diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1001 |
1 files changed, 655 insertions, 346 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index a8b08a72b71b..64510898eedd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -31,6 +31,10 @@ #include <linux/console.h> #include <linux/slab.h> #include <linux/iommu.h> +#include <linux/pci.h> +#include <linux/devcoredump.h> +#include <generated/utsrelease.h> +#include <linux/pci-p2pdma.h> #include <drm/drm_atomic_helper.h> #include <drm/drm_probe_helper.h> @@ -55,7 +59,6 @@ #include "soc15.h" #include "nv.h" #include "bif/bif_4_1_d.h" -#include <linux/pci.h> #include <linux/firmware.h> #include "amdgpu_vf_error.h" @@ -80,14 +83,11 @@ MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin"); MODULE_FIRMWARE("amdgpu/picasso_gpu_info.bin"); MODULE_FIRMWARE("amdgpu/raven2_gpu_info.bin"); MODULE_FIRMWARE("amdgpu/arcturus_gpu_info.bin"); -MODULE_FIRMWARE("amdgpu/renoir_gpu_info.bin"); -MODULE_FIRMWARE("amdgpu/navi10_gpu_info.bin"); -MODULE_FIRMWARE("amdgpu/navi14_gpu_info.bin"); MODULE_FIRMWARE("amdgpu/navi12_gpu_info.bin"); -MODULE_FIRMWARE("amdgpu/vangogh_gpu_info.bin"); -MODULE_FIRMWARE("amdgpu/yellow_carp_gpu_info.bin"); #define AMDGPU_RESUME_MS 2000 +#define AMDGPU_MAX_RETRY_LIMIT 2 +#define AMDGPU_RETRY_SRIOV_RESET(r) ((r) == -EBUSY || (r) == -ETIMEDOUT || (r) == -EINVAL) const char *amdgpu_asic_name[] = { "TAHITI", @@ -424,10 +424,10 @@ bool amdgpu_device_skip_hw_access(struct amdgpu_device *adev) * the lock. */ if (in_task()) { - if (down_read_trylock(&adev->reset_sem)) - up_read(&adev->reset_sem); + if (down_read_trylock(&adev->reset_domain->sem)) + up_read(&adev->reset_domain->sem); else - lockdep_assert_held(&adev->reset_sem); + lockdep_assert_held(&adev->reset_domain->sem); } #endif return false; @@ -453,9 +453,9 @@ uint32_t amdgpu_device_rreg(struct amdgpu_device *adev, if ((reg * 4) < adev->rmmio_size) { if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev) && - down_read_trylock(&adev->reset_sem)) { + down_read_trylock(&adev->reset_domain->sem)) { ret = amdgpu_kiq_rreg(adev, reg); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); } else { ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); } @@ -538,9 +538,9 @@ void amdgpu_device_wreg(struct amdgpu_device *adev, if ((reg * 4) < adev->rmmio_size) { if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev) && - down_read_trylock(&adev->reset_sem)) { + down_read_trylock(&adev->reset_domain->sem)) { amdgpu_kiq_wreg(adev, reg, v); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); } else { writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); } @@ -552,9 +552,13 @@ void amdgpu_device_wreg(struct amdgpu_device *adev, } /** - * amdgpu_mm_wreg_mmio_rlc - write register either with mmio or with RLC path if in range + * amdgpu_mm_wreg_mmio_rlc - write register either with direct/indirect mmio or with RLC path if in range * - * this function is invoked only the debugfs register access + * @adev: amdgpu_device pointer + * @reg: mmio/rlc register + * @v: value to write + * + * this function is invoked only for the debugfs register access */ void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev, uint32_t reg, uint32_t v) @@ -566,7 +570,9 @@ void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev, adev->gfx.rlc.funcs && adev->gfx.rlc.funcs->is_rlcg_access_range) { if (adev->gfx.rlc.funcs->is_rlcg_access_range(adev, reg)) - return adev->gfx.rlc.funcs->sriov_wreg(adev, reg, v, 0, 0); + return amdgpu_sriov_wreg(adev, reg, v, 0, 0); + } else if ((reg * 4) >= adev->rmmio_size) { + adev->pcie_wreg(adev, reg * 4, v); } else { writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); } @@ -910,7 +916,10 @@ static int amdgpu_device_asic_init(struct amdgpu_device *adev) { amdgpu_asic_pre_asic_init(adev); - return amdgpu_atom_asic_init(adev->mode_info.atom_context); + if (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(11, 0, 0)) + return amdgpu_atomfirmware_asic_init(adev, true); + else + return amdgpu_atom_asic_init(adev->mode_info.atom_context); } /** @@ -1038,19 +1047,25 @@ static int amdgpu_device_doorbell_init(struct amdgpu_device *adev) 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), - 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; + if (adev->enable_mes) { + adev->doorbell.num_doorbells = + adev->doorbell.size / sizeof(u32); + } else { + adev->doorbell.num_doorbells = + min_t(u32, adev->doorbell.size / sizeof(u32), + 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 * @@ -1310,6 +1325,31 @@ bool amdgpu_device_need_post(struct amdgpu_device *adev) return true; } +/** + * amdgpu_device_should_use_aspm - check if the device should program ASPM + * + * @adev: amdgpu_device pointer + * + * Confirm whether the module parameter and pcie bridge agree that ASPM should + * be set for this device. + * + * Returns true if it should be used or false if not. + */ +bool amdgpu_device_should_use_aspm(struct amdgpu_device *adev) +{ + switch (amdgpu_aspm) { + case -1: + break; + case 0: + return false; + case 1: + return true; + default: + return false; + } + return pcie_aspm_enabled(adev->pdev); +} + /* if we get transitioned to only one device, take VGA back */ /** * amdgpu_device_vga_set_decode - enable/disable vga decode @@ -1444,11 +1484,12 @@ static int amdgpu_device_init_apu_flags(struct amdgpu_device *adev) case CHIP_YELLOW_CARP: break; case CHIP_CYAN_SKILLFISH: - if (adev->pdev->device == 0x13FE) + if ((adev->pdev->device == 0x13FE) || + (adev->pdev->device == 0x143F)) adev->apu_flags |= AMD_APU_IS_CYAN_SKILLFISH2; break; default: - return -EINVAL; + break; } return 0; @@ -1505,6 +1546,11 @@ static int amdgpu_device_check_arguments(struct amdgpu_device *adev) amdgpu_sched_hw_submission = roundup_pow_of_two(amdgpu_sched_hw_submission); } + if (amdgpu_reset_method < -1 || amdgpu_reset_method > 4) { + dev_warn(adev->dev, "invalid option for reset method, reverting to default\n"); + amdgpu_reset_method = -1; + } + amdgpu_device_check_smu_prv_buffer_size(adev); amdgpu_device_check_vm_size(adev); @@ -1513,10 +1559,6 @@ static int amdgpu_device_check_arguments(struct amdgpu_device *adev) adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type); - amdgpu_gmc_tmz_set(adev); - - amdgpu_gmc_noretry_set(adev); - return 0; } @@ -1670,7 +1712,7 @@ int amdgpu_device_ip_set_powergating_state(void *dev, * clockgating is enabled. */ void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev, - u32 *flags) + u64 *flags) { int i; @@ -1893,11 +1935,9 @@ static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev) adev->firmware.gpu_info_fw = NULL; if (adev->mman.discovery_bin) { - amdgpu_discovery_get_gfx_info(adev); - /* * FIXME: The bounding box is still needed by Navi12, so - * temporarily read it from gpu_info firmware. Should be droped + * temporarily read it from gpu_info firmware. Should be dropped * when DAL no longer needs it. */ if (adev->asic_type != CHIP_NAVI12) @@ -1905,35 +1945,6 @@ static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev) } switch (adev->asic_type) { -#ifdef CONFIG_DRM_AMDGPU_SI - case CHIP_VERDE: - case CHIP_TAHITI: - case CHIP_PITCAIRN: - case CHIP_OLAND: - case CHIP_HAINAN: -#endif -#ifdef CONFIG_DRM_AMDGPU_CIK - case CHIP_BONAIRE: - case CHIP_HAWAII: - case CHIP_KAVERI: - case CHIP_KABINI: - case CHIP_MULLINS: -#endif - case CHIP_TOPAZ: - case CHIP_TONGA: - case CHIP_FIJI: - case CHIP_POLARIS10: - case CHIP_POLARIS11: - case CHIP_POLARIS12: - case CHIP_VEGAM: - case CHIP_CARRIZO: - case CHIP_STONEY: - case CHIP_VEGA20: - case CHIP_ALDEBARAN: - case CHIP_SIENNA_CICHLID: - case CHIP_NAVY_FLOUNDER: - case CHIP_DIMGREY_CAVEFISH: - case CHIP_BEIGE_GOBY: default: return 0; case CHIP_VEGA10: @@ -1953,27 +1964,9 @@ static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev) case CHIP_ARCTURUS: chip_name = "arcturus"; break; - case CHIP_RENOIR: - if (adev->apu_flags & AMD_APU_IS_RENOIR) - chip_name = "renoir"; - else - chip_name = "green_sardine"; - break; - case CHIP_NAVI10: - chip_name = "navi10"; - break; - case CHIP_NAVI14: - chip_name = "navi14"; - break; case CHIP_NAVI12: chip_name = "navi12"; break; - case CHIP_VANGOGH: - chip_name = "vangogh"; - break; - case CHIP_YELLOW_CARP: - chip_name = "yellow_carp"; - break; } snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_gpu_info.bin", chip_name); @@ -2071,6 +2064,8 @@ out: */ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) { + struct drm_device *dev = adev_to_drm(adev); + struct pci_dev *parent; int i, r; amdgpu_device_enable_virtual_display(adev); @@ -2135,6 +2130,18 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) break; } + if (amdgpu_has_atpx() && + (amdgpu_is_atpx_hybrid() || + amdgpu_has_atpx_dgpu_power_cntl()) && + ((adev->flags & AMD_IS_APU) == 0) && + !pci_is_thunderbolt_attached(to_pci_dev(dev->dev))) + adev->flags |= AMD_IS_PX; + + if (!(adev->flags & AMD_IS_APU)) { + parent = pci_upstream_bridge(adev->pdev); + adev->has_pr3 = parent ? pci_pr3_present(parent) : false; + } + amdgpu_amdkfd_device_probe(adev); adev->pm.pp_feature = amdgpu_pp_feature_mask; @@ -2285,6 +2292,49 @@ static int amdgpu_device_fw_loading(struct amdgpu_device *adev) return r; } +static int amdgpu_device_init_schedulers(struct amdgpu_device *adev) +{ + long timeout; + int r, i; + + for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + struct amdgpu_ring *ring = adev->rings[i]; + + /* No need to setup the GPU scheduler for rings that don't need it */ + if (!ring || ring->no_scheduler) + continue; + + switch (ring->funcs->type) { + case AMDGPU_RING_TYPE_GFX: + timeout = adev->gfx_timeout; + break; + case AMDGPU_RING_TYPE_COMPUTE: + timeout = adev->compute_timeout; + break; + case AMDGPU_RING_TYPE_SDMA: + timeout = adev->sdma_timeout; + break; + default: + timeout = adev->video_timeout; + break; + } + + r = drm_sched_init(&ring->sched, &amdgpu_sched_ops, + ring->num_hw_submission, amdgpu_job_hang_limit, + timeout, adev->reset_domain->wq, + ring->sched_score, ring->name, + adev->dev); + if (r) { + DRM_ERROR("Failed to create scheduler on ring %s.\n", + ring->name); + return r; + } + } + + return 0; +} + + /** * amdgpu_device_ip_init - run init for hardware IPs * @@ -2315,8 +2365,16 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) } adev->ip_blocks[i].status.sw = true; - /* need to do gmc hw init early so we can allocate gpu mem */ - if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { + if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON) { + /* need to do common hw init early so everything is set up for gmc */ + r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev); + if (r) { + DRM_ERROR("hw_init %d failed %d\n", i, r); + goto init_failed; + } + adev->ip_blocks[i].status.hw = true; + } else if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { + /* need to do gmc hw init early so we can allocate gpu mem */ /* Try to reserve bad pages early */ if (amdgpu_sriov_vf(adev)) amdgpu_virt_exchange_data(adev); @@ -2352,7 +2410,7 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) } if (amdgpu_sriov_vf(adev)) - amdgpu_virt_exchange_data(adev); + amdgpu_virt_init_data_exchange(adev); r = amdgpu_ib_pool_init(adev); if (r) { @@ -2396,8 +2454,32 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) if (r) goto init_failed; - if (adev->gmc.xgmi.num_physical_nodes > 1) - amdgpu_xgmi_add_device(adev); + /** + * In case of XGMI grab extra reference for reset domain for this device + */ + if (adev->gmc.xgmi.num_physical_nodes > 1) { + if (amdgpu_xgmi_add_device(adev) == 0) { + if (!amdgpu_sriov_vf(adev)) { + struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); + + if (!hive->reset_domain || + !amdgpu_reset_get_reset_domain(hive->reset_domain)) { + r = -ENOENT; + amdgpu_put_xgmi_hive(hive); + goto init_failed; + } + + /* Drop the early temporary reset domain we created for device */ + amdgpu_reset_put_reset_domain(adev->reset_domain); + adev->reset_domain = hive->reset_domain; + amdgpu_put_xgmi_hive(hive); + } + } + } + + r = amdgpu_device_init_schedulers(adev); + if (r) + goto init_failed; /* Don't init kfd if whole hive need to be reset during init */ if (!adev->gmc.xgmi.pending_reset) @@ -2608,6 +2690,12 @@ static int amdgpu_device_ip_late_init(struct amdgpu_device *adev) adev->ip_blocks[i].status.late_initialized = true; } + r = amdgpu_ras_late_init(adev); + if (r) { + DRM_ERROR("amdgpu_ras_late_init failed %d", r); + return r; + } + amdgpu_ras_set_error_query_ready(adev, true); amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE); @@ -2622,7 +2710,7 @@ static int amdgpu_device_ip_late_init(struct amdgpu_device *adev) /* For passthrough configuration on arcturus and aldebaran, enable special handling SBR */ if (amdgpu_passthrough(adev) && ((adev->asic_type == CHIP_ARCTURUS && adev->gmc.xgmi.num_physical_nodes > 1)|| adev->asic_type == CHIP_ALDEBARAN )) - smu_handle_passthrough_sbr(&adev->smu, true); + amdgpu_dpm_handle_passthrough_sbr(adev, true); if (adev->gmc.xgmi.num_physical_nodes > 1) { mutex_lock(&mgpu_info.mutex); @@ -2706,11 +2794,11 @@ static int amdgpu_device_ip_fini_early(struct amdgpu_device *adev) } } - amdgpu_amdkfd_suspend(adev, false); - amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); + amdgpu_amdkfd_suspend(adev, false); + /* Workaroud for ASICs need to disable SMC first */ amdgpu_device_smu_fini_early(adev); @@ -2840,6 +2928,14 @@ static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev) amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); + /* + * Per PMFW team's suggestion, driver needs to handle gfxoff + * and df cstate features disablement for gpu reset(e.g. Mode1Reset) + * scenario. Add the missing df cstate disablement here. + */ + if (amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_DISALLOW)) + dev_warn(adev->dev, "Failed to disallow df cstate"); + for (i = adev->num_ip_blocks - 1; i >= 0; i--) { if (!adev->ip_blocks[i].status.valid) continue; @@ -2879,7 +2975,7 @@ static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev) int i, r; if (adev->in_s0ix) - amdgpu_gfx_state_change_set(adev, sGpuChangeState_D3Entry); + amdgpu_dpm_gfx_state_change(adev, sGpuChangeState_D3Entry); for (i = adev->num_ip_blocks - 1; i >= 0; i--) { if (!adev->ip_blocks[i].status.valid) @@ -2974,8 +3070,8 @@ static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev) int i, r; static enum amd_ip_block_type ip_order[] = { - AMD_IP_BLOCK_TYPE_GMC, AMD_IP_BLOCK_TYPE_COMMON, + AMD_IP_BLOCK_TYPE_GMC, AMD_IP_BLOCK_TYPE_PSP, AMD_IP_BLOCK_TYPE_IH, }; @@ -3066,7 +3162,8 @@ static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev) continue; if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || - adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) { + adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH || + (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP && amdgpu_sriov_vf(adev))) { r = adev->ip_blocks[i].version->funcs->resume(adev); if (r) { @@ -3113,6 +3210,15 @@ static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev) return r; } adev->ip_blocks[i].status.hw = true; + + if (adev->in_s0ix && adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) { + /* disable gfxoff for IP resume. The gfxoff will be re-enabled in + * amdgpu_device_resume() after IP resume. + */ + amdgpu_gfx_off_ctrl(adev, false); + DRM_DEBUG("will disable gfxoff for re-initializing other blocks\n"); + } + } return 0; @@ -3214,38 +3320,12 @@ bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type) case CHIP_MULLINS: /* * We have systems in the wild with these ASICs that require - * LVDS and VGA support which is not supported with DC. + * VGA support which is not supported with DC. * * Fallback to the non-DC driver here by default so as not to * cause regressions. */ return amdgpu_dc > 0; - case CHIP_HAWAII: - case CHIP_CARRIZO: - case CHIP_STONEY: - case CHIP_POLARIS10: - case CHIP_POLARIS11: - case CHIP_POLARIS12: - case CHIP_VEGAM: - case CHIP_TONGA: - case CHIP_FIJI: - case CHIP_VEGA10: - case CHIP_VEGA12: - case CHIP_VEGA20: -#if defined(CONFIG_DRM_AMD_DC_DCN) - case CHIP_RAVEN: - case CHIP_NAVI10: - case CHIP_NAVI14: - case CHIP_NAVI12: - case CHIP_RENOIR: - case CHIP_CYAN_SKILLFISH: - case CHIP_SIENNA_CICHLID: - case CHIP_NAVY_FLOUNDER: - case CHIP_DIMGREY_CAVEFISH: - case CHIP_BEIGE_GOBY: - case CHIP_VANGOGH: - case CHIP_YELLOW_CARP: -#endif default: return amdgpu_dc != 0; #else @@ -3267,7 +3347,7 @@ bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type) */ bool amdgpu_device_has_dc_support(struct amdgpu_device *adev) { - if (amdgpu_sriov_vf(adev) || + if (amdgpu_sriov_vf(adev) || adev->enable_virtual_display || (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK)) return false; @@ -3305,9 +3385,9 @@ static void amdgpu_device_xgmi_reset_func(struct work_struct *__work) if (adev->asic_reset_res) goto fail; - if (adev->mmhub.ras_funcs && - adev->mmhub.ras_funcs->reset_ras_error_count) - adev->mmhub.ras_funcs->reset_ras_error_count(adev); + if (adev->mmhub.ras && adev->mmhub.ras->ras_block.hw_ops && + adev->mmhub.ras->ras_block.hw_ops->reset_ras_error_count) + adev->mmhub.ras->ras_block.hw_ops->reset_ras_error_count(adev); } else { task_barrier_full(&hive->tb); @@ -3449,6 +3529,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, adev->gmc.gart_size = 512 * 1024 * 1024; adev->accel_working = false; adev->num_rings = 0; + RCU_INIT_POINTER(adev->gang_submit, dma_fence_get_stub()); adev->mman.buffer_funcs = NULL; adev->mman.buffer_funcs_ring = NULL; adev->vm_manager.vm_pte_funcs = NULL; @@ -3491,14 +3572,12 @@ int amdgpu_device_init(struct amdgpu_device *adev, mutex_init(&adev->mn_lock); mutex_init(&adev->virt.vf_errors.lock); hash_init(adev->mn_hash); - atomic_set(&adev->in_gpu_reset, 0); - init_rwsem(&adev->reset_sem); mutex_init(&adev->psp.mutex); mutex_init(&adev->notifier_lock); + mutex_init(&adev->pm.stable_pstate_ctx_lock); + mutex_init(&adev->benchmark_mutex); - r = amdgpu_device_init_apu_flags(adev); - if (r) - return r; + amdgpu_device_init_apu_flags(adev); r = amdgpu_device_check_arguments(adev); if (r) @@ -3519,6 +3598,8 @@ int amdgpu_device_init(struct amdgpu_device *adev, INIT_LIST_HEAD(&adev->reset_list); + INIT_LIST_HEAD(&adev->ras_list); + INIT_DELAYED_WORK(&adev->delayed_init_work, amdgpu_device_delayed_init_work_handler); INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work, @@ -3527,6 +3608,8 @@ int amdgpu_device_init(struct amdgpu_device *adev, INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func); adev->gfx.gfx_off_req_count = 1; + adev->gfx.gfx_off_residency = 0; + adev->gfx.gfx_off_entrycount = 0; adev->pm.ac_power = power_supply_is_system_supplied() > 0; atomic_set(&adev->throttling_logging_enabled, 1); @@ -3565,8 +3648,14 @@ int amdgpu_device_init(struct amdgpu_device *adev, if (amdgpu_mcbp) DRM_INFO("MCBP is enabled\n"); - if (amdgpu_mes && adev->asic_type >= CHIP_NAVI10) - adev->enable_mes = true; + /* + * Reset domain needs to be present early, before XGMI hive discovered + * (if any) and intitialized to use reset sem and in_gpu reset flag + * early on during init and before calling to RREG32. + */ + adev->reset_domain = amdgpu_reset_create_reset_domain(SINGLE_DEVICE, "amdgpu-reset-dev"); + if (!adev->reset_domain) + return -ENOMEM; /* detect hw virtualization here */ amdgpu_detect_virtualization(adev); @@ -3582,6 +3671,10 @@ int amdgpu_device_init(struct amdgpu_device *adev, if (r) return r; + /* Enable TMZ based on IP_VERSION */ + amdgpu_gmc_tmz_set(adev); + + amdgpu_gmc_noretry_set(adev); /* Need to get xgmi info early to decide the reset behavior*/ if (adev->gmc.xgmi.supported) { r = adev->gfxhub.funcs->get_xgmi_info(adev); @@ -3592,7 +3685,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, /* enable PCIE atomic ops */ if (amdgpu_sriov_vf(adev)) adev->have_atomics_support = ((struct amd_sriov_msg_pf2vf_info *) - adev->virt.fw_reserve.p_pf2vf)->pcie_atomic_ops_enabled_flags == + adev->virt.fw_reserve.p_pf2vf)->pcie_atomic_ops_support_flags == (PCI_EXP_DEVCAP2_ATOMIC_COMP32 | PCI_EXP_DEVCAP2_ATOMIC_COMP64); else adev->have_atomics_support = @@ -3749,18 +3842,13 @@ fence_driver_init: } else adev->ucode_sysfs_en = true; - if ((amdgpu_testing & 1)) { - if (adev->accel_working) - amdgpu_test_moves(adev); - else - DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n"); - } - if (amdgpu_benchmarking) { - if (adev->accel_working) - amdgpu_benchmark(adev, amdgpu_benchmarking); - else - DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n"); - } + r = amdgpu_psp_sysfs_init(adev); + if (r) { + adev->psp_sysfs_en = false; + if (!amdgpu_sriov_vf(adev)) + DRM_ERROR("Creating psp sysfs failed\n"); + } else + adev->psp_sysfs_en = true; /* * Register gpu instance before amdgpu_device_enable_mgpu_fan_boost. @@ -3833,6 +3921,7 @@ failed: static void amdgpu_device_unmap_mmio(struct amdgpu_device *adev) { + /* Clear all CPU mappings pointing to this device */ unmap_mapping_range(adev->ddev.anon_inode->i_mapping, 0, 0, 1); @@ -3864,10 +3953,6 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) { dev_info(adev->dev, "amdgpu: finishing device.\n"); flush_delayed_work(&adev->delayed_init_work); - if (adev->mman.initialized) { - flush_delayed_work(&adev->mman.bdev.wq); - ttm_bo_lock_delayed_workqueue(&adev->mman.bdev); - } adev->shutdown = true; /* make sure IB test finished before entering exclusive mode @@ -3888,10 +3973,17 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) } amdgpu_fence_driver_hw_fini(adev); + if (adev->mman.initialized) { + flush_delayed_work(&adev->mman.bdev.wq); + ttm_bo_lock_delayed_workqueue(&adev->mman.bdev); + } + if (adev->pm_sysfs_en) amdgpu_pm_sysfs_fini(adev); if (adev->ucode_sysfs_en) amdgpu_ucode_sysfs_fini(adev); + if (adev->psp_sysfs_en) + amdgpu_psp_sysfs_fini(adev); sysfs_remove_files(&adev->dev->kobj, amdgpu_dev_attributes); /* disable ras feature must before hw fini */ @@ -3906,18 +3998,20 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) amdgpu_gart_dummy_page_fini(adev); - if (drm_dev_is_unplugged(adev_to_drm(adev))) - amdgpu_device_unmap_mmio(adev); + amdgpu_device_unmap_mmio(adev); } void amdgpu_device_fini_sw(struct amdgpu_device *adev) { + int idx; + amdgpu_fence_driver_sw_fini(adev); amdgpu_device_ip_fini(adev); release_firmware(adev->firmware.gpu_info_fw); adev->firmware.gpu_info_fw = NULL; adev->accel_working = false; + dma_fence_put(rcu_dereference_protected(adev->gang_submit, true)); amdgpu_reset_fini(adev); @@ -3937,11 +4031,22 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev) if ((adev->pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) vga_client_unregister(adev->pdev); + if (drm_dev_enter(adev_to_drm(adev), &idx)) { + + iounmap(adev->rmmio); + adev->rmmio = NULL; + amdgpu_device_doorbell_fini(adev); + drm_dev_exit(idx); + } + if (IS_ENABLED(CONFIG_PERF_EVENTS)) amdgpu_pmu_fini(adev); if (adev->mman.discovery_bin) amdgpu_discovery_fini(adev); + amdgpu_reset_put_reset_domain(adev->reset_domain); + adev->reset_domain = NULL; + kfree(adev->pci_state); } @@ -3955,15 +4060,18 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev) * at suspend time. * */ -static void amdgpu_device_evict_resources(struct amdgpu_device *adev) +static int amdgpu_device_evict_resources(struct amdgpu_device *adev) { - /* No need to evict vram on APUs for suspend to ram */ - if (adev->in_s3 && (adev->flags & AMD_IS_APU)) - return; + int ret; - if (amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM)) - DRM_WARN("evicting device resources failed\n"); + /* No need to evict vram on APUs for suspend to ram or s2idle */ + if ((adev->in_s3 || adev->in_s0ix) && (adev->flags & AMD_IS_APU)) + return 0; + ret = amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM); + if (ret) + DRM_WARN("evicting device resources failed\n"); + return ret; } /* @@ -3982,12 +4090,20 @@ static void amdgpu_device_evict_resources(struct amdgpu_device *adev) int amdgpu_device_suspend(struct drm_device *dev, bool fbcon) { struct amdgpu_device *adev = drm_to_adev(dev); + int r = 0; if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) return 0; adev->in_suspend = true; + if (amdgpu_sriov_vf(adev)) { + amdgpu_virt_fini_data_exchange(adev); + r = amdgpu_virt_request_full_gpu(adev, false); + if (r) + return r; + } + if (amdgpu_acpi_smart_shift_update(dev, AMDGPU_SS_DEV_D3)) DRM_WARN("smart shift update failed\n"); @@ -4005,16 +4121,16 @@ int amdgpu_device_suspend(struct drm_device *dev, bool fbcon) if (!adev->in_s0ix) amdgpu_amdkfd_suspend(adev, adev->in_runpm); - /* First evict vram memory */ - amdgpu_device_evict_resources(adev); + r = amdgpu_device_evict_resources(adev); + if (r) + return r; amdgpu_fence_driver_hw_fini(adev); amdgpu_device_ip_suspend_phase2(adev); - /* This second call to evict device resources is to evict - * the gart page table using the CPU. - */ - amdgpu_device_evict_resources(adev); + + if (amdgpu_sriov_vf(adev)) + amdgpu_virt_release_full_gpu(adev, false); return 0; } @@ -4034,11 +4150,17 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon) struct amdgpu_device *adev = drm_to_adev(dev); int r = 0; + if (amdgpu_sriov_vf(adev)) { + r = amdgpu_virt_request_full_gpu(adev, true); + if (r) + return r; + } + if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) return 0; if (adev->in_s0ix) - amdgpu_gfx_state_change_set(adev, sGpuChangeState_D0Entry); + amdgpu_dpm_gfx_state_change(adev, sGpuChangeState_D0Entry); /* post card */ if (amdgpu_device_need_post(adev)) { @@ -4048,6 +4170,13 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon) } r = amdgpu_device_ip_resume(adev); + + /* no matter what r is, always need to properly release full GPU */ + if (amdgpu_sriov_vf(adev)) { + amdgpu_virt_init_data_exchange(adev); + amdgpu_virt_release_full_gpu(adev, true); + } + if (r) { dev_err(adev->dev, "amdgpu_device_ip_resume failed (%d).