entry->ring_id) { - case 0: - amdgpu_fence_process(&adev->sdma.instance[1].ring); - break; - case 1: - /* XXX compute */ - break; - case 2: - /* XXX compute */ - break; - case 3: - /* XXX page queue*/ - break; - } + instance = 1; + break; + default: + return 0; + } + + switch (entry->ring_id) { + case 0: + amdgpu_fence_process(&adev->sdma.instance[instance].ring); + break; + case 1: + if (adev->asic_type == CHIP_VEGA20) + amdgpu_fence_process(&adev->sdma.instance[instance].page); + break; + case 2: + /* XXX compute */ + break; + case 3: + if (adev->asic_type != CHIP_VEGA20) + amdgpu_fence_process(&adev->sdma.instance[instance].page); break; } return 0; @@ -1512,12 +1718,29 @@ static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { + int instance; + DRM_ERROR("Illegal instruction in SDMA command stream\n"); - schedule_work(&adev->reset_work); + + switch (entry->client_id) { + case SOC15_IH_CLIENTID_SDMA0: + instance = 0; + break; + case SOC15_IH_CLIENTID_SDMA1: + instance = 1; + break; + default: + return 0; + } + + switch (entry->ring_id) { + case 0: + drm_sched_fault(&adev->sdma.instance[instance].ring.sched); + break; + } return 0; } - static void sdma_v4_0_update_medium_grain_clock_gating( struct amdgpu_device *adev, bool enable) @@ -1730,6 +1953,38 @@ static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = { .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, }; +static const struct amdgpu_ring_funcs sdma_v4_0_page_ring_funcs = { + .type = AMDGPU_RING_TYPE_SDMA, + .align_mask = 0xf, + .nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), + .support_64bit_ptrs = true, + .vmhub = AMDGPU_MMHUB, + .get_rptr = sdma_v4_0_ring_get_rptr, + .get_wptr = sdma_v4_0_page_ring_get_wptr, + .set_wptr = sdma_v4_0_page_ring_set_wptr, + .emit_frame_size = + 6 + /* sdma_v4_0_ring_emit_hdp_flush */ + 3 + /* hdp invalidate */ + 6 + /* sdma_v4_0_ring_emit_pipeline_sync */ + /* sdma_v4_0_ring_emit_vm_flush */ + SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + + SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 + + 10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */ + .emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */ + .emit_ib = sdma_v4_0_ring_emit_ib, + .emit_fence = sdma_v4_0_ring_emit_fence, + .emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync, + .emit_vm_flush = sdma_v4_0_ring_emit_vm_flush, + .emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush, + .test_ring = sdma_v4_0_ring_test_ring, + .test_ib = sdma_v4_0_ring_test_ib, + .insert_nop = sdma_v4_0_ring_insert_nop, + .pad_ib = sdma_v4_0_ring_pad_ib, + .emit_wreg = sdma_v4_0_ring_emit_wreg, + .emit_reg_wait = sdma_v4_0_ring_emit_reg_wait, + .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, +}; + static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev) { int i; @@ -1737,6 +1992,10 @@ static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev) for (i = 0; i < adev->sdma.num_instances; i++) { adev->sdma.instance[i].ring.funcs = &sdma_v4_0_ring_funcs; adev->sdma.instance[i].ring.me = i; + if (adev->sdma.has_page_queue) { + adev->sdma.instance[i].page.funcs = &sdma_v4_0_page_ring_funcs; + adev->sdma.instance[i].page.me = i; + } } } @@ -1818,7 +2077,10 @@ static const struct amdgpu_buffer_funcs sdma_v4_0_buffer_funcs = { static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev) { adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs; - adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring; + if (adev->sdma.has_page_queue) + adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page; + else + adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring; } static const struct amdgpu_vm_pte_funcs sdma_v4_0_vm_pte_funcs = { @@ -1836,7 +2098,10 @@ static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev) adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs; for (i = 0; i < adev->sdma.num_instances; i++) { - sched = &adev->sdma.instance[i].ring.sched; + if (adev->sdma.has_page_queue) + sched = &adev->sdma.instance[i].page.sched; + else + sched = &adev->sdma.instance[i].ring.sched; adev->vm_manager.vm_pte_rqs[i] = &sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL]; } diff --git a/drivers/gpu/drm/amd/amdgpu/si.c b/drivers/gpu/drm/amd/amdgpu/si.c index f8408f88cd37..9d8df68893b9 100644 --- a/drivers/gpu/drm/amd/amdgpu/si.c +++ b/drivers/gpu/drm/amd/amdgpu/si.c @@ -47,6 +47,7 @@ #include "dce/dce_6_0_d.h" #include "uvd/uvd_4_0_d.h" #include "bif/bif_3_0_d.h" +#include "bif/bif_3_0_sh_mask.h" static const u32 tahiti_golden_registers[] = { @@ -1258,6 +1259,11 @@ static bool si_need_full_reset(struct amdgpu_device *adev) return true; } +static bool si_need_reset_on_init(struct amdgpu_device *adev) +{ + return false; +} + static int si_get_pcie_lanes(struct amdgpu_device *adev) { u32 link_width_cntl; @@ -1323,6 +1329,52 @@ static void si_set_pcie_lanes(struct amdgpu_device *adev, int lanes) WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl); } +static void si_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0, + uint64_t *count1) +{ + uint32_t perfctr = 0; + uint64_t cnt0_of, cnt1_of; + int tmp; + + /* This reports 0 on APUs, so return to avoid writing/reading registers + * that may or may not be different from their GPU counterparts + */ + if (adev->flags & AMD_IS_APU) + return; + + /* Set the 2 events that we wish to watch, defined above */ + /* Reg 40 is # received msgs, Reg 104 is # of posted requests sent */ + perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40); + perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104); + + /* Write to enable desired perf counters */ + WREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK, perfctr); + /* Zero out and enable the perf counters + * Write 0x5: + * Bit 0 = Start all counters(1) + * Bit 2 = Global counter reset enable(1) + */ + WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000005); + + msleep(1000); + + /* Load the shadow and disable the perf counters + * Write 0x2: + * Bit 0 = Stop counters(0) + * Bit 1 = Load the shadow counters(1) + */ + WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000002); + + /* Read register values to get any >32bit overflow */ + tmp = RREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK); + cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER); + cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER); + + /* Get the values and add the overflow */ + *count0 = RREG32_PCIE(ixPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32); + *count1 = RREG32_PCIE(ixPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32); +} + static const struct amdgpu_asic_funcs si_asic_funcs = { .read_disabled_bios = &si_read_disabled_bios, @@ -1339,6 +1391,8 @@ static const struct amdgpu_asic_funcs si_asic_funcs = .flush_hdp = &si_flush_hdp, .invalidate_hdp = &si_invalidate_hdp, .need_full_reset = &si_need_full_reset, + .get_pcie_usage = &si_get_pcie_usage, + .need_reset_on_init = &si_need_reset_on_init, }; static uint32_t si_get_rev_id(struct amdgpu_device *adev) @@ -1382,7 +1436,7 @@ static int si_common_early_init(void *handle) AMD_CG_SUPPORT_UVD_MGCG | AMD_CG_SUPPORT_HDP_LS | AMD_CG_SUPPORT_HDP_MGCG; - adev->pg_flags = 0; + adev->pg_flags = 0; adev->external_rev_id = (adev->rev_id == 0) ? 1 : (adev->rev_id == 1) ? 5 : 6; break; diff --git a/drivers/gpu/drm/amd/amdgpu/si_dma.c b/drivers/gpu/drm/amd/amdgpu/si_dma.c index adbaea6da0d7..f15f196684ba 100644 --- a/drivers/gpu/drm/amd/amdgpu/si_dma.c +++ b/drivers/gpu/drm/amd/amdgpu/si_dma.c @@ -61,9 +61,11 @@ static void si_dma_ring_set_wptr(struct amdgpu_ring *ring) } static void si_dma_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); /* The indirect buffer packet must end on an 8 DW boundary in the DMA ring. * Pad as necessary with NOPs. */ @@ -122,7 +124,7 @@ static void si_dma_stop(struct amdgpu_device *adev) if (adev->mman.buffer_funcs_ring == ring) amdgpu_ttm_set_buffer_funcs_status(adev, false); - ring->ready = false; + ring->sched.ready = false; } } @@ -175,13 +177,11 @@ static int si_dma_start(struct amdgpu_device *adev) WREG32(DMA_RB_WPTR + sdma_offsets[i], lower_32_bits(ring->wptr) << 2); WREG32(DMA_RB_CNTL + sdma_offsets[i], rb_cntl | DMA_RB_ENABLE); - ring->ready = true; + ring->sched.ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) return r; - } if (adev->mman.buffer_funcs_ring == ring) amdgpu_ttm_set_buffer_funcs_status(adev, true); @@ -209,21 +209,16 @@ static int si_dma_ring_test_ring(struct amdgpu_ring *ring) u64 gpu_addr; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); r = amdgpu_ring_alloc(ring, 4); - if (r) { - DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); - amdgpu_device_wb_free(adev, index); - return r; - } + if (r) + goto error_free_wb; amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, 1)); amdgpu_ring_write(ring, lower_32_bits(gpu_addr)); @@ -238,15 +233,11 @@ static int si_dma_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } - amdgpu_device_wb_free(adev, index); + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; +error_free_wb: + amdgpu_device_wb_free(adev, index); return r; } @@ -269,20 +260,16 @@ static int si_dma_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); + if (r) return r; - } gpu_addr = adev->wb.gpu_addr + (index * 4); tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); memset(&ib, 0, sizeof(ib)); r = amdgpu_ib_get(adev, NULL, 256, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + if (r) goto err0; - } ib.ptr[0] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, 1); ib.ptr[1] = lower_32_bits(gpu_addr); @@ -295,21 +282,16 @@ static int si_dma_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = dma_fence_wait_timeout(f, false, timeout); if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out\n"); r = -ETIMEDOUT; goto err1; } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); goto err1; } tmp = le32_to_cpu(adev->wb.wb[index]); - if (tmp == 0xDEADBEEF) { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + if (tmp == 0xDEADBEEF) r = 0; - } else { - DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp); + else r = -EINVAL; - } err1: amdgpu_ib_free(adev, &ib, NULL); @@ -658,15 +640,6 @@ static int si_dma_process_trap_irq(struct amdgpu_device *adev, return 0; } -static int si_dma_process_illegal_inst_irq(struct amdgpu_device *adev, - struct amdgpu_irq_src *source, - struct amdgpu_iv_entry *entry) -{ - DRM_ERROR("Illegal instruction in SDMA command stream\n"); - schedule_work(&adev->reset_work); - return 0; -} - static int si_dma_set_clockgating_state(void *handle, enum amd_clockgating_state state) { @@ -781,15 +754,10 @@ static const struct amdgpu_irq_src_funcs si_dma_trap_irq_funcs = { .process = si_dma_process_trap_irq, }; -static const struct amdgpu_irq_src_funcs si_dma_illegal_inst_irq_funcs = { - .process = si_dma_process_illegal_inst_irq, -}; - static void si_dma_set_irq_funcs(struct amdgpu_device *adev) { adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; adev->sdma.trap_irq.funcs = &si_dma_trap_irq_funcs; - adev->sdma.illegal_inst_irq.funcs = &si_dma_illegal_inst_irq_funcs; } /** diff --git a/drivers/gpu/drm/amd/amdgpu/si_dpm.c b/drivers/gpu/drm/amd/amdgpu/si_dpm.c index da58040fdbdc..41e01a7f57a4 100644 --- a/drivers/gpu/drm/amd/amdgpu/si_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/si_dpm.c @@ -6216,10 +6216,12 @@ static void si_request_link_speed_change_before_state_change(struct amdgpu_devic si_pi->force_pcie_gen = AMDGPU_PCIE_GEN2; if (current_link_speed == AMDGPU_PCIE_GEN2) break; + /* fall through */ case AMDGPU_PCIE_GEN2: if (amdgpu_acpi_pcie_performance_request(adev, PCIE_PERF_REQ_PECI_GEN2, false) == 0) break; #endif + /* fall through */ default: si_pi->force_pcie_gen = si_get_current_pcie_speed(adev); break; diff --git a/drivers/gpu/drm/amd/amdgpu/si_ih.c b/drivers/gpu/drm/amd/amdgpu/si_ih.c index b3d7d9f83202..8c50c9cab455 100644 --- a/drivers/gpu/drm/amd/amdgpu/si_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/si_ih.c @@ -57,9 +57,9 @@ static void si_ih_disable_interrupts(struct amdgpu_device *adev) static int si_ih_irq_init(struct amdgpu_device *adev) { + struct amdgpu_ih_ring *ih = &adev->irq.ih; int rb_bufsz; u32 interrupt_cntl, ih_cntl, ih_rb_cntl; - u64 wptr_off; si_ih_disable_interrupts(adev); WREG32(INTERRUPT_CNTL2, adev->irq.ih.gpu_addr >> 8); @@ -76,9 +76,8 @@ static int si_ih_irq_init(struct amdgpu_device *adev) (rb_bufsz << 1) | IH_WPTR_WRITEBACK_ENABLE; - wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4); - WREG32(IH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off)); - WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF); + WREG32(IH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr)); + WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF); WREG32(IH_RB_CNTL, ih_rb_cntl); WREG32(IH_RB_RPTR, 0); WREG32(IH_RB_WPTR, 0); @@ -100,47 +99,36 @@ static void si_ih_irq_disable(struct amdgpu_device *adev) mdelay(1); } -static u32 si_ih_get_wptr(struct amdgpu_device *adev) +static u32 si_ih_get_wptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { u32 wptr, tmp; - wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]); + wptr = le32_to_cpu(*ih->wptr_cpu); if (wptr & IH_RB_WPTR__RB_OVERFLOW_MASK) { wptr &= ~IH_RB_WPTR__RB_OVERFLOW_MASK; dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", - wptr, adev->irq.ih.rptr, (wptr + 16) & adev->irq.ih.ptr_mask); - adev->irq.ih.rptr = (wptr + 16) & adev->irq.ih.ptr_mask; + wptr, ih->rptr, (wptr + 16) & ih->ptr_mask); + ih->rptr = (wptr + 16) & ih->ptr_mask; tmp = RREG32(IH_RB_CNTL); tmp |= IH_RB_CNTL__WPTR_OVERFLOW_CLEAR_MASK; WREG32(IH_RB_CNTL, tmp); } - return (wptr & adev->irq.ih.ptr_mask); -} - -/** - * si_ih_prescreen_iv - prescreen an interrupt vector - * - * @adev: amdgpu_device pointer - * - * Returns true if the interrupt vector should be further processed. - */ -static bool si_ih_prescreen_iv(struct amdgpu_device *adev) -{ - /* Process all interrupts */ - return true; + return (wptr & ih->ptr_mask); } static void si_ih_decode_iv(struct amdgpu_device *adev, - struct amdgpu_iv_entry *entry) + struct amdgpu_ih_ring *ih, + struct amdgpu_iv_entry *entry) { - u32 ring_index = adev->irq.ih.rptr >> 2; + u32 ring_index = ih->rptr >> 2; uint32_t dw[4]; - dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); - dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]); - dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]); - dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); + dw[0] = le32_to_cpu(ih->ring[ring_index + 0]); + dw[1] = le32_to_cpu(ih->ring[ring_index + 1]); + dw[2] = le32_to_cpu(ih->ring[ring_index + 2]); + dw[3] = le32_to_cpu(ih->ring[ring_index + 3]); entry->client_id = AMDGPU_IRQ_CLIENTID_LEGACY; entry->src_id = dw[0] & 0xff; @@ -148,12 +136,13 @@ static void si_ih_decode_iv(struct amdgpu_device *adev, entry->ring_id = dw[2] & 0xff; entry->vmid = (dw[2] >> 8) & 0xff; - adev->irq.ih.rptr += 16; + ih->rptr += 16; } -static void si_ih_set_rptr(struct amdgpu_device *adev) +static void si_ih_set_rptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { - WREG32(IH_RB_RPTR, adev->irq.ih.rptr); + WREG32(IH_RB_RPTR, ih->rptr); } static int si_ih_early_init(void *handle) @@ -301,7 +290,6 @@ static const struct amd_ip_funcs si_ih_ip_funcs = { static const struct amdgpu_ih_funcs si_ih_funcs = { .get_wptr = si_ih_get_wptr, - .prescreen_iv = si_ih_prescreen_iv, .decode_iv = si_ih_decode_iv, .set_rptr = si_ih_set_rptr }; diff --git a/drivers/gpu/drm/amd/amdgpu/soc15.c b/drivers/gpu/drm/amd/amdgpu/soc15.c index 4cc0dcb1a187..ed89a101f73f 100644 --- a/drivers/gpu/drm/amd/amdgpu/soc15.c +++ b/drivers/gpu/drm/amd/amdgpu/soc15.c @@ -43,6 +43,10 @@ #include "hdp/hdp_4_0_sh_mask.h" #include "smuio/smuio_9_0_offset.h" #include "smuio/smuio_9_0_sh_mask.h" +#include "nbio/nbio_7_0_default.h" +#include "nbio/nbio_7_0_sh_mask.h" +#include "nbio/nbio_7_0_smn.h" +#include "mp/mp_9_0_offset.h" #include "soc15.h" #include "soc15_common.h" @@ -385,14 +389,13 @@ void soc15_program_register_sequence(struct amdgpu_device *adev, } - -static int soc15_asic_reset(struct amdgpu_device *adev) +static int soc15_asic_mode1_reset(struct amdgpu_device *adev) { u32 i; amdgpu_atombios_scratch_regs_engine_hung(adev, true); - dev_info(adev->dev, "GPU reset\n"); + dev_info(adev->dev, "GPU mode1 reset\n"); /* disable BM */ pci_clear_master(adev->pdev); @@ -417,6 +420,62 @@ static int soc15_asic_reset(struct amdgpu_device *adev) return 0; } +static int soc15_asic_get_baco_capability(struct amdgpu_device *adev, bool *cap) +{ + void *pp_handle = adev->powerplay.pp_handle; + const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs; + + if (!pp_funcs || !pp_funcs->get_asic_baco_capability) { + *cap = false; + return -ENOENT; + } + + return pp_funcs->get_asic_baco_capability(pp_handle, cap); +} + +static int soc15_asic_baco_reset(struct amdgpu_device *adev) +{ + void *pp_handle = adev->powerplay.pp_handle; + const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs; + + if (!pp_funcs ||!pp_funcs->get_asic_baco_state ||!pp_funcs->set_asic_baco_state) + return -ENOENT; + + /* enter BACO state */ + if (pp_funcs->set_asic_baco_state(pp_handle, 1)) + return -EIO; + + /* exit BACO state */ + if (pp_funcs->set_asic_baco_state(pp_handle, 0)) + return -EIO; + + dev_info(adev->dev, "GPU BACO reset\n"); + + return 0; +} + +static int soc15_asic_reset(struct amdgpu_device *adev) +{ + int ret; + bool baco_reset; + + switch (adev->asic_type) { + case CHIP_VEGA10: + soc15_asic_get_baco_capability(adev, &baco_reset); + break; + default: + baco_reset = false; + break; + } + + if (baco_reset) + ret = soc15_asic_baco_reset(adev); + else + ret = soc15_asic_mode1_reset(adev); + + return ret; +} + /*static int soc15_set_uvd_clock(struct amdgpu_device *adev, u32 clock, u32 cntl_reg, u32 status_reg) { @@ -507,6 +566,9 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev) return -EINVAL; } + if (adev->asic_type == CHIP_VEGA20) + adev->gmc.xgmi.supported = true; + if (adev->flags & AMD_IS_APU) adev->nbio_funcs = &nbio_v7_0_funcs; else if (adev->asic_type == CHIP_VEGA20) @@ -532,10 +594,12 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev) amdgpu_device_ip_block_add(adev, &vega10_common_ip_block); amdgpu_device_ip_block_add(adev, &gmc_v9_0_ip_block); amdgpu_device_ip_block_add(adev, &vega10_ih_ip_block); - if (adev->asic_type == CHIP_VEGA20) - amdgpu_device_ip_block_add(adev, &psp_v11_0_ip_block); - else - amdgpu_device_ip_block_add(adev, &psp_v3_1_ip_block); + if (likely(adev->firmware.load_type == AMDGPU_FW_LOAD_PSP)) { + if (adev->asic_type == CHIP_VEGA20) + amdgpu_device_ip_block_add(adev, &psp_v11_0_ip_block); + else + amdgpu_device_ip_block_add(adev, &psp_v3_1_ip_block); + } amdgpu_device_ip_block_add(adev, &gfx_v9_0_ip_block); amdgpu_device_ip_block_add(adev, &sdma_v4_0_ip_block); if (!amdgpu_sriov_vf(adev)) @@ -557,7 +621,8 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev) amdgpu_device_ip_block_add(adev, &vega10_common_ip_block); amdgpu_device_ip_block_add(adev, &gmc_v9_0_ip_block); amdgpu_device_ip_block_add(adev, &vega10_ih_ip_block); - amdgpu_device_ip_block_add(adev, &psp_v10_0_ip_block); + if (likely(adev->firmware.load_type == AMDGPU_FW_LOAD_PSP)) + amdgpu_device_ip_block_add(adev, &psp_v10_0_ip_block); amdgpu_device_ip_block_add(adev, &gfx_v9_0_ip_block); amdgpu_device_ip_block_add(adev, &sdma_v4_0_ip_block); amdgpu_device_ip_block_add(adev, &pp_smu_ip_block); @@ -598,6 +663,68 @@ static bool soc15_need_full_reset(struct amdgpu_device *adev) /* change this when we implement soft reset */ return true; } +static void soc15_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0, + uint64_t *count1) +{ + uint32_t perfctr = 0; + uint64_t cnt0_of, cnt1_of; + int tmp; + + /* This reports 0 on APUs, so return to avoid writing/reading registers + * that may or may not be different from their GPU counterparts + */ + if (adev->flags & AMD_IS_APU) + return; + + /* Set the 2 events that we wish to watch, defined above */ + /* Reg 40 is # received msgs, Reg 104 is # of posted requests sent */ + perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40); + perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104); + + /* Write to enable desired perf counters */ + WREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK, perfctr); + /* Zero out and enable the perf counters + * Write 0x5: + * Bit 0 = Start all counters(1) + * Bit 2 = Global counter reset enable(1) + */ + WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000005); + + msleep(1000); + + /* Load the shadow and disable the perf counters + * Write 0x2: + * Bit 0 = Stop counters(0) + * Bit 1 = Load the shadow counters(1) + */ + WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000002); + + /* Read register values to get any >32bit overflow */ + tmp = RREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK); + cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER); + cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER); + + /* Get the values and add the overflow */ + *count0 = RREG32_PCIE(smnPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32); + *count1 = RREG32_PCIE(smnPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32); +} + +static bool soc15_need_reset_on_init(struct amdgpu_device *adev) +{ + u32 sol_reg; + + if (adev->flags & AMD_IS_APU) + return false; + + /* Check sOS sign of life register to confirm sys driver and sOS + * are already been loaded. + */ + sol_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_81); + if (sol_reg) + return true; + + return false; +} static const struct amdgpu_asic_funcs soc15_asic_funcs = { @@ -613,6 +740,28 @@ static const struct amdgpu_asic_funcs soc15_asic_funcs = .flush_hdp = &soc15_flush_hdp, .invalidate_hdp = &soc15_invalidate_hdp, .need_full_reset = &soc15_need_full_reset, + .init_doorbell_index = &vega10_doorbell_index_init, + .get_pcie_usage = &soc15_get_pcie_usage, + .need_reset_on_init = &soc15_need_reset_on_init, +}; + +static const struct amdgpu_asic_funcs vega20_asic_funcs = +{ + .read_disabled_bios = &soc15_read_disabled_bios, + .read_bios_from_rom = &soc15_read_bios_from_rom, + .read_register = &soc15_read_register, + .reset = &soc15_asic_reset, + .set_vga_state = &soc15_vga_set_state, + .get_xclk = &soc15_get_xclk, + .set_uvd_clocks = &soc15_set_uvd_clocks, + .set_vce_clocks = &soc15_set_vce_clocks, + .get_config_memsize = &soc15_get_config_memsize, + .flush_hdp = &soc15_flush_hdp, + .invalidate_hdp = &soc15_invalidate_hdp, + .need_full_reset = &soc15_need_full_reset, + .init_doorbell_index = &vega20_doorbell_index_init, + .get_pcie_usage = &soc15_get_pcie_usage, + .need_reset_on_init = &soc15_need_reset_on_init, }; static int soc15_common_early_init(void *handle) @@ -632,11 +781,11 @@ static int soc15_common_early_init(void *handle) adev->se_cac_rreg = &soc15_se_cac_rreg; adev->se_cac_wreg = &soc15_se_cac_wreg; - adev->asic_funcs = &soc15_asic_funcs; adev->external_rev_id = 0xFF; switch (adev->asic_type) { case CHIP_VEGA10: + adev->asic_funcs = &soc15_asic_funcs; adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS | AMD_CG_SUPPORT_GFX_RLC_LS | @@ -660,6 +809,7 @@ static int soc15_common_early_init(void *handle) adev->external_rev_id = 0x1; break; case CHIP_VEGA12: + adev->asic_funcs = &soc15_asic_funcs; adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS | AMD_CG_SUPPORT_GFX_CGCG | @@ -682,6 +832,7 @@ static int soc15_common_early_init(void *handle) adev->external_rev_id = adev->rev_id + 0x14; break; case CHIP_VEGA20: + adev->asic_funcs = &vega20_asic_funcs; adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS | AMD_CG_SUPPORT_GFX_CGCG | @@ -704,12 +855,15 @@ static int soc15_common_early_init(void *handle) adev->external_rev_id = adev->rev_id + 0x28; break; case CHIP_RAVEN: + adev->asic_funcs = &soc15_asic_funcs; if (adev->rev_id >= 0x8) - adev->external_rev_id = adev->rev_id + 0x81; + adev->external_rev_id = adev->rev_id + 0x79; else if (adev->pdev->device == 0x15d8) adev->external_rev_id = adev->rev_id + 0x41; + else if (adev->rev_id == 1) + adev->external_rev_id = adev->rev_id + 0x20; else - adev->external_rev_id = 0x1; + adev->external_rev_id = adev->rev_id + 0x01; if (adev->rev_id >= 0x8) { adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | @@ -816,6 +970,22 @@ static int soc15_common_sw_fini(void *handle) return 0; } +static void soc15_doorbell_range_init(struct amdgpu_device *adev) +{ + int i; + struct amdgpu_ring *ring; + + for (i = 0; i < adev->sdma.num_instances; i++) { + ring = &adev->sdma.instance[i].ring; + adev->nbio_funcs->sdma_doorbell_range(adev, i, + ring->use_doorbell, ring->doorbell_index, + adev->doorbell_index.sdma_doorbell_range); + } + + adev->nbio_funcs->ih_doorbell_range(adev, adev->irq.ih.use_doorbell, + adev->irq.ih.doorbell_index); +} + static int soc15_common_hw_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; @@ -828,6 +998,12 @@ static int soc15_common_hw_init(void *handle) adev->nbio_funcs->init_registers(adev); /* enable the doorbell aperture */ soc15_enable_doorbell_aperture(adev, true); + /* HW doorbell routing policy: doorbell writing not + * in SDMA/IH/MM/ACV range will be routed to CP. So + * we need to init SDMA/IH/MM/ACV doorbell range prior + * to CP ip block init and ring test. + */ + soc15_doorbell_range_init(adev); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/soc15.h b/drivers/gpu/drm/amd/amdgpu/soc15.h index f8ad7804dc40..a66c8bfbbaa6 100644 --- a/drivers/gpu/drm/amd/amdgpu/soc15.h +++ b/drivers/gpu/drm/amd/amdgpu/soc15.h @@ -58,4 +58,6 @@ void soc15_program_register_sequence(struct amdgpu_device *adev, int vega10_reg_base_init(struct amdgpu_device *adev); int vega20_reg_base_init(struct amdgpu_device *adev); +void vega10_doorbell_index_init(struct amdgpu_device *adev); +void vega20_doorbell_index_init(struct amdgpu_device *adev); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/soc15_common.h b/drivers/gpu/drm/amd/amdgpu/soc15_common.h index 958b10a57073..49c262540940 100644 --- a/drivers/gpu/drm/amd/amdgpu/soc15_common.h +++ b/drivers/gpu/drm/amd/amdgpu/soc15_common.h @@ -49,14 +49,19 @@ #define SOC15_WAIT_ON_RREG(ip, inst, reg, expected_value, mask, ret) \ do { \ + uint32_t old_ = 0; \ uint32_t tmp_ = RREG32(adev->reg_offset[ip##_HWIP][inst][reg##_BASE_IDX] + reg); \ uint32_t loop = adev->usec_timeout; \ while ((tmp_ & (mask)) != (expected_value)) { \ - udelay(2); \ + if (old_ != tmp_) { \ + loop = adev->usec_timeout; \ + old_ = tmp_; \ + } else \ + udelay(1); \ tmp_ = RREG32(adev->reg_offset[ip##_HWIP][inst][reg##_BASE_IDX] + reg); \ loop--; \ if (!loop) { \ - DRM_ERROR("Register(%d) [%s] failed to reach value 0x%08x != 0x%08x\n", \ + DRM_WARN("Register(%d) [%s] failed to reach value 0x%08x != 0x%08x\n", \ inst, #reg, (unsigned)expected_value, (unsigned)(tmp_ & (mask))); \ ret = -ETIMEDOUT; \ break; \ diff --git a/drivers/gpu/drm/amd/amdgpu/ta_xgmi_if.h b/drivers/gpu/drm/amd/amdgpu/ta_xgmi_if.h new file mode 100644 index 000000000000..ac2c27b7630c --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/ta_xgmi_if.h @@ -0,0 +1,130 @@ +/* + * Copyright 2018 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef _TA_XGMI_IF_H +#define _TA_XGMI_IF_H + +/* Responses have bit 31 set */ +#define RSP_ID_MASK (1U << 31) +#define RSP_ID(cmdId) (((uint32_t)(cmdId)) | RSP_ID_MASK) + +enum ta_command_xgmi { + TA_COMMAND_XGMI__INITIALIZE = 0x00, + TA_COMMAND_XGMI__GET_NODE_ID = 0x01, + TA_COMMAND_XGMI__GET_HIVE_ID = 0x02, + TA_COMMAND_XGMI__GET_GET_TOPOLOGY_INFO = 0x03, + TA_COMMAND_XGMI__SET_TOPOLOGY_INFO = 0x04 +}; + +/* XGMI related enumerations */ +/**********************************************************/; +enum ta_xgmi_connected_nodes { + TA_XGMI__MAX_CONNECTED_NODES = 64 +}; + +enum ta_xgmi_status { + TA_XGMI_STATUS__SUCCESS = 0x00, + TA_XGMI_STATUS__GENERIC_FAILURE = 0x01, + TA_XGMI_STATUS__NULL_POINTER = 0x02, + TA_XGMI_STATUS__INVALID_PARAMETER = 0x03, + TA_XGMI_STATUS__NOT_INITIALIZED = 0x04, + TA_XGMI_STATUS__INVALID_NODE_NUM = 0x05, + TA_XGMI_STATUS__INVALID_NODE_ID = 0x06, + TA_XGMI_STATUS__INVALID_TOPOLOGY = 0x07, + TA_XGMI_STATUS__FAILED_ID_GEN = 0x08, + TA_XGMI_STATUS__FAILED_TOPOLOGY_INIT = 0x09, + TA_XGMI_STATUS__SET_SHARING_ERROR = 0x0A +}; + +enum ta_xgmi_assigned_sdma_engine { + TA_XGMI_ASSIGNED_SDMA_ENGINE__NOT_ASSIGNED = -1, + TA_XGMI_ASSIGNED_SDMA_ENGINE__SDMA0 = 0, + TA_XGMI_ASSIGNED_SDMA_ENGINE__SDMA1 = 1, + TA_XGMI_ASSIGNED_SDMA_ENGINE__SDMA2 = 2, + TA_XGMI_ASSIGNED_SDMA_ENGINE__SDMA3 = 3, + TA_XGMI_ASSIGNED_SDMA_ENGINE__SDMA4 = 4, + TA_XGMI_ASSIGNED_SDMA_ENGINE__SDMA5 = 5 +}; + +/* input/output structures for XGMI commands */ +/**********************************************************/ +struct ta_xgmi_node_info { + uint64_t node_id; + uint8_t num_hops; + uint8_t is_sharing_enabled; + enum ta_xgmi_assigned_sdma_engine sdma_engine; +}; + +struct ta_xgmi_cmd_initialize_output { + uint32_t status; +}; + +struct ta_xgmi_cmd_get_node_id_output { + uint64_t node_id; +}; + +struct ta_xgmi_cmd_get_hive_id_output { + uint64_t hive_id; +}; + +struct ta_xgmi_cmd_get_topology_info_input { + uint32_t num_nodes; + struct ta_xgmi_node_info nodes[TA_XGMI__MAX_CONNECTED_NODES]; +}; + +struct ta_xgmi_cmd_get_topology_info_output { + uint32_t num_nodes; + struct ta_xgmi_node_info nodes[TA_XGMI__MAX_CONNECTED_NODES]; +}; + +struct ta_xgmi_cmd_set_topology_info_input { + uint32_t num_nodes; + struct ta_xgmi_node_info nodes[TA_XGMI__MAX_CONNECTED_NODES]; +}; + +/**********************************************************/ +/* Common input structure for XGMI callbacks */ +union ta_xgmi_cmd_input { + struct ta_xgmi_cmd_get_topology_info_input get_topology_info; + struct ta_xgmi_cmd_set_topology_info_input set_topology_info; +}; + +/* Common output structure for XGMI callbacks */ +union ta_xgmi_cmd_output { + struct ta_xgmi_cmd_initialize_output initialize; + struct ta_xgmi_cmd_get_node_id_output get_node_id; + struct ta_xgmi_cmd_get_hive_id_output get_hive_id; + struct ta_xgmi_cmd_get_topology_info_output get_topology_info; +}; +/**********************************************************/ + +struct ta_xgmi_shared_memory { + uint32_t cmd_id; + uint32_t resp_id; + enum ta_xgmi_status xgmi_status; + uint32_t reserved; + union ta_xgmi_cmd_input xgmi_in_message; + union ta_xgmi_cmd_output xgmi_out_message; +}; + +#endif //_TA_XGMI_IF_H diff --git a/drivers/gpu/drm/amd/amdgpu/tonga_ih.c b/drivers/gpu/drm/amd/amdgpu/tonga_ih.c index 3abffd06b5c7..a20b711a6756 100644 --- a/drivers/gpu/drm/amd/amdgpu/tonga_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/tonga_ih.c @@ -99,9 +99,9 @@ static void tonga_ih_disable_interrupts(struct amdgpu_device *adev) */ static int tonga_ih_irq_init(struct amdgpu_device *adev) { - int rb_bufsz; u32 interrupt_cntl, ih_rb_cntl, ih_doorbell_rtpr; - u64 wptr_off; + struct amdgpu_ih_ring *ih = &adev->irq.ih; + int rb_bufsz; /* disable irqs */ tonga_ih_disable_interrupts(adev); @@ -118,10 +118,7 @@ static int tonga_ih_irq_init(struct amdgpu_device *adev) WREG32(mmINTERRUPT_CNTL, interrupt_cntl); /* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/ - if (adev->irq.ih.use_bus_addr) - WREG32(mmIH_RB_BASE, adev->irq.ih.rb_dma_addr >> 8); - else - WREG32(mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8); + WREG32(mmIH_RB_BASE, ih->gpu_addr >> 8); rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4); ih_rb_cntl = REG_SET_FIELD(0, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); @@ -136,12 +133,8 @@ static int tonga_ih_irq_init(struct amdgpu_device *adev) WREG32(mmIH_RB_CNTL, ih_rb_cntl); /* set the writeback address whether it's enabled or