diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu')
107 files changed, 4109 insertions, 3056 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile index 138cb787d27e..f76bcb9c45e4 100644 --- a/drivers/gpu/drm/amd/amdgpu/Makefile +++ b/drivers/gpu/drm/amd/amdgpu/Makefile @@ -53,7 +53,7 @@ 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 \ @@ -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..bcef6ea4bcf9 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. */ @@ -653,6 +540,8 @@ 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); }; /* @@ -831,7 +720,6 @@ 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 +864,9 @@ struct amdgpu_device { /* GDS */ struct amdgpu_gds gds; + /* KFD */ + struct amdgpu_kfd_dev kfd; + /* display related functionality */ struct amdgpu_display_manager dm; @@ -989,9 +880,6 @@ 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 +911,10 @@ 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 +939,6 @@ 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 +1000,6 @@ 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 +1041,7 @@ 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)) /* Common functions */ bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev); @@ -1219,12 +1102,6 @@ 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 +1129,9 @@ 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..2dfaf158ef07 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -26,15 +26,26 @@ #include "amdgpu.h" #include "amdgpu_gfx.h" #include <linux/module.h> +#include <linux/dma-buf.h> const struct kgd2kfd_calls *kgd2kfd; 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) @@ -73,9 +84,11 @@ 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 +98,11 @@ 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; } /** @@ -126,7 +142,8 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) { int i, n; 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, @@ -144,7 +161,7 @@ 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, @@ -165,7 +182,7 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) &gpu_resources.doorbell_start_offset); if (adev->asic_type < CHIP_VEGA10) { - kgd2kfd->device_init(adev->kfd, &gpu_resources); + kgd2kfd->device_init(adev->kfd.dev, &gpu_resources); return; } @@ -179,25 +196,14 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) * 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); - } + gpu_resources.sdma_doorbell[0][i] = + adev->doorbell_index.sdma_engine0 + (i >> 1); + gpu_resources.sdma_doorbell[0][i+1] = + adev->doorbell_index.sdma_engine0 + 0x200 + (i >> 1); + gpu_resources.sdma_doorbell[1][i] = + adev->doorbell_index.sdma_engine1 + (i >> 1); + gpu_resources.sdma_doorbell[1][i+1] = + adev->doorbell_index.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. @@ -205,37 +211,37 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) gpu_resources.reserved_doorbell_mask = 0x1e0; gpu_resources.reserved_doorbell_val = 0x0e0; - 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 +250,8 @@ 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 +260,8 @@ 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 +274,9 @@ 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 +346,7 @@ 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 +357,8 @@ 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 +389,7 @@ 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 +398,7 @@ 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 +411,7 @@ 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 +434,62 @@ 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 +572,7 @@ 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 +586,7 @@ 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) { } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index 8e0d4f7196b4..70429f7aa9a8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -27,7 +27,6 @@ #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> @@ -35,6 +34,7 @@ #include "amdgpu_vm.h" extern const struct kgd2kfd_calls *kgd2kfd; +extern uint64_t amdgpu_amdkfd_total_mem_size; struct amdgpu_device; @@ -77,6 +77,11 @@ 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 +139,21 @@ 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 +205,13 @@ 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); #endif /* AMDGPU_AMDKFD_H_INCLUDED */ 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..be1ab43473c6 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; @@ -154,35 +168,39 @@ 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); } @@ -395,23 +413,6 @@ 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 +423,7 @@ 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 @@ -536,7 +537,7 @@ 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 +678,7 @@ 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 +742,7 @@ 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); @@ -826,7 +827,7 @@ static int unmap_bo_from_gpuvm(struct amdgpu_device *adev, /* 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 +852,7 @@ 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 +887,24 @@ 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 +920,26 @@ 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) { @@ -1149,6 +1188,8 @@ 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 +1218,25 @@ 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 +1252,8 @@ 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 +1269,10 @@ 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 +1283,7 @@ 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 +1291,10 @@ 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 +1324,17 @@ 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; } @@ -1341,6 +1404,14 @@ 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 +1476,8 @@ 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)) { @@ -1642,6 +1714,60 @@ 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 @@ -1808,7 +1934,7 @@ 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 */ @@ -2027,7 +2153,7 @@ 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 +2170,10 @@ 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 +2189,11 @@ 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 +2214,7 @@ 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 +2233,7 @@ 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_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_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 0acc8dee2cb8..cf4e190c0a72 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); @@ -598,6 +599,10 @@ 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 +722,14 @@ 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 +966,6 @@ 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) { @@ -1104,7 +1111,7 @@ 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; @@ -1193,7 +1200,7 @@ 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 +1267,7 @@ 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 +1291,7 @@ 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; } 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..d85184b5b35c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -248,7 +248,7 @@ 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_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 30bc345d6fdf..b60afeade50a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -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)); @@ -1656,7 +1670,9 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) /* 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; @@ -1681,7 +1697,8 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) if (r) return r; - 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)) @@ -1848,6 +1865,9 @@ 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 +1910,7 @@ 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); } @@ -2337,6 +2357,19 @@ 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 +2468,8 @@ 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 +2490,6 @@ 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 +2508,9 @@ 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 */ @@ -3148,86 +3183,6 @@ 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 @@ -3295,40 +3250,46 @@ 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) { @@ -3348,10 +3309,144 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, 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 (!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]; @@ -3363,7 +3458,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, * or all rings (in the case @job is NULL) * after above amdgpu_reset accomplished */ - if ((!job || job->base.sched == &ring->sched) && !r) + if ((!job || job->base.sched == &ring->sched) && !adev->asic_reset_res) drm_sched_job_recovery(&ring->sched); kthread_unpark(ring->sched.thread); @@ -3373,21 +3468,142 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *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 */ + amdgpu_amdkfd_pre_reset(adev); +} +static void amdgpu_device_unlock_adev(struct amdgpu_device *adev) +{ /*unlock kfd */ 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); + if (hive && adev->gmc.xgmi.num_physical_nodes > 1 && + !mutex_trylock(&hive->hive_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->hive_lock); + + if (r) + dev_info(adev->dev, "GPU reset end with ret = %d\n", r); return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c index 686a26de50f9..15ce7e681d67 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c @@ -631,6 +631,11 @@ 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 +862,12 @@ 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..be620b29f4aa --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h @@ -0,0 +1,243 @@ +/* + * 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_engine0; + uint32_t sdma_engine1; + uint32_t sdma_engine2; + uint32_t sdma_engine3; + uint32_t sdma_engine4; + uint32_t sdma_engine5; + uint32_t sdma_engine6; + uint32_t sdma_engine7; + 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 max_assignment; +}; + +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_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_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_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 74b611e8a1b1..9c77eaa45982 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -454,9 +454,10 @@ 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); /** @@ -1227,9 +1228,6 @@ 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 +1257,14 @@ 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 +1276,7 @@ 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_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 7b3d1ebda9df..f4f00217546e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -169,7 +169,7 @@ 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); @@ -604,7 +604,10 @@ 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..c48207b377bc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -146,7 +146,7 @@ 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; } @@ -221,8 +221,7 @@ 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); + amdgpu_ring_emit_ib(ring, job, ib, need_ctx_switch); need_ctx_switch = false; } @@ -347,19 +346,14 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev) tmo_gfx = 8 * AMDGPU_IB_TEST_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 +368,23 @@ 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.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h index 9ce8c93ec19b..f877bb78d10a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h @@ -51,14 +51,12 @@ struct amdgpu_ih_ring { 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, struct amdgpu_iv_entry *entry); void (*set_rptr)(struct amdgpu_device *adev); }; #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)) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c index 52c17f6219a7..b7968f426862 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 @@ -162,13 +145,6 @@ static void amdgpu_irq_callback(struct amdgpu_device *adev, 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); @@ -262,15 +238,12 @@ int amdgpu_irq_init(struct amdgpu_device *adev) amdgpu_hotplug_work_func); } - INIT_WORK(&adev->reset_work, amdgpu_irq_reset_work_func); - adev->irq.installed = true; r = drm_irq_install(adev->ddev, adev->ddev->pdev->irq); if (r) { 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 +272,6 @@ 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) { @@ -392,39 +364,38 @@ void