1 /* 2 * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved. 3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved. 4 * 5 * This software is available to you under a choice of one of two 6 * licenses. You may choose to be licensed under the terms of the GNU 7 * General Public License (GPL) Version 2, available from the file 8 * COPYING in the main directory of this source tree, or the 9 * OpenIB.org BSD license below: 10 * 11 * Redistribution and use in source and binary forms, with or 12 * without modification, are permitted provided that the following 13 * conditions are met: 14 * 15 * - Redistributions of source code must retain the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer. 18 * 19 * - Redistributions in binary form must reproduce the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer in the documentation and/or other materials 22 * provided with the distribution. 23 * 24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 31 * SOFTWARE. 32 */ 33 34 #include <linux/workqueue.h> 35 #include <linux/module.h> 36 37 #include "mlx4.h" 38 39 enum { 40 MLX4_CATAS_POLL_INTERVAL = 5 * HZ, 41 }; 42 43 44 45 int mlx4_internal_err_reset = 1; 46 module_param_named(internal_err_reset, mlx4_internal_err_reset, int, 0644); 47 MODULE_PARM_DESC(internal_err_reset, 48 "Reset device on internal errors if non-zero (default 1)"); 49 50 static int read_vendor_id(struct mlx4_dev *dev) 51 { 52 u16 vendor_id = 0; 53 int ret; 54 55 ret = pci_read_config_word(dev->persist->pdev, 0, &vendor_id); 56 if (ret) { 57 mlx4_err(dev, "Failed to read vendor ID, ret=%d\n", ret); 58 return ret; 59 } 60 61 if (vendor_id == 0xffff) { 62 mlx4_err(dev, "PCI can't be accessed to read vendor id\n"); 63 return -EINVAL; 64 } 65 66 return 0; 67 } 68 69 static int mlx4_reset_master(struct mlx4_dev *dev) 70 { 71 int err = 0; 72 73 if (mlx4_is_master(dev)) 74 mlx4_report_internal_err_comm_event(dev); 75 76 if (!pci_channel_offline(dev->persist->pdev)) { 77 err = read_vendor_id(dev); 78 /* If PCI can't be accessed to read vendor ID we assume that its 79 * link was disabled and chip was already reset. 80 */ 81 if (err) 82 return 0; 83 84 err = mlx4_reset(dev); 85 if (err) 86 mlx4_err(dev, "Fail to reset HCA\n"); 87 } 88 89 return err; 90 } 91 92 static int mlx4_reset_slave(struct mlx4_dev *dev) 93 { 94 #define COM_CHAN_RST_REQ_OFFSET 0x10 95 #define COM_CHAN_RST_ACK_OFFSET 0x08 96 97 u32 comm_flags; 98 u32 rst_req; 99 u32 rst_ack; 100 unsigned long end; 101 struct mlx4_priv *priv = mlx4_priv(dev); 102 103 if (pci_channel_offline(dev->persist->pdev)) 104 return 0; 105 106 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm + 107 MLX4_COMM_CHAN_FLAGS)); 108 if (comm_flags == 0xffffffff) { 109 mlx4_err(dev, "VF reset is not needed\n"); 110 return 0; 111 } 112 113 if (!