Since its inception, the GPU scheduler can leak memory if the driver calls drm_sched_fini() while there are still jobs in flight.
The simplest way to solve this in a backwards compatible manner is by adding a new callback, drm_sched_backend_ops.cancel_job(), which instructs the driver to signal the hardware fence associated with the job. Afterwards, the scheduler can safely use the established free_job() callback for freeing the job. Implement the new backend_ops callback cancel_job(). Suggested-by: Tvrtko Ursulin <tvrtko.ursu...@igalia.com> Link: https://lore.kernel.org/dri-devel/20250418113211.69956-1-tvrtko.ursu...@igalia.com/ Signed-off-by: Philipp Stanner <pha...@kernel.org> Reviewed-by: Maíra Canal <mca...@igalia.com> --- drivers/gpu/drm/scheduler/sched_main.c | 34 ++++++++++++++++---------- include/drm/gpu_scheduler.h | 18 ++++++++++++++ 2 files changed, 39 insertions(+), 13 deletions(-) diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c index 81ad40d9582b..a971f0c9e6e0 100644 --- a/drivers/gpu/drm/scheduler/sched_main.c +++ b/drivers/gpu/drm/scheduler/sched_main.c @@ -1352,6 +1352,18 @@ int drm_sched_init(struct drm_gpu_scheduler *sched, const struct drm_sched_init_ } EXPORT_SYMBOL(drm_sched_init); +static void drm_sched_cancel_remaining_jobs(struct drm_gpu_scheduler *sched) +{ + struct drm_sched_job *job, *tmp; + + /* All other accessors are stopped. No locking necessary. */ + list_for_each_entry_safe_reverse(job, tmp, &sched->pending_list, list) { + sched->ops->cancel_job(job); + list_del(&job->list); + sched->ops->free_job(job); + } +} + /** * drm_sched_fini - Destroy a gpu scheduler * @@ -1359,19 +1371,11 @@ EXPORT_SYMBOL(drm_sched_init); * * Tears down and cleans up the scheduler. * - * This stops submission of new jobs to the hardware through - * drm_sched_backend_ops.run_job(). Consequently, drm_sched_backend_ops.free_job() - * will not be called for all jobs still in drm_gpu_scheduler.pending_list. - * There is no solution for this currently. Thus, it is up to the driver to make - * sure that: - * - * a) drm_sched_fini() is only called after for all submitted jobs - * drm_sched_backend_ops.free_job() has been called or that - * b) the jobs for which drm_sched_backend_ops.free_job() has not been called - * after drm_sched_fini() ran are freed manually. - * - * FIXME: Take care of the above problem and prevent this function from leaking - * the jobs in drm_gpu_scheduler.pending_list under any circumstances. + * This stops submission of new jobs to the hardware through &struct + * drm_sched_backend_ops.run_job. If &struct drm_sched_backend_ops.cancel_job + * is implemented, all jobs will be canceled through it and afterwards cleaned + * up through &struct drm_sched_backend_ops.free_job. If cancel_job is not + * implemented, memory could leak. */ void drm_sched_fini(struct drm_gpu_scheduler *sched) { @@ -1401,6 +1405,10 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched) /* Confirm no work left behind accessing device structures */ cancel_delayed_work_sync(&sched->work_tdr); + /* Avoid memory leaks if supported by the driver. */ + if (sched->ops->cancel_job) + drm_sched_cancel_remaining_jobs(sched); + if (sched->own_submit_wq) destroy_workqueue(sched->submit_wq); sched->ready = false; diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h index e62a7214e052..190844370f48 100644 --- a/include/drm/gpu_scheduler.h +++ b/include/drm/gpu_scheduler.h @@ -512,6 +512,24 @@ struct drm_sched_backend_ops { * and it's time to clean it up. */ void (*free_job)(struct drm_sched_job *sched_job); + + /** + * @cancel_job: Used by the scheduler to guarantee remaining jobs' fences + * get signaled in drm_sched_fini(). + * + * Used by the scheduler to cancel all jobs that have not been executed + * with &struct drm_sched_backend_ops.run_job by the time + * drm_sched_fini() gets invoked. + * + * Drivers need to signal the passed job's hardware fence with an + * appropriate error code (e.g., -ECANCELED) in this callback. They + * must not free the job. + * + * The scheduler will only call this callback once it stopped calling + * all other callbacks forever, with the exception of &struct + * drm_sched_backend_ops.free_job. + */ + void (*cancel_job)(struct drm_sched_job *sched_job); }; /** -- 2.49.0
