drm/sched: Fix memory leaks with cancel_job() callback

---
 drivers/gpu/drm/panfrost/panfrost_job.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

drivers/gpu/drm/panfrost/panfrost_job.c +1 -1

index 5657106c2f7d..15e2d505550f 100644
@@ -841,7 +841,6 @@ int panfrost_job_init(struct panfrost_device *pfdev)
 		.num_rqs = DRM_SCHED_PRIORITY_COUNT,
 		.credit_limit = 2,
 		.timeout = msecs_to_jiffies(JOB_TIMEOUT_MS),
-		.timeout_wq = pfdev->reset.wq,
 		.name = "pan_js",
 		.dev = pfdev->dev,
 	};

@@ -879,6 +878,7 @@ int panfrost_job_init(struct panfrost_device *pfdev)
 	pfdev->reset.wq = alloc_ordered_workqueue("panfrost-reset", 0);
 	if (!pfdev->reset.wq)
 		return -ENOMEM;
+	args.timeout_wq = pfdev->reset.wq;
 
 	for (j = 0; j < NUM_JOB_SLOTS; j++) {
 		js->queue[j].fence_context = dma_fence_context_alloc(1);

Signature · 1 line

2.49.0

Philipp Stanner

2025-07-10 12:58 UTC #2 lore raw

Re: [PATCH v4 1/8] drm/panfrost: Fix scheduler workqueue bug

Quoted 10 lines · Philipp Stanner

When the GPU scheduler was ported to using a struct for its initialization parameters, it was overlooked that panfrost creates a distinct workqueue for timeout handling.
The pointer to this new workqueue is not initialized to the struct, resulting in NULL being passed to the scheduler, which then uses the system_wq for timeout handling.
Set the correct workqueue to the init args struct.

Quoted 8 lines

Cc [email protected] # 6.15+
Fixes 796a9f55a8d1 ("drm/sched: Use struct for drm_sched_init()
params")
Reported-by Tvrtko Ursulin <[email protected]>
Closes:
https://lore.kernel.org/dri-devel/[email protected]/
Signed-off-by Philipp Stanner <[email protected]>
---

aaaarrrgh, how did that one get here!

Ignore that. Will not be merged through this series.

Quoted 28 lines

drivers/gpu/drm/panfrost/panfrost_job.c | 2 +-

 1 file changed, 1 insertion(+), 1 deletion(-)

drivers/gpu/drm/panfrost/panfrost_job.c +0 -1

index 5657106c2f7d..15e2d505550f 100644
@@ -841,7 +841,6 @@ int panfrost_job_init(struct panfrost_device
 *pfdev)
  		.num_rqs = DRM_SCHED_PRIORITY_COUNT,
  		.credit_limit = 2,
  		.timeout = msecs_to_jiffies(JOB_TIMEOUT_MS),
-		.timeout_wq = pfdev->reset.wq,
  		.name = "pan_js",
  		.dev = pfdev->dev,

};

drivers/gpu/drm/panfrost/panfrost_job.c +1 -0

@@ -879,6 +878,7 @@ int panfrost_job_init(struct panfrost_device
 *pfdev)
  	pfdev->reset.wq = alloc_ordered_workqueue("panfrost-reset",
 0);
  	if (!pfdev->reset.wq)
  		return -ENOMEM;
+	args.timeout_wq = pfdev->reset.wq;

for (j = 0; j < NUM_JOB_SLOTS; j++) {
js->queue[j].fence_context =
dma_fence_context_alloc(1);

Philipp Stanner

author maintainer

2025-07-10 12:54 UTC #3 lore raw

PATCH v4 2/8 drm/sched: Avoid memory leaks with cancel_job() callback

Link https://lore.kernel.org/dri-devel/[email protected]/ open here

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().