\n", r); return r; @@ -4070,6 +4199,13 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon) /* Make sure IB tests flushed */ flush_delayed_work(&adev->delayed_init_work); + if (adev->in_s0ix) { + /* re-enable gfxoff after IP resume. This re-enables gfxoff after + * it was disabled for IP resume in amdgpu_device_ip_resume_phase2(). + */ + amdgpu_gfx_off_ctrl(adev, true); + DRM_DEBUG("will enable gfxoff for the mission mode\n"); + } if (fbcon) drm_fb_helper_set_suspend_unlocked(adev_to_drm(adev)->fb_helper, false); @@ -4341,9 +4477,9 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, { int r; struct amdgpu_hive_info *hive = NULL; + int retry_limit = 0; - amdgpu_amdkfd_pre_reset(adev); - +retry: amdgpu_amdkfd_pre_reset(adev); if (from_hypervisor) @@ -4359,8 +4495,6 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, goto error; amdgpu_virt_init_data_exchange(adev); - /* we need recover gart prior to run SMC/CP/SDMA resume */ - amdgpu_gtt_mgr_recover(ttm_manager_type(&adev->mman.bdev, TTM_PL_TT)); r = amdgpu_device_fw_loading(adev); if (r) @@ -4382,6 +4516,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, if (!r) { amdgpu_irq_gpu_reset_resume_helper(adev); r = amdgpu_ib_ring_tests(adev); + amdgpu_amdkfd_post_reset(adev); } @@ -4392,6 +4527,14 @@ error: } amdgpu_virt_release_full_gpu(adev, true); + if (AMDGPU_RETRY_SRIOV_RESET(r)) { + if (retry_limit < AMDGPU_MAX_RETRY_LIMIT) { + retry_limit++; + goto retry; + } else + DRM_ERROR("GPU reset retry is beyond the retry limit\n"); + } + return r; } @@ -4433,46 +4576,38 @@ bool amdgpu_device_has_job_running(struct amdgpu_device *adev) */ bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev) { - if (!amdgpu_device_ip_check_soft_reset(adev)) { - dev_info(adev->dev, "Timeout, but no hardware hang detected.\n"); - return false; - } if (amdgpu_gpu_recovery == 0) goto disabled; + if (!amdgpu_device_ip_check_soft_reset(adev)) { + dev_info(adev->dev,"Timeout, but no hardware hang detected.\n"); + return false; + } + if (amdgpu_sriov_vf(adev)) return true; 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: - case CHIP_RAVEN: - case CHIP_ARCTURUS: - case CHIP_RENOIR: - case CHIP_NAVI10: - case CHIP_NAVI14: - case CHIP_NAVI12: - case CHIP_SIENNA_CICHLID: - case CHIP_NAVY_FLOUNDER: - case CHIP_DIMGREY_CAVEFISH: - case CHIP_BEIGE_GOBY: - case CHIP_VANGOGH: - case CHIP_ALDEBARAN: - break; - default: +#ifdef CONFIG_DRM_AMDGPU_SI + case CHIP_VERDE: + case CHIP_TAHITI: + case CHIP_PITCAIRN: + case CHIP_OLAND: + case CHIP_HAINAN: +#endif +#ifdef CONFIG_DRM_AMDGPU_CIK + case CHIP_KAVERI: + case CHIP_KABINI: + case CHIP_MULLINS: +#endif + case CHIP_CARRIZO: + case CHIP_STONEY: + case CHIP_CYAN_SKILLFISH: goto disabled; + default: + break; } } @@ -4539,6 +4674,8 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, amdgpu_virt_fini_data_exchange(adev); } + amdgpu_fence_driver_isr_toggle(adev, true); + /* block all schedulers and reset given job's ring */ for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { struct amdgpu_ring *ring = adev->rings[i]; @@ -4554,6 +4691,8 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, amdgpu_fence_driver_force_completion(ring); } + amdgpu_fence_driver_isr_toggle(adev, false); + if (job && job->vm) drm_sched_increase_karma(&job->base); @@ -4570,7 +4709,7 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, if (!need_full_reset) need_full_reset = amdgpu_device_ip_need_full_reset(adev); - if (!need_full_reset) { + if (!need_full_reset && amdgpu_gpu_recovery) { amdgpu_device_ip_pre_soft_reset(adev); r = amdgpu_device_ip_soft_reset(adev); amdgpu_device_ip_post_soft_reset(adev); @@ -4592,16 +4731,88 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, return r; } +static int amdgpu_reset_reg_dumps(struct amdgpu_device *adev) +{ + int i; + + lockdep_assert_held(&adev->reset_domain->sem); + + for (i = 0; i < adev->num_regs; i++) { + adev->reset_dump_reg_value[i] = RREG32(adev->reset_dump_reg_list[i]); + trace_amdgpu_reset_reg_dumps(adev->reset_dump_reg_list[i], + adev->reset_dump_reg_value[i]); + } + + return 0; +} + +#ifdef CONFIG_DEV_COREDUMP +static ssize_t amdgpu_devcoredump_read(char *buffer, loff_t offset, + size_t count, void *data, size_t datalen) +{ + struct drm_printer p; + struct amdgpu_device *adev = data; + struct drm_print_iterator iter; + int i; + + iter.data = buffer; + iter.offset = 0; + iter.start = offset; + iter.remain = count; + + p = drm_coredump_printer(&iter); + + drm_printf(&p, "**** AMDGPU Device Coredump ****\n"); + drm_printf(&p, "kernel: " UTS_RELEASE "\n"); + drm_printf(&p, "module: " KBUILD_MODNAME "\n"); + drm_printf(&p, "time: %lld.%09ld\n", adev->reset_time.tv_sec, adev->reset_time.tv_nsec); + if (adev->reset_task_info.pid) + drm_printf(&p, "process_name: %s PID: %d\n", + adev->reset_task_info.process_name, + adev->reset_task_info.pid); + + if (adev->reset_vram_lost) + drm_printf(&p, "VRAM is lost due to GPU reset!\n"); + if (adev->num_regs) { + drm_printf(&p, "AMDGPU register dumps:\nOffset: Value:\n"); + + for (i = 0; i < adev->num_regs; i++) + drm_printf(&p, "0x%08x: 0x%08x\n", + adev->reset_dump_reg_list[i], + adev->reset_dump_reg_value[i]); + } + + return count - iter.remain; +} + +static void amdgpu_devcoredump_free(void *data) +{ +} + +static void amdgpu_reset_capture_coredumpm(struct amdgpu_device *adev) +{ + struct drm_device *dev = adev_to_drm(adev); + + ktime_get_ts64(&adev->reset_time); + dev_coredumpm(dev->dev, THIS_MODULE, adev, 0, GFP_KERNEL, + amdgpu_devcoredump_read, amdgpu_devcoredump_free); +} +#endif + int amdgpu_do_asic_reset(struct list_head *device_list_handle, struct amdgpu_reset_context *reset_context) { struct amdgpu_device *tmp_adev = NULL; bool need_full_reset, skip_hw_reset, vram_lost = false; int r = 0; + bool gpu_reset_for_dev_remove = 0; /* Try reset handler method first */ tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, reset_list); + amdgpu_reset_reg_dumps(tmp_adev); + + reset_context->reset_device_list = device_list_handle; r = amdgpu_reset_perform_reset(tmp_adev, reset_context); /* If reset handler not implemented, continue; otherwise return */ if (r == -ENOSYS) @@ -4614,6 +4825,10 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle, test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); skip_hw_reset = test_bit(AMDGPU_SKIP_HW_RESET, &reset_context->flags); + gpu_reset_for_dev_remove = + test_bit(AMDGPU_RESET_FOR_DEVICE_REMOVE, &reset_context->flags) && + test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); + /* * ASIC reset has to be done on all XGMI hive nodes ASAP * to allow proper links negotiation in FW (within 1 sec) @@ -4650,14 +4865,26 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle, if (!r && amdgpu_ras_intr_triggered()) { list_for_each_entry(tmp_adev, device_list_handle, reset_list) { - if (tmp_adev->mmhub.ras_funcs && - tmp_adev->mmhub.ras_funcs->reset_ras_error_count) - tmp_adev->mmhub.ras_funcs->reset_ras_error_count(tmp_adev); + if (tmp_adev->mmhub.ras && tmp_adev->mmhub.ras->ras_block.hw_ops && + tmp_adev->mmhub.ras->ras_block.hw_ops->reset_ras_error_count) + tmp_adev->mmhub.ras->ras_block.hw_ops->reset_ras_error_count(tmp_adev); } amdgpu_ras_intr_cleared(); } + /* Since the mode1 reset affects base ip blocks, the + * phase1 ip blocks need to be resumed. Otherwise there + * will be a BIOS signature error and the psp bootloader + * can't load kdb on the next amdgpu install. + */ + if (gpu_reset_for_dev_remove) { + list_for_each_entry(tmp_adev, device_list_handle, reset_list) + amdgpu_device_ip_resume_phase1(tmp_adev); + + goto end; + } + list_for_each_entry(tmp_adev, device_list_handle, reset_list) { if (need_full_reset) { /* post card */ @@ -4675,15 +4902,20 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle, goto out; vram_lost = amdgpu_device_check_vram_lost(tmp_adev); +#ifdef CONFIG_DEV_COREDUMP + tmp_adev->reset_vram_lost = vram_lost; + memset(&tmp_adev->reset_task_info, 0, + sizeof(tmp_adev->reset_task_info)); + if (reset_context->job && reset_context->job->vm) + tmp_adev->reset_task_info = + reset_context->job->vm->task_info; + amdgpu_reset_capture_coredumpm(tmp_adev); +#endif if (vram_lost) { DRM_INFO("VRAM is lost due to GPU reset!\n"); amdgpu_inc_vram_lost(tmp_adev); } - r = amdgpu_gtt_mgr_recover(ttm_manager_type(&tmp_adev->mman.bdev, TTM_PL_TT)); - if (r) - goto out; - r = amdgpu_device_fw_loading(tmp_adev); if (r) return r; @@ -4763,17 +4995,8 @@ end: return r; } -static bool amdgpu_device_lock_adev(struct amdgpu_device *adev, - struct amdgpu_hive_info *hive) +static void amdgpu_device_set_mp1_state(struct amdgpu_device *adev) { - if (atomic_cmpxchg(&adev->in_gpu_reset, 0, 1) != 0) - return false; - - if (hive) { - down_write_nest_lock(&adev->reset_sem, &hive->hive_lock); - } else { - down_write(&adev->reset_sem); - } switch (amdgpu_asic_reset_method(adev)) { case AMD_RESET_METHOD_MODE1: @@ -4786,56 +5009,12 @@ static bool amdgpu_device_lock_adev(struct amdgpu_device *adev, adev->mp1_state = PP_MP1_STATE_NONE; break; } - - return true; } -static void amdgpu_device_unlock_adev(struct amdgpu_device *adev) +static void amdgpu_device_unset_mp1_state(struct amdgpu_device *adev) { amdgpu_vf_error_trans_all(adev); adev->mp1_state = PP_MP1_STATE_NONE; - atomic_set(&adev->in_gpu_reset, 0); - up_write(&adev->reset_sem); -} - -/* - * to lockup a list of amdgpu devices in a hive safely, if not a hive - * with multiple nodes, it will be similar as amdgpu_device_lock_adev. - * - * unlock won't require roll back. - */ -static int amdgpu_device_lock_hive_adev(struct amdgpu_device *adev, struct amdgpu_hive_info *hive) -{ - struct amdgpu_device *tmp_adev = NULL; - - if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1)) { - if (!hive) { - dev_err(adev->dev, "Hive is NULL while device has multiple xgmi nodes"); - return -ENODEV; - } - list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) { - if (!amdgpu_device_lock_adev(tmp_adev, hive)) - goto roll_back; - } - } else if (!amdgpu_device_lock_adev(adev, hive)) - return -EAGAIN; - - return 0; -roll_back: - if (!list_is_first(&tmp_adev->gmc.xgmi.head, &hive->device_list)) { - /* - * if the lockup iteration break in the middle of a hive, - * it may means there may has a race issue, - * or a hive device locked up independently. - * we may be in trouble and may not, so will try to roll back - * the lock and give out a warnning. - */ - dev_warn(tmp_adev->dev, "Hive lock iteration broke in the middle. Rolling back to unlock"); - list_for_each_entry_continue_reverse(tmp_adev, &hive->device_list, gmc.xgmi.head) { - amdgpu_device_unlock_adev(tmp_adev); - } - } - return -EAGAIN; } static void amdgpu_device_resume_display_audio(struct amdgpu_device *adev) @@ -4917,18 +5096,35 @@ static void amdgpu_device_recheck_guilty_jobs( /* clear job's guilty and depend the folowing step to decide the real one */ drm_sched_reset_karma(s_job); - /* for the real bad job, it will be resubmitted twice, adding a dma_fence_get - * to make sure fence is balanced */ - dma_fence_get(s_job->s_fence->parent); drm_sched_resubmit_jobs_ext(&ring->sched, 1); + if (!s_job->s_fence->parent) { + DRM_WARN("Failed to get a HW fence for job!"); + continue; + } + ret = dma_fence_wait_timeout(s_job->s_fence->parent, false, ring->sched.timeout); if (ret == 0) { /* timeout */ DRM_ERROR("Found the real bad job! ring:%s, job_id:%llx\n", ring->sched.name, s_job->id); + + amdgpu_fence_driver_isr_toggle(adev, true); + + /* Clear this failed job from fence array */ + amdgpu_fence_driver_clear_job_fences(ring); + + amdgpu_fence_driver_isr_toggle(adev, false); + + /* Since the job won't signal and we go for + * another resubmit drop this parent pointer + */ + dma_fence_put(s_job->s_fence->parent); + s_job->s_fence->parent = NULL; + /* set guilty */ drm_sched_increase_karma(s_job); + amdgpu_reset_prepare_hwcontext(adev, reset_context); retry: /* do hw reset */ if (amdgpu_sriov_vf(adev)) { @@ -4955,7 +5151,6 @@ retry: /* got the hw fence, signal finished fence */ atomic_dec(ring->sched.score); - dma_fence_put(s_job->s_fence->parent); dma_fence_get(&s_job->s_fence->finished); dma_fence_signal(&s_job->s_fence->finished); dma_fence_put(&s_job->s_fence->finished); @@ -4968,6 +5163,27 @@ retry: } } +static inline void amdgpu_device_stop_pending_resets(struct amdgpu_device *adev) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + +#if defined(CONFIG_DEBUG_FS) + if (!amdgpu_sriov_vf(adev)) + cancel_work(&adev->reset_work); +#endif + + if (adev->kfd.dev) + cancel_work(&adev->kfd.reset_work); + + if (amdgpu_sriov_vf(adev)) + cancel_work(&adev->virt.flr_work); + + if (con && adev->ras_enabled) + cancel_work(&con->recovery_work); + +} + + /** * amdgpu_device_gpu_recover - reset the asic and recover scheduler * @@ -4980,7 +5196,8 @@ retry: */ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, - struct amdgpu_job *job) + struct amdgpu_job *job, + struct amdgpu_reset_context *reset_context) { struct list_head device_list, *device_list_handle = NULL; bool job_signaled = false; @@ -4990,9 +5207,11 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, bool need_emergency_restart = false; bool audio_suspended = false; int tmp_vram_lost_counter; - struct amdgpu_reset_context reset_context; + bool gpu_reset_for_dev_remove = false; - memset(&reset_context, 0, sizeof(reset_context)); + gpu_reset_for_dev_remove = + test_bit(AMDGPU_RESET_FOR_DEVICE_REMOVE, &reset_context->flags) && + test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); /* * Special case: RAS triggered and full reset isn't supported @@ -5013,49 +5232,13 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, dev_info(adev->dev, "GPU %s begin!