diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c | 20 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h | 7 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 6 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 3 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 67 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c | 23 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 15 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 166 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 116 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 106 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c | 133 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 129 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 9 |
18 files changed, 742 insertions, 79 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 7fe41a3c2541..e095572458cd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1164,6 +1164,7 @@ struct amdgpu_device { bool debug_disable_soft_recovery; bool debug_use_vram_fw_buf; bool debug_enable_ras_aca; + bool debug_exp_resets; bool enforce_isolation[MAX_XCP]; /* Added this mutex for cleaner shader isolation between GFX and compute processes */ 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 c63528a4e894..1254a43ec96b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c @@ -1151,6 +1151,10 @@ uint64_t kgd_gfx_v9_hqd_get_pq_addr(struct amdgpu_device *adev, uint32_t low, high; uint64_t queue_addr = 0; + if (!adev->debug_exp_resets && + !adev->gfx.num_gfx_rings) + return 0; + kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst); amdgpu_gfx_rlc_enter_safe_mode(adev, inst); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c index cf2b4dd4d865..5ac59b62020c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c @@ -28,8 +28,8 @@ #include "atom.h" #ifndef CONFIG_DEV_COREDUMP -void amdgpu_coredump(struct amdgpu_device *adev, bool vram_lost, - struct amdgpu_reset_context *reset_context) +void amdgpu_coredump(struct amdgpu_device *adev, bool skip_vram_check, + bool vram_lost, struct amdgpu_job *job) { } #else @@ -315,7 +315,9 @@ amdgpu_devcoredump_read(char *buffer, loff_t offset, size_t count, } } - if (coredump->reset_vram_lost) + if (coredump->skip_vram_check) + drm_printf(&p, "VRAM lost check is skipped!\n"); + else if (coredump->reset_vram_lost) drm_printf(&p, "VRAM is lost due to GPU reset!\n"); return count - iter.remain; @@ -326,12 +328,11 @@ static void amdgpu_devcoredump_free(void *data) kfree(data); } -void amdgpu_coredump(struct amdgpu_device *adev, bool vram_lost, - struct amdgpu_reset_context *reset_context) +void amdgpu_coredump(struct amdgpu_device *adev, bool skip_vram_check, + bool vram_lost, struct amdgpu_job *job) { - struct amdgpu_coredump_info *coredump; struct drm_device *dev = adev_to_drm(adev); - struct amdgpu_job *job = reset_context->job; + struct amdgpu_coredump_info *coredump; struct drm_sched_job *s_job; coredump = kzalloc(sizeof(*coredump), GFP_NOWAIT); @@ -341,11 +342,12 @@ void amdgpu_coredump(struct amdgpu_device *adev, bool vram_lost, return; } + coredump->skip_vram_check = skip_vram_check; coredump->reset_vram_lost = vram_lost; - if (reset_context->job && reset_context->job->vm) { + if (job && job->vm) { + struct amdgpu_vm *vm = job->vm; struct amdgpu_task_info *ti; - struct amdgpu_vm *vm = reset_context->job->vm; ti = amdgpu_vm_get_task_info_vm(vm); if (ti) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h index 52459512cb2b..ef9772c6bcc9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h @@ -26,7 +26,6 @@ #define __AMDGPU_DEV_COREDUMP_H__ #include "amdgpu.h" -#include "amdgpu_reset.h" #ifdef CONFIG_DEV_COREDUMP @@ -36,12 +35,12 @@ struct amdgpu_coredump_info { struct amdgpu_device *adev; struct amdgpu_task_info reset_task_info; struct timespec64 reset_time; + bool skip_vram_check; bool reset_vram_lost; struct amdgpu_ring *ring; }; #endif -void amdgpu_coredump(struct amdgpu_device *adev, bool vram_lost, - struct amdgpu_reset_context *reset_context); - +void amdgpu_coredump(struct amdgpu_device *adev, bool skip_vram_check, + bool vram_lost, struct amdgpu_job *job); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 49ef22dcf7fb..f4628412dac4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4531,6 +4531,9 @@ 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) + drain_workqueue(adev->mman.bdev.wq); adev->shutdown = true; /* make sure IB test finished before entering exclusive mode @@ -4551,9 +4554,6 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) } amdgpu_fence_driver_hw_fini(adev); - if (adev->mman.initialized) - drain_workqueue(adev->mman.bdev.wq); - if (adev->pm.sysfs_initialized) amdgpu_pm_sysfs_fini(adev); if (adev->ucode_sysfs_en) @@ -5489,7 +5489,7 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle, vram_lost = amdgpu_device_check_vram_lost(tmp_adev); if (!test_bit(AMDGPU_SKIP_COREDUMP, &reset_context->flags)) - amdgpu_coredump(tmp_adev, vram_lost, reset_context); + amdgpu_coredump(tmp_adev, false, vram_lost, reset_context->job); if (vram_lost) { DRM_INFO("VRAM is lost due to GPU reset!