Reviewed-by Maíra Canal <[email protected]>

---
 drivers/gpu/drm/scheduler/sched_main.c | 34 ++++++++++++++++----------
 include/drm/gpu_scheduler.h            | 18 ++++++++++++++
 2 files changed, 39 insertions(+), 13 deletions(-)

drivers/gpu/drm/scheduler/sched_main.c +21 ~~-13~~

index 81ad40d9582b..a971f0c9e6e0 100644
@@ -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;

include/drm/gpu_scheduler.h +18 -0

index e62a7214e052..190844370f48 100644
@@ -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);
 };
 
 /**

Signature · 1 line

2.49.0

Philipp Stanner

author maintainer

2025-07-10 12:54 UTC #4 lore raw

PATCH v4 3/8 drm/sched/tests: Implement cancel_job() callback

The GPU Scheduler now supports a new callback, cancel_job(), which lets the scheduler cancel all jobs which might not yet be freed when drm_sched_fini() runs. Using this callback allows for significantly simplifying the mock scheduler teardown code.

Implement the cancel_job() callback and adjust the code where necessary.

---
 .../gpu/drm/scheduler/tests/mock_scheduler.c  | 68 +++++++------------
 drivers/gpu/drm/scheduler/tests/sched_tests.h |  1 -
 2 files changed, 25 insertions(+), 44 deletions(-)

drivers/gpu/drm/scheduler/tests/mock_scheduler.c +25 ~~-43~~

index 49d067fecd67..0d1d57213e05 100644
@@ -63,7 +63,7 @@ static void drm_mock_sched_job_complete(struct drm_mock_sched_job *job)
 	lockdep_assert_held(&sched->lock);
 
 	job->flags |= DRM_MOCK_SCHED_JOB_DONE;
-	list_move_tail(&job->link, &sched->done_list);
+	list_del(&job->link);
 	dma_fence_signal_locked(&job->hw_fence);
 	complete(&job->done);
 }

@@ -236,26 +236,41 @@ mock_sched_timedout_job(struct drm_sched_job *sched_job)
 
 static void mock_sched_free_job(struct drm_sched_job *sched_job)
 {
-	struct drm_mock_scheduler *sched =
-			drm_sched_to_mock_sched(sched_job->sched);
 	struct drm_mock_sched_job *job = drm_sched_job_to_mock_job(sched_job);
-	unsigned long flags;
 
-	/* Remove from the scheduler done list. */
-	spin_lock_irqsave(&sched->lock, flags);
-	list_del(&job->link);
-	spin_unlock_irqrestore(&sched->lock, flags);
 	dma_fence_put(&job->hw_fence);
-
 	drm_sched_job_cleanup(sched_job);
 
 	/* Mock job itself is freed by the kunit framework. */
 }
 
+static void mock_sched_cancel_job(struct drm_sched_job *sched_job)
+{
+	struct drm_mock_scheduler *sched = drm_sched_to_mock_sched(sched_job->sched);
+	struct drm_mock_sched_job *job = drm_sched_job_to_mock_job(sched_job);
+	unsigned long flags;
+
+	hrtimer_cancel(&job->timer);
+
+	spin_lock_irqsave(&sched->lock, flags);
+	if (!dma_fence_is_signaled_locked(&job->hw_fence)) {
+		list_del(&job->link);
+		dma_fence_set_error(&job->hw_fence, -ECANCELED);
+		dma_fence_signal_locked(&job->hw_fence);
+	}
+	spin_unlock_irqrestore(&sched->lock, flags);
+
+	/*
+	 * The GPU Scheduler will call drm_sched_backend_ops.free_job(), still.
+	 * Mock job itself is freed by the kunit framework.
+	 */
+}
+
 static const struct drm_sched_backend_ops drm_mock_scheduler_ops = {
 	.run_job = mock_sched_run_job,
 	.timedout_job = mock_sched_timedout_job,
-	.free_job = mock_sched_free_job
+	.free_job = mock_sched_free_job,
+	.cancel_job = mock_sched_cancel_job,
 };
 
