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 <linux/slab.h>
  27 #include "aerdrv.h"
  28
  29 static int forceload;
  30 static int nosourceid;
  31 module_param(forceload, bool, 0);
  32 module_param(nosourceid, bool, 0);
  33
  34 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
  35 {
  36         u16 reg16 = 0;
  37         int pos;
  38
  39         if (dev->aer_firmware_first)
  40                 return -EIO;
  41
  42         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
  43         if (!pos)
  44                 return -EIO;
  45
  46         pos = pci_pcie_cap(dev);
  47         if (!pos)
  48                 return -EIO;
  49
  50         pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
  51         reg16 = reg16 |
  52                 PCI_EXP_DEVCTL_CERE |
  53                 PCI_EXP_DEVCTL_NFERE |
  54                 PCI_EXP_DEVCTL_FERE |
  55                 PCI_EXP_DEVCTL_URRE;
  56         pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
  57
  58         return 0;
  59 }
  60 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
  61
  62 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
  63 {
  64         u16 reg16 = 0;
  65         int pos;
  66
  67         if (dev->aer_firmware_first)
  68                 return -EIO;
  69
  70         pos = pci_pcie_cap(dev);
  71         if (!pos)
  72                 return -EIO;
  73
  74         pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
  75         reg16 = reg16 & ~(PCI_EXP_DEVCTL_CERE |
  76                         PCI_EXP_DEVCTL_NFERE |
  77                         PCI_EXP_DEVCTL_FERE |
  78                         PCI_EXP_DEVCTL_URRE);
  79         pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
  80
  81         return 0;
  82 }
  83 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
  84
  85 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
  86 {
  87         int pos;
  88         u32 status;
  89
  90         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
  91         if (!pos)
  92                 return -EIO;
  93
  94         pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
  95         if (status)
  96                 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
  97
  98         return 0;
  99 }
 100 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
 101
 102 static int set_device_error_reporting(struct pci_dev *dev, void *data)
 103 {
 104         bool enable = *((bool *)data);
 105
 106         if ((dev->pcie_type == PCI_EXP_TYPE_ROOT_PORT) ||
 107             (dev->pcie_type == PCI_EXP_TYPE_UPSTREAM) ||
 108             (dev->pcie_type == PCI_EXP_TYPE_DOWNSTREAM)) {
 109                 if (enable)
 110                         pci_enable_pcie_error_reporting(dev);
 111                 else
 112                         pci_disable_pcie_error_reporting(dev);
 113         }
 114
 115         if (enable)
 116                 pcie_set_ecrc_checking(dev);
 117
 118         return 0;
 119 }
 120
 121 /**
 122  * set_downstream_devices_error_reporting - enable/disable the error reporting  bits on the root port and its downstream ports.
 123  * @dev: pointer to root port's pci_dev data structure
 124  * @enable: true = enable error reporting, false = disable error reporting.
 125  */
 126 static void set_downstream_devices_error_reporting(struct pci_dev *dev,
 127                                                    bool enable)
 128 {
 129         set_device_error_reporting(dev, &enable);
 130
 131         if (!dev->subordinate)
 132                 return;
 133         pci_walk_bus(dev->subordinate, set_device_error_reporting, &enable);
 134 }
 135
 136 static inline int compare_device_id(struct pci_dev *dev,
 137                         struct aer_err_info *e_info)
 138 {
 139         if (e_info->id == ((dev->bus->number << 8) | dev->devfn)) {
 140                 /*
 141                  * Device ID match
 142                  */
 143                 return 1;
 144         }
 145
 146         return 0;
 147 }
 148
 149 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
 150 {
 151         if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
 152                 e_info->dev[e_info->error_dev_num] = dev;
 153                 e_info->error_dev_num++;
 154                 return 1;
 155         }
 156
 157         return 0;
 158 }
 159
 160
 161 #define PCI_BUS(x)      (((x) >> 8) & 0xff)
 162
 163 static int find_device_iter(struct pci_dev *dev, void *data)
 164 {
 165         int pos;
 166         u32 status;
 167         u32 mask;
 168         u16 reg16;
 169         int result;
 170         struct aer_err_info *e_info = (struct aer_err_info *)data;
 171
 172         /*
 173          * When bus id is equal to 0, it might be a bad id
 174          * reported by root port.
