From 7d64136e0f08c108434962d5a4c0c038c034caaa Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Ma=C3=ADra=20Canal?= Date: Thu, 4 Jun 2026 17:32:25 -0300 Subject: [PATCH] drm/v3d: Ensure atomic submissions in v3d_submit_jobs() MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Currently, v3d_submit_jobs() arms and pushes each job one at a time, wiring dependencies between consecutive jobs after each push. If drm_sched_job_add_dependency() fails midway, the already-pushed jobs are scheduler-owned and will be submitted to the GPU for execution, even though the subsequent jobs won't be submitted. This breaks the atomicity of the submissions, as only some of the jobs from a submission would be submitted to the hardware, while the other part fails. Restructure v3d_submit_jobs() into three phases: (1) arm all jobs belonging to a given submission, (2) wire inter-job dependencies, and (3) push all jobs to the scheduler unconditionally. Phase (2) can fail; on failure, it marks every armed job finished fence with an error, so that run_job() callbacks skip hardware execution. This guarantees that every armed job is always pushed, either to run or to be skipped, and it also ensures the atomicity of a submission. Suggested-by: Tvrtko Ursulin Reviewed-by: Tvrtko Ursulin Link: https://patch.msgid.link/20260604-v3d-sched-misc-fixes-v4-12-c068f5bf5ccf@igalia.com Signed-off-by: Maíra Canal --- drivers/gpu/drm/v3d/v3d_sched.c | 6 +++ drivers/gpu/drm/v3d/v3d_submit.c | 65 ++++++++++++++++---------------- 2 files changed, 39 insertions(+), 32 deletions(-) diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index 7f9a71022e5e..53deae1a2eda 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -657,6 +657,9 @@ v3d_cpu_job_run(struct drm_sched_job *sched_job) struct v3d_cpu_job *job = to_cpu_job(sched_job); struct v3d_dev *v3d = job->base.v3d; + if (unlikely(job->base.base.s_fence->finished.error)) + return NULL; + if (job->job_type >= ARRAY_SIZE(cpu_job_function)) { drm_dbg(&v3d->drm, "Unknown CPU job: %d\n", job->job_type); return NULL; @@ -680,6 +683,9 @@ v3d_cache_clean_job_run(struct drm_sched_job *sched_job) struct v3d_job *job = to_v3d_job(sched_job); struct v3d_dev *v3d = job->v3d; + if (unlikely(job->base.s_fence->finished.error)) + return NULL; + v3d_job_start_stats(job); v3d_clean_caches(v3d); diff --git a/drivers/gpu/drm/v3d/v3d_submit.c b/drivers/gpu/drm/v3d/v3d_submit.c index cee48acd25d9..ee2ac2540ed5 100644 --- a/drivers/gpu/drm/v3d/v3d_submit.c +++ b/drivers/gpu/drm/v3d/v3d_submit.c @@ -368,19 +368,6 @@ v3d_submit_process_post_deps(struct v3d_submit *submit, struct drm_syncobj *sync } } -static void -v3d_push_job(struct v3d_job *job) -{ - drm_sched_job_arm(&job->base); - - job->done_fence = dma_fence_get(&job->base.s_fence->finished); - - /* put by scheduler job completion */ - kref_get(&job->refcount); - - drm_sched_entity_push_job(&job->base); -} - static int v3d_submit_jobs(struct v3d_submit *submit, struct drm_syncobj *sync_out, struct v3d_submit_ext *se) @@ -393,16 +380,23 @@ v3d_submit_jobs(struct v3d_submit *submit, struct drm_syncobj *sync_out, for (int i = 0; i < submit->job_count; i++) { struct v3d_job *job = submit->jobs[i]; - v3d_push_job(job); + drm_sched_job_arm(&job->base); + job->done_fence = dma_fence_get(&job->base.s_fence->finished); - if (i + 1 < submit->job_count) { - ret = drm_sched_job_add_dependency(&submit->jobs[i + 1]->base, - dma_fence_get(job->done_fence)); - if (ret) - goto err; - } + /* put by scheduler job completion */ + kref_get(&job->refcount); } + for (int i = 1; i < submit->job_count; i++) { + ret = drm_sched_job_add_dependency(&submit->jobs[i]->base, + dma_fence_get(submit->jobs[i - 1]->done_fence)); + if (ret) + goto err; + } + + for (int i = 0; i < submit->job_count; i++) + drm_sched_entity_push_job(&submit->jobs[i]->base); + mutex_unlock(&v3d->sched_lock); v3d_submit_attach_object_fences(submit); @@ -414,7 +408,18 @@ v3d_submit_jobs(struct v3d_submit *submit, struct drm_syncobj *sync_out, return 0; err: + /* Mark every armed job as failed so run_job() skips execution */ + for (int i = 0; i < submit->job_count; i++) + dma_fence_set_error(&submit->jobs[i]->base.s_fence->finished, ret); + + for (int i = 0; i < submit->job_count; i++) + drm_sched_entity_push_job(&submit->jobs[i]->base); + mutex_unlock(&v3d->sched_lock); + + v3d_submit_unlock_reservations(submit); + v3d_submit_put_jobs(submit); + return ret; } @@ -1102,14 +1107,13 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data, ret = v3d_submit_jobs(&submit, sync_out, &se); if (ret) - goto fail_unreserve; + goto fail_submit; return 0; -fail_unreserve: - v3d_submit_unlock_reservations(&submit); fail: v3d_submit_cleanup_jobs(&submit); +fail_submit: v3d_submit_put_post_deps(sync_out, &se); return ret; @@ -1200,14 +1204,13 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data, ret = v3d_submit_jobs(&submit, sync_out, &se); if (ret) - goto fail_unreserve; + goto fail_submit; return 0; -fail_unreserve: - v3d_submit_unlock_reservations(&submit); fail: v3d_submit_cleanup_jobs(&submit); +fail_submit: v3d_submit_put_post_deps(sync_out, &se); return ret; @@ -1276,14 +1279,13 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data, ret = v3d_submit_jobs(&submit, sync_out, &se); if (ret) - goto fail_unreserve; + goto fail_submit; return 0; -fail_unreserve: - v3d_submit_unlock_reservations(&submit); fail: v3d_submit_cleanup_jobs(&submit); +fail_submit: v3d_submit_put_post_deps(sync_out, &se); return ret; @@ -1384,14 +1386,13 @@ v3d_submit_cpu_ioctl(struct drm_device *dev, void *data, ret = v3d_submit_jobs(&submit, NULL, &se); if (ret) - goto fail_unreserve; + goto fail_submit; return 0; -fail_unreserve: - v3d_submit_unlock_reservations(&submit); fail: v3d_submit_cleanup_jobs(&submit); +fail_submit: v3d_submit_put_post_deps(NULL, &se); return ret;