arch/powerpc/kernel/eeh_driver.c

   1 /*
   2  * PCI Error Recovery Driver for RPA-compliant PPC64 platform.
   3  * Copyright IBM Corp. 2004 2005
   4  * Copyright Linas Vepstas <linas@linas.org> 2004, 2005
   5  *
   6  * All rights reserved.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or (at
  11  * your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful, but
  14  * WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  16  * NON INFRINGEMENT.  See the GNU General Public License for more
  17  * details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  *
  23  * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com>
  24  */
  25 #include <linux/delay.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/irq.h>
  28 #include <linux/module.h>
  29 #include <linux/pci.h>
  30 #include <asm/eeh.h>
  31 #include <asm/eeh_event.h>
  32 #include <asm/ppc-pci.h>
  33 #include <asm/pci-bridge.h>
  34 #include <asm/prom.h>
  35 #include <asm/rtas.h>
  36
  37 /**
  38  * eeh_pcid_name - Retrieve name of PCI device driver
  39  * @pdev: PCI device
  40  *
  41  * This routine is used to retrieve the name of PCI device driver
  42  * if that's valid.
  43  */
  44 static inline const char *eeh_pcid_name(struct pci_dev *pdev)
  45 {
  46         if (pdev && pdev->dev.driver)
  47                 return pdev->dev.driver->name;
  48         return "";
  49 }
  50
  51 /**
  52  * eeh_pcid_get - Get the PCI device driver
  53  * @pdev: PCI device
  54  *
  55  * The function is used to retrieve the PCI device driver for
  56  * the indicated PCI device. Besides, we will increase the reference
  57  * of the PCI device driver to prevent that being unloaded on
  58  * the fly. Otherwise, kernel crash would be seen.
  59  */
  60 static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev)
  61 {
  62         if (!pdev || !pdev->driver)
  63                 return NULL;
  64
  65         if (!try_module_get(pdev->driver->driver.owner))
  66                 return NULL;
  67
  68         return pdev->driver;
  69 }
  70
  71 /**
  72  * eeh_pcid_put - Dereference on the PCI device driver
  73  * @pdev: PCI device
  74  *
  75  * The function is called to do dereference on the PCI device
  76  * driver of the indicated PCI device.
  77  */
  78 static inline void eeh_pcid_put(struct pci_dev *pdev)
  79 {
  80         if (!pdev || !pdev->driver)
  81                 return;
  82
  83         module_put(pdev->driver->driver.owner);
  84 }
  85
  86 #if 0
  87 static void print_device_node_tree(struct pci_dn *pdn, int dent)
  88 {
  89         int i;
  90         struct device_node *pc;
  91
  92         if (!pdn)
  93                 return;
  94         for (i = 0; i < dent; i++)
  95                 printk(" ");
  96         printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n",
  97                 pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr,
  98                 pdn->eeh_pe_config_addr, pdn->node->full_name);
  99         dent += 3;
 100         pc = pdn->node->child;
 101         while (pc) {
 102                 print_device_node_tree(PCI_DN(pc), dent);
 103                 pc = pc->sibling;
 104         }
 105 }
 106 #endif
 107
 108 /**
 109  * eeh_disable_irq - Disable interrupt for the recovering device
 110  * @dev: PCI device
 111  *
 112  * This routine must be called when reporting temporary or permanent
 113  * error to the particular PCI device to disable interrupt of that
 114  * device. If the device has enabled MSI or MSI-X interrupt, we needn't
 115  * do real work because EEH should freeze DMA transfers for those PCI
 116  * devices encountering EEH errors, which includes MSI or MSI-X.
 117  */
 118 static void eeh_disable_irq(struct pci_dev *dev)
 119 {
 120         struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
 121
 122         /* Don't disable MSI and MSI-X interrupts. They are
 123          * effectively disabled by the DMA Stopped state
 124          * when an EEH error occurs.