 175          */
 176         if (!nosourceid && (PCI_BUS(e_info->id) != 0)) {
 177                 result = compare_device_id(dev, e_info);
 178                 if (result)
 179                         add_error_device(e_info, dev);
 180
 181                 /*
 182                  * If there is no multiple error, we stop
 183                  * or continue based on the id comparing.
 184                  */
 185                 if (!e_info->multi_error_valid)
 186                         return result;
 187
 188                 /*
 189                  * If there are multiple errors and id does match,
 190                  * We need continue to search other devices under
 191                  * the root port. Return 0 means that.
 192                  */
 193                 if (result)
 194                         return 0;
 195         }
 196
 197         /*
 198          * When either
 199          *      1) nosourceid==y;
 200          *      2) bus id is equal to 0. Some ports might lose the bus
 201          *              id of error source id;
 202          *      3) There are multiple errors and prior id comparing fails;
 203          * We check AER status registers to find the initial reporter.
 204          */
 205         if (atomic_read(&dev->enable_cnt) == 0)
 206                 return 0;
 207         pos = pci_pcie_cap(dev);
 208         if (!pos)
 209                 return 0;
 210         /* Check if AER is enabled */
 211         pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
 212         if (!(reg16 & (
 213                 PCI_EXP_DEVCTL_CERE |
 214                 PCI_EXP_DEVCTL_NFERE |
 215                 PCI_EXP_DEVCTL_FERE |
 216                 PCI_EXP_DEVCTL_URRE)))
 217                 return 0;
 218         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 219         if (!pos)
 220                 return 0;
 221
 222         status = 0;
 223         mask = 0;
 224         if (e_info->severity == AER_CORRECTABLE) {
 225                 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
 226                 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, &mask);
 227                 if (status & ~mask) {
 228                         add_error_device(e_info, dev);
 229                         goto added;
 230                 }
 231         } else {
 232                 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
 233                 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
 234                 if (status & ~mask) {
 235                         add_error_device(e_info, dev);
 236                         goto added;
 237                 }
 238         }
 239
 240         return 0;
 241
 242 added:
 243         if (e_info->multi_error_valid)
 244                 return 0;
 245         else
 246                 return 1;
 247 }
 248
 249 /**
 250  * find_source_device - search through device hierarchy for source device
 251  * @parent: pointer to Root Port pci_dev data structure
 252  * @err_info: including detailed error information such like id
 253  *
 254  * Invoked when error is detected at the Root Port.
 255  */
 256 static void find_source_device(struct pci_dev *parent,
 257                 struct aer_err_info *e_info)
 258 {
 259         struct pci_dev *dev = parent;
 260         int result;
 261
 262         /* Is Root Port an agent that sends error message? */
 263         result = find_device_iter(dev, e_info);
 264         if (result)
 265                 return;
 266
 267         pci_walk_bus(parent->subordinate, find_device_iter, e_info);
 268 }
 269
 270 static int report_error_detected(struct pci_dev *dev, void *data)
 271 {
 272         pci_ers_result_t vote;
 273         struct pci_error_handlers *err_handler;
 274         struct aer_broadcast_data *result_data;
 275         result_data = (struct aer_broadcast_data *) data;
 276
 277         dev->error_state = result_data->state;
 278
 279         if (!dev->driver ||
 280                 !dev->driver->err_handler ||
 281                 !dev->driver->err_handler->error_detected) {
 282                 if (result_data->state == pci_channel_io_frozen &&
 283                         !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
 284                         /*
 285                          * In case of fatal recovery, if one of down-
 286                          * stream device has no driver. We might be
 287                          * unable to recover because a later insmod
 288                          * of a driver for this device is unaware of
 289                          * its hw state.
 290                          */
 291                         dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
 292                                    dev->driver ?