 /**

@@ -289,7 +304,6 @@ struct drm_mock_scheduler *drm_mock_sched_new(struct kunit *test, long timeout)
 	sched->hw_timeline.context = dma_fence_context_alloc(1);
 	atomic_set(&sched->hw_timeline.next_seqno, 0);
 	INIT_LIST_HEAD(&sched->job_list);
-	INIT_LIST_HEAD(&sched->done_list);
 	spin_lock_init(&sched->lock);
 
 	return sched;

@@ -304,38 +318,6 @@ struct drm_mock_scheduler *drm_mock_sched_new(struct kunit *test, long timeout)
  */
 void drm_mock_sched_fini(struct drm_mock_scheduler *sched)
 {
-	struct drm_mock_sched_job *job, *next;
-	unsigned long flags;
-	LIST_HEAD(list);
-
-	drm_sched_wqueue_stop(&sched->base);
-
-	/* Force complete all unfinished jobs. */
-	spin_lock_irqsave(&sched->lock, flags);
-	list_for_each_entry_safe(job, next, &sched->job_list, link)
-		list_move_tail(&job->link, &list);
-	spin_unlock_irqrestore(&sched->lock, flags);
-
-	list_for_each_entry(job, &list, link)
-		hrtimer_cancel(&job->timer);
-
-	spin_lock_irqsave(&sched->lock, flags);
-	list_for_each_entry_safe(job, next, &list, link)
-		drm_mock_sched_job_complete(job);
-	spin_unlock_irqrestore(&sched->lock, flags);
-
-	/*
-	 * Free completed jobs and jobs not yet processed by the DRM scheduler
-	 * free worker.
-	 */
-	spin_lock_irqsave(&sched->lock, flags);
-	list_for_each_entry_safe(job, next, &sched->done_list, link)
-		list_move_tail(&job->link, &list);
-	spin_unlock_irqrestore(&sched->lock, flags);
-
-	list_for_each_entry_safe(job, next, &list, link)
-		mock_sched_free_job(&job->base);
-
 	drm_sched_fini(&sched->base);
 }

drivers/gpu/drm/scheduler/tests/sched_tests.h +0 -1

index fbba38137f0c..0eddfb8d89e6 100644
@@ -49,7 +49,6 @@ struct drm_mock_scheduler {
 
 	spinlock_t		lock;
 	struct list_head	job_list;
-	struct list_head	done_list;
 
 	struct {
 		u64		context;

Signature · 1 line

2.49.0

Philipp Stanner

author maintainer

2025-07-10 12:54 UTC #5 lore raw

PATCH v4 4/8 drm/sched/tests: Add unit test for cancel_job()

The scheduler unit tests now provide a new callback, cancel_job(). This callback gets used by drm_sched_fini() for all still pending jobs to cancel them.

Implement a new unit test to test this.

Reviewed-by Tvrtko Ursulin <[email protected]>

---
 drivers/gpu/drm/scheduler/tests/tests_basic.c | 42 +++++++++++++++++++
 1 file changed, 42 insertions(+)

drivers/gpu/drm/scheduler/tests/tests_basic.c +42 -0

index 7230057e0594..b1ae10c6bb37 100644
@@ -204,6 +204,47 @@ static struct kunit_suite drm_sched_basic = {
 	.test_cases = drm_sched_basic_tests,
 };
 
+static void drm_sched_basic_cancel(struct kunit *test)
+{
+	struct drm_mock_sched_entity *entity;
+	struct drm_mock_scheduler *sched;
+	struct drm_mock_sched_job *job;
+	bool done;
+
+	/*
+	 * Check that drm_sched_fini() uses the cancel_job() callback to cancel
+	 * jobs that are still pending.
+	 */
+
+	sched = drm_mock_sched_new(test, MAX_SCHEDULE_TIMEOUT);
+	entity = drm_mock_sched_entity_new(test, DRM_SCHED_PRIORITY_NORMAL,
+					   sched);
+
+	job = drm_mock_sched_job_new(test, entity);
+
+	drm_mock_sched_job_submit(job);
+
+	done = drm_mock_sched_job_wait_scheduled(job, HZ);
+	KUNIT_ASSERT_TRUE(test, done);
+
+	drm_mock_sched_entity_free(entity);
+	drm_mock_sched_fini(sched);
+
+	KUNIT_ASSERT_EQ(test, job->hw_fence.error, -ECANCELED);
+}
+
+static struct kunit_case drm_sched_cancel_tests[] = {
+	KUNIT_CASE(drm_sched_basic_cancel),
+	{}
+};
+
+static struct kunit_suite drm_sched_cancel = {
+	.name = "drm_sched_basic_cancel_tests",
+	.init = drm_sched_basic_init,
+	.exit = drm_sched_basic_exit,
+	.test_cases = drm_sched_cancel_tests,
+};
+
 static void drm_sched_basic_timeout(struct kunit *test)
 {
 	struct drm_mock_scheduler *sched = test->priv;