\n", need_emergency_restart ? "jobs stop":"reset"); - /* - * Here we trylock to avoid chain of resets executing from - * either trigger by jobs on different adevs in XGMI hive or jobs on - * different schedulers for same device while this TO handler is running. - * We always reset all schedulers for device and all devices for XGMI - * hive so that should take care of them too. - */ if (!amdgpu_sriov_vf(adev)) hive = amdgpu_get_xgmi_hive(adev); - if (hive) { - if (atomic_cmpxchg(&hive->in_reset, 0, 1) != 0) { - DRM_INFO("Bailing on TDR for s_job:%llx, hive: %llx as another already in progress", - job ? job->base.id : -1, hive->hive_id); - amdgpu_put_xgmi_hive(hive); - if (job && job->vm) - drm_sched_increase_karma(&job->base); - return 0; - } + if (hive) mutex_lock(&hive->hive_lock); - } - - reset_context.method = AMD_RESET_METHOD_NONE; - reset_context.reset_req_dev = adev; - reset_context.job = job; - reset_context.hive = hive; - clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); - - /* - * lock the device before we try to operate the linked list - * if didn't get the device lock, don't touch the linked list since - * others may iterating it. - */ - r = amdgpu_device_lock_hive_adev(adev, hive); - if (r) { - dev_info(adev->dev, "Bailing on TDR for s_job:%llx, as another already in progress", - job ? job->base.id : -1); - - /* even we skipped this reset, still need to set the job to guilty */ - if (job && job->vm) - drm_sched_increase_karma(&job->base); - goto skip_recovery; - } + reset_context->job = job; + reset_context->hive = hive; /* * Build list of devices to reset. * In case we are in XGMI hive mode, resort the device list @@ -5063,8 +5246,11 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, */ INIT_LIST_HEAD(&device_list); if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1)) { - list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) + list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) { list_add_tail(&tmp_adev->reset_list, &device_list); + if (gpu_reset_for_dev_remove && adev->shutdown) + tmp_adev->shutdown = true; + } if (!list_is_first(&adev->reset_list, &device_list)) list_rotate_to_front(&adev->reset_list, &device_list); device_list_handle = &device_list; @@ -5073,8 +5259,16 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, device_list_handle = &device_list; } + /* We need to lock reset domain only once both for XGMI and single device */ + tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, + reset_list); + amdgpu_device_lock_reset_domain(tmp_adev->reset_domain); + /* block all schedulers and reset given job's ring */ list_for_each_entry(tmp_adev, device_list_handle, reset_list) { + + amdgpu_device_set_mp1_state(tmp_adev); + /* * Try to put the audio codec into suspend state * before gpu reset started. @@ -5101,7 +5295,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, */ amdgpu_unregister_gpu_instance(tmp_adev); - drm_fb_helper_set_suspend_unlocked(adev_to_drm(adev)->fb_helper, true); + drm_fb_helper_set_suspend_unlocked(adev_to_drm(tmp_adev)->fb_helper, true); /* disable ras on ALL IPs */ if (!need_emergency_restart && @@ -5131,8 +5325,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, * * job->base holds a reference to parent fence */ - if (job && job->base.s_fence->parent && - dma_fence_is_signaled(job->base.s_fence->parent)) { + if (job && dma_fence_is_signaled(&job->hw_fence)) { job_signaled = true; dev_info(adev->dev, "Guilty job already signaled, skipping HW reset"); goto skip_hw_reset; @@ -5140,13 +5333,23 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, retry: /* Rest of adevs pre asic reset from XGMI hive. */ list_for_each_entry(tmp_adev, device_list_handle, reset_list) { - r = amdgpu_device_pre_asic_reset(tmp_adev, &reset_context); + if (gpu_reset_for_dev_remove) { + /* Workaroud for ASICs need to disable SMC first */ + amdgpu_device_smu_fini_early(tmp_adev); + } + r = amdgpu_device_pre_asic_reset(tmp_adev, reset_context); /*TODO Should we stop ?*/ if (r) { dev_err(tmp_adev->dev, "GPU pre asic reset failed with err, %d for drm dev, %s ", r, adev_to_drm(tmp_adev)->unique); tmp_adev->asic_reset_res = r; } + + /* + * Drop all pending non scheduler resets. Scheduler resets + * were already dropped during drm_sched_stop + */ + amdgpu_device_stop_pending_resets(tmp_adev); } tmp_vram_lost_counter = atomic_read(&((adev)->vram_lost_counter)); @@ -5156,10 +5359,17 @@ retry: /* Rest of adevs pre asic reset from XGMI hive. */ r = amdgpu_device_reset_sriov(adev, job ? false : true); if (r) adev->asic_reset_res = r; + + /* Aldebaran supports ras in SRIOV, so need resume ras during reset */ + if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)) + amdgpu_ras_resume(adev); } else { - r = amdgpu_do_asic_reset(device_list_handle, &reset_context); + r = amdgpu_do_asic_reset(device_list_handle, reset_context); if (r && r == -EAGAIN) goto retry; + + if (!r && gpu_reset_for_dev_remove) + goto recover_end; } skip_hw_reset: @@ -5177,7 +5387,7 @@ skip_hw_reset: if (amdgpu_gpu_recovery == 2 && !