 125          */
 126         if (dev->msi_enabled || dev->msix_enabled)
 127                 return;
 128
 129         if (!irq_has_action(dev->irq))
 130                 return;
 131
 132         edev->mode |= EEH_DEV_IRQ_DISABLED;
 133         disable_irq_nosync(dev->irq);
 134 }
 135
 136 /**
 137  * eeh_enable_irq - Enable interrupt for the recovering device
 138  * @dev: PCI device
 139  *
 140  * This routine must be called to enable interrupt while failed
 141  * device could be resumed.
 142  */
 143 static void eeh_enable_irq(struct pci_dev *dev)
 144 {
 145         struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
 146
 147         if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
 148                 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
 149                 /*
 150                  * FIXME !!!!!
 151                  *
 152                  * This is just ass backwards. This maze has
 153                  * unbalanced irq_enable/disable calls. So instead of
 154                  * finding the root cause it works around the warning
 155                  * in the irq_enable code by conditionally calling
 156                  * into it.
 157                  *
 158                  * That's just wrong.The warning in the core code is
 159                  * there to tell people to fix their assymetries in
 160                  * their own code, not by abusing the core information
 161                  * to avoid it.
 162                  *
 163                  * I so wish that the assymetry would be the other way
 164                  * round and a few more irq_disable calls render that
 165                  * shit unusable forever.
 166                  *
 167                  *      tglx
 168                  */
 169                 if (irqd_irq_disabled(irq_get_irq_data(dev->irq)))
 170                         enable_irq(dev->irq);
 171         }
 172 }
 173
 174 static bool eeh_dev_removed(struct eeh_dev *edev)
 175 {
 176         /* EEH device removed ? */
 177         if (!edev || (edev->mode & EEH_DEV_REMOVED))
 178                 return true;
 179
 180         return false;
 181 }
 182
 183 static void *eeh_dev_save_state(void *data, void *userdata)
 184 {
 185         struct eeh_dev *edev = data;
 186         struct pci_dev *pdev;
 187
 188         if (!edev)
 189                 return NULL;
 190
 191         pdev = eeh_dev_to_pci_dev(edev);
 192         if (!pdev)
 193                 return NULL;
 194
 195         pci_save_state(pdev);
 196         return NULL;
 197 }
 198
 199 /**
 200  * eeh_report_error - Report pci error to each device driver
 201  * @data: eeh device
 202  * @userdata: return value
 203  *
 204  * Report an EEH error to each device driver, collect up and
 205  * merge the device driver responses. Cumulative response
 206  * passed back in "userdata".
 207  */
 208 static void *eeh_report_error(void *data, void *userdata)
 209 {
 210         struct eeh_dev *edev = (struct eeh_dev *)data;
 211         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
 212         enum pci_ers_result rc, *res = userdata;
 213         struct pci_driver *driver;
 214
 215         if (!dev || eeh_dev_removed(edev))
 216                 return NULL;
 217         dev->error_state = pci_channel_io_frozen;
 218
 219         driver = eeh_pcid_get(dev);
 220         if (!driver) return NULL;
 221
 222         eeh_disable_irq(dev);
 223
 224         if (!driver->err_handler ||
 225             !driver->err_handler->error_detected) {
 226                 eeh_pcid_put(dev);
 227                 return NULL;
 228         }
 229
 230         rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
 231
 232         /* A driver that needs a reset trumps all others */
 233         if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 234         if (*res == PCI_ERS_RESULT_NONE) *res = rc;
 235
 236         eeh_pcid_put(dev);
 237         return NULL;
 238 }
 239
 240 /**
 241  * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
 242  * @data: eeh device
 243  * @userdata: return value
 244  *
 245  * Tells each device driver that IO ports, MMIO and config space I/O
 246  * are now enabled. Collects up and merges the device driver responses.
 247  * Cumulative response passed back in "userdata".