@@ -471,6 +512,7 @@ static struct kunit_suite drm_sched_credits = {
 
 kunit_test_suites(&drm_sched_basic,
 		  &drm_sched_timeout,
+		  &drm_sched_cancel,
 		  &drm_sched_priority,
 		  &drm_sched_modify_sched,
 		  &drm_sched_credits);

Signature · 1 line

2.49.0

Philipp Stanner

author maintainer

2025-07-10 12:54 UTC #6 lore raw

PATCH v4 5/8 drm/sched: Warn if pending_list is not empty

drm_sched_fini() can leak jobs under certain circumstances.

Warn if that happens.

---
 drivers/gpu/drm/scheduler/sched_main.c | 3 +++
 1 file changed, 3 insertions(+)

drivers/gpu/drm/scheduler/sched_main.c +3 -0

index a971f0c9e6e0..d94270cf8bf5 100644
@@ -1414,6 +1414,9 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched)
 	sched->ready = false;
 	kfree(sched->sched_rq);
 	sched->sched_rq = NULL;
+
+	if (!list_empty(&sched->pending_list))
+		dev_warn(sched->dev, "Tearing down scheduler while jobs are pending!\n");
 }
 EXPORT_SYMBOL(drm_sched_fini);

Signature · 1 line

2.49.0

Danilo Krummrich

maintainer Acked-by on v4 5/8

2025-07-10 13:04 UTC #7 lore raw

Re: [PATCH v4 5/8] drm/sched: Warn if pending_list is not empty

Quoted 5 lines · Philipp Stanner

drm_sched_fini() can leak jobs under certain circumstances.
Warn if that happens.
Signed-off-by Philipp Stanner <[email protected]>

Philipp Stanner

author maintainer

2025-07-10 12:54 UTC #8 lore raw

PATCH v4 6/8 drm/nouveau: Make fence container helper usable driver-wide

In order to implement a new DRM GPU scheduler callback in Nouveau, a helper for obtaining a nouveau_fence from a dma_fence is necessary. Such a helper exists already inside nouveau_fence.c, called from_fence().

Make that helper available to other C files with a more precise name.

---
 drivers/gpu/drm/nouveau/nouveau_fence.c | 20 +++++++-------------
 drivers/gpu/drm/nouveau/nouveau_fence.h |  6 ++++++
 2 files changed, 13 insertions(+), 13 deletions(-)

drivers/gpu/drm/nouveau/nouveau_fence.c +7 ~~-13~~

index d5654e26d5bc..869d4335c0f4 100644
@@ -38,12 +38,6 @@
 static const struct dma_fence_ops nouveau_fence_ops_uevent;
 static const struct dma_fence_ops nouveau_fence_ops_legacy;
 
-static inline struct nouveau_fence *
-from_fence(struct dma_fence *fence)
-{
-	return container_of(fence, struct nouveau_fence, base);
-}
-
 static inline struct nouveau_fence_chan *
 nouveau_fctx(struct nouveau_fence *fence)
 {

@@ -77,7 +71,7 @@ nouveau_local_fence(struct dma_fence *fence, struct nouveau_drm *drm)
 	    fence->ops != &nouveau_fence_ops_uevent)
 		return NULL;
 
