Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 3ba64623 authored by Francisco Jerez's avatar Francisco Jerez Committed by Ben Skeggs
Browse files

drm/nouveau: Take fence spinlock before reading the last sequence. 



It fixes a race between the TTM delayed work queue and the GEM IOCTLs
(fdo bug 29583) uncovered by the BKL removal.

Signed-off-by: default avatarFrancisco Jerez <currojerez@riseup.net>
Signed-off-by: default avatarBen Skeggs <bskeggs@redhat.com>
parent ab699ec6

drivers/gpu/drm/nouveau/nouveau_fence.c

+4 −2
Original line number Diff line number Diff line
@@ -64,16 +64,17 @@ nouveau_fence_update(struct nouveau_channel *chan) 
	struct nouveau_fence *fence;

	uint32_t sequence;



	spin_lock(&chan->fence.lock);



	if (USE_REFCNT)

		sequence = nvchan_rd32(chan, 0x48);

	else

		sequence = atomic_read(&chan->fence.last_sequence_irq);



	if (chan->fence.sequence_ack == sequence)

		return;

		goto out;

	chan->fence.sequence_ack = sequence;



	spin_lock(&chan->fence.lock);

	list_for_each_safe(entry, tmp, &chan->fence.pending) {

		fence = list_entry(entry, struct nouveau_fence, entry);
@@ -85,6 +86,7 @@ nouveau_fence_update(struct nouveau_channel *chan) 
		if (sequence == chan->fence.sequence_ack)

			break;

	}

out:

	spin_unlock(&chan->fence.lock);

}