drivers/pci/pcie/err.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * This file implements the error recovery as a core part of PCIe error
   4  * reporting. When a PCIe error is delivered, an error message will be
   5  * collected and printed to console, then, an error recovery procedure
   6  * will be executed by following the PCI error recovery rules.
   7  *
   8  * Copyright (C) 2006 Intel Corp.
   9  *      Tom Long Nguyen (tom.l.nguyen@intel.com)
  10  *      Zhang Yanmin (yanmin.zhang@intel.com)
  11  */
  12
  13 #include <linux/pci.h>
  14 #include <linux/module.h>
  15 #include <linux/pci.h>
  16 #include <linux/kernel.h>
  17 #include <linux/errno.h>
  18 #include <linux/aer.h>
  19 #include "portdrv.h"
  20 #include "../pci.h"
  21
  22 struct aer_broadcast_data {
  23         enum pci_channel_state state;
  24         enum pci_ers_result result;
  25 };
  26
  27 static pci_ers_result_t merge_result(enum pci_ers_result orig,
  28                                   enum pci_ers_result new)
  29 {
  30         if (new == PCI_ERS_RESULT_NO_AER_DRIVER)
  31                 return PCI_ERS_RESULT_NO_AER_DRIVER;
  32
  33         if (new == PCI_ERS_RESULT_NONE)
  34                 return orig;
  35
  36         switch (orig) {
  37         case PCI_ERS_RESULT_CAN_RECOVER:
  38         case PCI_ERS_RESULT_RECOVERED:
  39                 orig = new;
  40                 break;
  41         case PCI_ERS_RESULT_DISCONNECT:
  42                 if (new == PCI_ERS_RESULT_NEED_RESET)
  43                         orig = PCI_ERS_RESULT_NEED_RESET;
  44                 break;
  45         default:
  46                 break;
  47         }
  48
  49         return orig;
  50 }
  51
  52 static int report_error_detected(struct pci_dev *dev, void *data)
  53 {
  54         pci_ers_result_t vote;
  55         const struct pci_error_handlers *err_handler;
  56         struct aer_broadcast_data *result_data;
  57
  58         result_data = (struct aer_broadcast_data *) data;
  59
  60         device_lock(&dev->dev);
  61         dev->error_state = result_data->state;
  62
  63         if (!dev->driver ||
  64                 !dev->driver->err_handler ||
  65                 !dev->driver->err_handler->error_detected) {
  66                 /*
  67                  * If any device in the subtree does not have an error_detected
  68                  * callback, PCI_ERS_RESULT_NO_AER_DRIVER prevents subsequent
  69                  * error callbacks of "any" device in the subtree, and will
  70                  * exit in the disconnected error state.
  71                  */
  72                 if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE)
  73                         vote = PCI_ERS_RESULT_NO_AER_DRIVER;
  74                 else
  75                         vote = PCI_ERS_RESULT_NONE;
  76         } else {
  77                 err_handler = dev->driver->err_handler;
  78                 vote = err_handler->error_detected(dev, result_data->state);
  79                 pci_uevent_ers(dev, PCI_ERS_RESULT_NONE);
  80         }
  81
  82         result_data->result = merge_result(result_data->result, vote);
  83         device_unlock(&dev->dev);
  84         return 0;
  85 }
  86
  87 static int report_mmio_enabled(struct pci_dev *dev, void *data)
  88 {
  89         pci_ers_result_t vote;
  90         const struct pci_error_handlers *err_handler;
  91         struct aer_broadcast_data *result_data;
  92
  93         result_data = (struct aer_broadcast_data *) data;
  94
  95         device_lock(&dev->dev);
  96         if (!dev->driver ||
  97                 !dev->driver->err_handler ||
  98                 !dev->driver->err_handler->mmio_enabled)
  99                 goto out;
 100
 101         err_handler = dev->driver->err_handler;
 102         vote = err_handler->mmio_enabled(dev);
 103         result_data->result = merge_result(result_data->result, vote);
 104 out:
 105         device_unlock(&dev->dev);
 106         return 0;
 107 }
 108
 109 static int report_slot_reset(struct pci_dev *dev, void *data)
 110 {
 111         pci_ers_result_t vote;
 112         const struct pci_error_handlers *err_handler;
 113         struct aer_broadcast_data *result_data;
 114
