Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
drm/sched: Remove racy hack from drm_sched_fini()
drm_sched_fini() contained a hack to work around a race in amdgpu.
According to AMD, the hack should not be necessary anymore. In case
there should have been undetected users,
commit 975ca62a014c ("drm/sched: Add warning for removing hack in drm_sched_fini()")
had added a warning one release cycle ago.
Thus, it can be derived that the hack can be savely removed by now.
Remove the hack.
Acked-by: Danilo Krummrich <dakr@kernel.org>
Signed-off-by: Philipp Stanner <phasta@kernel.org>
Link: https://patch.msgid.link/20260108083019.63532-2-phasta@kernel.org
+1
-37
+1
-37
drivers/gpu/drm/scheduler/sched_main.c
+1
-37
drivers/gpu/drm/scheduler/sched_main.c
···
1418
1418
*/
1419
1419
void drm_sched_fini(struct drm_gpu_scheduler *sched)
1420
1420
{
1421
-
struct drm_sched_entity *s_entity;
1422
1421
int i;
1423
1422
1424
1423
drm_sched_wqueue_stop(sched);
1425
1424
1426
-
for (i = DRM_SCHED_PRIORITY_KERNEL; i < sched->num_rqs; i++) {
1427
-
struct drm_sched_rq *rq = sched->sched_rq[i];
1428
-
1429
-
spin_lock(&rq->lock);
1430
-
list_for_each_entry(s_entity, &rq->entities, list) {
1431
-
/*
1432
-
* Prevents reinsertion and marks job_queue as idle,
1433
-
* it will be removed from the rq in drm_sched_entity_fini()
1434
-
* eventually
1435
-
*
1436
-
* FIXME:
1437
-
* This lacks the proper spin_lock(&s_entity->lock) and
1438
-
* is, therefore, a race condition. Most notably, it
1439
-
* can race with drm_sched_entity_push_job(). The lock
1440
-
* cannot be taken here, however, because this would
1441
-
* lead to lock inversion -> deadlock.
1442
-
*
1443
-
* The best solution probably is to enforce the life
1444
-
* time rule of all entities having to be torn down
1445
-
* before their scheduler. Then, however, locking could
1446
-
* be dropped alltogether from this function.
1447
-
*
1448
-
* For now, this remains a potential race in all
1449
-
* drivers that keep entities alive for longer than
1450
-
* the scheduler.
1451
-
*
1452
-
* The READ_ONCE() is there to make the lockless read
1453
-
* (warning about the lockless write below) slightly
1454
-
* less broken...
1455
-
*/
1456
-
if (!READ_ONCE(s_entity->stopped))
1457
-
dev_warn(sched->dev, "Tearing down scheduler with active entities!\n");
1458
-
s_entity->stopped = true;
1459
-
}
1460
-
spin_unlock(&rq->lock);
1425
+
for (i = DRM_SCHED_PRIORITY_KERNEL; i < sched->num_rqs; i++)
1461
1426
kfree(sched->sched_rq[i]);
1462
-
}
1463
1427
1464
1428
/* Wakeup everyone stuck in drm_sched_entity_flush for this scheduler */
1465
1429
wake_up_all(&sched->job_scheduled);