(dev->caps.vf_caps & MLX4_VF_CAP_FLAG_RESET)) { 114 mlx4_err(dev, "VF reset is not supported\n"); 115 return -EOPNOTSUPP; 116 } 117 118 rst_req = (comm_flags & (u32)(1 << COM_CHAN_RST_REQ_OFFSET)) >> 119 COM_CHAN_RST_REQ_OFFSET; 120 rst_ack = (comm_flags & (u32)(1 << COM_CHAN_RST_ACK_OFFSET)) >> 121 COM_CHAN_RST_ACK_OFFSET; 122 if (rst_req != rst_ack) { 123 mlx4_err(dev, "Communication channel isn't sync, fail to send reset\n"); 124 return -EIO; 125 } 126 127 rst_req ^= 1; 128 mlx4_warn(dev, "VF is sending reset request to Firmware\n"); 129 comm_flags = rst_req << COM_CHAN_RST_REQ_OFFSET; 130 __raw_writel((__force u32)cpu_to_be32(comm_flags), 131 (__iomem char *)priv->mfunc.comm + MLX4_COMM_CHAN_FLAGS); 132 /* Make sure that our comm channel write doesn't 133 * get mixed in with writes from another CPU. 134 */ 135 mmiowb(); 136 137 end = msecs_to_jiffies(MLX4_COMM_TIME) + jiffies; 138 while (time_before(jiffies, end)) { 139 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm + 140 MLX4_COMM_CHAN_FLAGS)); 141 rst_ack = (comm_flags & (u32)(1 << COM_CHAN_RST_ACK_OFFSET)) >> 142 COM_CHAN_RST_ACK_OFFSET; 143 144 /* Reading rst_req again since the communication channel can 145 * be reset at any time by the PF and all its bits will be 146 * set to zero. 147 */ 148 rst_req = (comm_flags & (u32)(1 << COM_CHAN_RST_REQ_OFFSET)) >> 149 COM_CHAN_RST_REQ_OFFSET; 150 151 if (rst_ack == rst_req) { 152 mlx4_warn(dev, "VF Reset succeed\n"); 153 return 0; 154 } 155 cond_resched(); 156 } 157 mlx4_err(dev, "Fail to send reset over the communication channel\n"); 158 return -ETIMEDOUT; 159 } 160 161 int mlx4_comm_internal_err(u32 slave_read) 162 { 163 return (u32)COMM_CHAN_EVENT_INTERNAL_ERR == 164 (slave_read & (u32)COMM_CHAN_EVENT_INTERNAL_ERR) ? 1 : 0; 165 } 166 167 void mlx4_enter_error_state(struct mlx4_dev_persistent *persist) 168 { 169 int err; 170 struct mlx4_dev *dev; 171 172 if (!mlx4_internal_err_reset) 173 return; 174 175 mutex_lock(&persist->device_state_mutex); 176 if (persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) 177 goto out; 178 179 dev = persist->dev; 180 mlx4_err(dev, "device is going to be reset\n"); 181 if (mlx4_is_slave(dev)) { 182 err = mlx4_reset_slave(dev); 183 } else { 184 mlx4_crdump_collect(dev); 185 err = mlx4_reset_master(dev); 186 } 187 188 if (!err) { 189 mlx4_err(dev, "device was reset successfully\n"); 190 } else { 191 /* EEH could have disabled the PCI channel during reset. That's 192 * recoverable and the PCI error flow will handle it. 193 */ 194 if (!pci_channel_offline(dev->persist->pdev)) 195 BUG_ON(1); 196 } 197 dev->persist->state |= MLX4_DEVICE_STATE_INTERNAL_ERROR; 198 mutex_unlock(&persist->device_state_mutex); 199 200 /* At that step HW was already reset, now notify clients */ 201 mlx4_dispatch_event(dev, MLX4_DEV_EVENT_CATASTROPHIC_ERROR, 0); 202 mlx4_cmd_wake_completions(dev); 203 return; 204 205 out: 206 mutex_unlock(&persist->device_state_mutex); 207 } 208 209 static void mlx4_handle_error_state(struct mlx4_dev_persistent *persist) 210 { 211 int err = 0; 212 213 mlx4_enter_error_state(persist); 214 mutex_lock(&persist->interface_state_mutex); 215 if (persist->interface_state & MLX4_INTERFACE_STATE_UP && 216 !