not */ - if (adev->irq.ih.use_bus_addr) - wptr_off = adev->irq.ih.rb_dma_addr + (adev->irq.ih.wptr_offs * 4); - else - wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4); - WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off)); - WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF); + WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr)); + WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF); /* set rptr, wptr to 0 */ WREG32(mmIH_RB_RPTR, 0); @@ -193,14 +186,12 @@ static void tonga_ih_irq_disable(struct amdgpu_device *adev) * Used by cz_irq_process(VI). * Returns the value of the wptr. */ -static u32 tonga_ih_get_wptr(struct amdgpu_device *adev) +static u32 tonga_ih_get_wptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { u32 wptr, tmp; - if (adev->irq.ih.use_bus_addr) - wptr = le32_to_cpu(adev->irq.ih.ring[adev->irq.ih.wptr_offs]); - else - wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]); + wptr = le32_to_cpu(*ih->wptr_cpu); if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) { wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0); @@ -209,41 +200,13 @@ static u32 tonga_ih_get_wptr(struct amdgpu_device *adev) * this should allow us to catchup. */ dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", - wptr, adev->irq.ih.rptr, (wptr + 16) & adev->irq.ih.ptr_mask); - adev->irq.ih.rptr = (wptr + 16) & adev->irq.ih.ptr_mask; + wptr, ih->rptr, (wptr + 16) & ih->ptr_mask); + ih->rptr = (wptr + 16) & ih->ptr_mask; tmp = RREG32(mmIH_RB_CNTL); tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); WREG32(mmIH_RB_CNTL, tmp); } - return (wptr & adev->irq.ih.ptr_mask); -} - -/** - * tonga_ih_prescreen_iv - prescreen an interrupt vector - * - * @adev: amdgpu_device pointer - * - * Returns true if the interrupt vector should be further processed. - */ -static bool tonga_ih_prescreen_iv(struct amdgpu_device *adev) -{ - u32 ring_index = adev->irq.ih.rptr >> 2; - u16 pasid; - - switch (le32_to_cpu(adev->irq.ih.ring[ring_index]) & 0xff) { - case 146: - case 147: - pasid = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]) >> 16; - if (!pasid || amdgpu_vm_pasid_fault_credit(adev, pasid)) - return true; - break; - default: - /* Not a VM fault */ - return true; - } - - adev->irq.ih.rptr += 16; - return false; + return (wptr & ih->ptr_mask); } /** @@ -255,16 +218,17 @@ static bool tonga_ih_prescreen_iv(struct amdgpu_device *adev) * position and also advance the position. */ static void tonga_ih_decode_iv(struct amdgpu_device *adev, - struct amdgpu_iv_entry *entry) + struct amdgpu_ih_ring *ih, + struct amdgpu_iv_entry *entry) { /* wptr/rptr are in bytes! */ - u32 ring_index = adev->irq.ih.rptr >> 2; + u32 ring_index = ih->rptr >> 2; uint32_t dw[4]; - dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); - dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]); - dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]); - dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); + dw[0] = le32_to_cpu(ih->ring[ring_index + 0]); + dw[1] = le32_to_cpu(ih->ring[ring_index + 1]); + dw[2] = le32_to_cpu(ih->ring[ring_index + 2]); + dw[3] = le32_to_cpu(ih->ring[ring_index + 3]); entry->client_id = AMDGPU_IRQ_CLIENTID_LEGACY; entry->src_id = dw[0] & 0xff; @@ -274,7 +238,7 @@ static void tonga_ih_decode_iv(struct amdgpu_device *adev, entry->pasid = (dw[2] >> 16) & 0xffff; /* wptr/rptr are in bytes! */ - adev->irq.ih.rptr += 16; + ih->rptr += 16; } /** @@ -284,17 +248,15 @@ static void tonga_ih_decode_iv(struct amdgpu_device *adev, * * Set the IH ring buffer rptr. */ -static void tonga_ih_set_rptr(struct amdgpu_device *adev) +static void tonga_ih_set_rptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { - if (adev->irq.ih.use_doorbell) { + if (ih->use_doorbell) { /* XXX check if swapping is necessary on BE */ - if (adev->irq.ih.use_bus_addr) - adev->irq.ih.ring[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr; - else - adev->wb.wb[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr; - WDOORBELL32(adev->irq.ih.doorbell_index, adev->irq.ih.rptr); + *ih->rptr_cpu = ih->rptr; + WDOORBELL32(ih->doorbell_index, ih->rptr); } else { - WREG32(mmIH_RB_RPTR, adev->irq.ih.rptr); + WREG32(mmIH_RB_RPTR, ih->rptr); } } @@ -322,7 +284,7 @@ static int tonga_ih_sw_init(void *handle) return r; adev->irq.ih.use_doorbell = true; - adev->irq.ih.doorbell_index = AMDGPU_DOORBELL_IH; + adev->irq.ih.doorbell_index = adev->doorbell_index.ih; r = amdgpu_irq_init(adev); @@ -506,7 +468,6 @@ static const struct amd_ip_funcs tonga_ih_ip_funcs = { static const struct amdgpu_ih_funcs tonga_ih_funcs = { .get_wptr = tonga_ih_get_wptr, - .prescreen_iv = tonga_ih_prescreen_iv, .decode_iv = tonga_ih_decode_iv, .set_rptr = tonga_ih_set_rptr }; diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c index 1fc17bf39fed..c4fb58667fd4 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c @@ -116,16 +116,16 @@ static int uvd_v4_2_sw_init(void *handle) if (r) return r; - r = amdgpu_uvd_resume(adev); - if (r) - return r; - ring = &adev->uvd.inst->ring; sprintf(ring->name, "uvd"); r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst->irq, 0); if (r) return r; + r = amdgpu_uvd_resume(adev); + if (r) + return r; + r = amdgpu_uvd_entity_init(adev); return r; @@ -162,12 +162,9 @@ static int uvd_v4_2_hw_init(void *handle) uvd_v4_2_enable_mgcg(adev, true); amdgpu_asic_set_uvd_clocks(adev, 10000, 10000); - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } r = amdgpu_ring_alloc(ring, 10); if (r) { @@ -218,7 +215,7 @@ static int uvd_v4_2_hw_fini(void *handle) if (RREG32(mmUVD_STATUS) != 0) uvd_v4_2_stop(adev); - ring->ready = false; + ring->sched.ready = false; return 0; } @@ -484,11 +481,9 @@ static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring) WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", - ring->idx, r); + if (r) return r; - } + amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); amdgpu_ring_write(ring, 0xDEADBEEF); amdgpu_ring_commit(ring); @@ -499,14 +494,9 @@ static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + return r; } @@ -519,8 +509,9 @@ static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0)); amdgpu_ring_write(ring, ib->gpu_addr); diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c index fde6ad5ac9ab..52bd8a654734 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c @@ -113,16 +113,16 @@ static int uvd_v5_0_sw_init(void *handle) if (r) return r; - r = amdgpu_uvd_resume(adev); - if (r) - return r; - ring = &adev->uvd.inst->ring; sprintf(ring->name, "uvd"); r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst->irq, 0); if (r) return r; + r = amdgpu_uvd_resume(adev); + if (r) + return r; + r = amdgpu_uvd_entity_init(adev); return r; @@ -158,12 +158,9 @@ static int uvd_v5_0_hw_init(void *handle) uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_UNGATE); uvd_v5_0_enable_mgcg(adev, true); - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } r = amdgpu_ring_alloc(ring, 10); if (r) { @@ -215,7 +212,7 @@ static int uvd_v5_0_hw_fini(void *handle) if (RREG32(mmUVD_STATUS) != 0) uvd_v5_0_stop(adev); - ring->ready = false; + ring->sched.ready = false; return 0; } @@ -500,11 +497,8 @@ static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", - ring->idx, r); + if (r) return r; - } amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); amdgpu_ring_write(ring, 0xDEADBEEF); amdgpu_ring_commit(ring); @@ -515,14 +509,9 @@ static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + return r; } @@ -535,8 +524,9 @@ static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0)); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c index 7a5b40275e8e..c9edddf9f88a 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c @@ -175,11 +175,8 @@ static int uvd_v6_0_enc_ring_test_ring(struct amdgpu_ring *ring) int r; r = amdgpu_ring_alloc(ring, 16); - if (r) { - DRM_ERROR("amdgpu: uvd enc failed to lock ring %d (%d).\n", - ring->idx, r); + if (r) return r; - } amdgpu_ring_write(ring, HEVC_ENC_CMD_END); amdgpu_ring_commit(ring); @@ -189,14 +186,8 @@ static int uvd_v6_0_enc_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed\n", - ring->idx); + if (i >= adev->usec_timeout) r = -ETIMEDOUT; - } return r; } @@ -336,31 +327,24 @@ static int uvd_v6_0_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = uvd_v6_0_enc_get_create_msg(ring, 1, NULL); - if (r) { - DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r); + if (r) goto error; - } r = uvd_v6_0_enc_get_destroy_msg(ring, 1, &fence); - if (r) { - DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r); + if (r) goto error; - } r = dma_fence_wait_timeout(fence, false, timeout); - if (r == 0) { - DRM_ERROR("amdgpu: IB test timed out.\n"); + if (r == 0) r = -ETIMEDOUT; - } else if (r < 0) { - DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); - } else { - DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); + else if (r > 0) r = 0; - } + error: dma_fence_put(fence); return r; } + static int uvd_v6_0_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; @@ -416,16 +400,16 @@ static int uvd_v6_0_sw_init(void *handle) DRM_INFO("UVD ENC is disabled\n"); } - r = amdgpu_uvd_resume(adev); - if (r) - return r; - ring = &adev->uvd.inst->ring; sprintf(ring->name, "uvd"); r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst->irq, 0); if (r) return r; + r = amdgpu_uvd_resume(adev); + if (r) + return r; + if (uvd_v6_0_enc_support(adev)) { for (i = 0; i < adev->uvd.num_enc_rings; ++i) { ring = &adev->uvd.inst->ring_enc[i]; @@ -476,12 +460,9 @@ static int uvd_v6_0_hw_init(void *handle) uvd_v6_0_set_clockgating_state(adev, AMD_CG_STATE_UNGATE); uvd_v6_0_enable_mgcg(adev, true); - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } r = amdgpu_ring_alloc(ring, 10); if (r) { @@ -513,12 +494,9 @@ static int uvd_v6_0_hw_init(void *handle) if (uvd_v6_0_enc_support(adev)) { for (i = 0; i < adev->uvd.num_enc_rings; ++i) { ring = &adev->uvd.inst->ring_enc[i]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } } } @@ -548,7 +526,7 @@ static int uvd_v6_0_hw_fini(void *handle) if (RREG32(mmUVD_STATUS) != 0) uvd_v6_0_stop(adev); - ring->ready = false; + ring->sched.ready = false; return 0; } @@ -969,11 +947,9 @@ static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring) WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", - ring->idx, r); + if (r) return r; - } + amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); amdgpu_ring_write(ring, 0xDEADBEEF); amdgpu_ring_commit(ring); @@ -984,14 +960,9 @@ static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("ring test on %d succeeded in %d usecs\n", - ring->idx, i); - } else { - DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", - ring->idx, tmp); - r = -EINVAL; - } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + return r; } @@ -1004,9 +975,12 @@ static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_VMID, 0)); amdgpu_ring_write(ring, vmid); @@ -1027,8 +1001,12 @@ static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring, * Write enc ring commands to execute the indirect buffer */ static void uvd_v6_0_enc_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, unsigned int vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + amdgpu_ring_write(ring, HEVC_ENC_CMD_IB_VM); amdgpu_ring_write(ring, vmid); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c index 58b39afcfb86..dc461df48da0 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c @@ -183,11 +183,8 @@ static int uvd_v7_0_enc_ring_test_ring(struct amdgpu_ring *ring) return 0; r = amdgpu_ring_alloc(ring, 16); - if (r) { - DRM_ERROR("amdgpu: uvd enc failed to lock (%d)ring %d (%d).