The fence allocation needs to be protected by the GPU mutex, otherwise the fence seqnos of concurrent submits might not match the insertion order of the jobs in the kernel ring. This breaks the assumption that jobs complete with monotonically increasing fence seqnos.
Fixes: d9853490176c (drm/etnaviv: take GPU lock later in the submit process) CC: stable@vger.kernel.org #4.9+ Signed-off-by: Lucas Stach l.stach@pengutronix.de --- drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c index 130d7d517a19..da48819ff2e6 100644 --- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c +++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c @@ -1311,6 +1311,8 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu, goto out_pm_put; }
+ mutex_lock(&gpu->lock); + fence = etnaviv_gpu_fence_alloc(gpu); if (!fence) { event_free(gpu, event); @@ -1318,8 +1320,6 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu, goto out_pm_put; }
- mutex_lock(&gpu->lock); - gpu->event[event].fence = fence; submit->fence = fence->seqno; gpu->active_fence = submit->fence;
Make sure the GPU lock is taken, so that fence completion order matches seqno order.
Signed-off-by: Lucas Stach l.stach@pengutronix.de --- drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 6 ++++++ 1 file changed, 6 insertions(+)
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c index da48819ff2e6..f01bf17a54ef 100644 --- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c +++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c @@ -1051,6 +1051,12 @@ static struct dma_fence *etnaviv_gpu_fence_alloc(struct etnaviv_gpu *gpu) { struct etnaviv_fence *f;
+ /* + * GPU lock must already be held, otherwise fence completion order might + * not match the seqno order assigned here. + */ + lockdep_assert_held(&gpu->lock); + f = kzalloc(sizeof(*f), GFP_KERNEL); if (!f) return NULL;
dri-devel@lists.freedesktop.org
-
Lucas Stach