50 static int read_vendor_id(struct mlx4_dev *dev)  in read_vendor_id()  argument
55 	ret = pci_read_config_word(dev->persist->pdev, 0, &vendor_id);  in read_vendor_id()
57 		mlx4_err(dev, "Failed to read vendor ID, ret=%d\n", ret);  in read_vendor_id()
62 		mlx4_err(dev, "PCI can't be accessed to read vendor id\n");  in read_vendor_id()
69 static int mlx4_reset_master(struct mlx4_dev *dev)  in mlx4_reset_master()  argument
73 	if (mlx4_is_master(dev))  in mlx4_reset_master()
74 		mlx4_report_internal_err_comm_event(dev);  in mlx4_reset_master()
76 	if (!pci_channel_offline(dev->persist->pdev)) {  in mlx4_reset_master()
77 		err = read_vendor_id(dev);  in mlx4_reset_master()
84 		err = mlx4_reset(dev);  in mlx4_reset_master()
86 			mlx4_err(dev, "Fail to reset HCA\n");  in mlx4_reset_master()
92 static int mlx4_reset_slave(struct mlx4_dev *dev)  in mlx4_reset_slave()  argument
101 	struct mlx4_priv *priv = mlx4_priv(dev);  in mlx4_reset_slave()
103 	if (pci_channel_offline(dev->persist->pdev))  in mlx4_reset_slave()
109 		mlx4_err(dev, "VF reset is not needed\n");  in mlx4_reset_slave()
113 	if (!(dev->caps.vf_caps & MLX4_VF_CAP_FLAG_RESET)) {  in mlx4_reset_slave()
114 		mlx4_err(dev, "VF reset is not supported\n");  in mlx4_reset_slave()
123 		mlx4_err(dev, "Communication channel isn't sync, fail to send reset\n");  in mlx4_reset_slave()
128 	mlx4_warn(dev, "VF is sending reset request to Firmware\n");  in mlx4_reset_slave()
148 			mlx4_warn(dev, "VF Reset succeed\n");  in mlx4_reset_slave()
153 	mlx4_err(dev, "Fail to send reset over the communication channel\n");  in mlx4_reset_slave()
166 	struct mlx4_dev *dev;  in mlx4_enter_error_state()  local
175 	dev = persist->dev;  in mlx4_enter_error_state()
176 	mlx4_err(dev, "device is going to be reset\n");  in mlx4_enter_error_state()
177 	if (mlx4_is_slave(dev)) {  in mlx4_enter_error_state()
178 		err = mlx4_reset_slave(dev);  in mlx4_enter_error_state()
180 		mlx4_crdump_collect(dev);  in mlx4_enter_error_state()
181 		err = mlx4_reset_master(dev);  in mlx4_enter_error_state()
185 		mlx4_err(dev, "device was reset successfully\n");  in mlx4_enter_error_state()
190 		if (!pci_channel_offline(dev->persist->pdev))  in mlx4_enter_error_state()
193 	dev->persist->state |= MLX4_DEVICE_STATE_INTERNAL_ERROR;  in mlx4_enter_error_state()
197 	mlx4_dispatch_event(dev, MLX4_DEV_EVENT_CATASTROPHIC_ERROR, NULL);  in mlx4_enter_error_state()
198 	mlx4_cmd_wake_completions(dev);  in mlx4_enter_error_state()
207 	struct mlx4_dev *dev = persist->dev;  in mlx4_handle_error_state()  local
212 	devlink = priv_to_devlink(mlx4_priv(dev));  in mlx4_handle_error_state()
218 		mlx4_info(persist->dev, "mlx4_restart_one was ended, ret=%d\n",  in mlx4_handle_error_state()
225 static void dump_err_buf(struct mlx4_dev *dev)  in dump_err_buf()  argument
227 	struct mlx4_priv *priv = mlx4_priv(dev);  in dump_err_buf()
231 	mlx4_err(dev, "Internal error detected:\n");  in dump_err_buf()
233 		mlx4_err(dev, "  buf[%02x]: %08x\n",  in dump_err_buf()
240 	struct mlx4_dev *dev = &priv->dev;  in poll_catas()  local
243 	if (mlx4_is_slave(dev)) {  in poll_catas()
246 			mlx4_warn(dev, "Internal error detected on the communication channel\n");  in poll_catas()
250 		dump_err_buf(dev);  in poll_catas()
254 	if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {  in poll_catas()
255 		mlx4_warn(dev, "Internal error mark was detected on device\n");  in poll_catas()
265 		queue_work(dev->persist->catas_wq, &dev->persist->catas_work);  in poll_catas()
277 void mlx4_start_catas_poll(struct mlx4_dev *dev)  in mlx4_start_catas_poll()  argument
279 	struct mlx4_priv *priv = mlx4_priv(dev);  in mlx4_start_catas_poll()
286 	if (!mlx4_is_slave(dev)) {  in mlx4_start_catas_poll()
287 		addr = pci_resource_start(dev->persist->pdev,  in mlx4_start_catas_poll()
293 			mlx4_warn(dev, "Failed to map internal error buffer at 0x%llx\n",  in mlx4_start_catas_poll()
304 void mlx4_stop_catas_poll(struct mlx4_dev *dev)  in mlx4_stop_catas_poll()  argument
306 	struct mlx4_priv *priv = mlx4_priv(dev);  in mlx4_stop_catas_poll()
315 	if (dev->persist->interface_state & MLX4_INTERFACE_STATE_DELETION)  in mlx4_stop_catas_poll()
316 		flush_workqueue(dev->persist->catas_wq);  in mlx4_stop_catas_poll()
319 int  mlx4_catas_init(struct mlx4_dev *dev)  in mlx4_catas_init()  argument
321 	INIT_WORK(&dev->persist->catas_work, catas_reset);  in mlx4_catas_init()
322 	dev->persist->catas_wq = create_singlethread_workqueue("mlx4_health");  in mlx4_catas_init()
323 	if (!dev->persist->catas_wq)  in mlx4_catas_init()
329 void mlx4_catas_end(struct mlx4_dev *dev)  in mlx4_catas_end()  argument
331 	if (dev->persist->catas_wq) {  in mlx4_catas_end()
332 		destroy_workqueue(dev->persist->catas_wq);  in mlx4_catas_end()
333 		dev->persist->catas_wq = NULL;  in mlx4_catas_end()