drivers/pci/pcie/aer/aerdrv_core.c

   1 /*
   2  * drivers/pci/pcie/aer/aerdrv_core.c
   3  *
   4  * This file is subject to the terms and conditions of the GNU General Public
   5  * License.  See the file "COPYING" in the main directory of this archive
   6  * for more details.
   7  *
   8  * This file implements the core part of PCI-Express AER. When an pci-express
   9  * error is delivered, an error message will be collected and printed to
  10  * console, then, an error recovery procedure will be executed by following
  11  * the pci error recovery rules.
  12  *
  13  * Copyright (C) 2006 Intel Corp.
  14  *      Tom Long Nguyen (tom.l.nguyen@intel.com)
  15  *      Zhang Yanmin (yanmin.zhang@intel.com)
  16  *
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/pci.h>
  21 #include <linux/kernel.h>
  22 #include <linux/errno.h>
  23 #include <linux/pm.h>
  24 #include <linux/suspend.h>
  25 #include <linux/delay.h>
  26 #include "aerdrv.h"
  27
  28 static int forceload;
  29 static int nosourceid;
  30 module_param(forceload, bool, 0);
  31 module_param(nosourceid, bool, 0);
  32
  33 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
  34 {
  35         u16 reg16 = 0;
  36         int pos;
  37
  38         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
  39         if (!pos)
  40                 return -EIO;
  41
  42         pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
  43         if (!pos)
  44                 return -EIO;
  45
  46         pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
  47         reg16 = reg16 |
  48                 PCI_EXP_DEVCTL_CERE |
  49                 PCI_EXP_DEVCTL_NFERE |
  50                 PCI_EXP_DEVCTL_FERE |
  51                 PCI_EXP_DEVCTL_URRE;
  52         pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
  53
  54         return 0;
  55 }
  56 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
  57
  58 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
  59 {
  60         u16 reg16 = 0;
  61         int pos;
  62
  63         pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
  64         if (!pos)
  65                 return -EIO;
  66
  67         pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
  68         reg16 = reg16 & ~(PCI_EXP_DEVCTL_CERE |
  69                         PCI_EXP_DEVCTL_NFERE |
  70                         PCI_EXP_DEVCTL_FERE |
  71                         PCI_EXP_DEVCTL_URRE);
  72         pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
  73
  74         return 0;
  75 }
  76 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
  77
  78 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
  79 {
  80         int pos;
  81         u32 status;
  82
  83         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
  84         if (!pos)
  85                 return -EIO;
  86
  87         pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
  88         if (status)
  89                 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
  90
  91         return 0;
  92 }
  93 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
  94
  95 #if 0
  96 int pci_cleanup_aer_correct_error_status(struct pci_dev *dev)
  97 {
  98         int pos;
  99         u32 status;
 100
 101         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 102         if (!pos)
 103                 return -EIO;
 104
 105         pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
 106         pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
 107
 108         return 0;
 109 }
 110 #endif  /*  0  */
 111
 112 static int set_device_error_reporting(struct pci_dev *dev, void *data)
 113 {
 114         bool enable = *((bool *)data);
 115
 116         if (dev->pcie_type == PCIE_RC_PORT ||
 117             dev->pcie_type == PCIE_SW_UPSTREAM_PORT ||
 118             dev->pcie_type == PCIE_SW_DOWNSTREAM_PORT) {
 119                 if (enable)
 120                         pci_enable_pcie_error_reporting(dev);
 121                 else
 122                         pci_disable_pcie_error_reporting(dev);
 123         }
 124
 125         if (enable)
 126                 pcie_set_ecrc_checking(dev);
 127
 128         return 0;
 129 }
 130
 131 /**
 132  * set_downstream_devices_error_reporting - enable/disable the error reporting  bits on the root port and its downstream ports.
 133  * @dev: pointer to root port's pci_dev data structure
 134  * @enable: true = enable error reporting, false = disable error reporting.
