On 04/13/2018 10:31 AM, Christian König wrote:
> Am 13.04.2018 um 09:01 schrieb Emily Deng:
>> issue:
>> there are VMC page fault occurred if force APP kill during
>> 3dmark test, the cause is in entity_fini we manually signal
>> all those jobs in entity's queue which confuse the sync/dep
>> mechanism:
>>
>> 1)page fault occurred in sdma's clear job which operate on
>> shadow buffer, and shadow buffer's Gart table is cleaned by
>> ttm_bo_release since the fence in its reservation was fake signaled
>> by entity_fini() under the case of SIGKILL received.
>>
>> 2)page fault occurred in gfx' job because during the lifetime
>> of gfx job we manually fake signal all jobs from its entity
>> in entity_fini(), thus the unmapping/clear PTE job depend on those
>> result fence is satisfied and sdma start clearing the PTE and lead
>> to GFX page fault.
>>
>> fix:
>> 1)should at least wait all jobs already scheduled complete in
>> entity_fini()
>> if SIGKILL is the case.
>>
>> 2)if a fence signaled and try to clear some entity's dependency, should
>> set this entity guilty to prevent its job really run since the
>> dependency
>> is fake signaled.
>>
>> v2:
>> splitting drm_sched_entity_fini() into two functions:
>> 1)The first one is does the waiting, removes the entity from the
>> runqueue and returns an error when the process was killed.
>> 2)The second one then goes over the entity, install it as
>> completion signal for the remaining jobs and signals all jobs
>> with an error code.
>>
>> v3:
>> 1)Replace the fini1 and fini2 with better name
>> 2)Call the first part before the VM teardown in
>> amdgpu_driver_postclose_kms() and the second part
>> after the VM teardown
>> 3)Keep the original function drm_sched_entity_fini to
>> refine the code.
>>
>> v4:
>> 1)Rename entity->finished to entity->last_scheduled;
>> 2)Rename drm_sched_entity_fini_job_cb() to
>> drm_sched_entity_kill_jobs_cb();
>> 3)Pass NULL to drm_sched_entity_fini_job_cb() if -ENOENT;
>> 4)Replace the type of entity->fini_status with "int";
>> 5)Remove the check about entity->finished.
>>
>> Signed-off-by: Monk Liu <Monk.Liu at amd.com>
>> Signed-off-by: Emily Deng <Emily.Deng at amd.com>
>
> At least of hand that looks really good.
>
> Patch is Reviewed-by: Christian König <christian.koenig at amd.com>.
>
> Andrey, David you guys also recently hacked on the scheduler, so can
> please take a look as well?
Took a look at the patch and previous discussions, as far as I can
understand, looks good to me.
Andrey
>
> Thanks,
> Christian.
>
>> ---
>>  drivers/gpu/drm/amd/amdgpu/amdgpu.h      | 2 +
>>  drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c  | 64
>> ++++++++++++++++++++++++----
>>  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c  | 5 ++-
>> Â drivers/gpu/drm/scheduler/gpu_scheduler.c | 71
>> ++++++++++++++++++++++++++-----
>>  include/drm/gpu_scheduler.h              | 7 +++
>> Â 5 files changed, 128 insertions(+), 21 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> index 5734871..b3d047d 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> @@ -681,6 +681,8 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void
>> *data,
>> Â int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx, unsigned
>> ring_id);
>> Â Â void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr);