 248  */
 249 static void *eeh_report_mmio_enabled(void *data, void *userdata)
 250 {
 251         struct eeh_dev *edev = (struct eeh_dev *)data;
 252         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
 253         enum pci_ers_result rc, *res = userdata;
 254         struct pci_driver *driver;
 255
 256         if (!dev || eeh_dev_removed(edev))
 257                 return NULL;
 258
 259         driver = eeh_pcid_get(dev);
 260         if (!driver) return NULL;
 261
 262         if (!driver->err_handler ||
 263             !driver->err_handler->mmio_enabled ||
 264             (edev->mode & EEH_DEV_NO_HANDLER)) {
 265                 eeh_pcid_put(dev);
 266                 return NULL;
 267         }
 268
 269         rc = driver->err_handler->mmio_enabled(dev);
 270
 271         /* A driver that needs a reset trumps all others */
 272         if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 273         if (*res == PCI_ERS_RESULT_NONE) *res = rc;
 274
 275         eeh_pcid_put(dev);
 276         return NULL;
 277 }
 278
 279 /**
 280  * eeh_report_reset - Tell device that slot has been reset
 281  * @data: eeh device
 282  * @userdata: return value
 283  *
 284  * This routine must be called while EEH tries to reset particular
 285  * PCI device so that the associated PCI device driver could take
 286  * some actions, usually to save data the driver needs so that the
 287  * driver can work again while the device is recovered.
 288  */
 289 static void *eeh_report_reset(void *data, void *userdata)
 290 {
 291         struct eeh_dev *edev = (struct eeh_dev *)data;
 292         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
 293         enum pci_ers_result rc, *res = userdata;
 294         struct pci_driver *driver;
 295
 296         if (!dev || eeh_dev_removed(edev))
 297                 return NULL;
 298         dev->error_state = pci_channel_io_normal;
 299
 300         driver = eeh_pcid_get(dev);
 301         if (!driver) return NULL;
 302
 303         eeh_enable_irq(dev);
 304
 305         if (!driver->err_handler ||
 306             !driver->err_handler->slot_reset ||
 307             (edev->mode & EEH_DEV_NO_HANDLER)) {
 308                 eeh_pcid_put(dev);
 309                 return NULL;
 310         }
 311
 312         rc = driver->err_handler->slot_reset(dev);
 313         if ((*res == PCI_ERS_RESULT_NONE) ||
 314             (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
 315         if (*res == PCI_ERS_RESULT_DISCONNECT &&
 316              rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 317
 318         eeh_pcid_put(dev);
 319         return NULL;
 320 }
 321
 322 static void *eeh_dev_restore_state(void *data, void *userdata)
 323 {
 324         struct eeh_dev *edev = data;
 325         struct pci_dev *pdev;
 326
 327         if (!edev)
 328                 return NULL;
 329
 330         pdev = eeh_dev_to_pci_dev(edev);
 331         if (!pdev)
 332                 return NULL;
 333
 334         pci_restore_state(pdev);
 335         return NULL;
 336 }
 337
 338 /**
 339  * eeh_report_resume - Tell device to resume normal operations
 340  * @data: eeh device
 341  * @userdata: return value
 342  *
 343  * This routine must be called to notify the device driver that it
 344  * could resume so that the device driver can do some initialization
 345  * to make the recovered device work again.
 346  */
 347 static void *eeh_report_resume(void *data, void *userdata)
 348 {
 349         struct eeh_dev *edev = (struct eeh_dev *)data;
 350         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
 351         struct pci_driver *driver;
 352
 353         if (!dev || eeh_dev_removed(edev))
 354                 return NULL;
 355         dev->error_state = pci_channel_io_normal;
 356
 357         driver = eeh_pcid_get(dev);
 358         if (!driver) return NULL;
 359
 360         eeh_enable_irq(dev);
 361
 362         if (!driver->err_handler ||
 363             !driver->err_handler->resume ||
 364             (edev->mode & EEH_DEV_NO_HANDLER)) {
 365                 edev->mode &= ~EEH_DEV_NO_HANDLER;
 366                 eeh_pcid_put(dev);
 367                 return NULL;
 368         }
 369
 370         driver->err_handler->resume(dev);
 371
 372         eeh_pcid_put(dev);
 373         return NULL;
 374 }
 375
 376 /**
 377  * eeh_report_failure - Tell device driver that device is dead.
 378  * @data: eeh device
 379  * @userdata: return value
 380  *
 381  * This informs the device driver that the device is permanently
 382  * dead, and that no further recovery attempts will be made on it.