 293                                    "no AER-aware driver" : "no driver");
 294                 }
 295                 return 0;
 296         }
 297
 298         err_handler = dev->driver->err_handler;
 299         vote = err_handler->error_detected(dev, result_data->state);
 300         result_data->result = merge_result(result_data->result, vote);
 301         return 0;
 302 }
 303
 304 static int report_mmio_enabled(struct pci_dev *dev, void *data)
 305 {
 306         pci_ers_result_t vote;
 307         struct pci_error_handlers *err_handler;
 308         struct aer_broadcast_data *result_data;
 309         result_data = (struct aer_broadcast_data *) data;
 310
 311         if (!dev->driver ||
 312                 !dev->driver->err_handler ||
 313                 !dev->driver->err_handler->mmio_enabled)
 314                 return 0;
 315
 316         err_handler = dev->driver->err_handler;
 317         vote = err_handler->mmio_enabled(dev);
 318         result_data->result = merge_result(result_data->result, vote);
 319         return 0;
 320 }
 321
 322 static int report_slot_reset(struct pci_dev *dev, void *data)
 323 {
 324         pci_ers_result_t vote;
 325         struct pci_error_handlers *err_handler;
 326         struct aer_broadcast_data *result_data;
 327         result_data = (struct aer_broadcast_data *) data;
 328
 329         if (!dev->driver ||
 330                 !dev->driver->err_handler ||
 331                 !dev->driver->err_handler->slot_reset)
 332                 return 0;
 333
 334         err_handler = dev->driver->err_handler;
 335         vote = err_handler->slot_reset(dev);
 336         result_data->result = merge_result(result_data->result, vote);
 337         return 0;
 338 }
 339
 340 static int report_resume(struct pci_dev *dev, void *data)
 341 {
 342         struct pci_error_handlers *err_handler;
 343
 344         dev->error_state = pci_channel_io_normal;
 345
 346         if (!dev->driver ||
 347                 !dev->driver->err_handler ||
 348                 !dev->driver->err_handler->resume)
 349                 return 0;
 350
 351         err_handler = dev->driver->err_handler;
 352         err_handler->resume(dev);
 353         return 0;
 354 }
 355
 356 /**
 357  * broadcast_error_message - handle message broadcast to downstream drivers
 358  * @dev: pointer to from where in a hierarchy message is broadcasted down
 359  * @state: error state
 360  * @error_mesg: message to print
 361  * @cb: callback to be broadcasted
 362  *
 363  * Invoked during error recovery process. Once being invoked, the content
 364  * of error severity will be broadcasted to all downstream drivers in a
 365  * hierarchy in question.
 366  */
 367 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
 368         enum pci_channel_state state,
 369         char *error_mesg,
 370         int (*cb)(struct pci_dev *, void *))
 371 {
 372         struct aer_broadcast_data result_data;
 373
 374         dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
 375         result_data.state = state;
 376         if (cb == report_error_detected)
 377                 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
 378         else
 379                 result_data.result = PCI_ERS_RESULT_RECOVERED;
 380
 381         if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
 382                 /*
 383                  * If the error is reported by a bridge, we think this error
 384                  * is related to the downstream link of the bridge, so we
 385                  * do error recovery on all subordinates of the bridge instead
 386                  * of the bridge and clear the error status of the bridge.
 387                  */
 388                 if (cb == report_error_detected)
 389                         dev->error_state = state;
 390                 pci_walk_bus(dev->subordinate, cb, &result_data);
 391                 if (cb == report_resume) {
 392                         pci_cleanup_aer_uncorrect_error_status(dev);
 393                         dev->error_state = pci_channel_io_normal;
 394                 }
 395         } else {
 396                 /*
 397                  * If the error is reported by an end point, we think this
 398                  * error is related to the upstream link of the end point.