>> +void amdgpu_ctx_mgr_entity_cleanup(struct amdgpu_ctx_mgr *mgr);
>> +void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr);
>> Â void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
>> Â Â diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>> index 09d35051..eb80edf 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>> @@ -111,8 +111,9 @@ static int amdgpu_ctx_init(struct amdgpu_device
>> *adev,
>> Â Â Â Â Â return r;
>> Â }
>> Â -static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
>> +static void amdgpu_ctx_fini(struct kref *ref)
>> Â {
>> +Â Â Â struct amdgpu_ctx *ctx = container_of(ref, struct amdgpu_ctx,
>> refcount);
>> Â Â Â Â Â struct amdgpu_device *adev = ctx->adev;
>> Â Â Â Â Â unsigned i, j;
>> Â @@ -125,13 +126,11 @@ static void amdgpu_ctx_fini(struct amdgpu_ctx
>> *ctx)
>> Â Â Â Â Â kfree(ctx->fences);
>> Â Â Â Â Â ctx->fences = NULL;
>> Â -Â Â Â for (i = 0; i < adev->num_rings; i++)
>> -Â Â Â Â Â Â Â drm_sched_entity_fini(&adev->rings[i]->sched,
>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â &ctx->rings[i].entity);
>> -
>> Â Â Â Â Â amdgpu_queue_mgr_fini(adev, &ctx->queue_mgr);
>> Â Â Â Â Â Â mutex_destroy(&ctx->lock);
>> +
>> +Â Â Â kfree(ctx);
>> Â }
>> Â Â static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
>> @@ -170,12 +169,15 @@ static int amdgpu_ctx_alloc(struct
>> amdgpu_device *adev,
>> Â static void amdgpu_ctx_do_release(struct kref *ref)
>> Â {
>> Â Â Â Â Â struct amdgpu_ctx *ctx;
>> +Â Â Â u32 i;
>> Â Â Â Â Â Â ctx = container_of(ref, struct amdgpu_ctx, refcount);
>> Â -Â Â Â amdgpu_ctx_fini(ctx);
>> +Â Â Â for (i = 0; i < ctx->adev->num_rings; i++)
>> + drm_sched_entity_fini(&ctx->adev->rings[i]->sched,
>> +Â Â Â Â Â Â Â Â Â Â Â &ctx->rings[i].entity);
>> Â -Â Â Â kfree(ctx);
>> +Â Â Â amdgpu_ctx_fini(ref);
>> Â }
>> Â Â static int amdgpu_ctx_free(struct amdgpu_fpriv *fpriv, uint32_t id)
>> @@ -435,16 +437,62 @@ void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr
>> *mgr)
>> Â Â Â Â Â idr_init(&mgr->ctx_handles);
>> Â }
>> Â +void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr)
>> +{
>> +Â Â Â struct amdgpu_ctx *ctx;
>> +Â Â Â struct idr *idp;
>> +Â Â Â uint32_t id, i;
>> +
>> +Â Â Â idp = &mgr->ctx_handles;
>> +
>> +Â Â Â idr_for_each_entry(idp, ctx, id) {
>> +
>> +Â Â Â Â Â Â Â if (!ctx->adev)
>> +Â Â Â Â Â Â Â Â Â Â Â return;
>> +
>> +Â Â Â Â Â Â Â for (i = 0; i < ctx->adev->num_rings; i++)
>> +Â Â Â Â Â Â Â Â Â Â Â if (kref_read(&ctx->refcount) == 1)
>> + drm_sched_entity_do_release(&ctx->adev->rings[i]->sched,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â &ctx->rings[i].entity);
>> +Â Â Â Â Â Â Â Â Â Â Â else
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â DRM_ERROR("ctx %p is still alive\n", ctx);
>> +Â Â Â }
>> +}
>> +
>> +void amdgpu_ctx_mgr_entity_cleanup(struct amdgpu_ctx_mgr *mgr)
>> +{
>> +Â Â Â struct amdgpu_ctx *ctx;
>> +Â Â Â struct idr *idp;
>> +Â Â Â uint32_t id, i;
>> +
>> +Â Â Â idp = &mgr->ctx_handles;
>> +
>> +Â Â Â idr_for_each_entry(idp, ctx, id) {
>> +
>> +Â Â Â Â Â Â Â if (!ctx->adev)
>> +Â Â Â Â Â Â Â Â Â Â Â return;
>> +
>> +Â Â Â Â Â Â Â for (i = 0; i < ctx->adev->num_rings; i++)
>> +Â Â Â Â Â Â Â Â Â Â Â if (kref_read(&ctx->refcount) == 1)