 383  */
 384 static void *eeh_report_failure(void *data, void *userdata)
 385 {
 386         struct eeh_dev *edev = (struct eeh_dev *)data;
 387         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
 388         struct pci_driver *driver;
 389
 390         if (!dev || eeh_dev_removed(edev))
 391                 return NULL;
 392         dev->error_state = pci_channel_io_perm_failure;
 393
 394         driver = eeh_pcid_get(dev);
 395         if (!driver) return NULL;
 396
 397         eeh_disable_irq(dev);
 398
 399         if (!driver->err_handler ||
 400             !driver->err_handler->error_detected) {
 401                 eeh_pcid_put(dev);
 402                 return NULL;
 403         }
 404
 405         driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
 406
 407         eeh_pcid_put(dev);
 408         return NULL;
 409 }
 410
 411 static void *eeh_rmv_device(void *data, void *userdata)
 412 {
 413         struct pci_driver *driver;
 414         struct eeh_dev *edev = (struct eeh_dev *)data;
 415         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
 416         int *removed = (int *)userdata;
 417
 418         /*
 419          * Actually, we should remove the PCI bridges as well.
 420          * However, that's lots of complexity to do that,
 421          * particularly some of devices under the bridge might
 422          * support EEH. So we just care about PCI devices for
 423          * simplicity here.
 424          */
 425         if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
 426                 return NULL;
 427
 428         /*
 429          * We rely on count-based pcibios_release_device() to
 430          * detach permanently offlined PEs. Unfortunately, that's
 431          * not reliable enough. We might have the permanently
 432          * offlined PEs attached, but we needn't take care of
 433          * them and their child devices.
 434          */
 435         if (eeh_dev_removed(edev))
 436                 return NULL;
 437
 438         driver = eeh_pcid_get(dev);
 439         if (driver) {
 440                 eeh_pcid_put(dev);
 441                 if (driver->err_handler)
 442                         return NULL;
 443         }
 444
 445         /* Remove it from PCI subsystem */
 446         pr_debug("EEH: Removing %s without EEH sensitive driver\n",
 447                  pci_name(dev));
 448         edev->bus = dev->bus;
 449         edev->mode |= EEH_DEV_DISCONNECTED;
 450         (*removed)++;
 451
 452         pci_lock_rescan_remove();
 453         pci_stop_and_remove_bus_device(dev);
 454         pci_unlock_rescan_remove();
 455
 456         return NULL;
 457 }
 458
 459 static void *eeh_pe_detach_dev(void *data, void *userdata)
 460 {
 461         struct eeh_pe *pe = (struct eeh_pe *)data;
 462         struct eeh_dev *edev, *tmp;
 463
 464         eeh_pe_for_each_dev(pe, edev, tmp) {
 465                 if (!(edev->mode & EEH_DEV_DISCONNECTED))
 466                         continue;
 467
 468                 edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
 469                 eeh_rmv_from_parent_pe(edev);
 470         }
 471
 472         return NULL;
 473 }
 474
 475 /*
 476  * Explicitly clear PE's frozen state for PowerNV where
 477  * we have frozen PE until BAR restore is completed. It's
 478  * harmless to clear it for pSeries. To be consistent with
 479  * PE reset (for 3 times), we try to clear the frozen state
 480  * for 3 times as well.