 135  */
 136 static void set_downstream_devices_error_reporting(struct pci_dev *dev,
 137                                                    bool enable)
 138 {
 139         set_device_error_reporting(dev, &enable);
 140
 141         if (!dev->subordinate)
 142                 return;
 143         pci_walk_bus(dev->subordinate, set_device_error_reporting, &enable);
 144 }
 145
 146 static inline int compare_device_id(struct pci_dev *dev,
 147                         struct aer_err_info *e_info)
 148 {
 149         if (e_info->id == ((dev->bus->number << 8) | dev->devfn)) {
 150                 /*
 151                  * Device ID match
 152                  */
 153                 return 1;
 154         }
 155
 156         return 0;
 157 }
 158
 159 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
 160 {
 161         if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
 162                 e_info->dev[e_info->error_dev_num] = dev;
 163                 e_info->error_dev_num++;
 164                 return 1;
 165         }
 166
 167         return 0;
 168 }
 169
 170
 171 #define PCI_BUS(x)      (((x) >> 8) & 0xff)
 172
 173 static int find_device_iter(struct pci_dev *dev, void *data)
 174 {
 175         int pos;
 176         u32 status;
 177         u32 mask;
 178         u16 reg16;
 179         int result;
 180         struct aer_err_info *e_info = (struct aer_err_info *)data;
 181
 182         /*
 183          * When bus id is equal to 0, it might be a bad id
 184          * reported by root port.
 185          */
 186         if (!nosourceid && (PCI_BUS(e_info->id) != 0)) {
 187                 result = compare_device_id(dev, e_info);
 188                 if (result)
 189                         add_error_device(e_info, dev);
 190
 191                 /*
 192                  * If there is no multiple error, we stop
 193                  * or continue based on the id comparing.
 194                  */
 195                 if (!e_info->multi_error_valid)
 196                         return result;
 197
 198                 /*
 199                  * If there are multiple errors and id does match,
 200                  * We need continue to search other devices under
 201                  * the root port. Return 0 means that.
 202                  */
 203                 if (result)
 204                         return 0;
 205         }
 206
 207         /*
 208          * When either
 209          *      1) nosourceid==y;
 210          *      2) bus id is equal to 0. Some ports might lose the bus
 211          *              id of error source id;
 212          *      3) There are multiple errors and prior id comparing fails;
 213          * We check AER status registers to find the initial reporter.
 214          */
 215         if (atomic_read(&dev->enable_cnt) == 0)
 216                 return 0;
 217         pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
 218         if (!pos)
 219                 return 0;
 220         /* Check if AER is enabled */
 221         pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
 222         if (!(reg16 & (
 223                 PCI_EXP_DEVCTL_CERE |
 224                 PCI_EXP_DEVCTL_NFERE |
 225                 PCI_EXP_DEVCTL_FERE |
 226                 PCI_EXP_DEVCTL_URRE)))
 227                 return 0;
 228         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 229         if (!pos)
 230                 return 0;
 231
 232         status = 0;
 233         mask = 0;
 234         if (e_info->severity == AER_CORRECTABLE) {
 235                 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
 236                 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, &mask);
 237                 if (status & ~mask) {
 238                         add_error_device(e_info, dev);
 239                         goto added;
 240                 }
 241         } else {
 242                 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
 243                 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
 244                 if (status & ~mask) {
 245                         add_error_device(e_info, dev);
 246                         goto added;
 247                 }
 248         }
 249
 250         return 0;
 251
 252 added:
 253         if (e_info->multi_error_valid)
 254                 return 0;
 255         else
 256                 return 1;
 257 }
 258
 259 /**
 260  * find_source_device - search through device hierarchy for source device
 261  * @parent: pointer to Root Port pci_dev data structure
 262  * @err_info: including detailed error information such like id
 263  *
 264  * Invoked when error is detected at the Root Port.
