On 06/11/2025 13:23, Christian König wrote:
On 11/4/25 16:12, Tvrtko Ursulin wrote:
On 31/10/2025 13:16, Christian König wrote:
Just as proof of concept and minor cleanup.
Signed-off-by: Christian König <[email protected]>
---
drivers/gpu/drm/scheduler/sched_fence.c | 11 +++++------
include/drm/gpu_scheduler.h | 4 ----
2 files changed, 5 insertions(+), 10 deletions(-)
diff --git a/drivers/gpu/drm/scheduler/sched_fence.c
b/drivers/gpu/drm/scheduler/sched_fence.c
index 9391d6f0dc01..7a94e03341cb 100644
--- a/drivers/gpu/drm/scheduler/sched_fence.c
+++ b/drivers/gpu/drm/scheduler/sched_fence.c
@@ -156,19 +156,19 @@ static void drm_sched_fence_set_deadline_finished(struct
dma_fence *f,
struct dma_fence *parent;
unsigned long flags;
- spin_lock_irqsave(&fence->lock, flags);
+ dma_fence_lock(f, flags);
Moving to dma_fence_lock should either be a separate patch or squashed into the
one which converts many other drivers. Even a separate patch before that
previous patch would be better.
As far as I can see that won't work or would be at least rather tricky.
Previously from spin_lock_irqsave() locked drm_sched_fence->lock, but now it is
locking dma_fence->lock.
That only works because we switched to using the internal lock.
What I meant here is to add a patch before the current 5/20.
Because in 5/20 one we have a lot of:
- spin_lock_irqsave(fence->lock, flags);
+ dma_fence_lock(fence, flags);
If before it you insert a patch like "dma-fence: Add lock/unlock helper"
as standalone, then 5/20 simply touches the internal of the helper and
becomes smaller.
That new patch would also include the drm/sched part where it would be a
straight replacement. At that point dma_fence->lock ==
dma_sched_fence->lock.
Then at the point of this patch you remove dma_sched_fence->lock member
but dma_fence_lock_* already does the right thing before it, no?
Sorry if I got confused somehow, I am jumping between topics.
Regards,
Tvrtko
Naming wise, I however still think dma_fence_lock_irqsave would probably be
better to stick with the same pattern everyone is so used too.
Oh, that is a good idea. Going to apply this to the patch set.
Regards,
Christian.
Regards,
Tvrtko
/* If we already have an earlier deadline, keep it: */
if (test_bit(DRM_SCHED_FENCE_FLAG_HAS_DEADLINE_BIT, &f->flags) &&
ktime_before(fence->deadline, deadline)) {
- spin_unlock_irqrestore(&fence->lock, flags);
+ dma_fence_unlock(f, flags);
return;
}
fence->deadline = deadline;
set_bit(DRM_SCHED_FENCE_FLAG_HAS_DEADLINE_BIT, &f->flags);
- spin_unlock_irqrestore(&fence->lock, flags);
+ dma_fence_unlock(f, flags);
/*
* smp_load_aquire() to ensure that if we are racing another
@@ -217,7 +217,6 @@ struct drm_sched_fence *drm_sched_fence_alloc(struct
drm_sched_entity *entity,
fence->owner = owner;
fence->drm_client_id = drm_client_id;
- spin_lock_init(&fence->lock);
return fence;
}
@@ -230,9 +229,9 @@ void drm_sched_fence_init(struct drm_sched_fence *fence,
fence->sched = entity->rq->sched;
seq = atomic_inc_return(&entity->fence_seq);
dma_fence_init(&fence->scheduled, &drm_sched_fence_ops_scheduled,
- &fence->lock, entity->fence_context, seq);
+ NULL, entity->fence_context, seq);
dma_fence_init(&fence->finished, &drm_sched_fence_ops_finished,
- &fence->lock, entity->fence_context + 1, seq);
+ NULL, entity->fence_context + 1, seq);
}
module_init(drm_sched_fence_slab_init);
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index fb88301b3c45..b77f24a783e3 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -297,10 +297,6 @@ struct drm_sched_fence {
* belongs to.
*/
struct drm_gpu_scheduler *sched;
- /**
- * @lock: the lock used by the scheduled and the finished fences.
- */
- spinlock_t lock;
/**
* @owner: job owner for debugging
*/