>> + drm_sched_entity_cleanup(&ctx->adev->rings[i]->sched,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â &ctx->rings[i].entity);
>> +Â Â Â Â Â Â Â Â Â Â Â else
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â DRM_ERROR("ctx %p is still alive\n", ctx);
>> +Â Â Â }
>> +}
>> +
>> Â void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
>> Â {
>> Â Â Â Â Â struct amdgpu_ctx *ctx;
>> Â Â Â Â Â struct idr *idp;
>> Â Â Â Â Â uint32_t id;
>> Â +Â Â Â amdgpu_ctx_mgr_entity_cleanup(mgr);
>> +
>> Â Â Â Â Â idp = &mgr->ctx_handles;
>> Â Â Â Â Â Â idr_for_each_entry(idp, ctx, id) {
>> -Â Â Â Â Â Â Â if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1)
>> +Â Â Â Â Â Â Â if (kref_put(&ctx->refcount, amdgpu_ctx_fini) != 1)
>> Â Â Â Â Â Â Â Â Â Â Â Â Â DRM_ERROR("ctx %p is still alive\n", ctx);
>> Â Â Â Â Â }
>> Â diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>> index 487d39e..6cbb427 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>> @@ -913,8 +913,7 @@ void amdgpu_driver_postclose_kms(struct
>> drm_device *dev,
>> Â Â Â Â Â Â Â Â Â return;
>> Â Â Â Â Â Â pm_runtime_get_sync(dev->dev);
>> -
>> -Â Â Â amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
>> +Â Â Â amdgpu_ctx_mgr_entity_fini(&fpriv->ctx_mgr);
>> Â Â Â Â Â Â if (adev->asic_type != CHIP_RAVEN) {
>> Â Â Â Â Â Â Â Â Â amdgpu_uvd_free_handles(adev, file_priv);
>> @@ -935,6 +934,8 @@ void amdgpu_driver_postclose_kms(struct
>> drm_device *dev,
>> Â Â Â Â Â pd = amdgpu_bo_ref(fpriv->vm.root.base.bo);
>> Â Â Â Â Â Â amdgpu_vm_fini(adev, &fpriv->vm);
>> +Â Â Â amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
>> +
>> Â Â Â Â Â if (pasid)
>> Â Â Â Â Â Â Â Â Â amdgpu_pasid_free_delayed(pd->tbo.resv, pasid);
>> Â Â Â Â Â amdgpu_bo_unref(&pd);
>> diff --git a/drivers/gpu/drm/scheduler/gpu_scheduler.c
>> b/drivers/gpu/drm/scheduler/gpu_scheduler.c
>> index 310275e..44d2198 100644
>> --- a/drivers/gpu/drm/scheduler/gpu_scheduler.c
>> +++ b/drivers/gpu/drm/scheduler/gpu_scheduler.c
>> @@ -136,6 +136,8 @@ int drm_sched_entity_init(struct
>> drm_gpu_scheduler *sched,
>> Â Â Â Â Â entity->rq = rq;
>> Â Â Â Â Â entity->sched = sched;
>> Â Â Â Â Â entity->guilty = guilty;
>> +Â Â Â entity->fini_status = 0;
>> +Â Â Â entity->last_scheduled = NULL;
>> Â Â Â Â Â Â spin_lock_init(&entity->rq_lock);
>> Â Â Â Â Â spin_lock_init(&entity->queue_lock);
>> @@ -197,19 +199,30 @@ static bool drm_sched_entity_is_ready(struct
>> drm_sched_entity *entity)
>> Â Â Â Â Â return true;
>> Â }
>> Â +static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct dma_fence_cb *cb)
>> +{
>> +Â Â Â struct drm_sched_job *job = container_of(cb, struct drm_sched_job,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â finish_cb);
>> +Â Â Â drm_sched_fence_finished(job->s_fence);
>> +Â Â Â WARN_ON(job->s_fence->parent);
>> +Â Â Â dma_fence_put(&job->s_fence->finished);
>> +Â Â Â job->sched->ops->free_job(job);
>> +}
>> +
>> +
>> Â /**
>> Â Â * Destroy a context entity
>> Â Â *
>>   * @sched      Pointer to scheduler instance
>>   * @entity   The pointer to a valid scheduler entity
>> Â Â *
>> - * Cleanup and free the allocated resources.
>> + * Splitting drm_sched_entity_fini() into two functions, The first
>> one is does the waiting,
>> + * removes the entity from the runqueue and returns an error when
>> the process was killed.