 265  */
 266 static void find_source_device(struct pci_dev *parent,
 267                 struct aer_err_info *e_info)
 268 {
 269         struct pci_dev *dev = parent;
 270         int result;
 271
 272         /* Is Root Port an agent that sends error message? */
 273         result = find_device_iter(dev, e_info);
 274         if (result)
 275                 return;
 276
 277         pci_walk_bus(parent->subordinate, find_device_iter, e_info);
 278 }
 279
 280 static int report_error_detected(struct pci_dev *dev, void *data)
 281 {
 282         pci_ers_result_t vote;
 283         struct pci_error_handlers *err_handler;
 284         struct aer_broadcast_data *result_data;
 285         result_data = (struct aer_broadcast_data *) data;
 286
 287         dev->error_state = result_data->state;
 288
 289         if (!dev->driver ||
 290                 !dev->driver->err_handler ||
 291                 !dev->driver->err_handler->error_detected) {
 292                 if (result_data->state == pci_channel_io_frozen &&
 293                         !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
 294                         /*
 295                          * In case of fatal recovery, if one of down-
 296                          * stream device has no driver. We might be
 297                          * unable to recover because a later insmod
 298                          * of a driver for this device is unaware of
 299                          * its hw state.
 300                          */
 301                         dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
 302                                    dev->driver ?
 303                                    "no AER-aware driver" : "no driver");
 304                 }
 305                 return 0;
 306         }
 307
 308         err_handler = dev->driver->err_handler;
 309         vote = err_handler->error_detected(dev, result_data->state);
 310         result_data->result = merge_result(result_data->result, vote);
 311         return 0;
 312 }
 313
 314 static int report_mmio_enabled(struct pci_dev *dev, void *data)
 315 {
 316         pci_ers_result_t vote;
 317         struct pci_error_handlers *err_handler;
 318         struct aer_broadcast_data *result_data;
 319         result_data = (struct aer_broadcast_data *) data;
 320
 321         if (!dev->driver ||
 322                 !dev->driver->err_handler ||
 323                 !dev->driver->err_handler->mmio_enabled)
 324                 return 0;
 325
 326         err_handler = dev->driver->err_handler;
 327         vote = err_handler->mmio_enabled(dev);
 328         result_data->result = merge_result(result_data->result, vote);
 329         return 0;
 330 }
 331
 332 static int report_slot_reset(struct pci_dev *dev, void *data)
 333 {
 334         pci_ers_result_t vote;
 335         struct pci_error_handlers *err_handler;
 336         struct aer_broadcast_data *result_data;
 337         result_data = (struct aer_broadcast_data *) data;
 338
 339         if (!dev->driver ||
 340                 !dev->driver->err_handler ||
 341                 !dev->driver->err_handler->slot_reset)
 342                 return 0;
 343
 344         err_handler = dev->driver->err_handler;
 345         vote = err_handler->slot_reset(dev);
 346         result_data->result = merge_result(result_data->result, vote);
 347         return 0;
 348 }
 349
 350 static int report_resume(struct pci_dev *dev, void *data)
 351 {
 352         struct pci_error_handlers *err_handler;
 353
 354         dev->error_state = pci_channel_io_normal;
 355
 356         if (!dev->driver ||
 357                 !dev->driver->err_handler ||
 358                 !dev->driver->err_handler->resume)
 359                 return 0;
 360
 361         err_handler = dev->driver->err_handler;
 362         err_handler->resume(dev);
 363         return 0;
 364 }
 365
 366 /**
 367  * broadcast_error_message - handle message broadcast to downstream drivers
 368  * @dev: pointer to from where in a hierarchy message is broadcasted down
 369  * @state: error state
 370  * @error_mesg: message to print
 371  * @cb: callback to be broadcasted
 372  *
 373  * Invoked during error recovery process. Once being invoked, the content
 374  * of error severity will be broadcasted to all downstream drivers in a
 375  * hierarchy in question.
 376  */
 377 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
 378         enum pci_channel_state state,
 379         char *error_mesg,
 380         int (*cb)(struct pci_dev *, void *))
 381 {
 382         struct aer_broadcast_data result_data;
 383
 384         dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
 385         result_data.state = state;
 386         if (cb == report_error_detected)
 387                 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
 388         else
 389                 result_data.result = PCI_ERS_RESULT_RECOVERED;
 390
 391         if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
 392                 /*
 393                  * If the error is reported by a bridge, we think this error
 394                  * is related to the downstream link of the bridge, so we
 395                  * do error recovery on all subordinates of the bridge instead
 396                  * of the bridge and clear the error status of the bridge.