\n", - ring->me, ring->idx, r); + if (r) return r; - } amdgpu_ring_write(ring, HEVC_ENC_CMD_END); amdgpu_ring_commit(ring); @@ -197,14 +194,8 @@ static int uvd_v7_0_enc_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("(%d)ring test on %d succeeded in %d usecs\n", - ring->me, ring->idx, i); - } else { - DRM_ERROR("amdgpu: (%d)ring %d test failed\n", - ring->me, ring->idx); + if (i >= adev->usec_timeout) r = -ETIMEDOUT; - } return r; } @@ -343,27 +334,19 @@ static int uvd_v7_0_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; r = uvd_v7_0_enc_get_create_msg(ring, 1, NULL); - if (r) { - DRM_ERROR("amdgpu: (%d)failed to get create msg (%ld).\n", ring->me, r); + if (r) goto error; - } r = uvd_v7_0_enc_get_destroy_msg(ring, 1, &fence); - if (r) { - DRM_ERROR("amdgpu: (%d)failed to get destroy ib (%ld).\n", ring->me, r); + if (r) goto error; - } r = dma_fence_wait_timeout(fence, false, timeout); - if (r == 0) { - DRM_ERROR("amdgpu: (%d)IB test timed out.\n", ring->me); + if (r == 0) r = -ETIMEDOUT; - } else if (r < 0) { - DRM_ERROR("amdgpu: (%d)fence wait failed (%ld).\n", ring->me, r); - } else { - DRM_DEBUG("ib test on (%d)ring %d succeeded\n", ring->me, ring->idx); + else if (r > 0) r = 0; - } + error: dma_fence_put(fence); return r; @@ -447,16 +430,12 @@ static int uvd_v7_0_sw_init(void *handle) DRM_INFO("PSP loading UVD firmware\n"); } - r = amdgpu_uvd_resume(adev); - if (r) - return r; - for (j = 0; j < adev->uvd.num_uvd_inst; j++) { if (adev->uvd.harvest_config & (1 << j)) continue; if (!amdgpu_sriov_vf(adev)) { ring = &adev->uvd.inst[j].ring; - sprintf(ring->name, "uvd<%d>", j); + sprintf(ring->name, "uvd_%d", ring->me); r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst[j].irq, 0); if (r) return r; @@ -464,7 +443,7 @@ static int uvd_v7_0_sw_init(void *handle) for (i = 0; i < adev->uvd.num_enc_rings; ++i) { ring = &adev->uvd.inst[j].ring_enc[i]; - sprintf(ring->name, "uvd_enc%d<%d>", i, j); + sprintf(ring->name, "uvd_enc_%d.%d", ring->me, i); if (amdgpu_sriov_vf(adev)) { ring->use_doorbell = true; @@ -472,9 +451,9 @@ static int uvd_v7_0_sw_init(void *handle) * sriov, so set unused location for other unused rings. */ if (i == 0) - ring->doorbell_index = AMDGPU_DOORBELL64_UVD_RING0_1 * 2; + ring->doorbell_index = adev->doorbell_index.uvd_vce.uvd_ring0_1 * 2; else - ring->doorbell_index = AMDGPU_DOORBELL64_UVD_RING2_3 * 2 + 1; + ring->doorbell_index = adev->doorbell_index.uvd_vce.uvd_ring2_3 * 2 + 1; } r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst[j].irq, 0); if (r) @@ -482,6 +461,10 @@ static int uvd_v7_0_sw_init(void *handle) } } + r = amdgpu_uvd_resume(adev); + if (r) + return r; + r = amdgpu_uvd_entity_init(adev); if (r) return r; @@ -540,12 +523,9 @@ static int uvd_v7_0_hw_init(void *handle) ring = &adev->uvd.inst[j].ring; if (!amdgpu_sriov_vf(adev)) { - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } r = amdgpu_ring_alloc(ring, 10); if (r) { @@ -582,12 +562,9 @@ static int uvd_v7_0_hw_init(void *handle) for (i = 0; i < adev->uvd.num_enc_rings; ++i) { ring = &adev->uvd.inst[j].ring_enc[i]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } } } done: @@ -619,7 +596,7 @@ static int uvd_v7_0_hw_fini(void *handle) for (i = 0; i < adev->uvd.num_uvd_inst; ++i) { if (adev->uvd.harvest_config & (1 << i)) continue; - adev->uvd.inst[i].ring.ready = false; + adev->uvd.inst[i].ring.sched.ready = false; } return 0; @@ -1235,11 +1212,9 @@ static int uvd_v7_0_ring_test_ring(struct amdgpu_ring *ring) WREG32_SOC15(UVD, ring->me, mmUVD_CONTEXT_ID, 0xCAFEDEAD); r = amdgpu_ring_alloc(ring, 3); - if (r) { - DRM_ERROR("amdgpu: (%d)cp failed to lock ring %d (%d).\n", - ring->me, ring->idx, r); + if (r) return r; - } + amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, ring->me, mmUVD_CONTEXT_ID), 0)); amdgpu_ring_write(ring, 0xDEADBEEF); @@ -1251,14 +1226,9 @@ static int uvd_v7_0_ring_test_ring(struct amdgpu_ring *ring) DRM_UDELAY(1); } - if (i < adev->usec_timeout) { - DRM_DEBUG("(%d)ring test on %d succeeded in %d usecs\n", - ring->me, ring->idx, i); - } else { - DRM_ERROR("(%d)amdgpu: ring %d test failed (0x%08X)\n", - ring->me, ring->idx, tmp); - r = -EINVAL; - } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + return r; } @@ -1300,10 +1270,12 @@ static int uvd_v7_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p, * Write ring commands to execute the indirect buffer */ static void uvd_v7_0_ring_emit_ib(struct amdgpu_ring *ring, + struct amdgpu_job *job, struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + uint32_t flags) { struct amdgpu_device *adev = ring->adev; + unsigned vmid = AMDGPU_JOB_GET_VMID(job); amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, ring->me, mmUVD_LMI_RBC_IB_VMID), 0)); @@ -1329,8 +1301,12 @@ static void uvd_v7_0_ring_emit_ib(struct amdgpu_ring *ring, * Write enc ring commands to execute the indirect buffer */ static void uvd_v7_0_enc_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, unsigned int vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + amdgpu_ring_write(ring, HEVC_ENC_CMD_IB_VM); amdgpu_ring_write(ring, vmid); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c index ea28828360d3..bed78a778e3f 100644 --- a/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c @@ -463,15 +463,11 @@ static int vce_v2_0_hw_init(void *handle) amdgpu_asic_set_vce_clocks(adev, 10000, 10000); vce_v2_0_enable_mgcg(adev, true, false); - for (i = 0; i < adev->vce.num_rings; i++) - adev->vce.ring[i].ready = false; for (i = 0; i < adev->vce.num_rings; i++) { - r = amdgpu_ring_test_ring(&adev->vce.ring[i]); + r = amdgpu_ring_test_helper(&adev->vce.ring[i]); if (r) return r; - else - adev->vce.ring[i].ready = true; } DRM_INFO("VCE initialized successfully.\n"); diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c index 6dbd39730070..6ec65cf11112 100644 --- a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c @@ -37,7 +37,6 @@ #include "gca/gfx_8_0_d.h" #include "smu/smu_7_1_2_d.h" #include "smu/smu_7_1_2_sh_mask.h" -#include "gca/gfx_8_0_d.h" #include "gca/gfx_8_0_sh_mask.h" #include "ivsrcid/ivsrcid_vislands30.h" @@ -474,15 +473,10 @@ static int vce_v3_0_hw_init(void *handle) amdgpu_asic_set_vce_clocks(adev, 10000, 10000); - for (i = 0; i < adev->vce.num_rings; i++) - adev->vce.ring[i].ready = false; - for (i = 0; i < adev->vce.num_rings; i++) { - r = amdgpu_ring_test_ring(&adev->vce.ring[i]); + r = amdgpu_ring_test_helper(&adev->vce.ring[i]); if (r) return r; - else - adev->vce.ring[i].ready = true; } DRM_INFO("VCE initialized successfully.\n"); @@ -838,8 +832,12 @@ out: } static void vce_v3_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, unsigned int vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + amdgpu_ring_write(ring, VCE_CMD_IB_VM); amdgpu_ring_write(ring, vmid); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v4_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v4_0.c index 1c9471890bf7..aadc3e66ebd7 100644 --- a/drivers/gpu/drm/amd/amdgpu/vce_v4_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vce_v4_0.c @@ -466,9 +466,9 @@ static int vce_v4_0_sw_init(void *handle) * so set unused location for other unused rings. */ if (i == 0) - ring->doorbell_index = AMDGPU_DOORBELL64_VCE_RING0_1 * 2; + ring->doorbell_index = adev->doorbell_index.uvd_vce.vce_ring0_1 * 2; else - ring->doorbell_index = AMDGPU_DOORBELL64_VCE_RING2_3 * 2 + 1; + ring->doorbell_index = adev->doorbell_index.uvd_vce.vce_ring2_3 * 2 + 1; } r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0); if (r) @@ -519,15 +519,10 @@ static int vce_v4_0_hw_init(void *handle) if (r) return r; - for (i = 0; i < adev->vce.num_rings; i++) - adev->vce.ring[i].ready = false; - for (i = 0; i < adev->vce.num_rings; i++) { - r = amdgpu_ring_test_ring(&adev->vce.ring[i]); + r = amdgpu_ring_test_helper(&adev->vce.ring[i]); if (r) return r; - else - adev->vce.ring[i].ready = true; } DRM_INFO("VCE initialized successfully.