drm/i915: Propagate errors on awaiting already signaled fences

If we see an already signaled fence that we want to await on, we skip
adding to the i915_sw_fence. However, we should pay attention to whether
there was an error on that fence and if so propagate it for our future
request.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191206160428.1503343-2-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/i915_request.c

@@ -958,8 +958,10 @@ i915_request_await_dma_fence(struct i915_request *rq, struct dma_fence *fence)
 
 	do {
 		fence = *child++;
-		if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))
+		if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags)) {
+			i915_sw_fence_set_error_once(&rq->submit, fence->error);
 			continue;
+		}
 
 		/*
 		 * Requests on the same timeline are explicitly ordered, along
@@ -1015,8 +1017,10 @@ i915_request_await_execution(struct i915_request *rq,
 
 	do {
 		fence = *child++;
-		if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))
+		if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags)) {
+			i915_sw_fence_set_error_once(&rq->submit, fence->error);
 			continue;
+		}
 
 		/*
 		 * We don't squash repeated fence dependencies here as we