 397                  */
 398                 if (cb == report_error_detected)
 399                         dev->error_state = state;
 400                 pci_walk_bus(dev->subordinate, cb, &result_data);
 401                 if (cb == report_resume) {
 402                         pci_cleanup_aer_uncorrect_error_status(dev);
 403                         dev->error_state = pci_channel_io_normal;
 404                 }
 405         } else {
 406                 /*
 407                  * If the error is reported by an end point, we think this
 408                  * error is related to the upstream link of the end point.
 409                  */
 410                 pci_walk_bus(dev->bus, cb, &result_data);
 411         }
 412
 413         return result_data.result;
 414 }
 415
 416 struct find_aer_service_data {
 417         struct pcie_port_service_driver *aer_driver;
 418         int is_downstream;
 419 };
 420
 421 static int find_aer_service_iter(struct device *device, void *data)
 422 {
 423         struct device_driver *driver;
 424         struct pcie_port_service_driver *service_driver;
 425         struct find_aer_service_data *result;
 426
 427         result = (struct find_aer_service_data *) data;
 428
 429         if (device->bus == &pcie_port_bus_type) {
 430                 struct pcie_port_data *port_data;
 431
 432                 port_data = pci_get_drvdata(to_pcie_device(device)->port);
 433                 if (port_data->port_type == PCIE_SW_DOWNSTREAM_PORT)
 434                         result->is_downstream = 1;
 435
 436                 driver = device->driver;
 437                 if (driver) {
 438                         service_driver = to_service_driver(driver);
 439                         if (service_driver->service == PCIE_PORT_SERVICE_AER) {
 440                                 result->aer_driver = service_driver;
 441                                 return 1;
 442                         }
 443                 }
 444         }
 445
 446         return 0;
 447 }
 448
 449 static void find_aer_service(struct pci_dev *dev,
 450                 struct find_aer_service_data *data)
 451 {
 452         int retval;
 453         retval = device_for_each_child(&dev->dev, data, find_aer_service_iter);
 454 }
 455
 456 static pci_ers_result_t reset_link(struct pcie_device *aerdev,
 457                 struct pci_dev *dev)
 458 {
 459         struct pci_dev *udev;
 460         pci_ers_result_t status;
 461         struct find_aer_service_data data;
 462
 463         if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)
 464                 udev = dev;
 465         else
 466                 udev = dev->bus->self;
 467
 468         data.is_downstream = 0;
 469         data.aer_driver = NULL;
 470         find_aer_service(udev, &data);
 471
 472         /*
 473          * Use the aer driver of the error agent firstly.
 474          * If it hasn't the aer driver, use the root port's
 475          */
 476         if (!data.aer_driver || !data.aer_driver->reset_link) {
 477                 if (data.is_downstream &&
 478                         aerdev->device.driver &&
 479                         to_service_driver(aerdev->device.driver)->reset_link) {
 480                         data.aer_driver =
 481                                 to_service_driver(aerdev->device.driver);
 482                 } else {
 483                         dev_printk(KERN_DEBUG, &dev->dev, "no link-reset "
 484                                    "support\n");
 485                         return PCI_ERS_RESULT_DISCONNECT;
 486                 }
 487         }
 488
 489         status = data.aer_driver->reset_link(udev);
 490         if (status != PCI_ERS_RESULT_RECOVERED) {
 491                 dev_printk(KERN_DEBUG, &dev->dev, "link reset at upstream "
 492                            "device %s failed\n", pci_name(udev));
 493                 return PCI_ERS_RESULT_DISCONNECT;
 494         }
 495
 496         return status;
 497 }
 498
 499 /**
 500  * do_recovery - handle nonfatal/fatal error recovery process
 501  * @aerdev: pointer to a pcie_device data structure of root port
 502  * @dev: pointer to a pci_dev data structure of agent detecting an error
 503  * @severity: error severity type
 504  *
 505  * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
 506  * error detected message to all downstream drivers within a hierarchy in
 507  * question and return the returned code.