\n"); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 5dd39e6c6223..8dee7c62c801 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -131,6 +131,7 @@ enum AMDGPU_DEBUG_MASK { AMDGPU_DEBUG_DISABLE_GPU_SOFT_RECOVERY = BIT(2), AMDGPU_DEBUG_USE_VRAM_FW_BUF = BIT(3), AMDGPU_DEBUG_ENABLE_RAS_ACA = BIT(4), + AMDGPU_DEBUG_ENABLE_EXP_RESETS = BIT(5), }; unsigned int amdgpu_vram_limit = UINT_MAX; @@ -2199,6 +2200,11 @@ static void amdgpu_init_debug_options(struct amdgpu_device *adev) pr_info("debug: enable RAS ACA\n"); adev->debug_enable_ras_aca = true; } + + if (amdgpu_debug_mask & AMDGPU_DEBUG_ENABLE_EXP_RESETS) { + pr_info("debug: enable experimental reset features\n"); + adev->debug_exp_resets = true; + } } static unsigned long amdgpu_fix_asic_type(struct pci_dev *pdev, unsigned long flags) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index aad2027e5c7c..0e617dff8765 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -348,6 +348,9 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, return -EINVAL; } + /* always clear VRAM */ + flags |= AMDGPU_GEM_CREATE_VRAM_CLEARED; + /* create a gem object to contain this object in */ if (args->in.domains & (AMDGPU_GEM_DOMAIN_GDS | AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA)) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index b4efeef848de..b779d47a546a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -660,7 +660,7 @@ int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev, int xcc_id) uint64_t queue_mask = 0; int r, i, j; - if (adev->enable_mes) + if (adev->mes.enable_legacy_queue_map) return amdgpu_gfx_mes_enable_kcq(adev, xcc_id); if (!kiq->pmf || !kiq->pmf->kiq_map_queues || !kiq->pmf->kiq_set_resources) @@ -722,7 +722,7 @@ int amdgpu_gfx_enable_kgq(struct amdgpu_device *adev, int xcc_id) amdgpu_device_flush_hdp(adev, NULL); - if (adev->enable_mes) { + if (adev->mes.enable_legacy_queue_map) { for (i = 0; i < adev->gfx.num_gfx_rings; i++) { j = i + xcc_id * adev->gfx.num_gfx_rings; r = amdgpu_mes_map_legacy_queue(adev, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 597489dea114..ad6bf5d4e0a9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -30,6 +30,60 @@ #include "amdgpu.h" #include "amdgpu_trace.h" #include "amdgpu_reset.h" +#include "amdgpu_dev_coredump.h" +#include "amdgpu_xgmi.h" + +static void amdgpu_job_do_core_dump(struct amdgpu_device *adev, + struct amdgpu_job *job) +{ + int i; + + dev_info(adev->dev, "Dumping IP State\n"); + for (i = 0; i < adev->num_ip_blocks; i++) + if (adev->ip_blocks[i].version->funcs->dump_ip_state) + adev->ip_blocks[i].version->funcs + ->dump_ip_state((void *)adev); + dev_info(adev->dev, "Dumping IP State Completed\n"); + + amdgpu_coredump(adev, true, false, job); +} + +static void amdgpu_job_core_dump(struct amdgpu_device *adev, + struct amdgpu_job *job) +{ + struct list_head device_list, *device_list_handle = NULL; + struct amdgpu_device *tmp_adev = NULL; + struct amdgpu_hive_info *hive = NULL; + + if (!amdgpu_sriov_vf(adev)) + hive = amdgpu_get_xgmi_hive(adev); + if (hive) + mutex_lock(&hive->hive_lock); + /* + * Reuse the logic in amdgpu_device_gpu_recover() to build list of + * devices for code dump + */ + INIT_LIST_HEAD(&device_list); + if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1) && hive) { + list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) + list_add_tail(&tmp_adev->reset_list, &device_list); + if (!list_is_first(&adev->reset_list, &device_list)) + list_rotate_to_front(&adev->reset_list, &device_list); + device_list_handle = &device_list; + } else { + list_add_tail(&adev->reset_list, &device_list); + device_list_handle = &device_list; + } + + /* Do the coredump for each device */ + list_for_each_entry(tmp_adev, device_list_handle, reset_list) + amdgpu_job_do_core_dump(tmp_adev, job); + + if (hive) { + mutex_unlock(&hive->hive_lock); + amdgpu_put_xgmi_hive(hive); + } +} static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job) { @@ -48,9 +102,14 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job) return DRM_GPU_SCHED_STAT_ENODEV; } - adev->job_hang = true; + /* + * Do the coredump immediately after a job timeout to get a very + * close dump/snapshot/representation of GPU's current error status + */ + amdgpu_job_core_dump(adev, job); + if (amdgpu_gpu_recovery && amdgpu_ring_soft_recovery(ring, job->vmid, s_job->s_fence->parent)) { dev_err(adev->dev, "ring %s timeout, but soft recovered\n", @@ -101,6 +160,12 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job) reset_context.src = AMDGPU_RESET_SRC_JOB; clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); + /* + * To avoid an unnecessary extra coredump, as we have already + * got the very close representation of GPU's error status + */ + set_bit(AMDGPU_SKIP_COREDUMP, &reset_context.flags); + r = amdgpu_device_gpu_recover(ring->adev, job, &reset_context); if (r) dev_err(adev->dev, "GPU Recovery Failed: %d\n", r); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c index 44c74a08987d..f7d5d4f08a53 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c @@ -826,6 +826,24 @@ int amdgpu_mes_reset_hw_queue(struct amdgpu_device *adev, int queue_id) return 0; } +int amdgpu_mes_reset_hw_queue_mmio(struct amdgpu_device *adev, int queue_type, + int me_id, int pipe_id, int queue_id, int vmid) +{ + struct mes_reset_queue_input queue_input; + int r; + + queue_input.use_mmio = true; + queue_input.me_id = me_id; + queue_input.pipe_id = pipe_id; + queue_input.queue_id = queue_id; + queue_input.vmid = vmid; + r = adev->mes.funcs->reset_hw_queue(&adev->mes, &queue_input); + if (r) + DRM_ERROR("failed to reset hardware queue by mmio, queue id = %d\n", + queue_id); + return r; +} + int amdgpu_mes_map_legacy_queue(struct amdgpu_device *adev, struct amdgpu_ring *ring) { @@ -873,7 +891,8 @@ int amdgpu_mes_unmap_legacy_queue(struct amdgpu_device *adev, int amdgpu_mes_reset_legacy_queue(struct amdgpu_device *adev, struct amdgpu_ring *ring, - unsigned int vmid) + unsigned