[Mesa-dev] [PATCH 1/3] anv/device: return VK_ERROR_DEVICE_LOST for errors during queue submissions

[Iago Toral Quiroga]

So that we don't have to do things like rolling back address relocations in
case that we ran into OOM after computing them, etc

Also, make sure that if the queue submission comes with a fence, we set it up
correctly so it behaves according to the spec after returning
VK_ERROR_DEVICE_LOST.
---
 src/intel/vulkan/anv_device.c | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/src/intel/vulkan/anv_device.c b/src/intel/vulkan/anv_device.c
index 74aad45..a11cb32 100644
--- a/src/intel/vulkan/anv_device.c
+++ b/src/intel/vulkan/anv_device.c
@@ -1325,6 +1325,31 @@ VkResult anv_QueueSubmit(
    }
 
 out:
+   if (result != VK_SUCCESS) {
+      /* In the case that something has gone wrong we may end up with an
+       * inconsistent state from which it may not be trivial to recover.
+       * For example, we might have computed address relocations and
+       * any future attempt to re-submit this job will need to know about
+       * this and avoid computing relocation addresses again.
+       *
+       * To avoid this sort of issues, we assume that if something was
+       * wrong during submission we must already be in a really bad situation
+       * anyway (such us being out of memory) and return
+       * VK_ERROR_DEVICE_LOST to ensure that clients do not attempt to
+       * submit the same job again to this device.
+       */
+      result = VK_ERROR_DEVICE_LOST;
+
+      /* If we return VK_ERROR_DEVICE LOST here, we need to ensure that
+       * vkWaitForFences() and vkGetFenceStatus() return a valid result
+       * (VK_SUCCESS or VK_ERROR_DEVICE_LOST) in a finite amount of time.
+       * Setting the fence status to SIGNALED ensures this will happen in
+       * any case.
+       */
+      if (fence)
+         fence->state = ANV_FENCE_STATE_SIGNALED;
+   }
+
    pthread_mutex_unlock(&device->mutex);
 
    return result;
-- 
2.7.4