drm/amdkfd: fix the hang caused by the write reorder to fence_addr

make sure KFD_FENCE_INIT write to fence_addr before pm_send_query_status
called, to avoid qcm fence timeout caused by incorrect ordering.

Signed-off-by: Victor Zhao <Victor.Zhao@amd.com>
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c

@@ -2048,7 +2048,7 @@ int amdkfd_fence_wait_timeout(struct device_queue_manager *dqm,
 {
 	unsigned long end_jiffies = msecs_to_jiffies(timeout_ms) + jiffies;
 	struct device *dev = dqm->dev->adev->dev;
-	uint64_t *fence_addr =  dqm->fence_addr;
+	volatile uint64_t *fence_addr = dqm->fence_addr;
 
 	while (*fence_addr != fence_value) {
 		/* Fatal err detected, this response won't come */
@@ -2254,6 +2254,7 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm,
 		goto out;
 
 	*dqm->fence_addr = KFD_FENCE_INIT;
+	mb();
 	pm_send_query_status(&dqm->packet_mgr, dqm->fence_gpu_addr,
 				KFD_FENCE_COMPLETED);
 	/* should be timed out */

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h

@@ -260,7 +260,7 @@ struct device_queue_manager {
 	uint16_t		vmid_pasid[VMID_NUM];
 	uint64_t		pipelines_addr;
 	uint64_t		fence_gpu_addr;
-	uint64_t		*fence_addr;
+	volatile uint64_t	*fence_addr;
 	struct kfd_mem_obj	*fence_mem;
 	bool			active_runlist;
 	int			sched_policy;