net/mlx5: Drain fw_reset when removing device

In case fw sync reset is called in parallel to device removal, device
might stuck in the following deadlock:
         CPU 0                        CPU 1
         -----                        -----
                                  remove_one
                                   uninit_one (locks intf_state_mutex)
mlx5_sync_reset_now_event()
work in fw_reset->wq.
 mlx5_enter_error_state()
  mutex_lock (intf_state_mutex)
                                   cleanup_once
                                    fw_reset_cleanup()
                                     destroy_workqueue(fw_reset->wq)

Drain the fw_reset WQ, and make sure no new work is being queued, before
entering uninit_one().
The Drain is done before devlink_unregister() since fw_reset, in some
flows, is using devlink API devlink_remote_reload_actions_performed().

Fixes: 38b9f903 ("net/mlx5: Handle sync reset request event")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

drivers/net/ethernet/mellanox/mlx5/core/fw_reset.c

@@ -8,7 +8,8 @@
 enum {
 	MLX5_FW_RESET_FLAGS_RESET_REQUESTED,
 	MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST,
-	MLX5_FW_RESET_FLAGS_PENDING_COMP
+	MLX5_FW_RESET_FLAGS_PENDING_COMP,
+	MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS
 };
 
 struct mlx5_fw_reset {
@@ -208,7 +209,10 @@ static void poll_sync_reset(struct timer_list *t)
 
 	if (fatal_error) {
 		mlx5_core_warn(dev, "Got Device Reset\n");
+		if (!test_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags))
 			queue_work(fw_reset->wq, &fw_reset->reset_reload_work);
+		else
+			mlx5_core_err(dev, "Device is being removed, Drop new reset work\n");
 		return;
 	}
 
@@ -433,6 +437,9 @@ static int fw_reset_event_notifier(struct notifier_block *nb, unsigned long acti
 	struct mlx5_fw_reset *fw_reset = mlx5_nb_cof(nb, struct mlx5_fw_reset, nb);
 	struct mlx5_eqe *eqe = data;
 
+	if (test_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags))
+		return NOTIFY_DONE;
+
 	switch (eqe->sub_type) {
 	case MLX5_GENERAL_SUBTYPE_FW_LIVE_PATCH_EVENT:
 		queue_work(fw_reset->wq, &fw_reset->fw_live_patch_work);
@@ -479,6 +486,18 @@ void mlx5_fw_reset_events_stop(struct mlx5_core_dev *dev)
 	mlx5_eq_notifier_unregister(dev, &dev->priv.fw_reset->nb);
 }
 
+void mlx5_drain_fw_reset(struct mlx5_core_dev *dev)
+{
+	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
+
+	set_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags);
+	cancel_work_sync(&fw_reset->fw_live_patch_work);
+	cancel_work_sync(&fw_reset->reset_request_work);
+	cancel_work_sync(&fw_reset->reset_reload_work);
+	cancel_work_sync(&fw_reset->reset_now_work);
+	cancel_work_sync(&fw_reset->reset_abort_work);
+}
+
 int mlx5_fw_reset_init(struct mlx5_core_dev *dev)
 {
 	struct mlx5_fw_reset *fw_reset = kzalloc(sizeof(*fw_reset), GFP_KERNEL);

drivers/net/ethernet/mellanox/mlx5/core/fw_reset.h

@@ -16,6 +16,7 @@ int mlx5_fw_reset_set_live_patch(struct mlx5_core_dev *dev);
 int mlx5_fw_reset_wait_reset_done(struct mlx5_core_dev *dev);
 void mlx5_fw_reset_events_start(struct mlx5_core_dev *dev);
 void mlx5_fw_reset_events_stop(struct mlx5_core_dev *dev);
+void mlx5_drain_fw_reset(struct mlx5_core_dev *dev);
 int mlx5_fw_reset_init(struct mlx5_core_dev *dev);
 void mlx5_fw_reset_cleanup(struct mlx5_core_dev *dev);
 

drivers/net/ethernet/mellanox/mlx5/core/main.c

@@ -1627,6 +1627,10 @@ static void remove_one(struct pci_dev *pdev)
 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
 	struct devlink *devlink = priv_to_devlink(dev);
 
+	/* mlx5_drain_fw_reset() is using devlink APIs. Hence, we must drain
+	 * fw_reset before unregistering the devlink.
+	 */
+	mlx5_drain_fw_reset(dev);
 	devlink_unregister(devlink);
 	mlx5_sriov_disable(pdev);
 	mlx5_crdump_disable(dev);