 399                  */
 400                 pci_walk_bus(dev->bus, cb, &result_data);
 401         }
 402
 403         return result_data.result;
 404 }
 405
 406 struct find_aer_service_data {
 407         struct pcie_port_service_driver *aer_driver;
 408         int is_downstream;
 409 };
 410
 411 static int find_aer_service_iter(struct device *device, void *data)
 412 {
 413         struct device_driver *driver;
 414         struct pcie_port_service_driver *service_driver;
 415         struct find_aer_service_data *result;
 416
 417         result = (struct find_aer_service_data *) data;
 418
 419         if (device->bus == &pcie_port_bus_type) {
 420                 struct pcie_device *pcie = to_pcie_device(device);
 421
 422                 if (pcie->port->pcie_type == PCI_EXP_TYPE_DOWNSTREAM)
 423                         result->is_downstream = 1;
 424
 425                 driver = device->driver;
 426                 if (driver) {
 427                         service_driver = to_service_driver(driver);
 428                         if (service_driver->service == PCIE_PORT_SERVICE_AER) {
 429                                 result->aer_driver = service_driver;
 430                                 return 1;
 431                         }
 432                 }
 433         }
 434
 435         return 0;
 436 }
 437
 438 static void find_aer_service(struct pci_dev *dev,
 439                 struct find_aer_service_data *data)
 440 {
 441         int retval;
 442         retval = device_for_each_child(&dev->dev, data, find_aer_service_iter);
 443 }
 444
 445 static pci_ers_result_t reset_link(struct pcie_device *aerdev,
 446                 struct pci_dev *dev)
 447 {
 448         struct pci_dev *udev;
 449         pci_ers_result_t status;
 450         struct find_aer_service_data data;
 451
 452         if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)
 453                 udev = dev;
 454         else
 455                 udev = dev->bus->self;
 456
 457         data.is_downstream = 0;
 458         data.aer_driver = NULL;
 459         find_aer_service(udev, &data);
 460
 461         /*
 462          * Use the aer driver of the error agent firstly.
 463          * If it hasn't the aer driver, use the root port's
 464          */
 465         if (!data.aer_driver || !data.aer_driver->reset_link) {
 466                 if (data.is_downstream &&
 467                         aerdev->device.driver &&
 468                         to_service_driver(aerdev->device.driver)->reset_link) {
 469                         data.aer_driver =
 470                                 to_service_driver(aerdev->device.driver);
 471                 } else {
 472                         dev_printk(KERN_DEBUG, &dev->dev, "no link-reset "
 473                                    "support\n");
 474                         return PCI_ERS_RESULT_DISCONNECT;
 475                 }
 476         }
 477
 478         status = data.aer_driver->reset_link(udev);
 479         if (status != PCI_ERS_RESULT_RECOVERED) {
 480                 dev_printk(KERN_DEBUG, &dev->dev, "link reset at upstream "
 481                            "device %s failed\n", pci_name(udev));
 482                 return PCI_ERS_RESULT_DISCONNECT;
 483         }
 484
 485         return status;
 486 }
 487
 488 /**
 489  * do_recovery - handle nonfatal/fatal error recovery process
 490  * @aerdev: pointer to a pcie_device data structure of root port
 491  * @dev: pointer to a pci_dev data structure of agent detecting an error
 492  * @severity: error severity type
 493  *
 494  * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
 495  * error detected message to all downstream drivers within a hierarchy in
 496  * question and return the returned code.
 497  */
 498 static pci_ers_result_t do_recovery(struct pcie_device *aerdev,
 499                 struct pci_dev *dev,
 500                 int severity)
 501 {
 502         pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
 503         enum pci_channel_state state;
 504
 505         if (severity == AER_FATAL)
 506                 state = pci_channel_io_frozen;
 507         else
 508                 state = pci_channel_io_normal;
 509
 510         status = broadcast_error_message(dev,
 511                         state,
 512                         "error_detected",
 513                         report_error_detected);
 514
 515         if (severity == AER_FATAL) {
 516                 result = reset_link(aerdev, dev);
 517                 if (result != PCI_ERS_RESULT_RECOVERED) {
 518                         /* TODO: Should panic here? */
 519                         return result;
 520                 }
 521         }
 522
 523         if (status == PCI_ERS_RESULT_CAN_RECOVER)
 524                 status = broadcast_error_message(dev,
 525                                 state,
 526                                 "mmio_enabled",
 527                                 report_mmio_enabled);
 528
 529         if (status == PCI_ERS_RESULT_NEED_RESET) {
 530                 /*
 531                  * TODO: Should call platform-specific
 532                  * functions to reset slot before calling
 533                  * drivers' slot_reset callbacks?