(tmp_vram_lost_counter < atomic_read(&adev->vram_lost_counter))) amdgpu_device_recheck_guilty_jobs( - tmp_adev, device_list_handle, &reset_context); + tmp_adev, device_list_handle, reset_context); for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { struct amdgpu_ring *ring = tmp_adev->rings[i]; @@ -5192,10 +5402,16 @@ skip_hw_reset: drm_sched_start(&ring->sched, !tmp_adev->asic_reset_res); } + if (adev->enable_mes && adev->ip_versions[GC_HWIP][0] != IP_VERSION(11, 0, 3)) + amdgpu_mes_self_test(tmp_adev); + if (!drm_drv_uses_atomic_modeset(adev_to_drm(tmp_adev)) && !job_signaled) { drm_helper_resume_force_mode(adev_to_drm(tmp_adev)); } + if (tmp_adev->asic_reset_res) + r = tmp_adev->asic_reset_res; + tmp_adev->asic_reset_res = 0; if (r) { @@ -5223,18 +5439,24 @@ skip_sched_resume: if (audio_suspended) amdgpu_device_resume_display_audio(tmp_adev); - amdgpu_device_unlock_adev(tmp_adev); + + amdgpu_device_unset_mp1_state(tmp_adev); } -skip_recovery: +recover_end: + tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, + reset_list); + amdgpu_device_unlock_reset_domain(tmp_adev->reset_domain); + if (hive) { - atomic_set(&hive->in_reset, 0); mutex_unlock(&hive->hive_lock); amdgpu_put_xgmi_hive(hive); } - if (r && r != -EAGAIN) + if (r) dev_info(adev->dev, "GPU reset end with ret = %d\n", r); + + atomic_set(&adev->reset_domain->reset_res, r); return r; } @@ -5386,6 +5608,37 @@ static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev) } } +/** + * amdgpu_device_is_peer_accessible - Check peer access through PCIe BAR + * + * @adev: amdgpu_device pointer + * @peer_adev: amdgpu_device pointer for peer device trying to access @adev + * + * Return true if @peer_adev can access (DMA) @adev through the PCIe + * BAR, i.e. @adev is "large BAR" and the BAR matches the DMA mask of + * @peer_adev. + */ +bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev, + struct amdgpu_device *peer_adev) +{ +#ifdef CONFIG_HSA_AMD_P2P + uint64_t address_mask = peer_adev->dev->dma_mask ? + ~*peer_adev->dev->dma_mask : ~((1ULL << 32) - 1); + resource_size_t aper_limit = + adev->gmc.aper_base + adev->gmc.aper_size - 1; + bool p2p_access = + !adev->gmc.xgmi.connected_to_cpu && + !(pci_p2pdma_distance(adev->pdev, peer_adev->dev, false) < 0); + + return pcie_p2p && p2p_access && (adev->gmc.visible_vram_size && + adev->gmc.real_vram_size == adev->gmc.visible_vram_size && + !(adev->gmc.aper_base & address_mask || + aper_limit & address_mask)); +#else + return false; +#endif +} + int amdgpu_device_baco_enter(struct drm_device *dev) { struct amdgpu_device *adev = drm_to_adev(dev); @@ -5425,20 +5678,6 @@ int amdgpu_device_baco_exit(struct drm_device *dev) return 0; } -static void amdgpu_cancel_all_tdr(struct amdgpu_device *adev) -{ - int i; - - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - struct amdgpu_ring *ring = adev->rings[i]; - - if (!ring || !ring->sched.thread) - continue; - - cancel_delayed_work_sync(&ring->sched.work_tdr); - } -} - /** * amdgpu_pci_error_detected - Called when a PCI error is detected. * @pdev: PCI device struct @@ -5469,14 +5708,11 @@ pci_ers_result_t amdgpu_pci_error_detected(struct pci_dev *pdev, pci_channel_sta /* Fatal error, prepare for slot reset */ case pci_channel_io_frozen: /* - * Cancel and wait for all TDRs in progress if failing to - * set adev->in_gpu_reset in amdgpu_device_lock_adev - * - * Locking adev->reset_sem will prevent any external access + * Locking adev->reset_domain->sem will prevent any external access * to GPU during PCI error recovery */ - while (!amdgpu_device_lock_adev(adev, NULL)) - amdgpu_cancel_all_tdr(adev); + amdgpu_device_lock_reset_domain(adev->reset_domain); + amdgpu_device_set_mp1_state(adev); /* * Block any work scheduling as we do for regular GPU reset @@ -5583,7 +5819,8 @@ out: DRM_INFO("PCIe error recovery succeeded\n"); } else { DRM_ERROR("PCIe error recovery failed, err:%d", r); - amdgpu_device_unlock_adev(adev); + amdgpu_device_unset_mp1_state(adev); + amdgpu_device_unlock_reset_domain(adev->reset_domain); } return r ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; @@ -5620,7 +5857,8 @@ void amdgpu_pci_resume(struct pci_dev *pdev) drm_sched_start(&ring->sched, true); } - amdgpu_device_unlock_adev(adev); + amdgpu_device_unset_mp1_state(adev); + amdgpu_device_unlock_reset_domain(adev->reset_domain); } bool amdgpu_device_cache_pci_state(struct pci_dev *pdev) @@ -5672,7 +5910,7 @@ void amdgpu_device_flush_hdp(struct amdgpu_device *adev, struct amdgpu_ring *ring) { #ifdef CONFIG_X86_64 - if (adev->flags & AMD_IS_APU) + if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev)) return; #endif if (adev->gmc.xgmi.connected_to_cpu) @@ -5688,7 +5926,7 @@ void amdgpu_device_invalidate_hdp(struct amdgpu_device *adev, struct amdgpu_ring *ring) { #ifdef CONFIG_X86_64 - if (adev->flags & AMD_IS_APU) + if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev)) return; #endif if (adev->gmc.xgmi.connected_to_cpu) @@ -5697,6 +5935,11 @@ void amdgpu_device_invalidate_hdp(struct amdgpu_device *adev, amdgpu_asic_invalidate_hdp(adev, ring); } +int amdgpu_in_reset(struct amdgpu_device *adev) +{ + return atomic_read(&adev->reset_domain->in_gpu_reset); + } + /** * amdgpu_device_halt() - bring hardware to some kind of halt state * @@ -5735,3 +5978,69 @@ void amdgpu_device_halt(struct amdgpu_device *adev) pci_disable_device(pdev); pci_wait_for_pending_transaction(pdev); } + +u32 amdgpu_device_pcie_port_rreg(struct amdgpu_device *adev, + u32 reg) +{ + unsigned long flags, address, data; + u32 r; + + address = adev->nbio.funcs->get_pcie_port_index_offset(adev); + data = adev->nbio.funcs->get_pcie_port_data_offset(adev); + + spin_lock_irqsave(&adev->pcie_idx_lock, flags); + WREG32(address, reg * 4); + (void)RREG32(address); + r = RREG32(data); + spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); + return r; +} + +void amdgpu_device_pcie_port_wreg(struct amdgpu_device *adev, + u32 reg, u32 v) +{ + unsigned long flags, address, data; + + address = adev->nbio.funcs->get_pcie_port_index_offset(adev); + data = adev->nbio.funcs->get_pcie_port_data_offset(adev); + + spin_lock_irqsave(&adev->pcie_idx_lock, flags); + WREG32(address, reg * 4); + (void)RREG32(address); + WREG32(data, v); + (void)RREG32(data); + spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); +} + +/** + * amdgpu_device_switch_gang - switch to a new gang + * @adev: amdgpu_device pointer + * @gang: the gang to switch to + * + * Try to switch to a new gang. + * Returns: NULL if we switched to the new gang or a reference to the current + * gang leader. + */ +struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev, + struct dma_fence *gang) +{ + struct dma_fence *old = NULL; + + do { + dma_fence_put(old); + rcu_read_lock(); + old = dma_fence_get_rcu_safe(&adev->gang_submit); + rcu_read_unlock(); + + if (old == gang) + break; + + if (!dma_fence_is_signaled(old)) + return old; + + } while (cmpxchg((struct dma_fence __force **)&adev->gang_submit, + old, gang) != old); + + dma_fence_put(old); + return NULL; +} |