(persist->interface_state & MLX4_INTERFACE_STATE_DELETION)) { 217 err = mlx4_restart_one(persist->pdev, false, NULL); 218 mlx4_info(persist->dev, "mlx4_restart_one was ended, ret=%d\n", 219 err); 220 } 221 mutex_unlock(&persist->interface_state_mutex); 222 } 223 224 static void dump_err_buf(struct mlx4_dev *dev) 225 { 226 struct mlx4_priv *priv = mlx4_priv(dev); 227 228 int i; 229 230 mlx4_err(dev, "Internal error detected:\n"); 231 for (i = 0; i < priv->fw.catas_size; ++i) 232 mlx4_err(dev, " buf[%02x]: %08x\n", 233 i, swab32(readl(priv->catas_err.map + i))); 234 } 235 236 static void poll_catas(struct timer_list *t) 237 { 238 struct mlx4_priv *priv = from_timer(priv, t, catas_err.timer); 239 struct mlx4_dev *dev = &priv->dev; 240 u32 slave_read; 241 242 if (mlx4_is_slave(dev)) { 243 slave_read = swab32(readl(&priv->mfunc.comm->slave_read)); 244 if (mlx4_comm_internal_err(slave_read)) { 245 mlx4_warn(dev, "Internal error detected on the communication channel\n"); 246 goto internal_err; 247 } 248 } else if (readl(priv->catas_err.map)) { 249 dump_err_buf(dev); 250 goto internal_err; 251 } 252 253 if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) { 254 mlx4_warn(dev, "Internal error mark was detected on device\n"); 255 goto internal_err; 256 } 257 258 mod_timer(&priv->catas_err.timer, 259 round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL)); 260 return; 261 262 internal_err: 263 if (mlx4_internal_err_reset) 264 queue_work(dev->persist->catas_wq, &dev->persist->catas_work); 265 } 266 267 static void catas_reset(struct work_struct *work) 268 { 269 struct mlx4_dev_persistent *persist = 270 container_of(work, struct mlx4_dev_persistent, 271 catas_work); 272 273 mlx4_handle_error_state(persist); 274 } 275 276 void mlx4_start_catas_poll(struct mlx4_dev *dev) 277 { 278 struct mlx4_priv *priv = mlx4_priv(dev); 279 phys_addr_t addr; 280 281 INIT_LIST_HEAD(&priv->catas_err.list); 282 timer_setup(&priv->catas_err.timer, poll_catas, 0); 283 priv->catas_err.map = NULL; 284 285 if (!mlx4_is_slave(dev)) { 286 addr = pci_resource_start(dev->persist->pdev, 287 priv->fw.catas_bar) + 288 priv->fw.catas_offset; 289 290 priv->catas_err.map = ioremap(addr, priv->fw.catas_size * 4); 291 if (!priv->catas_err.map) { 292 mlx4_warn(dev, "Failed to map internal error buffer at 0x%llx\n", 293 (unsigned long long)addr); 294 return; 295 } 296 } 297 298 priv->catas_err.timer.expires = 299 round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL); 300 add_timer(&priv->catas_err.timer); 301 } 302 303 void mlx4_stop_catas_poll(struct mlx4_dev *dev) 304 { 305 struct mlx4_priv *priv = mlx4_priv(dev); 306 307 del_timer_sync(&priv->catas_err.timer); 308 309 if (priv->catas_err.map) { 310 iounmap(priv->catas_err.map); 311 priv->catas_err.map = NULL; 312 } 313 314 if (dev->persist->interface_state & MLX4_INTERFACE_STATE_DELETION) 315 flush_workqueue(dev->persist->catas_wq); 316 } 317 318 int mlx4_catas_init(struct mlx4_dev *dev) 319 { 320 INIT_WORK(&dev->persist->catas_work, catas_reset); 321 dev->persist->catas_wq = create_singlethread_workqueue("mlx4_health"); 322 if (!dev->persist->catas_wq) 323 return -ENOMEM; 324 325 return 0; 326 } 327 328 void mlx4_catas_end(struct mlx4_dev *dev) 329 { 330 if (dev->persist->catas_wq) { 331 destroy_workqueue(dev->persist->catas_wq); 332 dev->persist->catas_wq = NULL; 333 } 334 } 335