int vmid, + bool use_mmio) { struct mes_reset_legacy_queue_input queue_input; int r; @@ -882,11 +901,13 @@ int amdgpu_mes_reset_legacy_queue(struct amdgpu_device *adev, queue_input.queue_type = ring->funcs->type; queue_input.doorbell_offset = ring->doorbell_index; + queue_input.me_id = ring->me; queue_input.pipe_id = ring->pipe; queue_input.queue_id = ring->queue; queue_input.mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj); queue_input.wptr_addr = ring->wptr_gpu_addr; queue_input.vmid = vmid; + queue_input.use_mmio = use_mmio; r = adev->mes.funcs->reset_legacy_queue(&adev->mes, &queue_input); if (r) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h index a5b1ea60cac8..96788c0f42f1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h @@ -75,6 +75,7 @@ struct amdgpu_mes { uint32_t sched_version; uint32_t kiq_version; + bool enable_legacy_queue_map; uint32_t total_max_queue; uint32_t max_doorbell_slices; @@ -251,6 +252,13 @@ struct mes_remove_queue_input { struct mes_reset_queue_input { uint32_t doorbell_offset; uint64_t gang_context_addr; + bool use_mmio; + uint32_t queue_type; + uint32_t me_id; + uint32_t pipe_id; + uint32_t queue_id; + uint32_t xcc_id; + uint32_t vmid; }; struct mes_map_legacy_queue_input { @@ -287,6 +295,8 @@ struct mes_resume_gang_input { struct mes_reset_legacy_queue_input { uint32_t queue_type; uint32_t doorbell_offset; + bool use_mmio; + uint32_t me_id; uint32_t pipe_id; uint32_t queue_id; uint64_t mqd_addr; @@ -396,6 +406,8 @@ int amdgpu_mes_add_hw_queue(struct amdgpu_device *adev, int gang_id, int *queue_id); int amdgpu_mes_remove_hw_queue(struct amdgpu_device *adev, int queue_id); int amdgpu_mes_reset_hw_queue(struct amdgpu_device *adev, int queue_id); +int amdgpu_mes_reset_hw_queue_mmio(struct amdgpu_device *adev, int queue_type, + int me_id, int pipe_id, int queue_id, int vmid); int amdgpu_mes_map_legacy_queue(struct amdgpu_device *adev, struct amdgpu_ring *ring); @@ -405,7 +417,8 @@ int amdgpu_mes_unmap_legacy_queue(struct amdgpu_device *adev, u64 gpu_addr, u64 seq); int amdgpu_mes_reset_legacy_queue(struct amdgpu_device *adev, struct amdgpu_ring *ring, - unsigned int vmid); + unsigned int vmid, + bool use_mmio); uint32_t amdgpu_mes_rreg(struct amdgpu_device *adev, uint32_t reg); int amdgpu_mes_wreg(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c index ca983a014ba0..45ed97038df0 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c @@ -6692,13 +6692,13 @@ static int gfx_v10_0_gfx_mqd_init(struct amdgpu_device *adev, void *m, return 0; } -static int gfx_v10_0_gfx_init_queue(struct amdgpu_ring *ring) +static int gfx_v10_0_kgq_init_queue(struct amdgpu_ring *ring, bool reset) { struct amdgpu_device *adev = ring->adev; struct v10_gfx_mqd *mqd = ring->mqd_ptr; int mqd_idx = ring - &adev->gfx.gfx_ring[0]; - if (!amdgpu_in_reset(adev) && !adev->in_suspend) { + if (!reset && !amdgpu_in_reset(adev) && !adev->in_suspend) { memset((void *)mqd, 0, sizeof(*mqd)); mutex_lock(&adev->srbm_mutex); nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); @@ -6750,7 +6750,7 @@ static int gfx_v10_0_cp_async_gfx_ring_resume(struct amdgpu_device *adev) r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); if (!r) { - r = gfx_v10_0_gfx_init_queue(ring); + r = gfx_v10_0_kgq_init_queue(ring, false); amdgpu_bo_kunmap(ring->mqd_obj); ring->mqd_ptr = NULL; } @@ -7030,13 +7030,13 @@ static int gfx_v10_0_kiq_init_queue(struct amdgpu_ring *ring) return 0; } -static int gfx_v10_0_kcq_init_queue(struct amdgpu_ring *ring) +static int gfx_v10_0_kcq_init_queue(struct amdgpu_ring *ring, bool restore) { struct amdgpu_device *adev = ring->adev; struct v10_compute_mqd *mqd = ring->mqd_ptr; int mqd_idx = ring - &adev->gfx.compute_ring[0]; - if (!amdgpu_in_reset(adev) && !adev->in_suspend) { + if (!restore && !amdgpu_in_reset(adev) && !adev->in_suspend) { memset((void *)mqd, 0, sizeof(*mqd)); mutex_lock(&adev->srbm_mutex); nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); @@ -7098,7 +7098,7 @@ static int gfx_v10_0_kcq_resume(struct amdgpu_device *adev) goto done; r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); if (!r) { - r = gfx_v10_0_kcq_init_queue(ring); + r = gfx_v10_0_kcq_init_queue(ring, false); amdgpu_bo_kunmap(ring->mqd_obj); ring->mqd_ptr = NULL; } @@ -8949,7 +8949,9 @@ static void gfx_v10_0_ring_soft_recovery(struct amdgpu_ring *ring, value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); + amdgpu_gfx_rlc_enter_safe_mode(adev, 0); WREG32_SOC15(GC, 0, mmSQ_CMD, value); + amdgpu_gfx_rlc_exit_safe_mode(adev, 0); } static void @@ -9416,6 +9418,156 @@ static void gfx_v10_ring_insert_nop(struct amdgpu_ring *ring, uint32_t num_nop) amdgpu_ring_write(ring, ring->funcs->nop); } +static int gfx_v10_0_reset_kgq(struct amdgpu_ring *ring, unsigned int vmid) +{ + struct amdgpu_device *adev = ring->adev; + struct amdgpu_kiq *kiq = &adev->gfx.kiq[0]; + struct amdgpu_ring *kiq_ring = &kiq->ring; + unsigned long flags; + u32 tmp; + u64 addr; + int r; + + if (amdgpu_sriov_vf(adev)) + return -EINVAL; + + if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) + return -EINVAL; + + spin_lock_irqsave(&kiq->ring_lock, flags); + + if (amdgpu_ring_alloc(kiq_ring, 5 + 7 + 7 + kiq->pmf->map_queues_size)) { + spin_unlock_irqrestore(&kiq->ring_lock, flags); + return -ENOMEM; + } + + addr = amdgpu_bo_gpu_offset(ring->mqd_obj) + + offsetof(struct v10_gfx_mqd, cp_gfx_hqd_active); + tmp = REG_SET_FIELD(0, CP_VMID_RESET, RESET_REQUEST, 1 << vmid); + if (ring->pipe == 0) + tmp = REG_SET_FIELD(tmp, CP_VMID_RESET, PIPE0_QUEUES, 