 481  */
 482 static void *__eeh_clear_pe_frozen_state(void *data, void *flag)
 483 {
 484         struct eeh_pe *pe = (struct eeh_pe *)data;
 485         bool *clear_sw_state = flag;
 486         int i, rc = 1;
 487
 488         for (i = 0; rc && i < 3; i++)
 489                 rc = eeh_unfreeze_pe(pe, clear_sw_state);
 490
 491         /* Stop immediately on any errors */
 492         if (rc) {
 493                 pr_warn("%s: Failure %d unfreezing PHB#%x-PE#%x\n",
 494                         __func__, rc, pe->phb->global_number, pe->addr);
 495                 return (void *)pe;
 496         }
 497
 498         return NULL;
 499 }
 500
 501 static int eeh_clear_pe_frozen_state(struct eeh_pe *pe,
 502                                      bool clear_sw_state)
 503 {
 504         void *rc;
 505
 506         rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, &clear_sw_state);
 507         if (!rc)
 508                 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
 509
 510         return rc ? -EIO : 0;
 511 }
 512
 513 int eeh_pe_reset_and_recover(struct eeh_pe *pe)
 514 {
 515         int result, ret;
 516
 517         /* Bail if the PE is being recovered */
 518         if (pe->state & EEH_PE_RECOVERING)
 519                 return 0;
 520
 521         /* Put the PE into recovery mode */
 522         eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
 523
 524         /* Save states */
 525         eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL);
 526
 527         /* Report error */
 528         eeh_pe_dev_traverse(pe, eeh_report_error, &result);
 529
 530         /* Issue reset */
 531         ret = eeh_reset_pe(pe);
 532         if (ret) {
 533                 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
 534                 return ret;
 535         }
 536
 537         /* Unfreeze the PE */
 538         ret = eeh_clear_pe_frozen_state(pe, true);
 539         if (ret) {
 540                 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
 541                 return ret;
 542         }
 543
 544         /* Notify completion of reset */
 545         eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
 546
 547         /* Restore device state */
 548         eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL);
 549
 550         /* Resume */
 551         eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
 552
 553         /* Clear recovery mode */
 554         eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
 555
 556         return 0;
 557 }
 558
 559 /**
 560  * eeh_reset_device - Perform actual reset of a pci slot
 561  * @pe: EEH PE
 562  * @bus: PCI bus corresponding to the isolcated slot
 563  *
 564  * This routine must be called to do reset on the indicated PE.
 565  * During the reset, udev might be invoked because those affected
 566  * PCI devices will be removed and then added.
 567  */
 568 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
 569 {
 570         struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
 571         struct timeval tstamp;
 572         int cnt, rc, removed = 0;
 573
 574         /* pcibios will clear the counter; save the value */
 575         cnt = pe->freeze_count;
 576         tstamp = pe->tstamp;
 577
 578         /*
 579          * We don't remove the corresponding PE instances because
 580          * we need the information afterwords. The attached EEH
 581          * devices are expected to be attached soon when calling
 582          * into pcibios_add_pci_devices().
 583          */
 584         eeh_pe_state_mark(pe, EEH_PE_KEEP);
 585         if (bus) {
 586                 pci_lock_rescan_remove();
 587                 pcibios_remove_pci_devices(bus);
 588                 pci_unlock_rescan_remove();
 589         } else if (frozen_bus) {
 590                 eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
 591         }
 592
 593         /*
 594          * Reset the pci controller. (Asserts RST#; resets config space).
 595          * Reconfigure bridges and devices. Don't try to bring the system
 596          * up if the reset failed for some reason.
 597          *
 598          * During the reset, it's very dangerous to have uncontrolled PCI
 599          * config accesses. So we prefer to block them. However, controlled
 600          * PCI config accesses initiated from EEH itself are allowed.
 601          */
 602         rc = eeh_reset_pe(pe);
 603         if (rc)
 604                 return rc;
 605
 606         pci_lock_rescan_remove();
 607
 608         /* Restore PE */
 609         eeh_ops->configure_bridge(pe);
 610         eeh_pe_restore_bars(pe);
 611
 612         /* Clear frozen state */
 613         rc = eeh_clear_pe_frozen_state(pe, false);
 614         if (rc)
 615                 return rc;
 616
 617         /* Give the system 5 seconds to finish running the user-space
 618          * hotplug shutdown scripts, e.g. ifdown for ethernet.  Yes,
 619          * this is a hack, but if we don't do this, and try to bring
 620          * the device up before the scripts have taken it down,
 621          * potentially weird things happen.
 622          */
 623         if (bus) {
 624                 pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
 625                 ssleep(5);
 626
 627                 /*
 628                  * The EEH device is still connected with its parent
 629                  * PE. We should disconnect it so the binding can be
 630                  * rebuilt when adding PCI devices.