 508  */
 509 static pci_ers_result_t do_recovery(struct pcie_device *aerdev,
 510                 struct pci_dev *dev,
 511                 int severity)
 512 {
 513         pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
 514         enum pci_channel_state state;
 515
 516         if (severity == AER_FATAL)
 517                 state = pci_channel_io_frozen;
 518         else
 519                 state = pci_channel_io_normal;
 520
 521         status = broadcast_error_message(dev,
 522                         state,
 523                         "error_detected",
 524                         report_error_detected);
 525
 526         if (severity == AER_FATAL) {
 527                 result = reset_link(aerdev, dev);
 528                 if (result != PCI_ERS_RESULT_RECOVERED) {
 529                         /* TODO: Should panic here? */
 530                         return result;
 531                 }
 532         }
 533
 534         if (status == PCI_ERS_RESULT_CAN_RECOVER)
 535                 status = broadcast_error_message(dev,
 536                                 state,
 537                                 "mmio_enabled",
 538                                 report_mmio_enabled);
 539
 540         if (status == PCI_ERS_RESULT_NEED_RESET) {
 541                 /*
 542                  * TODO: Should call platform-specific
 543                  * functions to reset slot before calling
 544                  * drivers' slot_reset callbacks?
 545                  */
 546                 status = broadcast_error_message(dev,
 547                                 state,
 548                                 "slot_reset",
 549                                 report_slot_reset);
 550         }
 551
 552         if (status == PCI_ERS_RESULT_RECOVERED)
 553                 broadcast_error_message(dev,
 554                                 state,
 555                                 "resume",
 556                                 report_resume);
 557
 558         return status;
 559 }
 560
 561 /**
 562  * handle_error_source - handle logging error into an event log
 563  * @aerdev: pointer to pcie_device data structure of the root port
 564  * @dev: pointer to pci_dev data structure of error source device
 565  * @info: comprehensive error information
 566  *
 567  * Invoked when an error being detected by Root Port.
 568  */
 569 static void handle_error_source(struct pcie_device *aerdev,
 570         struct pci_dev *dev,
 571         struct aer_err_info *info)
 572 {
 573         pci_ers_result_t status = 0;
 574         int pos;
 575
 576         if (info->severity == AER_CORRECTABLE) {
 577                 /*
 578                  * Correctable error does not need software intevention.
 579                  * No need to go through error recovery process.
 580                  */
 581                 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 582                 if (pos)
 583                         pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
 584                                         info->status);
 585         } else {
 586                 status = do_recovery(aerdev, dev, info->severity);
 587                 if (status == PCI_ERS_RESULT_RECOVERED) {
 588                         dev_printk(KERN_DEBUG, &dev->dev, "AER driver "
 589                                    "successfully recovered\n");
 590                 } else {
 591                         /* TODO: Should kernel panic here? */
 592                         dev_printk(KERN_DEBUG, &dev->dev, "AER driver didn't "
 593                                    "recover\n");
 594                 }
 595         }
 596 }
 597
 598 /**
 599  * aer_enable_rootport - enable Root Port's interrupts when receiving messages
 600  * @rpc: pointer to a Root Port data structure
 601  *
 602  * Invoked when PCIE bus loads AER service driver.
 603  */
 604 void aer_enable_rootport(struct aer_rpc *rpc)
 605 {
 606         struct pci_dev *pdev = rpc->rpd->port;
 607         int pos, aer_pos;
 608         u16 reg16;
 609         u32 reg32;
 610
 611         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
 612         /* Clear PCIE Capability's Device Status */
 613         pci_read_config_word(pdev, pos+PCI_EXP_DEVSTA, &reg16);
 614         pci_write_config_word(pdev, pos+PCI_EXP_DEVSTA, reg16);
 615
 616         /* Disable system error generation in response to error messages */
 617         pci_read_config_word(pdev, pos + PCI_EXP_RTCTL, &reg16);
 618         reg16 &= ~(SYSTEM_ERROR_INTR_ON_MESG_MASK);
 619         pci_write_config_word(pdev, pos + PCI_EXP_RTCTL, reg16);
 620
 621         aer_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
 622         /* Clear error status */
 623         pci_read_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, &reg32);
 624         pci_write_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, reg32);
 625         pci_read_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, &reg32);
 626         pci_write_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, reg32);
 627         pci_read_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, &reg32);
 628         pci_write_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, reg32);
 629
 630         /*
 631          * Enable error reporting for the root port device and downstream port
 632          * devices.