amdgpu_irq_dispatch(struct amdgpu_device *adev, unsigned client_id = entry->client_id; unsigned src_id = entry->src_id; struct amdgpu_irq_src *src; + bool handled = false; int r; trace_amdgpu_iv(entry); 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); + } else if ((src = adev->irq.client[client_id].sources[src_id])) { r = src->funcs->process(adev, src, entry); - if (r) + 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_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 755f733bf0d9..e0af44fd6a0c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -112,6 +112,8 @@ 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..bc62bf41b7e9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -336,7 +336,7 @@ 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 +344,7 @@ 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 +352,7 @@ 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 +363,7 @@ 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 +372,7 @@ 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 +384,7 @@ 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 +392,7 @@ 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 +400,14 @@ 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 +978,10 @@ 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 +1051,8 @@ 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_mode.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h index d1b4d9b6aae0..aadd0fa42e43 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h @@ -38,7 +38,6 @@ #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 <linux/i2c.h> #include <linux/i2c-algo-bit.h> #include <linux/hrtimer.h> @@ -57,7 +56,6 @@ 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 +293,6 @@ 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 +316,7 @@ 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 +332,8 @@ 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; @@ -436,11 +429,6 @@ 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..fd271f9746a2 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); @@ -608,53 +608,6 @@ 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 * diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h index 7d3312d0da11..9291c2f837e9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h @@ -267,11 +267,6 @@ 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); 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..1f61ed95727c 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); @@ -1642,6 +1644,19 @@ 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 || @@ -1956,6 +1971,17 @@ 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) @@ -2129,7 +2155,7 @@ 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..71913a18d142 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c @@ -39,8 +39,6 @@ #include <drm/amdgpu_drm.h> #include <linux/dma-buf.h> -static const struct dma_buf_ops amdgpu_dmabuf_ops; - /** * amdgpu_gem_prime_get_sg_table - &drm_driver.gem_prime_get_sg_table * implementation @@ -332,15 +330,13 @@ 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..6759d898b3ab 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c @@ -90,6 +90,8 @@ static int psp_sw_fini(void *handle) adev->psp.sos_fw = NULL; release_firmware(adev->psp.asd_fw); adev->psp.asd_fw = NULL; + release_firmware(adev->psp.ta_fw); + adev->psp.ta_fw = NULL; return 0; } @@ -118,21 +120,25 @@ 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) { @@ -149,10 +155,22 @@ psp_cmd_submit_buf(struct psp_context *psp, return ret; } -static void psp_prep_tmr_cmd_buf(struct psp_gfx_cmd_resp *cmd, +bool psp_support_vmr_ring(struct psp_context *psp) +{ + if (amdgpu_sriov_vf(psp->adev) && psp->sos_fw_version > 0x80045) + return true; + else + return false; +} + +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 +204,12 @@ 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 +276,194 @@ 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->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; @@ -292,6 +491,15 @@ 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; } @@ -321,7 +529,7 @@ static int psp_np_fw_load(struct psp_context *psp) 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 +548,10 @@ 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_destroy(psp, PSP_RING_TYPE__KM); goto skip_memalloc; + } psp->cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL); if (!psp->cmd) @@ -452,6 +662,10 @@ static int psp_hw_fini(void *handle) 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); amdgpu_bo_free_kernel(&psp->tmr_bo, &psp->tmr_mc_addr, &psp->tmr_buf); @@ -479,6 +693,15 @@ static int psp_suspend(void *handle) 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) { DRM_ERROR("PSP ring stop failed\n"); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h index 8b8720e9c3f0..10decf70c9aa 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 @@ -80,12 +83,20 @@ 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_node_id)(struct psp_context *psp); uint64_t (*xgmi_get_hive_id)(struct psp_context *psp); 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); +}; + +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 +107,7 @@ 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 +145,16 @@ 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 { @@ -141,21 +162,17 @@ 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)) @@ -177,8 +194,8 @@ struct psp_xgmi_topology_info { ((psp)->funcs->smu_reload_quirk ? (psp)->funcs->smu_reload_quirk((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_node_id(psp) \ + ((psp)->funcs->xgmi_get_node_id ? (psp)->funcs->xgmi_get_node_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_topology_info(psp, num_device, topology) \ @@ -199,6 +216,9 @@ 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); +bool psp_support_vmr_ring(struct psp_context *psp); + 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..0beb01fef83f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h @@ -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); + bool ctx_switch); 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, c) ((r)->funcs->emit_ib((r), (job), (ib), (c))) #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_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..626abca770a0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -218,6 +218,7 @@ 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 +228,14 @@ 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 +367,20 @@ 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..c91ec3101d00 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; @@ -1758,14 +1664,10 @@ 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, @@ -1922,7 +1824,6 @@ 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 +1970,7 @@ 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..98a1b2ce2b9d 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, + bool ctx_switch) { 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..50293652af14 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, bool ctx_switch); 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..e2e42e3fbcf3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c @@ -425,11 +425,9 @@ 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 +439,9 @@ 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 +563,20 @@ 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 +589,9 @@ 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 +601,8 @@ 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 +717,19 @@ 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 +745,8 @@ 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 +760,8 @@ 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 +814,18 @@ 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 +834,10 @@ 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..e73d152659a2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -617,7 +617,8 @@ 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); } @@ -773,10 +774,6 @@ 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; @@ -1844,10 +1841,6 @@ 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; @@ -3028,6 +3021,10 @@ 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; + r = amdgpu_vm_clear_bo(adev, vm, root, adev->vm_manager.root_level, vm->pte_support_ats); @@ -3057,7 +3054,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, } INIT_KFIFO(vm->faults); - vm->fault_credit = 16; return 0; @@ -3270,42 +3266,6 @@ 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 diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h index 2a8898d19c8b..e8dcfd59fc93 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); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c index 897afbb348c1..0b263a9857c6 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); @@ -31,15 +31,16 @@ 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 i; struct amdgpu_hive_info *tmp; @@ -58,62 +59,99 @@ static struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev) tmp = &xgmi_hives[hive_count++]; tmp->hive_id = adev->gmc.xgmi.hive_id; INIT_LIST_HEAD(&tmp->device_list); + mutex_init(&tmp->hive_lock); + 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); + else + dev_info(adev->dev, "XGMI: Set topology for node %d, hive 0x%llx.\n", + adev->gmc.xgmi.physical_node_id, + adev->gmc.xgmi.hive_id); + + 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.node_id = psp_xgmi_get_node_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; + hive_topology = &hive->topology_info; + 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; + hive_topology->nodes[count++].node_id = entry->node_id; + hive->number_devices = count; - ret = psp_xgmi_get_topology_info(&adev->psp, count, tmp_topology); - 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); - goto exit; - } - /* Each psp need to set the latest topology */ + /* 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; + } 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; + + mutex_lock(&xgmi_mutex); + + hive = amdgpu_get_xgmi_hive(adev); + if (!hive) + goto exit; + + if (!(hive->number_devices--)) + mutex_destroy(&hive->hive_lock); + +exit: + mutex_unlock(&xgmi_mutex); +} 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..6151eb9c8ad3 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.h @@ -0,0 +1,40 @@ +/* + * 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; +}; + +struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev); +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/ci_dpm.c b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c index 79220a91abe3..86e14c754dd4 100644 --- a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c @@ -743,19 +743,19 @@ static int ci_enable_didt(struct amdgpu_device *adev, bool enable) 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); + amdgpu_gfx_rlc_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); + amdgpu_gfx_rlc_exit_safe_mode(adev); return ret; } } ci_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/cik.c b/drivers/gpu/drm/amd/amdgpu/cik.c index f41f5f57e9f3..71c50d8900e3 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik.c +++ b/drivers/gpu/drm/amd/amdgpu/cik.c @@ -1755,6 +1755,7 @@ 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, }; static int cik_common_early_init(void *handle) 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_ih.c b/drivers/gpu/drm/amd/amdgpu/cik_ih.c index b5775c6a857b..8a8b4967a101 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_ih.c @@ -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 * @@ -461,7 +433,6 @@ 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..45795191de1f 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) + bool ctx_switch) { + 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..9d3ea298e116 100644 --- a/drivers/gpu/drm/amd/amdgpu/cz_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/cz_ih.c @@ -208,34 +208,6 @@ static u32 cz_ih_get_wptr(struct amdgpu_device *adev) } /** - * 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; -} - -/** * cz_ih_decode_iv - decode an interrupt vector * * @adev: amdgpu_device pointer @@ -442,7 +414,6 @@ 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/gfx_v6_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c index d76eb27945dc..1dc3013ea1d5 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; } @@ -1845,9 +1840,11 @@ 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) + bool ctx_switch) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; /* insert