 534                  */
 535                 status = broadcast_error_message(dev,
 536                                 state,
 537                                 "slot_reset",
 538                                 report_slot_reset);
 539         }
 540
 541         if (status == PCI_ERS_RESULT_RECOVERED)
 542                 broadcast_error_message(dev,
 543                                 state,
 544                                 "resume",
 545                                 report_resume);
 546
 547         return status;
 548 }
 549
 550 /**
 551  * handle_error_source - handle logging error into an event log
 552  * @aerdev: pointer to pcie_device data structure of the root port
 553  * @dev: pointer to pci_dev data structure of error source device
 554  * @info: comprehensive error information
 555  *
 556  * Invoked when an error being detected by Root Port.
 557  */
 558 static void handle_error_source(struct pcie_device *aerdev,
 559         struct pci_dev *dev,
 560         struct aer_err_info *info)
 561 {
 562         pci_ers_result_t status = 0;
 563         int pos;
 564
 565         if (info->severity == AER_CORRECTABLE) {
 566                 /*
 567                  * Correctable error does not need software intevention.
 568                  * No need to go through error recovery process.
 569                  */
 570                 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 571                 if (pos)
 572                         pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
 573                                         info->status);
 574         } else {
 575                 status = do_recovery(aerdev, dev, info->severity);
 576                 if (status == PCI_ERS_RESULT_RECOVERED) {
 577                         dev_printk(KERN_DEBUG, &dev->dev, "AER driver "
 578                                    "successfully recovered\n");
 579                 } else {
 580                         /* TODO: Should kernel panic here? */
 581                         dev_printk(KERN_DEBUG, &dev->dev, "AER driver didn't "
 582                                    "recover\n");
 583                 }
 584         }
 585 }
 586
 587 /**
 588  * aer_enable_rootport - enable Root Port's interrupts when receiving messages
 589  * @rpc: pointer to a Root Port data structure
 590  *
 591  * Invoked when PCIe bus loads AER service driver.
 592  */
 593 void aer_enable_rootport(struct aer_rpc *rpc)
 594 {
 595         struct pci_dev *pdev = rpc->rpd->port;
 596         int pos, aer_pos;
 597         u16 reg16;
 598         u32 reg32;
 599
 600         pos = pci_pcie_cap(pdev);
 601         /* Clear PCIe Capability's Device Status */
 602         pci_read_config_word(pdev, pos+PCI_EXP_DEVSTA, &reg16);
 603         pci_write_config_word(pdev, pos+PCI_EXP_DEVSTA, reg16);
 604
 605         /* Disable system error generation in response to error messages */
 606         pci_read_config_word(pdev, pos + PCI_EXP_RTCTL, &reg16);
 607         reg16 &= ~(SYSTEM_ERROR_INTR_ON_MESG_MASK);
 608         pci_write_config_word(pdev, pos + PCI_EXP_RTCTL, reg16);
 609
 610         aer_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
 611         /* Clear error status */
 612         pci_read_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, &reg32);
 613         pci_write_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, reg32);
 614         pci_read_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, &reg32);
 615         pci_write_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, reg32);
 616         pci_read_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, &reg32);
 617         pci_write_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, reg32);
 618
 619         /*
 620          * Enable error reporting for the root port device and downstream port
 621          * devices.
 622          */
 623         set_downstream_devices_error_reporting(pdev, true);
 624
 625         /* Enable Root Port's interrupt in response to error messages */
 626         pci_write_config_dword(pdev,
 627                 aer_pos + PCI_ERR_ROOT_COMMAND,
 628                 ROOT_PORT_INTR_ON_MESG_MASK);
 629 }
 630
 631 /**
 632  * disable_root_aer - disable Root Port's interrupts when receiving messages
 633  * @rpc: pointer to a Root Port data structure
 634  *
 635  * Invoked when PCIe bus unloads AER service driver.
