This fixes a regression introduced by "drm/nouveau: rework to new fence interface" (commit 29ba89b2371d466).
The fence sequence should not be reset after creation, the old value is used instead. On destruction the final value is written, to prevent another source of accidental wraparound in case of a channel being destroyed after a hang, and unblocking any other channel that may wait on the about-to-be-deleted channel to signal.
I'm nothing if not optimistic about any hope of recovery from that. ;-)
Reported-by: Ted Percival ted@tedp.id.au Signed-off-by: Maarten Lankhorst maarten.lankhorst@canonical.com --- diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c index 7b372a68aa4e..4138db4d8291 100644 --- a/drivers/gpu/drm/nouveau/nv84_fence.c +++ b/drivers/gpu/drm/nouveau/nv84_fence.c @@ -120,6 +120,7 @@ nv84_fence_context_del(struct nouveau_channel *chan) nouveau_bo_vma_del(bo, &fctx->dispc_vma[i]); }
+ nouveau_bo_wr32(priv->bo, chan->chid * 16 / 4, fctx->base.sequence); nouveau_bo_vma_del(priv->bo, &fctx->vma_gart); nouveau_bo_vma_del(priv->bo, &fctx->vma); nouveau_fence_context_del(&fctx->base); @@ -159,8 +160,6 @@ nv84_fence_context_new(struct nouveau_channel *chan) ret = nouveau_bo_vma_add(bo, cli->vm, &fctx->dispc_vma[i]); }
- nouveau_bo_wr32(priv->bo, chan->chid * 16/4, 0x00000000); - if (ret) nv84_fence_context_del(chan); return ret;