SWITCH_BUFFER packet before first IB in the ring frame */ @@ -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..3a9fb6018c16 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; } @@ -2233,9 +2226,11 @@ 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, + bool ctx_switch) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; /* insert SWITCH_BUFFER packet before first IB in the ring frame */ @@ -2262,9 +2257,11 @@ 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) + bool ctx_switch) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); @@ -2316,17 +2313,15 @@ 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 +2333,16 @@ 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 +2392,7 @@ 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 +2602,9 @@ 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 +2661,7 @@ 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 +2767,7 @@ 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 +3092,7 @@ 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 +3251,10 @@ 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 +3281,23 @@ 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 +3378,12 @@ 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 +3405,7 @@ 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 +3482,13 @@ 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 +3519,7 @@ 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 +3721,12 @@ 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 +4165,17 @@ 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 +4363,7 @@ 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 +4426,7 @@ 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 +4490,7 @@ 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 +4513,7 @@ 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 +4531,7 @@ 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 +4616,7 @@ 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 +4845,36 @@ 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 +4884,7 @@ 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; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 617b0c8908a3..381f593b0cda 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, @@ -4379,10 +4277,8 @@ 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; + r = amdgpu_ring_test_helper(ring); return r; } @@ -4396,8 +4292,8 @@ 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); } @@ -4473,11 +4369,9 @@ 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) { + r = amdgpu_ring_test_helper(kiq_ring); + if (r) DRM_ERROR("KCQ enable failed\n"); - kiq_ring->ready = false; - } return r; } @@ -4755,8 +4649,8 @@ 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 +4675,7 @@ 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; } @@ -4820,10 +4714,7 @@ static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev) */ for (i = adev->gfx.num_compute_rings - 1; i >= 0; i--) { ring = &adev->gfx.compute_ring[i]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) - ring->ready = false; + r = amdgpu_ring_test_helper(ring); } done: @@ -4867,7 +4758,7 @@ 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 +4790,7 @@ 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 +4864,16 @@ 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 +4952,16 @@ 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 +5055,13 @@ 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 +5086,7 @@ 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); + adev->gfx.rlc.funcs->start(adev); return 0; } @@ -5445,7 +5334,7 @@ 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 +5388,7 @@ 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 +5482,53 @@ 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 +5538,17 @@ 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 +5556,7 @@ 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 +5652,7 @@ 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 +5662,7 @@ 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 +5745,7 @@ 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 +6025,11 @@ 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, + bool ctx_switch) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; if (ib->flags & AMDGPU_IB_FLAG_CE) @@ -6161,9 +6057,11 @@ 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) + bool ctx_switch) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); @@ -6738,12 +6636,39 @@ 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 +6677,7 @@ 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; } @@ -6976,10 +6901,8 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = { 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_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, diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index 21363b2b2ee5..7556716038d3 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"); @@ -396,18 +397,14 @@ 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 +416,11 @@ 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 +437,16 @@ 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 +460,17 @@ 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 +646,20 @@ 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 +1064,13 @@ 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) +static int gfx_v9_0_cp_jump_table_num(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) -{ - /* 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 +1079,18 @@ 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 +1164,7 @@ 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 +1535,8 @@ 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 +1580,7 @@ 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 +1648,7 @@ 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 +1669,7 @@ 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) @@ -2499,12 +2399,12 @@ 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); + adev->gfx.rlc.funcs->reset(adev); gfx_v9_0_init_pg(adev); @@ -2515,15 +2415,24 @@ 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 +2447,7 @@ 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 +2637,7 @@ 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 +2652,8 @@ 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 +2776,9 @@ 