drm/amdkfd: Improve HWS hang detection and handling

Move HWS hang detection into unmap_queues_cpsch to catch hangs in all
cases. If this happens during a reset, don't schedule another reset
because the reset already in progress is expected to take care of it.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Tested-by: Emily Deng <Emily.Deng@amd.com>
Reviewed-by: shaoyunl  <shaoyun.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -728,6 +728,9 @@ int kgd2kfd_pre_reset(struct kfd_dev *kfd)
 {
 	if (!kfd->init_complete)
 		return 0;
+
+	kfd->dqm->ops.pre_reset(kfd->dqm);
+
 	kgd2kfd_suspend(kfd);
 
 	kfd_signal_reset_event(kfd);

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

@@ -952,6 +952,13 @@ static int stop_nocpsch(struct device_queue_manager *dqm)
 	return 0;
 }
 
+static void pre_reset(struct device_queue_manager *dqm)
+{
+	dqm_lock(dqm);
+	dqm->is_resetting = true;
+	dqm_unlock(dqm);
+}
+
 static int allocate_sdma_queue(struct device_queue_manager *dqm,
 				struct queue *q)
 {
@@ -1099,6 +1106,7 @@ static int start_cpsch(struct device_queue_manager *dqm)
 	dqm_lock(dqm);
 	/* clear hang status when driver try to start the hw scheduler */
 	dqm->is_hws_hang = false;
+	dqm->is_resetting = false;
 	dqm->sched_running = true;
 	execute_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0);
 	dqm_unlock(dqm);
@@ -1351,8 +1359,17 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm,
 	/* should be timed out */
 	retval = amdkfd_fence_wait_timeout(dqm->fence_addr, KFD_FENCE_COMPLETED,
 				queue_preemption_timeout_ms);
-	if (retval)
+	if (retval) {
+		pr_err("The cp might be in an unrecoverable state due to an unsuccessful queues preemption\n");
+		dqm->is_hws_hang = true;
+		/* It's possible we're detecting a HWS hang in the
+		 * middle of a GPU reset. No need to schedule another
+		 * reset in this case.
+		 */
+		if (!dqm->is_resetting)
+			schedule_work(&dqm->hw_exception_work);
 		return retval;
+	}
 
 	pm_release_ib(&dqm->packets);
 	dqm->active_runlist = false;
@@ -1370,12 +1387,8 @@ static int execute_queues_cpsch(struct device_queue_manager *dqm,
 	if (dqm->is_hws_hang)
 		return -EIO;
 	retval = unmap_queues_cpsch(dqm, filter, filter_param);
-	if (retval) {
-		pr_err("The cp might be in an unrecoverable state due to an unsuccessful queues preemption\n");
-		dqm->is_hws_hang = true;
-		schedule_work(&dqm->hw_exception_work);
+	if (retval)
 		return retval;
-	}
 
 	return map_queues_cpsch(dqm);
 }
@@ -1769,6 +1782,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 		dqm->ops.initialize = initialize_cpsch;
 		dqm->ops.start = start_cpsch;
 		dqm->ops.stop = stop_cpsch;
+		dqm->ops.pre_reset = pre_reset;
 		dqm->ops.destroy_queue = destroy_queue_cpsch;
 		dqm->ops.update_queue = update_queue;
 		dqm->ops.register_process = register_process;
@@ -1787,6 +1801,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 		/* initialize dqm for no cp scheduling */
 		dqm->ops.start = start_nocpsch;
 		dqm->ops.stop = stop_nocpsch;
+		dqm->ops.pre_reset = pre_reset;
 		dqm->ops.create_queue = create_queue_nocpsch;
 		dqm->ops.destroy_queue = destroy_queue_nocpsch;
 		dqm->ops.update_queue = update_queue;

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

@@ -104,6 +104,7 @@ struct device_queue_manager_ops {
 	int	(*initialize)(struct device_queue_manager *dqm);
 	int	(*start)(struct device_queue_manager *dqm);
 	int	(*stop)(struct device_queue_manager *dqm);
+	void	(*pre_reset)(struct device_queue_manager *dqm);
 	void	(*uninitialize)(struct device_queue_manager *dqm);
 	int	(*create_kernel_queue)(struct device_queue_manager *dqm,
 					struct kernel_queue *kq,
@@ -198,6 +199,7 @@ struct device_queue_manager {
 
 	/* hw exception  */
 	bool			is_hws_hang;
+	bool			is_resetting;
 	struct work_struct	hw_exception_work;
 	struct kfd_mem_obj	hiq_sdma_mqd;
 	bool			sched_running;