 636  */
 637 static void disable_root_aer(struct aer_rpc *rpc)
 638 {
 639         struct pci_dev *pdev = rpc->rpd->port;
 640         u32 reg32;
 641         int pos;
 642
 643         /*
 644          * Disable error reporting for the root port device and downstream port
 645          * devices.
 646          */
 647         set_downstream_devices_error_reporting(pdev, false);
 648
 649         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
 650         /* Disable Root's interrupt in response to error messages */
 651         pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_COMMAND, 0);
 652
 653         /* Clear Root's error status reg */
 654         pci_read_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, &reg32);
 655         pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, reg32);
 656 }
 657
 658 /**
 659  * get_e_source - retrieve an error source
 660  * @rpc: pointer to the root port which holds an error
 661  *
 662  * Invoked by DPC handler to consume an error.
 663  */
 664 static struct aer_err_source *get_e_source(struct aer_rpc *rpc)
 665 {
 666         struct aer_err_source *e_source;
 667         unsigned long flags;
 668
 669         /* Lock access to Root error producer/consumer index */
 670         spin_lock_irqsave(&rpc->e_lock, flags);
 671         if (rpc->prod_idx == rpc->cons_idx) {
 672                 spin_unlock_irqrestore(&rpc->e_lock, flags);
 673                 return NULL;
 674         }
 675         e_source = &rpc->e_sources[rpc->cons_idx];
 676         rpc->cons_idx++;
 677         if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
 678                 rpc->cons_idx = 0;
 679         spin_unlock_irqrestore(&rpc->e_lock, flags);
 680
 681         return e_source;
 682 }
 683
 684 /**
 685  * get_device_error_info - read error status from dev and store it to info
 686  * @dev: pointer to the device expected to have a error record
 687  * @info: pointer to structure to store the error record
 688  *
 689  * Return 1 on success, 0 on error.
 690  */
 691 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
 692 {
 693         int pos, temp;
 694
 695         info->status = 0;
 696         info->tlp_header_valid = 0;
 697
 698         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 699
 700         /* The device might not support AER */
 701         if (!pos)
 702                 return 1;
 703
 704         if (info->severity == AER_CORRECTABLE) {
 705                 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
 706                         &info->status);
 707                 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
 708                         &info->mask);
 709                 if (!(info->status & ~info->mask))
 710                         return 0;
 711         } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
 712                 info->severity == AER_NONFATAL) {
 713
 714                 /* Link is still healthy for IO reads */
 715                 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
 716                         &info->status);
 717                 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
 718                         &info->mask);
 719                 if (!(info->status & ~info->mask))
 720                         return 0;
 721
 722                 /* Get First Error Pointer */
 723                 pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
 724                 info->first_error = PCI_ERR_CAP_FEP(temp);
 725
 726                 if (info->status & AER_LOG_TLP_MASKS) {
 727                         info->tlp_header_valid = 1;
 728                         pci_read_config_dword(dev,
 729                                 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
 730                         pci_read_config_dword(dev,
 731                                 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
 732                         pci_read_config_dword(dev,
 733                                 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
 734                         pci_read_config_dword(dev,
 735                                 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
 736                 }
 737         }
 738
 739         return 1;
 740 }
 741
 742 static inline void aer_process_err_devices(struct pcie_device *p_device,
 743                         struct aer_err_info *e_info)
 744 {
 745         int i;
 746
 747         if (!e_info->dev[0]) {
 748                 dev_printk(KERN_DEBUG, &p_device->port->dev,
 749                                 "can't find device of ID%04x\n",
 750                                 e_info->id);
 751         }
 752
 753         /* Report all before handle them, not to lost records by reset etc. */
 754         for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
 755                 if (get_device_error_info(e_info->dev[i], e_info))
 756                         aer_print_error(e_info->dev[i], e_info);
 757         }
 758         for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
 759                 if (get_device_error_info(e_info->dev[i], e_info))
 760                         handle_error_source(p_device, e_info->dev[i], e_info);
 761         }
 762 }
 763
 764 /**
 765  * aer_isr_one_error - consume an error detected by root port
 766  * @p_device: pointer to error root port service device
 767  * @e_src: pointer to an error source
 768  */
 769 static void aer_isr_one_error(struct pcie_device *p_device,
 770                 struct aer_err_source *e_src)
 771 {
 772         struct aer_err_info *e_info;
 773         int i;
 774
 775         /* struct aer_err_info might be big, so we allocate it with slab */
 776         e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL);
 777         if (e_info == NULL) {
 778                 dev_printk(KERN_DEBUG, &p_device->port->dev,
 779                         "Can't allocate mem when processing AER errors\n");
 780                 return;
 781         }
 782
 783         /*
 784          * There is a possibility that both correctable error and
 785          * uncorrectable error being logged. Report correctable error first.