>> Â Â */
>> -void drm_sched_entity_fini(struct drm_gpu_scheduler *sched,
>> +void drm_sched_entity_do_release(struct drm_gpu_scheduler *sched,
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_entity *entity)
>> Â {
>> -Â Â Â int r;
>> -
>> Â Â Â Â Â if (!drm_sched_entity_is_initialized(sched, entity))
>> Â Â Â Â Â Â Â Â Â return;
>> Â Â Â Â Â /**
>> @@ -217,13 +230,28 @@ void drm_sched_entity_fini(struct
>> drm_gpu_scheduler *sched,
>> Â Â Â Â Â Â * queued IBs or discard them on SIGKILL
>> Â Â Â Â Â */
>> Â Â Â Â Â if ((current->flags & PF_SIGNALED) && current->exit_code ==
>> SIGKILL)
>> -Â Â Â Â Â Â Â r = -ERESTARTSYS;
>> +Â Â Â Â Â Â Â entity->fini_status = -ERESTARTSYS;
>> Â Â Â Â Â else
>> -Â Â Â Â Â Â Â r = wait_event_killable(sched->job_scheduled,
>> +Â Â Â Â Â Â Â entity->fini_status = wait_event_killable(sched->job_scheduled,
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â drm_sched_entity_is_idle(entity));
>> Â Â Â Â Â drm_sched_entity_set_rq(entity, NULL);
>> -Â Â Â if (r) {
>> +}
>> +EXPORT_SYMBOL(drm_sched_entity_do_release);
>> +
>> +/**
>> + * Destroy a context entity
>> + *
>> + * @sched      Pointer to scheduler instance
>> + * @entity   The pointer to a valid scheduler entity
>> + *
>> + * The second one then goes over the entity and signals all jobs
>> with an error code.
>> + */
>> +void drm_sched_entity_cleanup(struct drm_gpu_scheduler *sched,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_entity *entity)
>> +{
>> +Â Â Â if (entity->fini_status) {
>> Â Â Â Â Â Â Â Â Â struct drm_sched_job *job;
>> +Â Â Â Â Â Â Â int r;
>> Â Â Â Â Â Â Â Â Â Â /* Park the kernel for a moment to make sure it isn't
>> processing
>> Â Â Â Â Â Â Â Â Â Â * our enity.
>> @@ -241,13 +269,26 @@ void drm_sched_entity_fini(struct
>> drm_gpu_scheduler *sched,
>> Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_fence *s_fence = job->s_fence;
>> Â Â Â Â Â Â Â Â Â Â Â Â Â drm_sched_fence_scheduled(s_fence);
>> Â Â Â Â Â Â Â Â Â Â Â Â Â dma_fence_set_error(&s_fence->finished, -ESRCH);
>> -Â Â Â Â Â Â Â Â Â Â Â drm_sched_fence_finished(s_fence);
>> -Â Â Â Â Â Â Â Â Â Â Â WARN_ON(s_fence->parent);
>> -Â Â Â Â Â Â Â Â Â Â Â dma_fence_put(&s_fence->finished);
>> -Â Â Â Â Â Â Â Â Â Â Â sched->ops->free_job(job);
>> +Â Â Â Â Â Â Â Â Â Â Â r = dma_fence_add_callback(entity->last_scheduled,
>> &job->finish_cb,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â drm_sched_entity_kill_jobs_cb);
>> +Â Â Â Â Â Â Â Â Â Â Â if (r == -ENOENT)
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â drm_sched_entity_kill_jobs_cb(NULL, &job->finish_cb);
>> +Â Â Â Â Â Â Â Â Â Â Â else if (r)
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â DRM_ERROR("fence add callback failed (%d)\n", r);
>> Â Â Â Â Â Â Â Â Â }
>> +
>> +Â Â Â Â Â Â Â dma_fence_put(entity->last_scheduled);
>> +Â Â Â Â Â Â Â entity->last_scheduled = NULL;
>> Â Â Â Â Â }
>> Â }
>> +EXPORT_SYMBOL(drm_sched_entity_cleanup);
>> +
>> +void drm_sched_entity_fini(struct drm_gpu_scheduler *sched,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_entity *entity)
>> +{
>> +Â Â Â drm_sched_entity_do_release(sched, entity);
>> +Â Â Â drm_sched_entity_cleanup(sched, entity);