 115         result_data = (struct aer_broadcast_data *) data;
 116
 117         device_lock(&dev->dev);
 118         if (!dev->driver ||
 119                 !dev->driver->err_handler ||
 120                 !dev->driver->err_handler->slot_reset)
 121                 goto out;
 122
 123         err_handler = dev->driver->err_handler;
 124         vote = err_handler->slot_reset(dev);
 125         result_data->result = merge_result(result_data->result, vote);
 126 out:
 127         device_unlock(&dev->dev);
 128         return 0;
 129 }
 130
 131 static int report_resume(struct pci_dev *dev, void *data)
 132 {
 133         const struct pci_error_handlers *err_handler;
 134
 135         device_lock(&dev->dev);
 136         dev->error_state = pci_channel_io_normal;
 137
 138         if (!dev->driver ||
 139                 !dev->driver->err_handler ||
 140                 !dev->driver->err_handler->resume)
 141                 goto out;
 142
 143         err_handler = dev->driver->err_handler;
 144         err_handler->resume(dev);
 145         pci_uevent_ers(dev, PCI_ERS_RESULT_RECOVERED);
 146 out:
 147         device_unlock(&dev->dev);
 148         return 0;
 149 }
 150
 151 /**
 152  * default_reset_link - default reset function
 153  * @dev: pointer to pci_dev data structure
 154  *
 155  * Invoked when performing link reset on a Downstream Port or a
 156  * Root Port with no aer driver.
 157  */
 158 static pci_ers_result_t default_reset_link(struct pci_dev *dev)
 159 {
 160         int rc;
 161
 162         rc = pci_bus_error_reset(dev);
 163         pci_printk(KERN_DEBUG, dev, "downstream link has been reset\n");
 164         return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
 165 }
 166
 167 static pci_ers_result_t reset_link(struct pci_dev *dev, u32 service)
 168 {
 169         pci_ers_result_t status;
 170         struct pcie_port_service_driver *driver = NULL;
 171
 172         driver = pcie_port_find_service(dev, service);
 173         if (driver && driver->reset_link) {
 174                 status = driver->reset_link(dev);
 175         } else if (dev->has_secondary_link) {
 176                 status = default_reset_link(dev);
 177         } else {
 178                 pci_printk(KERN_DEBUG, dev, "no link-reset support at upstream device %s\n",
 179                         pci_name(dev));
 180                 return PCI_ERS_RESULT_DISCONNECT;
 181         }
 182
 183         if (status != PCI_ERS_RESULT_RECOVERED) {
 184                 pci_printk(KERN_DEBUG, dev, "link reset at upstream device %s failed\n",
 185                         pci_name(dev));
 186                 return PCI_ERS_RESULT_DISCONNECT;
 187         }
 188
 189         return status;
 190 }
 191
 192 /**
 193  * broadcast_error_message - handle message broadcast to downstream drivers
 194  * @dev: pointer to from where in a hierarchy message is broadcasted down
 195  * @state: error state
 196  * @error_mesg: message to print
 197  * @cb: callback to be broadcasted
 198  *
 199  * Invoked during error recovery process. Once being invoked, the content
 200  * of error severity will be broadcasted to all downstream drivers in a
 201  * hierarchy in question.
 202  */
 203 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
 204         enum pci_channel_state state,
 205         char *error_mesg,
 206         int (*cb)(struct pci_dev *, void *))
 207 {
 208         struct aer_broadcast_data result_data;
 209
 210         pci_printk(KERN_DEBUG, dev, "broadcast %s message\n", error_mesg);
 211         result_data.state = state;
 212         if (cb == report_error_detected)
 213                 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
 214         else
 215                 result_data.result = PCI_ERS_RESULT_RECOVERED;
 216
 217         pci_walk_bus(dev->subordinate, cb, &result_data);
 218         return result_data.result;
 219 }
 220
 221 /**
 222  * pcie_do_fatal_recovery - handle fatal error recovery process
 223  * @dev: pointer to a pci_dev data structure of agent detecting an error
 224  *
 225  * Invoked when an error is fatal. Once being invoked, removes the devices
 226  * beneath this AER agent, followed by reset link e.g. secondary bus reset
 227  * followed by re-enumeration of devices.