 633          */
 634         set_downstream_devices_error_reporting(pdev, true);
 635
 636         /* Enable Root Port's interrupt in response to error messages */
 637         pci_write_config_dword(pdev,
 638                 aer_pos + PCI_ERR_ROOT_COMMAND,
 639                 ROOT_PORT_INTR_ON_MESG_MASK);
 640 }
 641
 642 /**
 643  * disable_root_aer - disable Root Port's interrupts when receiving messages
 644  * @rpc: pointer to a Root Port data structure
 645  *
 646  * Invoked when PCIE bus unloads AER service driver.
 647  */
 648 static void disable_root_aer(struct aer_rpc *rpc)
 649 {
 650         struct pci_dev *pdev = rpc->rpd->port;
 651         u32 reg32;
 652         int pos;
 653
 654         /*
 655          * Disable error reporting for the root port device and downstream port
 656          * devices.
 657          */
 658         set_downstream_devices_error_reporting(pdev, false);
 659
 660         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
 661         /* Disable Root's interrupt in response to error messages */
 662         pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_COMMAND, 0);
 663
 664         /* Clear Root's error status reg */
 665         pci_read_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, &reg32);
 666         pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, reg32);
 667 }
 668
 669 /**
 670  * get_e_source - retrieve an error source
 671  * @rpc: pointer to the root port which holds an error
 672  *
 673  * Invoked by DPC handler to consume an error.
 674  */
 675 static struct aer_err_source *get_e_source(struct aer_rpc *rpc)
 676 {
 677         struct aer_err_source *e_source;
 678         unsigned long flags;
 679
 680         /* Lock access to Root error producer/consumer index */
 681         spin_lock_irqsave(&rpc->e_lock, flags);
 682         if (rpc->prod_idx == rpc->cons_idx) {
 683                 spin_unlock_irqrestore(&rpc->e_lock, flags);
 684                 return NULL;
 685         }
 686         e_source = &rpc->e_sources[rpc->cons_idx];
 687         rpc->cons_idx++;
 688         if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
 689                 rpc->cons_idx = 0;
 690         spin_unlock_irqrestore(&rpc->e_lock, flags);
 691
 692         return e_source;
 693 }
 694
 695 /**
 696  * get_device_error_info - read error status from dev and store it to info
 697  * @dev: pointer to the device expected to have a error record
 698  * @info: pointer to structure to store the error record
 699  *
 700  * Return 1 on success, 0 on error.
 701  */
 702 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
 703 {
 704         int pos, temp;
 705
 706         info->status = 0;
 707         info->tlp_header_valid = 0;
 708
 709         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 710
 711         /* The device might not support AER */
 712         if (!pos)
 713                 return 1;
 714
 715         if (info->severity == AER_CORRECTABLE) {
 716                 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
 717                         &info->status);
 718                 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
 719                         &info->mask);
 720                 if (!(info->status & ~info->mask))
 721                         return 0;
 722         } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
 723                 info->severity == AER_NONFATAL) {
 724
 725                 /* Link is still healthy for IO reads */
 726                 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
 727                         &info->status);
 728                 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
 729                         &info->mask);
 730                 if (!(info->status & ~info->mask))
 731                         return 0;
 732
 733                 /* Get First Error Pointer */
 734                 pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
 735                 info->first_error = PCI_ERR_CAP_FEP(temp);
 736
 737                 if (info->status & AER_LOG_TLP_MASKS) {
 738                         info->tlp_header_valid = 1;
 739                         pci_read_config_dword(dev,
 740                                 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
 741                         pci_read_config_dword(dev,
 742                                 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
 743                         pci_read_config_dword(dev,
 744                                 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
 745                         pci_read_config_dword(dev,
 746                                 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
