diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_job.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 86 |
1 files changed, 64 insertions, 22 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index a21c510c408e..acb21fc8b3ce 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -102,8 +102,6 @@ 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 @@ -132,29 +130,48 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job) amdgpu_vm_put_task_info(ti); } - dma_fence_set_error(&s_job->s_fence->finished, -ETIME); - /* attempt a per ring reset */ - if (amdgpu_gpu_recovery && - ring->funcs->reset) { + if (unlikely(adev->debug_disable_gpu_ring_reset)) { + dev_err(adev->dev, "Ring reset disabled by debug mask\n"); + } else if (amdgpu_gpu_recovery && ring->funcs->reset) { + bool is_guilty; + dev_err(adev->dev, "Starting %s ring reset\n", s_job->sched->name); /* stop the scheduler, but don't mess with the * bad job yet because if ring reset fails * we'll fall back to full GPU reset. */ drm_sched_wqueue_stop(&ring->sched); + + /* for engine resets, we need to reset the engine, + * but individual queues may be unaffected. + * check here to make sure the accounting is correct. + */ + if (ring->funcs->is_guilty) + is_guilty = ring->funcs->is_guilty(ring); + else + is_guilty = true; + + if (is_guilty) + dma_fence_set_error(&s_job->s_fence->finished, -ETIME); + r = amdgpu_ring_reset(ring, job->vmid); if (!r) { if (amdgpu_ring_sched_ready(ring)) drm_sched_stop(&ring->sched, s_job); - atomic_inc(&ring->adev->gpu_reset_counter); - amdgpu_fence_driver_force_completion(ring); + if (is_guilty) { + atomic_inc(&ring->adev->gpu_reset_counter); + amdgpu_fence_driver_force_completion(ring); + } if (amdgpu_ring_sched_ready(ring)) drm_sched_start(&ring->sched, 0); + dev_err(adev->dev, "Ring %s reset succeeded\n", ring->sched.name); + drm_dev_wedged_event(adev_to_drm(adev), DRM_WEDGE_RECOVERY_NONE); goto exit; } dev_err(adev->dev, "Ring %s reset failure\n", ring->sched.name); } + dma_fence_set_error(&s_job->s_fence->finished, -ETIME); if (amdgpu_device_should_recover_gpu(ring->adev)) { struct amdgpu_reset_context reset_context; @@ -181,7 +198,6 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job) } exit: - adev->job_hang = false; drm_dev_exit(idx); return DRM_GPU_SCHED_STAT_NOMINAL; } @@ -197,11 +213,6 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, struct amdgpu_vm *vm, if (!*job) return -ENOMEM; - /* - * Initialize the scheduler to at least some ring so that we always - * have a pointer to adev. - */ - (*job)->base.sched = &adev->rings[0]->sched; (*job)->vm = vm; amdgpu_sync_create(&(*job)->explicit_sync); @@ -267,7 +278,7 @@ void amdgpu_job_free_resources(struct amdgpu_job *job) f = NULL; for (i = 0; i < job->num_ibs; ++i) - amdgpu_ib_free(NULL, &job->ibs[i], f); + amdgpu_ib_free(&job->ibs[i], f); } static void amdgpu_job_free_cb(struct drm_sched_job *s_job) @@ -350,25 +361,40 @@ amdgpu_job_prepare_job(struct drm_sched_job *sched_job, { struct amdgpu_ring *ring = to_amdgpu_ring(s_entity->rq->sched); struct amdgpu_job *job = to_amdgpu_job(sched_job); - struct dma_fence *fence = NULL; + struct dma_fence *fence; int r; r = drm_sched_entity_error(s_entity); if (r) goto error; - if (job->gang_submit) + if (job->gang_submit) { fence = amdgpu_device_switch_gang(ring->adev, job->gang_submit); + if (fence) + return fence; + } + + fence = amdgpu_device_enforce_isolation(ring->adev, ring, job); + if (fence) + return fence; - if (!fence && job->vm && !job->vmid) { + if (job->vm && !job->vmid) { r = amdgpu_vmid_grab(job->vm, ring, job, &fence); if (r) { dev_err(ring->adev->dev, "Error getting VM ID (%d)\n", r); goto error; } + /* + * The VM structure might be released after the VMID is + * assigned, we had multiple problems with people trying to use + * the VM pointer so better set it to NULL. + */ + if (!fence) + job->vm = NULL; + return fence; } - return fence; + return NULL; error: dma_fence_set_error(&job->base.s_fence->finished, r); @@ -412,8 +438,24 @@ static struct dma_fence *amdgpu_job_run(struct drm_sched_job *sched_job) return fence; } -#define to_drm_sched_job(sched_job) \ - container_of((sched_job), struct drm_sched_job, queue_node) +/* + * This is a duplicate function from DRM scheduler sched_internal.h. + * Plan is to remove it when amdgpu_job_stop_all_jobs_on_sched is removed, due + * latter being incorrect and racy. + * + * See https://lore.kernel.org/amd-gfx/44edde63-7181-44fb-a4f7-94e50514f539@amd.com/ + */ +static struct drm_sched_job * +drm_sched_entity_queue_pop(struct drm_sched_entity *entity) +{ + struct spsc_node *node; + + node = spsc_queue_pop(&entity->job_queue); + if (!node) + return NULL; + + return container_of(node, struct drm_sched_job, queue_node); +} void amdgpu_job_stop_all_jobs_on_sched(struct drm_gpu_scheduler *sched) { @@ -426,7 +468,7 @@ void amdgpu_job_stop_all_jobs_on_sched(struct drm_gpu_scheduler *sched) struct drm_sched_rq *rq = sched->sched_rq[i]; spin_lock(&rq->lock); list_for_each_entry(s_entity, &rq->entities, list) { - while ((s_job = to_drm_sched_job(spsc_queue_pop(&s_entity->job_queue)))) { + while ((s_job = drm_sched_entity_queue_pop(s_entity))) { struct drm_sched_fence *s_fence = s_job->s_fence; dma_fence_signal(&s_fence->scheduled); |