1 << ring->queue); + else + tmp = REG_SET_FIELD(tmp, CP_VMID_RESET, PIPE1_QUEUES, 1 << ring->queue); + + gfx_v10_0_ring_emit_wreg(kiq_ring, + SOC15_REG_OFFSET(GC, 0, mmCP_VMID_RESET), tmp); + gfx_v10_0_wait_reg_mem(kiq_ring, 0, 1, 0, + lower_32_bits(addr), upper_32_bits(addr), + 0, 1, 0x20); + gfx_v10_0_ring_emit_reg_wait(kiq_ring, + SOC15_REG_OFFSET(GC, 0, mmCP_VMID_RESET), 0, 0xffffffff); + kiq->pmf->kiq_map_queues(kiq_ring, ring); + amdgpu_ring_commit(kiq_ring); + + spin_unlock_irqrestore(&kiq->ring_lock, flags); + + r = amdgpu_ring_test_ring(kiq_ring); + if (r) + return r; + + r = amdgpu_bo_reserve(ring->mqd_obj, false); + if (unlikely(r != 0)) { + DRM_ERROR("fail to resv mqd_obj\n"); + return r; + } + r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); + if (!r) { + r = gfx_v10_0_kgq_init_queue(ring, true); + amdgpu_bo_kunmap(ring->mqd_obj); + ring->mqd_ptr = NULL; + } + amdgpu_bo_unreserve(ring->mqd_obj); + if (r) { + DRM_ERROR("fail to unresv mqd_obj\n"); + return r; + } + + return amdgpu_ring_test_ring(ring); +} + +static int gfx_v10_0_reset_kcq(struct amdgpu_ring *ring, + unsigned int vmid) +{ + struct amdgpu_device *adev = ring->adev; + struct amdgpu_kiq *kiq = &adev->gfx.kiq[0]; + struct amdgpu_ring *kiq_ring = &kiq->ring; + unsigned long flags; + int i, r; + + if (amdgpu_sriov_vf(adev)) + return -EINVAL; + + if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) + return -EINVAL; + + spin_lock_irqsave(&kiq->ring_lock, flags); + + if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) { + spin_unlock_irqrestore(&kiq->ring_lock, flags); + return -ENOMEM; + } + + kiq->pmf->kiq_unmap_queues(kiq_ring, ring, RESET_QUEUES, + 0, 0); + amdgpu_ring_commit(kiq_ring); + spin_unlock_irqrestore(&kiq->ring_lock, flags); + + r = amdgpu_ring_test_ring(kiq_ring); + if (r) + return r; + + /* make sure dequeue is complete*/ + amdgpu_gfx_rlc_enter_safe_mode(adev, 0); + mutex_lock(&adev->srbm_mutex); + nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); + for (i = 0; i < adev->usec_timeout; i++) { + if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1)) + break; + udelay(1); + } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + nv_grbm_select(adev, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); + amdgpu_gfx_rlc_exit_safe_mode(adev, 0); + if (r) { + dev_err(adev->dev, "fail to wait on hqd deactivate\n"); + return r; + } + + r = amdgpu_bo_reserve(ring->mqd_obj, false); + if (unlikely(r != 0)) { + dev_err(adev->dev, "fail to resv mqd_obj\n"); + return r; + } + r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); + if (!r) { + r = gfx_v10_0_kcq_init_queue(ring, true); + amdgpu_bo_kunmap(ring->mqd_obj); + ring->mqd_ptr = NULL; + } + amdgpu_bo_unreserve(ring->mqd_obj); + if (r) { + dev_err(adev->dev, "fail to unresv mqd_obj\n"); + return r; + } + + spin_lock_irqsave(&kiq->ring_lock, flags); + if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size)) { + spin_unlock_irqrestore(&kiq->ring_lock, flags); + return -ENOMEM; + } + kiq->pmf->kiq_map_queues(kiq_ring, ring); + amdgpu_ring_commit(kiq_ring); + spin_unlock_irqrestore(&kiq->ring_lock, flags); + + r = amdgpu_ring_test_ring(kiq_ring); + if (r) + return r; + + return amdgpu_ring_test_ring(ring); +} + static void gfx_v10_ip_print(void *handle, struct drm_printer *p) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; @@ -9619,6 +9771,7 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_gfx = { .emit_reg_write_reg_wait = gfx_v10_0_ring_emit_reg_write_reg_wait, .soft_recovery = gfx_v10_0_ring_soft_recovery, .emit_mem_sync = gfx_v10_0_emit_mem_sync, + .reset = gfx_v10_0_reset_kgq, }; static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_compute = { @@ -9655,6 +9808,7 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_compute = { .emit_reg_write_reg_wait = gfx_v10_0_ring_emit_reg_write_reg_wait, .soft_recovery = gfx_v10_0_ring_soft_recovery, .emit_mem_sync = gfx_v10_0_emit_mem_sync, + .reset = gfx_v10_0_reset_kcq, }; static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_kiq = { diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c index ee8604722467..d3e8be82a172 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c @@ -3984,13 +3984,13 @@ static int gfx_v11_0_gfx_mqd_init(struct amdgpu_device *adev, void *m, return 0; } -static int gfx_v11_0_gfx_init_queue(struct amdgpu_ring *ring) +static int gfx_v11_0_kgq_init_queue(struct amdgpu_ring *ring, bool reset) { struct amdgpu_device *adev = ring->adev; struct v11_gfx_mqd *mqd = ring->mqd_ptr; int mqd_idx = ring - &adev->gfx.gfx_ring[0]; - if (!amdgpu_in_reset(adev) && !adev->in_suspend) { + if (!reset && !amdgpu_in_reset(adev) && !adev->in_suspend) { memset((void *)mqd, 0, sizeof(*mqd)); mutex_lock(&adev->srbm_mutex); soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); @@ -4026,7 +4026,7 @@ static int gfx_v11_0_cp_async_gfx_ring_resume(struct amdgpu_device *adev) r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); if (!r) { - r = gfx_v11_0_gfx_init_queue(ring); + r = gfx_v11_0_kgq_init_queue(ring, false); amdgpu_bo_kunmap(ring->mqd_obj); ring->mqd_ptr = NULL; } @@ -4321,13 +4321,13 @@ static int gfx_v11_0_kiq_init_queue(struct amdgpu_ring *ring) return 0; } -static int gfx_v11_0_kcq_init_queue(struct amdgpu_ring *ring) +static int gfx_v11_0_kcq_init_queue(struct amdgpu_ring *ring, bool reset) { struct amdgpu_device *adev = ring->adev; struct v11_compute_mqd *mqd = ring->mqd_ptr; int mqd_idx = ring - &adev->gfx.compute_ring[0]; - if (!amdgpu_in_reset(adev) && !adev->in_suspend) { + if (!reset && !amdgpu_in_reset(adev) && !adev->in_suspend) { memset((void *)mqd, 0, sizeof(*mqd)); mutex_lock(&adev->srbm_mutex); soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); @@ -4391,7 +4391,7 @@ static int gfx_v11_0_kcq_resume(struct amdgpu_device *adev) goto done; r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); if (!r) { - r = gfx_v11_0_kcq_init_queue(ring); + r = gfx_v11_0_kcq_init_queue(ring, false); amdgpu_bo_kunmap(ring->mqd_obj); ring->mqd_ptr = NULL; } @@ -4781,7 +4781,7 @@ static int gfx_v11_0_soft_reset(void *handle) int r, i, j, k; struct amdgpu_device *adev = (struct amdgpu_device *)handle; - gfx_v11_0_set_safe_mode(adev, 0); + amdgpu_gfx_rlc_enter_safe_mode(adev, 0); tmp = RREG32_SOC15(GC, 0, regCP_INT_CNTL); tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CMP_BUSY_INT_ENABLE, 0); @@ -4900,7 +4900,7 @@ static int gfx_v11_0_soft_reset(void *handle) tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, GFX_IDLE_INT_ENABLE, 1); WREG32_SOC15(GC, 0, regCP_INT_CNTL, tmp); - gfx_v11_0_unset_safe_mode(adev, 0); + amdgpu_gfx_rlc_exit_safe_mode(adev, 0); return gfx_v11_0_cp_resume(adev); } @@ -5923,6 +5923,9 @@ static int gfx_v11_0_ring_preempt_ib(struct amdgpu_ring *ring) struct amdgpu_ring *kiq_ring = &kiq->ring; unsigned long flags; + if (adev->enable_mes) + return -EINVAL; + if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) return -EINVAL; @@ -6088,7 +6091,9 @@ static void gfx_v11_0_ring_soft_recovery(struct amdgpu_ring *ring, value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); + amdgpu_gfx_rlc_enter_safe_mode(adev, 0); WREG32_SOC15(GC, 0, regSQ_CMD, value); + amdgpu_gfx_rlc_exit_safe_mode(adev, 0); } static void @@ -6541,6 +6546,99 @@ static void gfx_v11_0_emit_mem_sync(struct amdgpu_ring *ring) amdgpu_ring_write(ring, gcr_cntl); /* GCR_CNTL */ } +static int gfx_v11_0_reset_kgq(struct amdgpu_ring *ring, unsigned int vmid) +{ + struct amdgpu_device *adev = ring->adev; + int r; + + if (amdgpu_sriov_vf(adev)) + return -EINVAL; + + r = amdgpu_mes_reset_legacy_queue(ring->adev, ring, vmid, false); + if (r) + return r; + + r = amdgpu_bo_reserve(ring->mqd_obj, false); + if (unlikely(r != 0)) { + dev_err(adev->dev, "fail to resv mqd_obj\n"); + return r; + } + r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); + if (!r) { + r = gfx_v11_0_kgq_init_queue(ring, true); + amdgpu_bo_kunmap(ring->mqd_obj); + ring->mqd_ptr = NULL; + } + amdgpu_bo_unreserve(ring->mqd_obj); + if (r) { + dev_err(adev->dev, "fail to unresv mqd_obj\n"); + return r; + } + + r = amdgpu_mes_map_legacy_queue(adev, ring); + if (r) { + dev_err(adev->dev, "failed to remap kgq\n"); + return r; + } + + return amdgpu_ring_test_ring(ring); +} + +static int gfx_v11_0_reset_kcq(struct amdgpu_ring *ring, unsigned int vmid) +{ + struct amdgpu_device *adev = ring->adev; + int i, r = 0; + + if (amdgpu_sriov_vf(adev)) + return -EINVAL; + + amdgpu_gfx_rlc_enter_safe_mode(adev, 0); + mutex_lock(&adev->srbm_mutex); + soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); + WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 0x2); + WREG32_SOC15(GC, 0, regSPI_COMPUTE_QUEUE_RESET, 0x1); + + /* make sure dequeue is complete*/ + for (i = 0; i < adev->usec_timeout; i++) { + if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1)) + break; + udelay(1); + } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + soc21_grbm_select(adev, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); + amdgpu_gfx_rlc_exit_safe_mode(adev, 0); + if (r) { + dev_err(adev->dev, "fail to wait on hqd deactivate\n"); + return r; + } + + r = amdgpu_bo_reserve(ring->mqd_obj, false); + if (unlikely(r != 0)) { + dev_err(adev->dev, "fail to resv mqd_obj\n"); + return r; + } + r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); + if (!r) { + r = gfx_v11_0_kcq_init_queue(ring, true); + amdgpu_bo_kunmap(ring->mqd_obj); + ring->mqd_ptr = NULL; + } + amdgpu_bo_unreserve(ring->mqd_obj); + if (r) { + dev_err(adev->dev, "fail to unresv mqd_obj\n"); + return r; + } + r = amdgpu_mes_map_legacy_queue(adev, ring); + if (r) { + dev_err(adev->dev, "failed to remap kcq\n"); + return r; + } + + return amdgpu_ring_test_ring(ring); +} + static void gfx_v11_ip_print(void *handle, struct drm_printer *p) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; @@ -6742,6 +6840,7 @@ static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_gfx = { .emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait, .soft_recovery = gfx_v11_0_ring_soft_recovery, .emit_mem_sync = gfx_v11_0_emit_mem_sync, + .reset = gfx_v11_0_reset_kgq, }; static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_compute = { @@ -6779,6 +6878,7 @@ static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_compute = { .emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait, .soft_recovery = gfx_v11_0_ring_soft_recovery, .emit_mem_sync = gfx_v11_0_emit_mem_sync, + .reset = gfx_v11_0_reset_kcq, }; static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_kiq = { diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c index 54059cbcfc08..d1357c01eb39 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c @@ -2916,13 +2916,13 @@ static int gfx_v12_0_gfx_mqd_init(struct amdgpu_device *adev, void *m, return 0; } -static int gfx_v12_0_gfx_init_queue(struct amdgpu_ring *ring) +static int gfx_v12_0_kgq_init_queue(struct amdgpu_ring *ring, bool reset) { struct amdgpu_device *adev = ring->adev; struct v12_gfx_mqd *mqd = ring->mqd_ptr; int mqd_idx = ring - &adev->gfx.gfx_ring[0]; - if (!amdgpu_in_reset(adev) && !adev->in_suspend) { + if (!reset && !amdgpu_in_reset(adev) && !adev->in_suspend) { memset((void *)mqd, 0, sizeof(*mqd)); mutex_lock(&adev->srbm_mutex); soc24_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); @@ -2958,7 +2958,7 @@ static int gfx_v12_0_cp_async_gfx_ring_resume(struct amdgpu_device *adev) r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); if (!r) { - r = gfx_v12_0_gfx_init_queue(ring); + r = gfx_v12_0_kgq_init_queue(ring, false); amdgpu_bo_kunmap(ring->mqd_obj); ring->mqd_ptr = NULL; } @@ -3262,13 +3262,13 @@ static int gfx_v12_0_kiq_init_queue(struct amdgpu_ring *ring) return 0; } -static int gfx_v12_0_kcq_init_queue(struct amdgpu_ring *ring) +static int gfx_v12_0_kcq_init_queue(struct amdgpu_ring *ring, bool reset) { struct amdgpu_device *adev = ring->adev; struct v12_compute_mqd *mqd = ring->mqd_ptr; int mqd_idx = ring - &adev->gfx.compute_ring[0]; - if (!amdgpu_in_reset(adev) && !adev->in_suspend) { + if (!reset && !amdgpu_in_reset(adev) && !adev->in_suspend) { memset((void *)mqd, 0, sizeof(*mqd)); mutex_lock(&adev->srbm_mutex); soc24_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); @@ -3332,7 +3332,7 @@ static int gfx_v12_0_kcq_resume(struct amdgpu_device *adev) goto done; r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); if (!r) { - r = gfx_v12_0_kcq_init_queue(ring); + r = gfx_v12_0_kcq_init_queue(ring, false); amdgpu_bo_kunmap(ring->mqd_obj); ring->mqd_ptr = NULL; } @@ -4501,6 +4501,9 @@ static int gfx_v12_0_ring_preempt_ib(struct amdgpu_ring *ring) struct amdgpu_ring *kiq_ring = &kiq->ring; unsigned long flags; + if (adev->enable_mes) + return -EINVAL; + if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) return -EINVAL; @@ -4617,7 +4620,9 @@ static void gfx_v12_0_ring_soft_recovery(struct amdgpu_ring *ring, value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); + amdgpu_gfx_rlc_enter_safe_mode(adev, 0); WREG32_SOC15(GC, 0, regSQ_CMD, value); + amdgpu_gfx_rlc_exit_safe_mode(adev, 0); } static void @@ -5155,6 +5160,93 @@ static void gfx_v12_ip_dump(void *handle) amdgpu_gfx_off_ctrl(adev, true); } +static int gfx_v12_0_reset_kgq(struct amdgpu_ring *ring, unsigned int vmid) +{ + struct amdgpu_device *adev = ring->adev; + int r; + + if (amdgpu_sriov_vf(adev)) + return -EINVAL; + + r = amdgpu_mes_reset_legacy_queue(ring->adev, ring, vmid, false); + if (r) { + dev_err(adev->dev, "reset via MES failed %d\n", r); + return r; + } + + r = amdgpu_bo_reserve(ring->mqd_obj, false); + if (unlikely(r != 0)) { + dev_err(adev->dev, "fail to resv mqd_obj\n"); + return r; + } + r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); + if (!r) { + r = gfx_v12_0_kgq_init_queue(ring, true); + amdgpu_bo_kunmap(ring->mqd_obj); + ring->mqd_ptr = NULL; + } + amdgpu_bo_unreserve(ring->mqd_obj); + if (r) { + DRM_ERROR("fail to unresv mqd_obj\n"); + return r; + } + + r = amdgpu_mes_map_legacy_queue(adev, ring); + if (r) { + dev_err(adev->dev, "failed to remap kgq\n"); + return r; + } + + return amdgpu_ring_test_ring(ring); +} + +static int gfx_v12_0_reset_kcq(struct amdgpu_ring *ring, unsigned int vmid) +{ + struct amdgpu_device *adev = ring->adev; + int r, i; + + if (amdgpu_sriov_vf(adev)) + return -EINVAL; + + amdgpu_gfx_rlc_enter_safe_mode(adev, 0); + mutex_lock(&adev->srbm_mutex); + soc24_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); + WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 0x2); + WREG32_SOC15(GC, 0, regSPI_COMPUTE_QUEUE_RESET, 0x1); + for (i = 0; i < adev->usec_timeout; i++) { + if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1)) + break; + udelay(1); + } + soc24_grbm_select(adev, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); + amdgpu_gfx_rlc_exit_safe_mode(adev, 0); + + r = amdgpu_bo_reserve(ring->mqd_obj, false); + if (unlikely(r != 0)) { + DRM_ERROR("fail to resv mqd_obj\n"); + return r; + } + r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); + if (!r) { + r = gfx_v12_0_kcq_init_queue(ring, true); + amdgpu_bo_kunmap(ring->mqd_obj); + ring->mqd_ptr = NULL; + } + amdgpu_bo_unreserve(ring->mqd_obj); + if (r) { + DRM_ERROR("fail to unresv mqd_obj\n"); + return r; + } + r = amdgpu_mes_map_legacy_queue(adev, ring); + if (r) { + dev_err(adev->dev, "failed to remap kcq\n"); + return r; + } + + return amdgpu_ring_test_ring(ring); +} + static const struct amd_ip_funcs gfx_v12_0_ip_funcs = { .name = "gfx_v12_0", .early_init = gfx_v12_0_early_init, @@ -5217,6 +5309,7 @@ static const struct amdgpu_ring_funcs gfx_v12_0_ring_funcs_gfx = { .emit_reg_write_reg_wait = gfx_v12_0_ring_emit_reg_write_reg_wait, .soft_recovery = gfx_v12_0_ring_soft_recovery, .emit_mem_sync = gfx_v12_0_emit_mem_sync, + .reset = gfx_v12_0_reset_kgq, }; static const struct amdgpu_ring_funcs gfx_v12_0_ring_funcs_compute = { @@ -5251,6 +5344,7 @@ static const struct amdgpu_ring_funcs gfx_v12_0_ring_funcs_compute = { .emit_reg_write_reg_wait = gfx_v12_0_ring_emit_reg_write_reg_wait, .soft_recovery = gfx_v12_0_ring_soft_recovery, .emit_mem_sync = gfx_v12_0_emit_mem_sync, + .reset = gfx_v12_0_reset_kcq, }; static const struct amdgpu_ring_funcs gfx_v12_0_ring_funcs_kiq = { diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index 21089aadbb7b..8cf5d7925b51 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -7233,6 +7233,10 @@ static int gfx_v9_0_reset_kcq(struct amdgpu_ring *ring, unsigned long flags; int i, r; + if (!adev->debug_exp_resets && + !adev->gfx.num_gfx_rings) + return -EINVAL; + if (amdgpu_sriov_vf(adev)) return -EINVAL; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c index 2067f26d3a9d..408e5600bb61 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c @@ -3052,6 +3052,9 @@ static void gfx_v9_4_3_ring_soft_recovery(struct amdgpu_ring *ring, struct amdgpu_device *adev = ring->adev; uint32_t value = 0; + if (!adev->debug_exp_resets) + return; + value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03); value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); @@ -3466,6 +3469,98 @@ static void gfx_v9_4_3_emit_wave_limit(struct amdgpu_ring *ring, bool enable) } } +static int gfx_v9_4_3_unmap_done(struct amdgpu_device *adev, uint32_t me, + uint32_t pipe, uint32_t queue, + uint32_t xcc_id) +{ + int i, r; + /* make sure dequeue is complete*/ + gfx_v9_4_3_xcc_set_safe_mode(adev, xcc_id); + mutex_lock(&adev->srbm_mutex); + soc15_grbm_select(adev, me, pipe, queue, 0, GET_INST(GC, xcc_id)); + for (i = 0; i < adev->usec_timeout; i++) { + if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1)) + break; + udelay(1); + } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + else + r = 0; + soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); + mutex_unlock(&adev->srbm_mutex); + gfx_v9_4_3_xcc_unset_safe_mode(adev, xcc_id); + + return r; + +} + +static bool gfx_v9_4_3_pipe_reset_support(struct amdgpu_device *adev) +{ + /*TODO: Need check gfx9.4.4 mec fw whether supports pipe reset as well.*/ + if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) && + adev->gfx.mec_fw_version >= 0x0000009b) + return true; + else + dev_warn_once(adev->dev, "Please use the latest MEC version to see whether support pipe reset\n"); + + return false; +} + +static int gfx_v9_4_3_reset_hw_pipe(struct amdgpu_ring *ring) +{ + struct amdgpu_device *adev = ring->adev; + uint32_t reset_pipe, clean_pipe; + int r; + + if (!gfx_v9_4_3_pipe_reset_support(adev)) + return -EINVAL; + + gfx_v9_4_3_xcc_set_safe_mode(adev, ring->xcc_id); + mutex_lock(&adev->srbm_mutex); + + reset_pipe = RREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL); + clean_pipe = reset_pipe; + + if (ring->me == 1) { + switch (ring->pipe) { + case 0: + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME1_PIPE0_RESET, 1); + break; + case 1: + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME1_PIPE1_RESET, 1); + break; + case 2: + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME1_PIPE2_RESET, 1); + break; + case 3: + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME1_PIPE3_RESET, 1); + break; + default: + break; + } + } else { + if (ring->pipe) + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME2_PIPE1_RESET, 1); + else + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME2_PIPE0_RESET, 1); + } + + WREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL, reset_pipe); + WREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL, clean_pipe); + mutex_unlock(&adev->srbm_mutex); + gfx_v9_4_3_xcc_unset_safe_mode(adev, ring->xcc_id); + + r = gfx_v9_4_3_unmap_done(adev, ring->me, ring->pipe, ring->queue, ring->xcc_id); + return r; +} + static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring, unsigned int vmid) { @@ -3473,7 +3568,10 @@ static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring, struct amdgpu_kiq *kiq = &adev->gfx.kiq[ring->xcc_id]; struct amdgpu_ring *kiq_ring = &kiq->ring; unsigned long flags; - int r, i; + int r; + + if (!adev->debug_exp_resets) + return -EINVAL; if (amdgpu_sriov_vf(adev)) return -EINVAL; @@ -3495,26 +3593,23 @@ static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring, spin_unlock_irqrestore(&kiq->ring_lock, flags); r = amdgpu_ring_test_ring(kiq_ring); + if (r) { + dev_err(adev->dev, "kiq ring test failed after ring: %s queue reset\n", + ring->name); + goto pipe_reset; + } + + r = gfx_v9_4_3_unmap_done(adev, ring->me, ring->pipe, ring->queue, ring->xcc_id); if (r) - return r; + dev_err(adev->dev, "fail to wait on hqd deactive and will try pipe reset\n"); - /* make sure dequeue is complete*/ - amdgpu_gfx_rlc_enter_safe_mode(adev, ring->xcc_id); - mutex_lock(&adev->srbm_mutex); - soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, GET_INST(GC, ring->xcc_id)); - for (i = 0; i < adev->usec_timeout; i++) { - if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1)) - break; - udelay(1); - } - if (i >= adev->usec_timeout) - r = -ETIMEDOUT; - soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, ring->xcc_id)); - mutex_unlock(&adev->srbm_mutex); - amdgpu_gfx_rlc_exit_safe_mode(adev, ring->xcc_id); - if (r) { - dev_err(adev->dev, "fail to wait on hqd deactive\n"); - return r; +pipe_reset: + if(r) { + r = gfx_v9_4_3_reset_hw_pipe(ring); + dev_info(adev->dev, "ring: %s pipe reset :%s\n", ring->name, + r ? "failed" : "successfully"); + if (r) + return r; } r = amdgpu_bo_reserve(ring->mqd_obj, false); diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c index 8edcd85a1261..0f055d1b1da6 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c @@ -26,6 +26,7 @@ #include "amdgpu.h" #include "soc15_common.h" #include "soc21.h" +#include "gfx_v11_0.h" #include "gc/gc_11_0_0_offset.h" #include "gc/gc_11_0_0_sh_mask.h" #include "gc/gc_11_0_0_default.h" @@ -360,9 +361,83 @@ static int mes_v11_0_remove_hw_queue(struct amdgpu_mes *mes, offsetof(union MESAPI__REMOVE_QUEUE, api_status)); } +static int mes_v11_0_reset_queue_mmio(struct amdgpu_mes *mes, uint32_t queue_type, + uint32_t me_id, uint32_t pipe_id, + uint32_t queue_id, uint32_t vmid) +{ + struct amdgpu_device *adev = mes->adev; + uint32_t value; + int i, r = 