>> +}
>> Â EXPORT_SYMBOL(drm_sched_entity_fini);
>> Â Â static void drm_sched_entity_wakeup(struct dma_fence *f, struct
>> dma_fence_cb *cb)
>> @@ -530,6 +571,10 @@ void drm_sched_job_recovery(struct
>> drm_gpu_scheduler *sched)
>> Â Â Â Â Â Â Â Â Â spin_unlock(&sched->job_list_lock);
>> Â Â Â Â Â Â Â Â Â fence = sched->ops->run_job(s_job);
>> Â Â Â Â Â Â Â Â Â atomic_inc(&sched->hw_rq_count);
>> +
>> +Â Â Â Â Â Â Â dma_fence_put(s_job->entity->last_scheduled);
>> +Â Â Â Â Â Â Â s_job->entity->last_scheduled =
>> dma_fence_get(&s_fence->finished);
>> +
>> Â Â Â Â Â Â Â Â Â if (fence) {
>> Â Â Â Â Â Â Â Â Â Â Â Â Â s_fence->parent = dma_fence_get(fence);
>> Â Â Â Â Â Â Â Â Â Â Â Â Â r = dma_fence_add_callback(fence, &s_fence->cb,
>> @@ -556,6 +601,7 @@ int drm_sched_job_init(struct drm_sched_job *job,
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â void *owner)
>> Â {
>> Â Â Â Â Â job->sched = sched;
>> +Â Â Â job->entity = entity;
>> Â Â Â Â Â job->s_priority = entity->rq - sched->sched_rq;
>> Â Â Â Â Â job->s_fence = drm_sched_fence_create(entity, owner);
>> Â Â Â Â Â if (!job->s_fence)
>> @@ -669,6 +715,9 @@ static int drm_sched_main(void *param)
>> Â Â Â Â Â Â Â Â Â fence = sched->ops->run_job(sched_job);
>> Â Â Â Â Â Â Â Â Â drm_sched_fence_scheduled(s_fence);
>> Â +Â Â Â Â Â Â Â dma_fence_put(entity->last_scheduled);
>> +Â Â Â Â Â Â Â entity->last_scheduled = dma_fence_get(&s_fence->finished);
>> +
>> Â Â Â Â Â Â Â Â Â if (fence) {
>> Â Â Â Â Â Â Â Â Â Â Â Â Â s_fence->parent = dma_fence_get(fence);
>> Â Â Â Â Â Â Â Â Â Â Â Â Â r = dma_fence_add_callback(fence, &s_fence->cb,
>> diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
>> index c053a32..350a62c 100644
>> --- a/include/drm/gpu_scheduler.h
>> +++ b/include/drm/gpu_scheduler.h
>> @@ -65,6 +65,8 @@ struct drm_sched_entity {
>>      struct dma_fence       *dependency;
>>      struct dma_fence_cb       cb;
>>      atomic_t           *guilty; /* points to ctx's guilty */
>> +   int           fini_status;
>> +   struct dma_fence   *last_scheduled;
>> Â };
>> Â Â /**
>> @@ -119,6 +121,7 @@ struct drm_sched_job {
>>      uint64_t           id;
>>      atomic_t           karma;
>>      enum drm_sched_priority       s_priority;
>> +   struct drm_sched_entity *entity;
>> Â };
>> Â Â static inline bool drm_sched_invalidate_job(struct drm_sched_job
>> *s_job,
>> @@ -186,6 +189,10 @@ int drm_sched_entity_init(struct
>> drm_gpu_scheduler *sched,
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_entity *entity,
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_rq *rq,
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â uint32_t jobs, atomic_t *guilty);
>> +void drm_sched_entity_do_release(struct drm_gpu_scheduler *sched,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_entity *entity);
>> +void drm_sched_entity_cleanup(struct drm_gpu_scheduler *sched,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_entity *entity);
>> Â void drm_sched_entity_fini(struct drm_gpu_scheduler *sched,
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â struct drm_sched_entity *entity);
>> Â void drm_sched_entity_push_job(struct drm_sched_job *sched_job,
>