 786          */
 787         for (i = 1; i & ROOT_ERR_STATUS_MASKS ; i <<= 2) {
 788                 if (i > 4)
 789                         break;
 790                 if (!(e_src->status & i))
 791                         continue;
 792
 793                 memset(e_info, 0, sizeof(struct aer_err_info));
 794
 795                 /* Init comprehensive error information */
 796                 if (i & PCI_ERR_ROOT_COR_RCV) {
 797                         e_info->id = ERR_COR_ID(e_src->id);
 798                         e_info->severity = AER_CORRECTABLE;
 799                 } else {
 800                         e_info->id = ERR_UNCOR_ID(e_src->id);
 801                         e_info->severity = ((e_src->status >> 6) & 1);
 802                 }
 803                 if (e_src->status &
 804                         (PCI_ERR_ROOT_MULTI_COR_RCV |
 805                          PCI_ERR_ROOT_MULTI_UNCOR_RCV))
 806                         e_info->multi_error_valid = 1;
 807
 808                 aer_print_port_info(p_device->port, e_info);
 809
 810                 find_source_device(p_device->port, e_info);
 811                 aer_process_err_devices(p_device, e_info);
 812         }
 813
 814         kfree(e_info);
 815 }
 816
 817 /**
 818  * aer_isr - consume errors detected by root port
 819  * @work: definition of this work item
 820  *
 821  * Invoked, as DPC, when root port records new detected error
 822  */
 823 void aer_isr(struct work_struct *work)
 824 {
 825         struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
 826         struct pcie_device *p_device = rpc->rpd;
 827         struct aer_err_source *e_src;
 828
 829         mutex_lock(&rpc->rpc_mutex);
 830         e_src = get_e_source(rpc);
 831         while (e_src) {
 832                 aer_isr_one_error(p_device, e_src);
 833                 e_src = get_e_source(rpc);
 834         }
 835         mutex_unlock(&rpc->rpc_mutex);
 836
 837         wake_up(&rpc->wait_release);
 838 }
 839
 840 /**
 841  * aer_delete_rootport - disable root port aer and delete service data
 842  * @rpc: pointer to a root port device being deleted
 843  *
 844  * Invoked when AER service unloaded on a specific Root Port
 845  */
 846 void aer_delete_rootport(struct aer_rpc *rpc)
 847 {
 848         /* Disable root port AER itself */
 849         disable_root_aer(rpc);
 850
 851         kfree(rpc);
 852 }
 853
 854 /**
 855  * aer_init - provide AER initialization
 856  * @dev: pointer to AER pcie device
 857  *
 858  * Invoked when AER service driver is loaded.
 859  */
 860 int aer_init(struct pcie_device *dev)
 861 {
 862         if (dev->port->aer_firmware_first) {
 863                 dev_printk(KERN_DEBUG, &dev->device,
 864                            "PCIe errors handled by platform firmware.\n");
 865                 goto out;
 866         }
 867
 868         if (aer_osc_setup(dev))
 869                 goto out;
 870
 871         return 0;
 872 out:
 873         if (forceload) {
 874                 dev_printk(KERN_DEBUG, &dev->device,
 875                            "aerdrv forceload requested.\n");
 876                 dev->port->aer_firmware_first = 0;
 877                 return 0;
 878         }
 879         return -ENXIO;
 880 }