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 +2996,9 @@ 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 +3157,7 @@ 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 +3222,13 @@ 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 +3255,7 @@ 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 +3293,7 @@ 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 +3335,7 @@ 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 +3410,7 @@ 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 +3509,47 @@ 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 +3560,7 @@ 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, @@ -3774,7 +3658,7 @@ 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 +3698,7 @@ 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 +3706,7 @@ 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 +3746,7 @@ 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 +3775,17 @@ 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 +3966,11 @@ 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, + bool ctx_switch) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; if (ib->flags & AMDGPU_IB_FLAG_CE) @@ -4104,20 +3999,22 @@ 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) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + bool ctx_switch) { - u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); - 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 +4593,39 @@ 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 +4634,7 @@ 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; } @@ -4837,10 +4761,8 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = { 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_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, 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..1ad7e6b8ed1d 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); diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c index f35d7a554ad5..ce150de723c9 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c @@ -244,6 +244,62 @@ 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) @@ -255,6 +311,9 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev, 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); @@ -293,14 +352,14 @@ 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 +371,6 @@ 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 +379,50 @@ 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; - } - - spin_lock(&adev->gmc.invalidate_lock); - - WREG32_NO_KIQ(hub->vm_inv_eng0_req + eng, tmp); + uint32_t req = hub->vm_inv_eng0_req + eng; + uint32_t ack = hub->vm_inv_eng0_ack + eng; - /* 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 +432,7 @@ 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), @@ -739,9 +742,8 @@ static int gmc_v9_0_late_init(void *handle) 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); + dev_info(adev->dev, "ring %s uses VM inv eng %u on hub %u\n", + ring->name, ring->vm_inv_eng, ring->funcs->vmhub); } /* Engine 16 is used for KFD and 17 for GART flushes */ @@ -959,6 +961,9 @@ 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 +996,7 @@ 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 +1127,7 @@ 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/iceland_ih.c b/drivers/gpu/drm/amd/amdgpu/iceland_ih.c index cf0fc61aebe6..a3984d10b604 100644 --- a/drivers/gpu/drm/amd/amdgpu/iceland_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/iceland_ih.c @@ -208,34 +208,6 @@ static u32 iceland_ih_get_wptr(struct amdgpu_device *adev) } /** - * 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; -} - -/** * iceland_ih_decode_iv - decode an interrupt vector * * @adev: amdgpu_device pointer @@ -440,7 +412,6 @@ 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..d0d966d6080a 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)); 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_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/psp_gfx_if.h b/drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h index 882bd83a28c4..0de00fbe9233 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 */ }; diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v10_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v10_0.c index 295c2205485a..d78b4306a36f 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v10_0.c @@ -240,12 +240,9 @@ 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); diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c index 3f3fac2d50cd..6c9a1b748ca7 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c @@ -34,6 +34,7 @@ #include "nbio/nbio_7_4_offset.h" MODULE_FIRMWARE("amdgpu/vega20_sos.bin"); +MODULE_FIRMWARE("amdgpu/vega20_ta.bin"); /* address block */ #define smnMP1_FIRMWARE_FLAGS 0x3010024 @@ -98,7 +99,8 @@ static int psp_v11_0_init_microcode(struct psp_context *psp) 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 ta_firmware_header_v1_0 *ta_hdr; DRM_DEBUG("\n"); @@ -119,16 +121,32 @@ 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); + le32_to_cpu(sos_hdr->sos_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) + goto out; + + err = amdgpu_ucode_validate(adev->psp.ta_fw); + if (err) + goto out; + + 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; out: if (err) { @@ -153,8 +171,11 @@ 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 +188,7 @@ 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 +229,7 @@ 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; @@ -278,26 +299,47 @@ 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_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 +350,24 @@ 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_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_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 +406,10 @@ 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_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 +438,11 @@ 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_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 +587,7 @@ 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); @@ -552,24 +610,110 @@ 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) { - return 0; + 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 u64 psp_v11_0_xgmi_get_hive_id(struct psp_context *psp) { - u64 hive_id = 0; + 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)); - /* Remove me when we can get correct hive_id through PSP */ - if (psp->adev->gmc.xgmi.num_physical_nodes) - hive_id = 0x123456789abcdef; + xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_HIVE_ID; - return hive_id; + /* Invoke xgmi ta to get hive id */ + ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id); + if (ret) + return 0; + else + return xgmi_cmd->xgmi_out_message.get_hive_id.hive_id; +} + +static u64 psp_v11_0_xgmi_get_node_id(struct psp_context *psp) +{ + 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_NODE_ID; + + /* Invoke xgmi ta to get the node id */ + ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id); + if (ret) + return 0; + else + return xgmi_cmd->xgmi_out_message.get_node_id.node_id; } static const struct psp_funcs psp_v11_0_funcs = { @@ -587,6 +731,7 @@ 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, }; 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..7357fd56e614 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v3_1.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v3_1.c @@ -194,7 +194,7 @@ 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; @@ -254,7 +254,7 @@ 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; @@ -356,12 +356,9 @@ 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); @@ -593,9 +590,9 @@ 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); diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index 2d4770e173dd..9f3cb2aec7c2 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) + bool ctx_switch) { + 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..1bccc5fe2d9d 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) + bool ctx_switch) { + 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; @@ -1163,7 +1146,7 @@ static int sdma_v3_0_sw_init(void *handle) if (!amdgpu_sriov_vf(adev)) { ring->use_doorbell = true; ring->doorbell_index = (i == 0) ? - AMDGPU_DOORBELL_sDMA_ENGINE0 : AMDGPU_DOORBELL_sDMA_ENGINE1; + adev->doorbell_index.sdma_engine0 : adev->doorbell_index.sdma_engine1; } else { ring->use_pollmem = true; } @@ -1440,8 +1423,14 @@ 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..4b6d3e5c821f 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); @@ -367,16 +372,11 @@ 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 +417,67 @@ 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 +497,12 @@ 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, + bool ctx_switch) { + 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 +624,16 @@ 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 +649,39 @@ 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 +719,15 @@ 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 +748,215 @@ 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); - - 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(i, mmSDMA0_GFX_DOORBELL); + doorbell_offset = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET); - 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); + adev->nbio_funcs->sdma_doorbell_range(adev, i, ring->use_doorbell, + ring->doorbell_index); + + 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 +1067,14 @@ 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 +1090,78 @@ 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); } - if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { - r = sdma_v4_0_load_microcode(adev); + /* 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 (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; + + r = amdgpu_ring_test_helper(ring); + if (r) + return r; - /* 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); + 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 +1185,16 @@ 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 +1211,11 @@ 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 +1238,16 @@ 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 +1266,17 @@ 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 +1381,7 @@ 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 +1447,46 @@ 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 >= 115;*/ + return false; + 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 +1495,6 @@ 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 +1513,6 @@ 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 +1521,10 @@ 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 = (i == 0) ? + (adev->doorbell_index.sdma_engine0 << 1) + : (adev->doorbell_index.sdma_engine1 << 1); sprintf(ring->name, "sdma%d", i); r = amdgpu_ring_init(adev, ring, 1024, @@ -1340,6 +1534,29 @@ 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 = (i == 0) ? + (adev->doorbell_index.sdma_engine0 << 1) + : (adev->doorbell_index.sdma_engine1 << 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 +1567,11 @@ 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 +1634,7 @@ 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 +1650,8 @@ 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 +1672,13 @@ 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 +1687,32 @@ 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: + /* XXX compute */ + break; + case 2: + /* XXX compute */ + break; + case 3: + amdgpu_fence_process(&adev->sdma.instance[instance].page); break; } return 0; @@ -1512,12 +1722,29 @@ 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 +1957,38 @@ 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 +1996,10 @@ 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 +2081,10 @@ 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 +2102,10 @@ 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_dma.c b/drivers/gpu/drm/amd/amdgpu/si_dma.c index adbaea6da0d7..b6e473134e19 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) + bool ctx_switch) { + 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_ih.c b/drivers/gpu/drm/amd/amdgpu/si_ih.c index b3d7d9f83202..2938fb9f17cc 100644 --- a/drivers/gpu/drm/amd/amdgpu/si_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/si_ih.c @@ -118,19 +118,6 @@ static u32 si_ih_get_wptr(struct amdgpu_device *adev) 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; -} - static void si_ih_decode_iv(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry) { @@ -301,7 +288,6 @@ 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..8849b74078d6 100644 --- a/drivers/gpu/drm/amd/amdgpu/soc15.c +++ b/drivers/gpu/drm/amd/amdgpu/soc15.c @@ -507,6 +507,9 @@ 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) @@ -613,6 +616,24 @@ 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, +}; + +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, }; static int soc15_common_early_init(void *handle) @@ -632,11 +653,11 @@ 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 +681,7 @@ 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 +704,7 @@ 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 | @@ -704,6 +727,7 @@ 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; else if (adev->pdev->device == 0x15d8) 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/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..15da06ddeb75 100644 --- a/drivers/gpu/drm/amd/amdgpu/tonga_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/tonga_ih.c @@ -219,34 +219,6 @@ static u32 tonga_ih_get_wptr(struct amdgpu_device *adev) } /** - * 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; -} - -/** * tonga_ih_decode_iv - decode an interrupt vector * * @adev: amdgpu_device pointer @@ -322,7 +294,7 @@ 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 +478,6 @@ 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..d69c8f6daaf8 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) + bool ctx_switch) { 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..ee8cd06ddc38 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) + bool ctx_switch) { 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..d4f4a66f8324 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) + bool ctx_switch) { + 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, + bool ctx_switch) { + 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..089645e78f98 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; @@ -447,10 +430,6 @@ 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; @@ -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) + bool ctx_switch) { 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, + bool ctx_switch) { + 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..2668effadd27 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, + bool ctx_switch) { + 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..9fb34b7d8e03 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, bool ctx_switch) { + 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..4f8352044563 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) @@ -225,7 +217,7 @@ static int vcn_v1_0_hw_fini(void *handle) if (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; } @@ -1367,10 +1359,12 @@ 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, + bool ctx_switch) { 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 +1519,12 @@ 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, + bool ctx_switch) { + 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 +1724,12 @@ 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, + bool ctx_switch) { 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..2c250b01a903 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c @@ -220,90 +220,6 @@ static u32 vega10_ih_get_wptr(struct amdgpu_device *adev) } /** - * 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; - - 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]); - - /* 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; - } - - pasid = dw3 & 0xffff; - /* No PASID, can't identify faulting process */ - if (!pasid) - return true; - - /* Not a retry fault, check fault credit */ - if (!(dw5 & 0x80)) { - if (!amdgpu_vm_pasid_fault_credit(adev, pasid)) - goto ignore_iv; - 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, 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; - } - - spin_unlock(&adev->vm_manager.pasid_lock); - /* It's the first fault for this address, process it normally */ - return true; - -ignore_iv: - adev->irq.ih.rptr += 32; - return false; -} - -/** * vega10_ih_decode_iv - decode an interrupt vector * * @adev: amdgpu_device pointer @@ -385,7 +301,7 @@ static int vega10_ih_sw_init(void *handle) return r; 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); @@ -487,7 +403,6 @@ 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..422674bb3cdf 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c @@ -56,4 +56,32 @@ 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_engine0 = AMDGPU_DOORBELL64_sDMA_ENGINE0; + adev->doorbell_index.sdma_engine1 = 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; + /* In unit of dword doorbell */ + adev->doorbell_index.max_assignment = AMDGPU_DOORBELL64_MAX_ASSIGNMENT << 1; +} diff --git a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c index d13fc4fcb517..edce413fda9a 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c @@ -54,4 +54,37 @@ 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_engine0 = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0; + adev->doorbell_index.sdma_engine1 = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE1; + adev->doorbell_index.sdma_engine2 = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE2; + adev->doorbell_index.sdma_engine3 = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE3; + adev->doorbell_index.sdma_engine4 = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE4; + adev->doorbell_index.sdma_engine5 = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE5; + adev->doorbell_index.sdma_engine6 = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE6; + adev->doorbell_index.sdma_engine7 = 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.max_assignment = AMDGPU_VEGA20_DOORBELL_MAX_ASSIGNMENT << 1; +} diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c index 07880d35e9de..ff2906c215fa 100644 --- a/drivers/gpu/drm/amd/amdgpu/vi.c +++ b/drivers/gpu/drm/amd/amdgpu/vi.c @@ -955,6 +955,7 @@ 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, }; #define CZ_REV_BRISTOL(rev) \ @@ -1712,3 +1713,21 @@ 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_engine0 = AMDGPU_DOORBELL_sDMA_ENGINE0; + adev->doorbell_index.sdma_engine1 = 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 |