\n"); @@ -549,7 +544,7 @@ static int vce_v4_0_hw_fini(void *handle) } for (i = 0; i < adev->vce.num_rings; i++) - adev->vce.ring[i].ready = false; + adev->vce.ring[i].sched.ready = false; return 0; } @@ -951,9 +946,11 @@ static int vce_v4_0_set_powergating_state(void *handle, } #endif -static void vce_v4_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, unsigned int vmid, bool ctx_switch) +static void vce_v4_0_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_job *job, + struct amdgpu_ib *ib, uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + amdgpu_ring_write(ring, VCE_CMD_IB_VM); amdgpu_ring_write(ring, vmid); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c index 322e09b5b448..3dbc51f9d3b9 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c @@ -177,30 +177,22 @@ static int vcn_v1_0_hw_init(void *handle) struct amdgpu_ring *ring = &adev->vcn.ring_dec; int i, r; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } for (i = 0; i < adev->vcn.num_enc_rings; ++i) { ring = &adev->vcn.ring_enc[i]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + ring->sched.ready = true; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } } ring = &adev->vcn.ring_jpeg; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; + r = amdgpu_ring_test_helper(ring); + if (r) goto done; - } done: if (!r) @@ -222,10 +214,11 @@ static int vcn_v1_0_hw_fini(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; struct amdgpu_ring *ring = &adev->vcn.ring_dec; - if (RREG32_SOC15(VCN, 0, mmUVD_STATUS)) + if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) || + RREG32_SOC15(VCN, 0, mmUVD_STATUS)) vcn_v1_0_set_powergating_state(adev, AMD_PG_STATE_GATE); - ring->ready = false; + ring->sched.ready = false; return 0; } @@ -1095,7 +1088,8 @@ static int vcn_v1_0_start_dpg_mode(struct amdgpu_device *adev) WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); - /* initialize wptr */ + /* initialize JPEG wptr */ + ring = &adev->vcn.ring_jpeg; ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_WPTR); /* copy patch commands to the jpeg ring */ @@ -1167,21 +1161,29 @@ static int vcn_v1_0_stop_spg_mode(struct amdgpu_device *adev) static int vcn_v1_0_stop_dpg_mode(struct amdgpu_device *adev) { int ret_code = 0; + uint32_t tmp; /* Wait for power status to be UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF */ SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); - if (!ret_code) { - int tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF; - /* wait for read ptr to be equal to write ptr */ - SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code); + /* wait for read ptr to be equal to write ptr */ + tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF, ret_code); - SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, - UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, - UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); - } + tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF, ret_code); + + tmp = RREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_WPTR); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_JRBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code); + + tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF; + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code); + + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, + UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); /* disable dynamic power gating mode */ WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0, @@ -1367,10 +1369,12 @@ static void vcn_v1_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 * Write ring commands to execute the indirect buffer */ static void vcn_v1_0_dec_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { struct amdgpu_device *adev = ring->adev; + unsigned vmid = AMDGPU_JOB_GET_VMID(job); amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_VMID), 0)); @@ -1525,8 +1529,12 @@ static void vcn_v1_0_enc_ring_insert_end(struct amdgpu_ring *ring) * Write enc ring commands to execute the indirect buffer */ static void vcn_v1_0_enc_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, unsigned int vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + amdgpu_ring_write(ring, VCN_ENC_CMD_IB); amdgpu_ring_write(ring, vmid); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); @@ -1726,10 +1734,12 @@ static void vcn_v1_0_jpeg_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u6 * Write ring commands to execute the indirect buffer. */ static void vcn_v1_0_jpeg_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, - unsigned vmid, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { struct amdgpu_device *adev = ring->adev; + unsigned vmid = AMDGPU_JOB_GET_VMID(job); amdgpu_ring_write(ring, PACKETJ(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_JRBC_IB_VMID), 0, 0, PACKETJ_TYPE0)); diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c index a0fda6f9252a..6d1f804277f8 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c @@ -50,6 +50,22 @@ static void vega10_ih_enable_interrupts(struct amdgpu_device *adev) ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 1); WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL, ih_rb_cntl); adev->irq.ih.enabled = true; + + if (adev->irq.ih1.ring_size) { + ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL_RING1, + RB_ENABLE, 1); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1, ih_rb_cntl); + adev->irq.ih1.enabled = true; + } + + if (adev->irq.ih2.ring_size) { + ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING2); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL_RING2, + RB_ENABLE, 1); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING2, ih_rb_cntl); + adev->irq.ih2.enabled = true; + } } /** @@ -71,6 +87,53 @@ static void vega10_ih_disable_interrupts(struct amdgpu_device *adev) WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR, 0); adev->irq.ih.enabled = false; adev->irq.ih.rptr = 0; + + if (adev->irq.ih1.ring_size) { + ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL_RING1, + RB_ENABLE, 0); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1, ih_rb_cntl); + /* set rptr, wptr to 0 */ + WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR_RING1, 0); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_RING1, 0); + adev->irq.ih1.enabled = false; + adev->irq.ih1.rptr = 0; + } + + if (adev->irq.ih2.ring_size) { + ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING2); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL_RING2, + RB_ENABLE, 0); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING2, ih_rb_cntl); + /* set rptr, wptr to 0 */ + WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR_RING2, 0); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_RING2, 0); + adev->irq.ih2.enabled = false; + adev->irq.ih2.rptr = 0; + } +} + +static uint32_t vega10_ih_rb_cntl(struct amdgpu_ih_ring *ih, uint32_t ih_rb_cntl) +{ + int rb_bufsz = order_base_2(ih->ring_size / 4); + + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, + MC_SPACE, ih->use_bus_addr ? 1 : 4); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, + WPTR_OVERFLOW_CLEAR, 1); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, + WPTR_OVERFLOW_ENABLE, 1); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz); + /* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register + * value is written to memory + */ + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, + WPTR_WRITEBACK_ENABLE, 1); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SNOOP, 1); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_RO, 0); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_VMID, 0); + + return ih_rb_cntl; } /** @@ -86,50 +149,32 @@ static void vega10_ih_disable_interrupts(struct amdgpu_device *adev) */ static int vega10_ih_irq_init(struct amdgpu_device *adev) { + struct amdgpu_ih_ring *ih; int ret = 0; - int rb_bufsz; u32 ih_rb_cntl, ih_doorbell_rtpr; u32 tmp; - u64 wptr_off; /* disable irqs */ vega10_ih_disable_interrupts(adev); adev->nbio_funcs->ih_control(adev); - ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL); + ih = &adev->irq.ih; /* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/ - if (adev->irq.ih.use_bus_addr) { - WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE, adev->irq.ih.rb_dma_addr >> 8); - WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI, ((u64)adev->irq.ih.rb_dma_addr >> 40) & 0xff); - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 1); - } else { - WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8); - WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI, (adev->irq.ih.gpu_addr >> 40) & 0xff); - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 4); - } - rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4); - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 1); - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz); - /* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register value is written to memory */ - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1); - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SNOOP, 1); - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_RO, 0); - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_VMID, 0); - - if (adev->irq.msi_enabled) - ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RPTR_REARM, 1); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE, ih->gpu_addr >> 8); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI, (ih->gpu_addr >> 40) & 0xff); + ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL); + ih_rb_cntl = vega10_ih_rb_cntl(ih, ih_rb_cntl); + ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RPTR_REARM, + !!adev->irq.msi_enabled); WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL, ih_rb_cntl); /* set the writeback address whether it's enabled or not */ - if (adev->irq.ih.use_bus_addr) - wptr_off = adev->irq.ih.rb_dma_addr + (adev->irq.ih.wptr_offs * 4); - else - wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4); - WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off)); - WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFFFF); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO, + lower_32_bits(ih->wptr_addr)); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI, + upper_32_bits(ih->wptr_addr) & 0xFFFF); /* set rptr, wptr to 0 */ WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR, 0); @@ -137,17 +182,48 @@ static int vega10_ih_irq_init(struct amdgpu_device *adev) ih_doorbell_rtpr = RREG32_SOC15(OSSSYS, 0, mmIH_DOORBELL_RPTR); if (adev->irq.ih.use_doorbell) { - ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR, - OFFSET, adev->irq.ih.doorbell_index); - ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR, + ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, + IH_DOORBELL_RPTR, OFFSET, + adev->irq.ih.doorbell_index); + ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, + IH_DOORBELL_RPTR, ENABLE, 1); } else { - ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR, + ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, + IH_DOORBELL_RPTR, ENABLE, 0); } WREG32_SOC15(OSSSYS, 0, mmIH_DOORBELL_RPTR, ih_doorbell_rtpr); - adev->nbio_funcs->ih_doorbell_range(adev, adev->irq.ih.use_doorbell, - adev->irq.ih.doorbell_index); + + ih = &adev->irq.ih1; + if (ih->ring_size) { + WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_RING1, ih->gpu_addr >> 8); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI_RING1, + (ih->gpu_addr >> 40) & 0xff); + + ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1); + ih_rb_cntl = vega10_ih_rb_cntl(ih, ih_rb_cntl); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1, ih_rb_cntl); + + /* set rptr, wptr to 0 */ + WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR_RING1, 0); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_RING1, 0); + } + + ih = &adev->irq.ih2; + if (ih->ring_size) { + WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_RING2, ih->gpu_addr >> 8); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI_RING2, + (ih->gpu_addr >> 40) & 0xff); + + ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1); + ih_rb_cntl = vega10_ih_rb_cntl(ih, ih_rb_cntl); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING2, ih_rb_cntl); + + /* set rptr, wptr to 0 */ + WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR_RING2, 0); + WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_RING2, 0); + } tmp = RREG32_SOC15(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL); tmp = REG_SET_FIELD(tmp, IH_STORM_CLIENT_LIST_CNTL, @@ -191,116 +267,58 @@ static void vega10_ih_irq_disable(struct amdgpu_device *adev) * ring buffer overflow and deal with it. * Returns the value of the wptr. */ -static u32 vega10_ih_get_wptr(struct amdgpu_device *adev) +static u32 vega10_ih_get_wptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { - u32 wptr, tmp; + u32 wptr, reg, tmp; - if (adev->irq.ih.use_bus_addr) - wptr = le32_to_cpu(adev->irq.ih.ring[adev->irq.ih.wptr_offs]); - else - wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]); - - if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) { - wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0); - - /* When a ring buffer overflow happen start parsing interrupt - * from the last not overwritten vector (wptr + 32). Hopefully - * this should allow us to catchup. - */ - tmp = (wptr + 32) & adev->irq.ih.ptr_mask; - dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", - wptr, adev->irq.ih.rptr, tmp); - adev->irq.ih.rptr = tmp; - - tmp = RREG32_NO_KIQ(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL)); - tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); - WREG32_NO_KIQ(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), tmp); - } - return (wptr & adev->irq.ih.ptr_mask); -} + wptr = le32_to_cpu(*ih->wptr_cpu); -/** - * vega10_ih_prescreen_iv - prescreen an interrupt vector - * - * @adev: amdgpu_device pointer - * - * Returns true if the interrupt vector should be further processed. - */ -static bool vega10_ih_prescreen_iv(struct amdgpu_device *adev) -{ - u32 ring_index = adev->irq.ih.rptr >> 2; - u32 dw0, dw3, dw4, dw5; - u16 pasid; - u64 addr, key; - struct amdgpu_vm *vm; - int r; + if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) + goto out; - dw0 = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); - dw3 = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); - dw4 = le32_to_cpu(adev->irq.ih.ring[ring_index + 4]); - dw5 = le32_to_cpu(adev->irq.ih.ring[ring_index + 5]); + /* Double check that the overflow wasn't already cleared. */ - /* Filter retry page faults, let only the first one pass. If - * there are too many outstanding faults, ignore them until - * some faults get cleared. - */ - switch (dw0 & 0xff) { - case SOC15_IH_CLIENTID_VMC: - case SOC15_IH_CLIENTID_UTCL2: - break; - default: - /* Not a VM fault */ - return true; - } + if (ih == &adev->irq.ih) + reg = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR); + else if (ih == &adev->irq.ih1) + reg = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_RING1); + else if (ih == &adev->irq.ih2) + reg = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_RING2); + else + BUG(); - pasid = dw3 & 0xffff; - /* No PASID, can't identify faulting process */ - if (!pasid) - return true; + wptr = RREG32_NO_KIQ(reg); + if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) + goto out; - /* Not a retry fault, check fault credit */ - if (!(dw5 & 0x80)) { - if (!amdgpu_vm_pasid_fault_credit(adev, pasid)) - goto ignore_iv; - return true; - } + wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0); - /* Track retry faults in per-VM fault FIFO. */ - spin_lock(&adev->vm_manager.pasid_lock); - vm = idr_find(&adev->vm_manager.pasid_idr, pasid); - addr = ((u64)(dw5 & 0xf) << 44) | ((u64)dw4 << 12); - key = AMDGPU_VM_FAULT(pasid, addr); - if (!vm) { - /* VM not found, process it normally */ - spin_unlock(&adev->vm_manager.pasid_lock); - return true; - } else { - r = amdgpu_vm_add_fault(vm->fault_hash, key); - - /* Hash table is full or the fault is already being processed, - * ignore further page faults - */ - if (r != 0) { - spin_unlock(&adev->vm_manager.pasid_lock); - goto ignore_iv; - } - } - /* No locking required with single writer and single reader */ - r = kfifo_put(&vm->faults, key); - if (!r) { - /* FIFO is full. Ignore it until there is space */ - amdgpu_vm_clear_fault(vm->fault_hash, key); - spin_unlock(&adev->vm_manager.pasid_lock); - goto ignore_iv; - } + /* When a ring buffer overflow happen start parsing interrupt + * from the last not overwritten vector (wptr + 32). Hopefully + * this should allow us to catchup. + */ + tmp = (wptr + 32) & ih->ptr_mask; + dev_warn(adev->dev, "IH ring buffer overflow " + "(0x%08X, 0x%08X, 0x%08X)\n", + wptr, ih->rptr, tmp); + ih->rptr = tmp; + + if (ih == &adev->irq.ih) + reg = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL); + else if (ih == &adev->irq.ih1) + reg = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL_RING1); + else if (ih == &adev->irq.ih2) + reg = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL_RING2); + else + BUG(); - spin_unlock(&adev->vm_manager.pasid_lock); - /* It's the first fault for this address, process it normally */ - return true; + tmp = RREG32_NO_KIQ(reg); + tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); + WREG32_NO_KIQ(reg, tmp); -ignore_iv: - adev->irq.ih.rptr += 32; - return false; +out: + return (wptr & ih->ptr_mask); } /** @@ -312,20 +330,21 @@ ignore_iv: * position and also advance the position. */ static void vega10_ih_decode_iv(struct amdgpu_device *adev, - struct amdgpu_iv_entry *entry) + struct amdgpu_ih_ring *ih, + struct amdgpu_iv_entry *entry) { /* wptr/rptr are in bytes! */ - u32 ring_index = adev->irq.ih.rptr >> 2; + u32 ring_index = ih->rptr >> 2; uint32_t dw[8]; - dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); - dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]); - dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]); - dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); - dw[4] = le32_to_cpu(adev->irq.ih.ring[ring_index + 4]); - dw[5] = le32_to_cpu(adev->irq.ih.ring[ring_index + 5]); - dw[6] = le32_to_cpu(adev->irq.ih.ring[ring_index + 6]); - dw[7] = le32_to_cpu(adev->irq.ih.ring[ring_index + 7]); + dw[0] = le32_to_cpu(ih->ring[ring_index + 0]); + dw[1] = le32_to_cpu(ih->ring[ring_index + 1]); + dw[2] = le32_to_cpu(ih->ring[ring_index + 2]); + dw[3] = le32_to_cpu(ih->ring[ring_index + 3]); + dw[4] = le32_to_cpu(ih->ring[ring_index + 4]); + dw[5] = le32_to_cpu(ih->ring[ring_index + 5]); + dw[6] = le32_to_cpu(ih->ring[ring_index + 6]); + dw[7] = le32_to_cpu(ih->ring[ring_index + 7]); entry->client_id = dw[0] & 0xff; entry->src_id = (dw[0] >> 8) & 0xff; @@ -341,9 +360,8 @@ static void vega10_ih_decode_iv(struct amdgpu_device *adev, entry->src_data[2] = dw[6]; entry->src_data[3] = dw[7]; - /* wptr/rptr are in bytes! */ - adev->irq.ih.rptr += 32; + ih->rptr += 32; } /** @@ -353,18 +371,59 @@ static void vega10_ih_decode_iv(struct amdgpu_device *adev, * * Set the IH ring buffer rptr. */ -static void vega10_ih_set_rptr(struct amdgpu_device *adev) +static void vega10_ih_set_rptr(struct amdgpu_device *adev, + struct amdgpu_ih_ring *ih) { - if (adev->irq.ih.use_doorbell) { + if (ih->use_doorbell) { /* XXX check if swapping is necessary on BE */ - if (adev->irq.ih.use_bus_addr) - adev->irq.ih.ring[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr; - else - adev->wb.wb[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr; - WDOORBELL32(adev->irq.ih.doorbell_index, adev->irq.ih.rptr); - } else { - WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR, adev->irq.ih.rptr); + *ih->rptr_cpu = ih->rptr; + WDOORBELL32(ih->doorbell_index, ih->rptr); + } else if (ih == &adev->irq.ih) { + WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR, ih->rptr); + } else if (ih == &adev->irq.ih1) { + WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR_RING1, ih->rptr); + } else if (ih == &adev->irq.ih2) { + WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR_RING2, ih->rptr); + } +} + +/** + * vega10_ih_self_irq - dispatch work for ring 1 and 2 + * + * @adev: amdgpu_device pointer + * @source: irq source + * @entry: IV with WPTR update + * + * Update the WPTR from the IV and schedule work to handle the entries. + */ +static int vega10_ih_self_irq(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + struct amdgpu_iv_entry *entry) +{ + uint32_t wptr = cpu_to_le32(entry->src_data[0]); + + switch (entry->ring_id) { + case 1: + *adev->irq.ih1.wptr_cpu = wptr; + schedule_work(&adev->irq.ih1_work); + break; + case 2: + *adev->irq.ih2.wptr_cpu = wptr; + schedule_work(&adev->irq.ih2_work); + break; + default: break; } + return 0; +} + +static const struct amdgpu_irq_src_funcs vega10_ih_self_irq_funcs = { + .process = vega10_ih_self_irq, +}; + +static void vega10_ih_set_self_irq_funcs(struct amdgpu_device *adev) +{ + adev->irq.self_irq.num_types = 0; + adev->irq.self_irq.funcs = &vega10_ih_self_irq_funcs; } static int vega10_ih_early_init(void *handle) @@ -372,20 +431,37 @@ static int vega10_ih_early_init(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; vega10_ih_set_interrupt_funcs(adev); + vega10_ih_set_self_irq_funcs(adev); return 0; } static int vega10_ih_sw_init(void *handle) { - int r; struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int r; + + r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_IH, 0, + &adev->irq.self_irq); + if (r) + return r; r = amdgpu_ih_ring_init(adev, &adev->irq.ih, 256 * 1024, true); if (r) return r; + if (adev->asic_type == CHIP_VEGA10) { + r = amdgpu_ih_ring_init(adev, &adev->irq.ih1, PAGE_SIZE, true); + if (r) + return r; + + r = amdgpu_ih_ring_init(adev, &adev->irq.ih2, PAGE_SIZE, true); + if (r) + return r; + } + + /* TODO add doorbell for IH1 & IH2 as well */ adev->irq.ih.use_doorbell = true; - adev->irq.ih.doorbell_index = AMDGPU_DOORBELL64_IH << 1; + adev->irq.ih.doorbell_index = adev->doorbell_index.ih << 1; r = amdgpu_irq_init(adev); @@ -397,6 +473,8 @@ static int vega10_ih_sw_fini(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; amdgpu_irq_fini(adev); + amdgpu_ih_ring_fini(adev, &adev->irq.ih2); + amdgpu_ih_ring_fini(adev, &adev->irq.ih1); amdgpu_ih_ring_fini(adev, &adev->irq.ih); return 0; @@ -487,7 +565,6 @@ const struct amd_ip_funcs vega10_ih_ip_funcs = { static const struct amdgpu_ih_funcs vega10_ih_funcs = { .get_wptr = vega10_ih_get_wptr, - .prescreen_iv = vega10_ih_prescreen_iv, .decode_iv = vega10_ih_decode_iv, .set_rptr = vega10_ih_set_rptr }; diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c index c5c9b2bc190d..a8e92638a2e8 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c @@ -56,4 +56,37 @@ int vega10_reg_base_init(struct amdgpu_device *adev) return 0; } +void vega10_doorbell_index_init(struct amdgpu_device *adev) +{ + adev->doorbell_index.kiq = AMDGPU_DOORBELL64_KIQ; + adev->doorbell_index.mec_ring0 = AMDGPU_DOORBELL64_MEC_RING0; + adev->doorbell_index.mec_ring1 = AMDGPU_DOORBELL64_MEC_RING1; + adev->doorbell_index.mec_ring2 = AMDGPU_DOORBELL64_MEC_RING2; + adev->doorbell_index.mec_ring3 = AMDGPU_DOORBELL64_MEC_RING3; + adev->doorbell_index.mec_ring4 = AMDGPU_DOORBELL64_MEC_RING4; + adev->doorbell_index.mec_ring5 = AMDGPU_DOORBELL64_MEC_RING5; + adev->doorbell_index.mec_ring6 = AMDGPU_DOORBELL64_MEC_RING6; + adev->doorbell_index.mec_ring7 = AMDGPU_DOORBELL64_MEC_RING7; + adev->doorbell_index.userqueue_start = AMDGPU_DOORBELL64_USERQUEUE_START; + adev->doorbell_index.userqueue_end = AMDGPU_DOORBELL64_USERQUEUE_END; + adev->doorbell_index.gfx_ring0 = AMDGPU_DOORBELL64_GFX_RING0; + adev->doorbell_index.sdma_engine[0] = AMDGPU_DOORBELL64_sDMA_ENGINE0; + adev->doorbell_index.sdma_engine[1] = AMDGPU_DOORBELL64_sDMA_ENGINE1; + adev->doorbell_index.ih = AMDGPU_DOORBELL64_IH; + adev->doorbell_index.uvd_vce.uvd_ring0_1 = AMDGPU_DOORBELL64_UVD_RING0_1; + adev->doorbell_index.uvd_vce.uvd_ring2_3 = AMDGPU_DOORBELL64_UVD_RING2_3; + adev->doorbell_index.uvd_vce.uvd_ring4_5 = AMDGPU_DOORBELL64_UVD_RING4_5; + adev->doorbell_index.uvd_vce.uvd_ring6_7 = AMDGPU_DOORBELL64_UVD_RING6_7; + adev->doorbell_index.uvd_vce.vce_ring0_1 = AMDGPU_DOORBELL64_VCE_RING0_1; + adev->doorbell_index.uvd_vce.vce_ring2_3 = AMDGPU_DOORBELL64_VCE_RING2_3; + adev->doorbell_index.uvd_vce.vce_ring4_5 = AMDGPU_DOORBELL64_VCE_RING4_5; + adev->doorbell_index.uvd_vce.vce_ring6_7 = AMDGPU_DOORBELL64_VCE_RING6_7; + + adev->doorbell_index.first_non_cp = AMDGPU_DOORBELL64_FIRST_NON_CP; + adev->doorbell_index.last_non_cp = AMDGPU_DOORBELL64_LAST_NON_CP; + + /* In unit of dword doorbell */ + adev->doorbell_index.max_assignment = AMDGPU_DOORBELL64_MAX_ASSIGNMENT << 1; + adev->doorbell_index.sdma_doorbell_range = 4; +} diff --git a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c index d13fc4fcb517..0db84386252a 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c @@ -54,4 +54,42 @@ int vega20_reg_base_init(struct amdgpu_device *adev) return 0; } +void vega20_doorbell_index_init(struct amdgpu_device *adev) +{ + adev->doorbell_index.kiq = AMDGPU_VEGA20_DOORBELL_KIQ; + adev->doorbell_index.mec_ring0 = AMDGPU_VEGA20_DOORBELL_MEC_RING0; + adev->doorbell_index.mec_ring1 = AMDGPU_VEGA20_DOORBELL_MEC_RING1; + adev->doorbell_index.mec_ring2 = AMDGPU_VEGA20_DOORBELL_MEC_RING2; + adev->doorbell_index.mec_ring3 = AMDGPU_VEGA20_DOORBELL_MEC_RING3; + adev->doorbell_index.mec_ring4 = AMDGPU_VEGA20_DOORBELL_MEC_RING4; + adev->doorbell_index.mec_ring5 = AMDGPU_VEGA20_DOORBELL_MEC_RING5; + adev->doorbell_index.mec_ring6 = AMDGPU_VEGA20_DOORBELL_MEC_RING6; + adev->doorbell_index.mec_ring7 = AMDGPU_VEGA20_DOORBELL_MEC_RING7; + adev->doorbell_index.userqueue_start = AMDGPU_VEGA20_DOORBELL_USERQUEUE_START; + adev->doorbell_index.userqueue_end = AMDGPU_VEGA20_DOORBELL_USERQUEUE_END; + adev->doorbell_index.gfx_ring0 = AMDGPU_VEGA20_DOORBELL_GFX_RING0; + adev->doorbell_index.sdma_engine[0] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0; + adev->doorbell_index.sdma_engine[1] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE1; + adev->doorbell_index.sdma_engine[2] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE2; + adev->doorbell_index.sdma_engine[3] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE3; + adev->doorbell_index.sdma_engine[4] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE4; + adev->doorbell_index.sdma_engine[5] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE5; + adev->doorbell_index.sdma_engine[6] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE6; + adev->doorbell_index.sdma_engine[7] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE7; + adev->doorbell_index.ih = AMDGPU_VEGA20_DOORBELL_IH; + adev->doorbell_index.uvd_vce.uvd_ring0_1 = AMDGPU_VEGA20_DOORBELL64_UVD_RING0_1; + adev->doorbell_index.uvd_vce.uvd_ring2_3 = AMDGPU_VEGA20_DOORBELL64_UVD_RING2_3; + adev->doorbell_index.uvd_vce.uvd_ring4_5 = AMDGPU_VEGA20_DOORBELL64_UVD_RING4_5; + adev->doorbell_index.uvd_vce.uvd_ring6_7 = AMDGPU_VEGA20_DOORBELL64_UVD_RING6_7; + adev->doorbell_index.uvd_vce.vce_ring0_1 = AMDGPU_VEGA20_DOORBELL64_VCE_RING0_1; + adev->doorbell_index.uvd_vce.vce_ring2_3 = AMDGPU_VEGA20_DOORBELL64_VCE_RING2_3; + adev->doorbell_index.uvd_vce.vce_ring4_5 = AMDGPU_VEGA20_DOORBELL64_VCE_RING4_5; + adev->doorbell_index.uvd_vce.vce_ring6_7 = AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7; + + adev->doorbell_index.first_non_cp = AMDGPU_VEGA20_DOORBELL64_FIRST_NON_CP; + adev->doorbell_index.last_non_cp = AMDGPU_VEGA20_DOORBELL64_LAST_NON_CP; + + adev->doorbell_index.max_assignment = AMDGPU_VEGA20_DOORBELL_MAX_ASSIGNMENT << 1; + adev->doorbell_index.sdma_doorbell_range = 20; +} diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c index 07880d35e9de..5e5b42a0744a 100644 --- a/drivers/gpu/drm/amd/amdgpu/vi.c +++ b/drivers/gpu/drm/amd/amdgpu/vi.c @@ -87,9 +87,9 @@ static u32 vi_pcie_rreg(struct amdgpu_device *adev, u32 reg) u32 r; spin_lock_irqsave(&adev->pcie_idx_lock, flags); - WREG32(mmPCIE_INDEX, reg); - (void)RREG32(mmPCIE_INDEX); - r = RREG32(mmPCIE_DATA); + WREG32_NO_KIQ(mmPCIE_INDEX, reg); + (void)RREG32_NO_KIQ(mmPCIE_INDEX); + r = RREG32_NO_KIQ(mmPCIE_DATA); spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); return r; } @@ -99,10 +99,10 @@ static void vi_pcie_wreg(struct amdgpu_device *adev, u32 reg, u32 v) unsigned long flags; spin_lock_irqsave(&adev->pcie_idx_lock, flags); - WREG32(mmPCIE_INDEX, reg); - (void)RREG32(mmPCIE_INDEX); - WREG32(mmPCIE_DATA, v); - (void)RREG32(mmPCIE_DATA); + WREG32_NO_KIQ(mmPCIE_INDEX, reg); + (void)RREG32_NO_KIQ(mmPCIE_INDEX); + WREG32_NO_KIQ(mmPCIE_DATA, v); + (void)RREG32_NO_KIQ(mmPCIE_DATA); spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); } @@ -123,8 +123,8 @@ static void vi_smc_wreg(struct amdgpu_device *adev, u32 reg, u32 v) unsigned long flags; spin_lock_irqsave(&adev->smc_idx_lock, flags); - WREG32(mmSMC_IND_INDEX_11, (reg)); - WREG32(mmSMC_IND_DATA_11, (v)); + WREG32_NO_KIQ(mmSMC_IND_INDEX_11, (reg)); + WREG32_NO_KIQ(mmSMC_IND_DATA_11, (v)); spin_unlock_irqrestore(&adev->smc_idx_lock, flags); } @@ -941,6 +941,69 @@ static bool vi_need_full_reset(struct amdgpu_device *adev) } } +static void vi_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0, + uint64_t *count1) +{ + uint32_t perfctr = 0; + uint64_t cnt0_of, cnt1_of; + int tmp; + + /* This reports 0 on APUs, so return to avoid writing/reading registers + * that may or may not be different from their GPU counterparts + */ + if (adev->flags & AMD_IS_APU) + return; + + /* Set the 2 events that we wish to watch, defined above */ + /* Reg 40 is # received msgs, Reg 104 is # of posted requests sent */ + perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40); + perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104); + + /* Write to enable desired perf counters */ + WREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK, perfctr); + /* Zero out and enable the perf counters + * Write 0x5: + * Bit 0 = Start all counters(1) + * Bit 2 = Global counter reset enable(1) + */ + WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000005); + + msleep(1000); + + /* Load the shadow and disable the perf counters + * Write 0x2: + * Bit 0 = Stop counters(0) + * Bit 1 = Load the shadow counters(1) + */ + WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000002); + + /* Read register values to get any >32bit overflow */ + tmp = RREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK); + cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER); + cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER); + + /* Get the values and add the overflow */ + *count0 = RREG32_PCIE(ixPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32); + *count1 = RREG32_PCIE(ixPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32); +} + +static bool vi_need_reset_on_init(struct amdgpu_device *adev) +{ + u32 clock_cntl, pc; + + if (adev->flags & AMD_IS_APU) + return false; + + /* check if the SMC is already running */ + clock_cntl = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0); + pc = RREG32_SMC(ixSMC_PC_C); + if ((0 == REG_GET_FIELD(clock_cntl, SMC_SYSCON_CLOCK_CNTL_0, ck_disable)) && + (0x20100 <= pc)) + return true; + + return false; +} + static const struct amdgpu_asic_funcs vi_asic_funcs = { .read_disabled_bios = &vi_read_disabled_bios, @@ -955,6 +1018,9 @@ static const struct amdgpu_asic_funcs vi_asic_funcs = .flush_hdp = &vi_flush_hdp, .invalidate_hdp = &vi_invalidate_hdp, .need_full_reset = &vi_need_full_reset, + .init_doorbell_index = &legacy_doorbell_index_init, + .get_pcie_usage = &vi_get_pcie_usage, + .need_reset_on_init = &vi_need_reset_on_init, }; #define CZ_REV_BRISTOL(rev) \ @@ -1712,3 +1778,21 @@ int vi_set_ip_blocks(struct amdgpu_device *adev) return 0; } + +void legacy_doorbell_index_init(struct amdgpu_device *adev) +{ + adev->doorbell_index.kiq = AMDGPU_DOORBELL_KIQ; + adev->doorbell_index.mec_ring0 = AMDGPU_DOORBELL_MEC_RING0; + adev->doorbell_index.mec_ring1 = AMDGPU_DOORBELL_MEC_RING1; + adev->doorbell_index.mec_ring2 = AMDGPU_DOORBELL_MEC_RING2; + adev->doorbell_index.mec_ring3 = AMDGPU_DOORBELL_MEC_RING3; + adev->doorbell_index.mec_ring4 = AMDGPU_DOORBELL_MEC_RING4; + adev->doorbell_index.mec_ring5 = AMDGPU_DOORBELL_MEC_RING5; + adev->doorbell_index.mec_ring6 = AMDGPU_DOORBELL_MEC_RING6; + adev->doorbell_index.mec_ring7 = AMDGPU_DOORBELL_MEC_RING7; + adev->doorbell_index.gfx_ring0 = AMDGPU_DOORBELL_GFX_RING0; + adev->doorbell_index.sdma_engine[0] = AMDGPU_DOORBELL_sDMA_ENGINE0; + adev->doorbell_index.sdma_engine[1] = AMDGPU_DOORBELL_sDMA_ENGINE1; + adev->doorbell_index.ih = AMDGPU_DOORBELL_IH; + adev->doorbell_index.max_assignment = AMDGPU_DOORBELL_MAX_ASSIGNMENT; +} diff --git a/drivers/gpu/drm/amd/amdgpu/vi.h b/drivers/gpu/drm/amd/amdgpu/vi.h index 0429fe332269..8de0772f986c 100644 --- a/drivers/gpu/drm/amd/amdgpu/vi.h +++ b/drivers/gpu/drm/amd/amdgpu/vi.h @@ -30,4 +30,5 @@ void vi_srbm_select(struct amdgpu_device *adev, u32 me, u32 pipe, u32 queue, u32 vmid); int vi_set_ip_blocks(struct amdgpu_device *adev); +void legacy_doorbell_index_init(struct amdgpu_device *adev); #endif |