 747                 }
 748         }
 749
 750         return 1;
 751 }
 752
 753 static inline void aer_process_err_devices(struct pcie_device *p_device,
 754                         struct aer_err_info *e_info)
 755 {
 756         int i;
 757
 758         if (!e_info->dev[0]) {
 759                 dev_printk(KERN_DEBUG, &p_device->port->dev,
 760                                 "can't find device of ID%04x\n",
 761                                 e_info->id);
 762         }
 763
 764         /* Report all before handle them, not to lost records by reset etc. */
 765         for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
 766                 if (get_device_error_info(e_info->dev[i], e_info))
 767                         aer_print_error(e_info->dev[i], e_info);
 768         }
 769         for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
 770                 if (get_device_error_info(e_info->dev[i], e_info))
 771                         handle_error_source(p_device, e_info->dev[i], e_info);
 772         }
 773 }
 774
 775 /**
 776  * aer_isr_one_error - consume an error detected by root port
 777  * @p_device: pointer to error root port service device
 778  * @e_src: pointer to an error source
 779  */
 780 static void aer_isr_one_error(struct pcie_device *p_device,
 781                 struct aer_err_source *e_src)
 782 {
 783         struct aer_err_info *e_info;
 784         int i;
 785
 786         /* struct aer_err_info might be big, so we allocate it with slab */
 787         e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL);
 788         if (e_info == NULL) {
 789                 dev_printk(KERN_DEBUG, &p_device->port->dev,
 790                         "Can't allocate mem when processing AER errors\n");
 791                 return;
 792         }
 793
 794         /*
 795          * There is a possibility that both correctable error and
 796          * uncorrectable error being logged. Report correctable error first.
 797          */
 798         for (i = 1; i & ROOT_ERR_STATUS_MASKS ; i <<= 2) {
 799                 if (i > 4)
 800                         break;
 801                 if (!(e_src->status & i))
 802                         continue;
 803
 804                 memset(e_info, 0, sizeof(struct aer_err_info));
 805
 806                 /* Init comprehensive error information */
 807                 if (i & PCI_ERR_ROOT_COR_RCV) {
 808                         e_info->id = ERR_COR_ID(e_src->id);
 809                         e_info->severity = AER_CORRECTABLE;
 810                 } else {
 811                         e_info->id = ERR_UNCOR_ID(e_src->id);
 812                         e_info->severity = ((e_src->status >> 6) & 1);
 813                 }
 814                 if (e_src->status &
 815                         (PCI_ERR_ROOT_MULTI_COR_RCV |
 816                          PCI_ERR_ROOT_MULTI_UNCOR_RCV))
 817                         e_info->multi_error_valid = 1;
 818
 819                 aer_print_port_info(p_device->port, e_info);
 820
 821                 find_source_device(p_device->port, e_info);
 822                 aer_process_err_devices(p_device, e_info);
 823         }
 824
 825         kfree(e_info);
 826 }
 827
 828 /**
 829  * aer_isr - consume errors detected by root port
 830  * @work: definition of this work item
 831  *
 832  * Invoked, as DPC, when root port records new detected error
 833  */
 834 void aer_isr(struct work_struct *work)
 835 {
 836         struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
 837         struct pcie_device *p_device = rpc->rpd;
 838         struct aer_err_source *e_src;
 839
 840         mutex_lock(&rpc->rpc_mutex);
 841         e_src = get_e_source(rpc);
 842         while (e_src) {
 843                 aer_isr_one_error(p_device, e_src);
 844                 e_src = get_e_source(rpc);
 845         }
 846         mutex_unlock(&rpc->rpc_mutex);
 847
 848         wake_up(&rpc->wait_release);
 849 }
 850
 851 /**
 852  * aer_delete_rootport - disable root port aer and delete service data
 853  * @rpc: pointer to a root port device being deleted
 854  *
 855  * Invoked when AER service unloaded on a specific Root Port
 856  */
 857 void aer_delete_rootport(struct aer_rpc *rpc)
 858 {
 859         /* Disable root port AER itself */
 860         disable_root_aer(rpc);
 861
 862         kfree(rpc);
 863 }
 864
 865 /**
 866  * aer_init - provide AER initialization
 867  * @dev: pointer to AER pcie device
 868  *
 869  * Invoked when AER service driver is loaded.
 870  */
 871 int aer_init(struct pcie_device *dev)
 872 {
 873         if (aer_osc_setup(dev) && !forceload)
 874                 return -ENXIO;
 875
 876         return 0;
 877 }