0; + + amdgpu_gfx_rlc_enter_safe_mode(adev, 0); + + if (queue_type == AMDGPU_RING_TYPE_GFX) { + dev_info(adev->dev, "reset gfx queue (%d:%d:%d: vmid:%d)\n", + me_id, pipe_id, queue_id, vmid); + + mutex_lock(&adev->gfx.reset_sem_mutex); + gfx_v11_0_request_gfx_index_mutex(adev, true); + /* all se allow writes */ + WREG32_SOC15(GC, 0, regGRBM_GFX_INDEX, + (uint32_t)(0x1 << GRBM_GFX_INDEX__SE_BROADCAST_WRITES__SHIFT)); + value = REG_SET_FIELD(0, CP_VMID_RESET, RESET_REQUEST, 1 << vmid); + if (pipe_id == 0) + value = REG_SET_FIELD(value, CP_VMID_RESET, PIPE0_QUEUES, 1 << queue_id); + else + value = REG_SET_FIELD(value, CP_VMID_RESET, PIPE1_QUEUES, 1 << queue_id); + WREG32_SOC15(GC, 0, regCP_VMID_RESET, value); + gfx_v11_0_request_gfx_index_mutex(adev, false); + mutex_unlock(&adev->gfx.reset_sem_mutex); + + mutex_lock(&adev->srbm_mutex); + soc21_grbm_select(adev, me_id, pipe_id, queue_id, 0); + /* wait till dequeue take effects */ + for (i = 0; i < adev->usec_timeout; i++) { + if (!(RREG32_SOC15(GC, 0, regCP_GFX_HQD_ACTIVE) & 1)) + break; + udelay(1); + } + if (i >= adev->usec_timeout) { + dev_err(adev->dev, "failed to wait on gfx hqd deactivate\n"); + r = -ETIMEDOUT; + } + + soc21_grbm_select(adev, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); + } else if (queue_type == AMDGPU_RING_TYPE_COMPUTE) { + dev_info(adev->dev, "reset compute queue (%d:%d:%d)\n", + me_id, pipe_id, queue_id); + mutex_lock(&adev->srbm_mutex); + soc21_grbm_select(adev, me_id, pipe_id, queue_id, 0); + WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 0x2); + WREG32_SOC15(GC, 0, regSPI_COMPUTE_QUEUE_RESET, 0x1); + + /* wait till dequeue take effects */ + for (i = 0; i < adev->usec_timeout; i++) { + if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1)) + break; + udelay(1); + } + if (i >= adev->usec_timeout) { + dev_err(adev->dev, "failed to wait on hqd deactivate\n"); + r = -ETIMEDOUT; + } + soc21_grbm_select(adev, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); + } + + amdgpu_gfx_rlc_exit_safe_mode(adev, 0); + return r; +} + static int mes_v11_0_reset_hw_queue(struct amdgpu_mes *mes, struct mes_reset_queue_input *input) { + if (input->use_mmio) + return mes_v11_0_reset_queue_mmio(mes, input->queue_type, + input->me_id, input->pipe_id, + input->queue_id, input->vmid); + union MESAPI__RESET mes_reset_queue_pkt; memset(&mes_reset_queue_pkt, 0, sizeof(mes_reset_queue_pkt)); @@ -648,6 +723,11 @@ static int mes_v11_0_reset_legacy_queue(struct amdgpu_mes *mes, { union MESAPI__RESET mes_reset_queue_pkt; + if (input->use_mmio) + return mes_v11_0_reset_queue_mmio(mes, input->queue_type, + input->me_id, input->pipe_id, + input->queue_id, input->vmid); + memset(&mes_reset_queue_pkt, 0, sizeof(mes_reset_queue_pkt)); mes_reset_queue_pkt.header.type = MES_API_TYPE_SCHEDULER; @@ -775,6 +855,28 @@ static void mes_v11_0_free_ucode_buffers(struct amdgpu_device *adev, (void **)&adev->mes.ucode_fw_ptr[pipe]); } +static void mes_v11_0_get_fw_version(struct amdgpu_device *adev) +{ + int pipe; + + /* get MES scheduler/KIQ versions */ + mutex_lock(&adev->srbm_mutex); + + for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++) { + soc21_grbm_select(adev, 3, pipe, 0, 0); + + if (pipe == AMDGPU_MES_SCHED_PIPE) + adev->mes.sched_version = + RREG32_SOC15(GC, 0, regCP_MES_GP3_LO); + else if (pipe == AMDGPU_MES_KIQ_PIPE && adev->enable_mes_kiq) + adev->mes.kiq_version = + RREG32_SOC15(GC, 0, regCP_MES_GP3_LO); + } + + soc21_grbm_select(adev, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); +} + static void mes_v11_0_enable(struct amdgpu_device *adev, bool enable) { uint64_t ucode_addr; @@ -1144,18 +1246,6 @@ static int mes_v11_0_queue_init(struct amdgpu_device *adev, mes_v11_0_queue_init_register(ring); } - /* get MES scheduler/KIQ versions */ - mutex_lock(&adev->srbm_mutex); - soc21_grbm_select(adev, 3, pipe, 0, 0); - - if (pipe == AMDGPU_MES_SCHED_PIPE) - adev->mes.sched_version = RREG32_SOC15(GC, 0, regCP_MES_GP3_LO); - else if (pipe == AMDGPU_MES_KIQ_PIPE && adev->enable_mes_kiq) - adev->mes.kiq_version = RREG32_SOC15(GC, 0, regCP_MES_GP3_LO); - - soc21_grbm_select(adev, 0, 0, 0, 0); - mutex_unlock(&adev->srbm_mutex); - return 0; } @@ -1402,15 +1492,24 @@ static int mes_v11_0_kiq_hw_init(struct amdgpu_device *adev) mes_v11_0_enable(adev, true); + mes_v11_0_get_fw_version(adev); + mes_v11_0_kiq_setting(&adev->gfx.kiq[0].ring); r = mes_v11_0_queue_init(adev, AMDGPU_MES_KIQ_PIPE); if (r) goto failure; - r = mes_v11_0_hw_init(adev); - if (r) - goto failure; + if ((adev->mes.sched_version & AMDGPU_MES_VERSION_MASK) >= 0x47) + adev->mes.enable_legacy_queue_map = true; + else + adev->mes.enable_legacy_queue_map = false; + + if (adev->mes.enable_legacy_queue_map) { + r = mes_v11_0_hw_init(adev); + if (r) + goto failure; + } return r; diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c index 47a73f6ae4da..e499b2857a01 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c @@ -1332,6 +1332,7 @@ static int mes_v12_0_sw_init(void *handle) adev->mes.funcs = &mes_v12_0_funcs; adev->mes.kiq_hw_init = &mes_v12_0_kiq_hw_init; adev->mes.kiq_hw_fini = &mes_v12_0_kiq_hw_fini; + adev->mes.enable_legacy_queue_map = true; adev->mes.event_log_size = AMDGPU_MES_LOG_BUFFER_SIZE; @@ -1488,9 +1489,11 @@ static int mes_v12_0_kiq_hw_init(struct amdgpu_device *adev) mes_v12_0_set_hw_resources_1(&adev->mes, AMDGPU_MES_KIQ_PIPE); } - r = mes_v12_0_hw_init(adev); - if (r) - goto failure; + if (adev->mes.enable_legacy_queue_map) { + r = mes_v12_0_hw_init(adev); + if (r) + goto failure; + } return r; |