15  *      - Redistributions of source code must retain the above
19  *      - Redistributions in binary form must reproduce the above
48 		 "Reset device on internal errors if non-zero (default 1)");
55 	ret = pci_read_config_word(dev->persist->pdev, 0, &vendor_id);  in read_vendor_id()
63 		return -EINVAL;  in read_vendor_id()
76 	if (!pci_channel_offline(dev->persist->pdev)) {  in mlx4_reset_master()
79 		 * link was disabled and chip was already reset.  in mlx4_reset_master()
86 			mlx4_err(dev, "Fail to reset HCA\n");  in mlx4_reset_master()
103 	if (pci_channel_offline(dev->persist->pdev))  in mlx4_reset_slave()
106 	comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +  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()
115 		return -EOPNOTSUPP;  in mlx4_reset_slave()
123 		mlx4_err(dev, "Communication channel isn't sync, fail to send reset\n");  in mlx4_reset_slave()
124 		return -EIO;  in mlx4_reset_slave()
128 	mlx4_warn(dev, "VF is sending reset request to Firmware\n");  in mlx4_reset_slave()
131 		     (__iomem char *)priv->mfunc.comm + MLX4_COMM_CHAN_FLAGS);  in mlx4_reset_slave()
135 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +  in mlx4_reset_slave()
141 		 * be reset at any time by the PF and all its bits will be  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()
154 	return -ETIMEDOUT;  in mlx4_reset_slave()
171 	mutex_lock(&persist->device_state_mutex);  in mlx4_enter_error_state()
172 	if (persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)  in mlx4_enter_error_state()
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()
185 		mlx4_err(dev, "device was reset successfully\n");  in mlx4_enter_error_state()
187 		/* EEH could have disabled the PCI channel during reset. That's  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()
194 	mutex_unlock(&persist->device_state_mutex);  in mlx4_enter_error_state()
196 	/* At that step HW was already reset, now notify clients */  in mlx4_enter_error_state()
202 	mutex_unlock(&persist->device_state_mutex);  in mlx4_enter_error_state()
207 	struct mlx4_dev *dev = persist->dev;  in mlx4_handle_error_state()
214 	mutex_lock(&persist->interface_state_mutex);  in mlx4_handle_error_state()
215 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP &&  in mlx4_handle_error_state()
216 	    !(persist->interface_state & MLX4_INTERFACE_STATE_DELETION)) {  in mlx4_handle_error_state()
217 		err = mlx4_restart_one(persist->pdev);  in mlx4_handle_error_state()
218 		mlx4_info(persist->dev, "mlx4_restart_one was ended, ret=%d\n",  in mlx4_handle_error_state()
221 	mutex_unlock(&persist->interface_state_mutex);  in mlx4_handle_error_state()
232 	for (i = 0; i < priv->fw.catas_size; ++i)  in dump_err_buf()
234 			 i, swab32(readl(priv->catas_err.map + i)));  in dump_err_buf()
240 	struct mlx4_dev *dev = &priv->dev;  in poll_catas()
244 		slave_read = swab32(readl(&priv->mfunc.comm->slave_read));  in poll_catas()
246 			mlx4_warn(dev, "Internal error detected on the communication channel\n");  in poll_catas()
249 	} else if (readl(priv->catas_err.map)) {  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()
259 	mod_timer(&priv->catas_err.timer,  in poll_catas()
265 		queue_work(dev->persist->catas_wq, &dev->persist->catas_work);  in poll_catas()
282 	INIT_LIST_HEAD(&priv->catas_err.list);  in mlx4_start_catas_poll()
283 	timer_setup(&priv->catas_err.timer, poll_catas, 0);  in mlx4_start_catas_poll()
284 	priv->catas_err.map = NULL;  in mlx4_start_catas_poll()
287 		addr = pci_resource_start(dev->persist->pdev,  in mlx4_start_catas_poll()
288 					  priv->fw.catas_bar) +  in mlx4_start_catas_poll()
289 					  priv->fw.catas_offset;  in mlx4_start_catas_poll()
291 		priv->catas_err.map = ioremap(addr, priv->fw.catas_size * 4);  in mlx4_start_catas_poll()
292 		if (!priv->catas_err.map) {  in mlx4_start_catas_poll()
299 	priv->catas_err.timer.expires  =  in mlx4_start_catas_poll()
301 	add_timer(&priv->catas_err.timer);  in mlx4_start_catas_poll()
308 	del_timer_sync(&priv->catas_err.timer);  in mlx4_stop_catas_poll()
310 	if (priv->catas_err.map) {  in mlx4_stop_catas_poll()
311 		iounmap(priv->catas_err.map);  in mlx4_stop_catas_poll()
312 		priv->catas_err.map = NULL;  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()
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()
324 		return -ENOMEM;  in mlx4_catas_init()
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()