 228  */
 229 void pcie_do_fatal_recovery(struct pci_dev *dev, u32 service)
 230 {
 231         struct pci_dev *udev;
 232         struct pci_bus *parent;
 233         struct pci_dev *pdev, *temp;
 234         pci_ers_result_t result;
 235
 236         if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
 237                 udev = dev;
 238         else
 239                 udev = dev->bus->self;
 240
 241         parent = udev->subordinate;
 242         pci_lock_rescan_remove();
 243         pci_dev_get(dev);
 244         list_for_each_entry_safe_reverse(pdev, temp, &parent->devices,
 245                                          bus_list) {
 246                 pci_dev_get(pdev);
 247                 pci_dev_set_disconnected(pdev, NULL);
 248                 if (pci_has_subordinate(pdev))
 249                         pci_walk_bus(pdev->subordinate,
 250                                      pci_dev_set_disconnected, NULL);
 251                 pci_stop_and_remove_bus_device(pdev);
 252                 pci_dev_put(pdev);
 253         }
 254
 255         result = reset_link(udev, service);
 256
 257         if ((service == PCIE_PORT_SERVICE_AER) &&
 258             (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)) {
 259                 /*
 260                  * If the error is reported by a bridge, we think this error
 261                  * is related to the downstream link of the bridge, so we
 262                  * do error recovery on all subordinates of the bridge instead
 263                  * of the bridge and clear the error status of the bridge.
 264                  */
 265                 pci_aer_clear_fatal_status(dev);
 266                 pci_aer_clear_device_status(dev);
 267         }
 268
 269         if (result == PCI_ERS_RESULT_RECOVERED) {
 270                 if (pcie_wait_for_link(udev, true))
 271                         pci_rescan_bus(udev->bus);
 272                 pci_info(dev, "Device recovery from fatal error successful\n");
 273         } else {
 274                 pci_uevent_ers(dev, PCI_ERS_RESULT_DISCONNECT);
 275                 pci_info(dev, "Device recovery from fatal error failed\n");
 276         }
 277
 278         pci_dev_put(dev);
 279         pci_unlock_rescan_remove();
 280 }
 281
 282 /**
 283  * pcie_do_nonfatal_recovery - handle nonfatal error recovery process
 284  * @dev: pointer to a pci_dev data structure of agent detecting an error
 285  *
 286  * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
 287  * error detected message to all downstream drivers within a hierarchy in
 288  * question and return the returned code.
 289  */
 290 void pcie_do_nonfatal_recovery(struct pci_dev *dev)
 291 {
 292         pci_ers_result_t status;
 293         enum pci_channel_state state;
 294
 295         state = pci_channel_io_normal;
 296
 297         /*
 298          * Error recovery runs on all subordinates of the first downstream port.
 299          * If the downstream port detected the error, it is cleared at the end.
 300          */
 301         if (!(pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT ||
 302               pci_pcie_type(dev) == PCI_EXP_TYPE_DOWNSTREAM))
 303                 dev = dev->bus->self;
 304
 305         status = broadcast_error_message(dev,
 306                         state,
 307                         "error_detected",
 308                         report_error_detected);
 309
 310         if (status == PCI_ERS_RESULT_CAN_RECOVER)
 311                 status = broadcast_error_message(dev,
 312                                 state,
 313                                 "mmio_enabled",
 314                                 report_mmio_enabled);
 315
 316         if (status == PCI_ERS_RESULT_NEED_RESET) {
 317                 /*
 318                  * TODO: Should call platform-specific
 319                  * functions to reset slot before calling
 320                  * drivers' slot_reset callbacks?
 321                  */
 322                 status = broadcast_error_message(dev,
 323                                 state,
 324                                 "slot_reset",
 325                                 report_slot_reset);
 326         }
 327
 328         if (status != PCI_ERS_RESULT_RECOVERED)
 329                 goto failed;
 330
 331         broadcast_error_message(dev,
 332                                 state,
 333                                 "resume",
 334                                 report_resume);
 335
 336         pci_aer_clear_device_status(dev);
 337         pci_cleanup_aer_uncorrect_error_status(dev);
 338         pci_info(dev, "AER: Device recovery successful\n");
 339         return;
 340
 341 failed:
 342         pci_uevent_ers(dev, PCI_ERS_RESULT_DISCONNECT);
 343
 344         /* TODO: Should kernel panic here? */
 345         pci_info(dev, "AER: Device recovery failed\n");
 346 }