-	return from_fence(fence);
+	return to_nouveau_fence(fence);
 }
 
 void

@@ -268,7 +262,7 @@ nouveau_fence_done(struct nouveau_fence *fence)
 static long
 nouveau_fence_wait_legacy(struct dma_fence *f, bool intr, long wait)
 {
-	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence *fence = to_nouveau_fence(f);
 	unsigned long sleep_time = NSEC_PER_MSEC / 1000;
 	unsigned long t = jiffies, timeout = t + wait;

@@ -448,7 +442,7 @@ static const char *nouveau_fence_get_get_driver_name(struct dma_fence *fence)
 
 static const char *nouveau_fence_get_timeline_name(struct dma_fence *f)
 {
-	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence *fence = to_nouveau_fence(f);
 	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
 
 	return !fctx->dead ? fctx->name : "dead channel";

@@ -462,7 +456,7 @@ static const char *nouveau_fence_get_timeline_name(struct dma_fence *f)
  */
 static bool nouveau_fence_is_signaled(struct dma_fence *f)
 {
-	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence *fence = to_nouveau_fence(f);
 	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
 	struct nouveau_channel *chan;
 	bool ret = false;

@@ -478,7 +472,7 @@ static bool nouveau_fence_is_signaled(struct dma_fence *f)
 
 static bool nouveau_fence_no_signaling(struct dma_fence *f)
 {
-	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence *fence = to_nouveau_fence(f);
 
 	/*
 	 * caller should have a reference on the fence,

@@ -503,7 +497,7 @@ static bool nouveau_fence_no_signaling(struct dma_fence *f)
 
 static void nouveau_fence_release(struct dma_fence *f)
 {
-	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence *fence = to_nouveau_fence(f);
 	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
 
 	kref_put(&fctx->fence_ref, nouveau_fence_context_put);

@@ -521,7 +515,7 @@ static const struct dma_fence_ops nouveau_fence_ops_legacy = {
 
 static bool nouveau_fence_enable_signaling(struct dma_fence *f)
 {
-	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence *fence = to_nouveau_fence(f);
 	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
 	bool ret;

drivers/gpu/drm/nouveau/nouveau_fence.h +6 -0

index 6a983dd9f7b9..183dd43ecfff 100644
@@ -17,6 +17,12 @@ struct nouveau_fence {
 	unsigned long timeout;
 };
 
+static inline struct nouveau_fence *
+to_nouveau_fence(struct dma_fence *fence)
+{
+	return container_of(fence, struct nouveau_fence, base);
+}
+
 int  nouveau_fence_create(struct nouveau_fence **, struct nouveau_channel *);
 int  nouveau_fence_new(struct nouveau_fence **, struct nouveau_channel *);
 void nouveau_fence_unref(struct nouveau_fence **);

Signature · 1 line

2.49.0

Philipp Stanner

author maintainer

2025-07-10 12:54 UTC #9 lore raw

PATCH v4 7/8 drm/nouveau: Add new callback for scheduler teardown

There is a new callback for always tearing the scheduler down in a leak-free, deadlock-free manner.

Port Nouveau as its first user by providing the scheduler with a callback that ensures the fence context gets killed in drm_sched_fini().

---
 drivers/gpu/drm/nouveau/nouveau_fence.c | 15 +++++++++++++++
 drivers/gpu/drm/nouveau/nouveau_fence.h |  1 +
 drivers/gpu/drm/nouveau/nouveau_sched.c | 15 ++++++++++++++-
 3 files changed, 30 insertions(+), 1 deletion(-)

drivers/gpu/drm/nouveau/nouveau_fence.c +15 -0

index 869d4335c0f4..9f345a008717 100644
@@ -240,6 +240,21 @@ nouveau_fence_emit(struct nouveau_fence *fence)
 	return ret;
 }
 
+void
+nouveau_fence_cancel(struct nouveau_fence *fence)
+{
+	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
+	unsigned long flags;
+
+	spin_lock_irqsave(&fctx->lock, flags);
+	if (!dma_fence_is_signaled_locked(&fence->base)) {
+		dma_fence_set_error(&fence->base, -ECANCELED);
+		if (nouveau_fence_signal(fence))
+			nvif_event_block(&fctx->event);
+	}
+	spin_unlock_irqrestore(&fctx->lock, flags);
+}
+
 bool
 nouveau_fence_done(struct nouveau_fence *fence)
 {

drivers/gpu/drm/nouveau/nouveau_fence.h +1 -0

index 183dd43ecfff..9957a919bd38 100644
@@ -29,6 +29,7 @@ void nouveau_fence_unref(struct nouveau_fence **);
 
 int  nouveau_fence_emit(struct nouveau_fence *);
 bool nouveau_fence_done(struct nouveau_fence *);
+void nouveau_fence_cancel(struct nouveau_fence *fence);
 int  nouveau_fence_wait(struct nouveau_fence *, bool lazy, bool intr);
 int  nouveau_fence_sync(struct nouveau_bo *, struct nouveau_channel *, bool exclusive, bool intr);

drivers/gpu/drm/nouveau/nouveau_sched.c +14 -1

index 460a5fb02412..2ec62059c351 100644
@@ -11,6 +11,7 @@
 #include "nouveau_exec.h"
 #include "nouveau_abi16.h"
 #include "nouveau_sched.h"
+#include "nouveau_chan.h"
 
 #define NOUVEAU_SCHED_JOB_TIMEOUT_MS		10000

@@ -393,10 +394,23 @@ nouveau_sched_free_job(struct drm_sched_job *sched_job)
 	nouveau_job_fini(job);
 }
 
+static void
+nouveau_sched_cancel_job(struct drm_sched_job *sched_job)
+{
+	struct nouveau_fence *fence;
+	struct nouveau_job *job;
+
+	job = to_nouveau_job(sched_job);
+	fence = to_nouveau_fence(job->done_fence);
+
+	nouveau_fence_cancel(fence);
+}
+
 static const struct drm_sched_backend_ops nouveau_sched_ops = {
 	.run_job = nouveau_sched_run_job,
 	.timedout_job = nouveau_sched_timedout_job,
 	.free_job = nouveau_sched_free_job,
+	.cancel_job = nouveau_sched_cancel_job,
 };
 
 static int

@@ -482,7 +496,6 @@ nouveau_sched_create(struct nouveau_sched **psched, struct nouveau_drm *drm,
 	return 0;
 }
 
-
 static void
 nouveau_sched_fini(struct nouveau_sched *sched)
 {

Signature · 1 line

2.49.0

Philipp Stanner

author maintainer

2025-07-10 12:55 UTC #10 lore raw

PATCH v4 8/8 drm/nouveau: Remove waitque for sched teardown

struct nouveau_sched contains a waitque needed to prevent drm_sched_fini() from being called while there are still jobs pending. Doing so so far would have caused memory leaks.

With the new memleak-free mode of operation switched on in drm_sched_fini() by providing the callback nouveau_sched_cancel_job() the waitque is not necessary anymore.

Remove the waitque.

---
 drivers/gpu/drm/nouveau/nouveau_sched.c | 20 +++++++-------------
 drivers/gpu/drm/nouveau/nouveau_sched.h |  9 +++------
 drivers/gpu/drm/nouveau/nouveau_uvmm.c  |  8 ++++----
 3 files changed, 14 insertions(+), 23 deletions(-)

drivers/gpu/drm/nouveau/nouveau_sched.c +7 ~~-13~~

index 2ec62059c351..7d9c3418e76b 100644
@@ -122,11 +122,9 @@ nouveau_job_done(struct nouveau_job *job)
 {
 	struct nouveau_sched *sched = job->sched;
 
-	spin_lock(&sched->job.list.lock);
+	spin_lock(&sched->job_list.lock);
 	list_del(&job->entry);
-	spin_unlock(&sched->job.list.lock);
-
-	wake_up(&sched->job.wq);
+	spin_unlock(&sched->job_list.lock);
 }
 
 void

@@ -307,9 +305,9 @@ nouveau_job_submit(struct nouveau_job *job)
 	}
 
 	/* Submit was successful; add the job to the schedulers job list. */
-	spin_lock(&sched->job.list.lock);
-	list_add(&job->entry, &sched->job.list.head);
-	spin_unlock(&sched->job.list.lock);
+	spin_lock(&sched->job_list.lock);
+	list_add(&job->entry, &sched->job_list.head);
+	spin_unlock(&sched->job_list.lock);
 
 	drm_sched_job_arm(&job->base);
 	job->done_fence = dma_fence_get(&job->base.s_fence->finished);

@@ -460,9 +458,8 @@ nouveau_sched_init(struct nouveau_sched *sched, struct nouveau_drm *drm,
 		goto fail_sched;
 
 	mutex_init(&sched->mutex);
-	spin_lock_init(&sched->job.list.lock);
-	INIT_LIST_HEAD(&sched->job.list.head);
-	init_waitqueue_head(&sched->job.wq);
+	spin_lock_init(&sched->job_list.lock);
+	INIT_LIST_HEAD(&sched->job_list.head);
 
 	return 0;

@@ -502,9 +499,6 @@ nouveau_sched_fini(struct nouveau_sched *sched)
 	struct drm_gpu_scheduler *drm_sched = &sched->base;
 	struct drm_sched_entity *entity = &sched->entity;
 
-	rmb(); /* for list_empty to work without lock */
-	wait_event(sched->job.wq, list_empty(&sched->job.list.head));
-
 	drm_sched_entity_fini(entity);
 	drm_sched_fini(drm_sched);

drivers/gpu/drm/nouveau/nouveau_sched.h +3 -6

index 20cd1da8db73..b98c3f0bef30 100644
@@ -103,12 +103,9 @@ struct nouveau_sched {
 	struct mutex mutex;
 
 	struct {
-		struct {
-			struct list_head head;
-			spinlock_t lock;
-		} list;
-		struct wait_queue_head wq;
-	} job;
+		struct list_head head;
+		spinlock_t lock;
+	} job_list;
 };
 
 int nouveau_sched_create(struct nouveau_sched **psched, struct nouveau_drm *drm,

drivers/gpu/drm/nouveau/nouveau_uvmm.c +4 -4

index 48f105239f42..ddfc46bc1b3e 100644
@@ -1019,8 +1019,8 @@ bind_validate_map_sparse(struct nouveau_job *job, u64 addr, u64 range)
 	u64 end = addr + range;
 
 again:
-	spin_lock(&sched->job.list.lock);
-	list_for_each_entry(__job, &sched->job.list.head, entry) {
+	spin_lock(&sched->job_list.lock);
+	list_for_each_entry(__job, &sched->job_list.head, entry) {
 		struct nouveau_uvmm_bind_job *bind_job = to_uvmm_bind_job(__job);
 
 		list_for_each_op(op, &bind_job->ops) {

@@ -1030,7 +1030,7 @@ bind_validate_map_sparse(struct nouveau_job *job, u64 addr, u64 range)
 
 				if (!(end <= op_addr || addr >= op_end)) {
 					nouveau_uvmm_bind_job_get(bind_job);
-					spin_unlock(&sched->job.list.lock);
+					spin_unlock(&sched->job_list.lock);
 					wait_for_completion(&bind_job->complete);
 					nouveau_uvmm_bind_job_put(bind_job);
 					goto again;

@@ -1038,7 +1038,7 @@ bind_validate_map_sparse(struct nouveau_job *job, u64 addr, u64 range)
 			}
 		}
 	}
-	spin_unlock(&sched->job.list.lock);
+	spin_unlock(&sched->job_list.lock);
 }
 
 static int

Signature · 1 line

2.49.0

Philipp Stanner

Applied

2025-07-10 15:12 UTC #11 lore raw