 631                  */
 632                 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
 633                 pcibios_add_pci_devices(bus);
 634         } else if (frozen_bus && removed) {
 635                 pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
 636                 ssleep(5);
 637
 638                 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
 639                 pcibios_add_pci_devices(frozen_bus);
 640         }
 641         eeh_pe_state_clear(pe, EEH_PE_KEEP);
 642
 643         pe->tstamp = tstamp;
 644         pe->freeze_count = cnt;
 645
 646         pci_unlock_rescan_remove();
 647         return 0;
 648 }
 649
 650 /* The longest amount of time to wait for a pci device
 651  * to come back on line, in seconds.
 652  */
 653 #define MAX_WAIT_FOR_RECOVERY 300
 654
 655 static void eeh_handle_normal_event(struct eeh_pe *pe)
 656 {
 657         struct pci_bus *frozen_bus;
 658         int rc = 0;
 659         enum pci_ers_result result = PCI_ERS_RESULT_NONE;
 660
 661         frozen_bus = eeh_pe_bus_get(pe);
 662         if (!frozen_bus) {
 663                 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
 664                         __func__, pe->phb->global_number, pe->addr);
 665                 return;
 666         }
 667
 668         eeh_pe_update_time_stamp(pe);
 669         pe->freeze_count++;
 670         if (pe->freeze_count > eeh_max_freezes)
 671                 goto excess_failures;
 672         pr_warn("EEH: This PCI device has failed %d times in the last hour\n",
 673                 pe->freeze_count);
 674
 675         /* Walk the various device drivers attached to this slot through
 676          * a reset sequence, giving each an opportunity to do what it needs
 677          * to accomplish the reset.  Each child gets a report of the
 678          * status ... if any child can't handle the reset, then the entire
 679          * slot is dlpar removed and added.
 680          */
 681         pr_info("EEH: Notify device drivers to shutdown\n");
 682         eeh_pe_dev_traverse(pe, eeh_report_error, &result);
 683
 684         /* Get the current PCI slot state. This can take a long time,
 685          * sometimes over 3 seconds for certain systems.
 686          */
 687         rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
 688         if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
 689                 pr_warn("EEH: Permanent failure\n");
 690                 goto hard_fail;
 691         }
 692
 693         /* Since rtas may enable MMIO when posting the error log,
 694          * don't post the error log until after all dev drivers
 695          * have been informed.
 696          */
 697         pr_info("EEH: Collect temporary log\n");
 698         eeh_slot_error_detail(pe, EEH_LOG_TEMP);
 699
 700         /* If all device drivers were EEH-unaware, then shut
 701          * down all of the device drivers, and hope they
 702          * go down willingly, without panicing the system.
 703          */
 704         if (result == PCI_ERS_RESULT_NONE) {
 705                 pr_info("EEH: Reset with hotplug activity\n");
 706                 rc = eeh_reset_device(pe, frozen_bus);
 707                 if (rc) {
 708                         pr_warn("%s: Unable to reset, err=%d\n",
 709                                 __func__, rc);
 710                         goto hard_fail;
 711                 }
 712         }
 713
 714         /* If all devices reported they can proceed, then re-enable MMIO */
 715         if (result == PCI_ERS_RESULT_CAN_RECOVER) {
 716                 pr_info("EEH: Enable I/O for affected devices\n");
 717                 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
 718
 719                 if (rc < 0)
 720                         goto hard_fail;
 721                 if (rc) {
 722                         result = PCI_ERS_RESULT_NEED_RESET;
 723                 } else {
 724                         pr_info("EEH: Notify device drivers to resume I/O\n");
 725                         eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
 726                 }
 727         }
 728
 729         /* If all devices reported they can proceed, then re-enable DMA */
 730         if (result == PCI_ERS_RESULT_CAN_RECOVER) {
 731                 pr_info("EEH: Enabled DMA for affected devices\n");
 732                 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
 733
 734                 if (rc < 0)
 735                         goto hard_fail;
 736                 if (rc) {
 737                         result = PCI_ERS_RESULT_NEED_RESET;
 738                 } else {
 739                         /*
 740                          * We didn't do PE reset for the case. The PE
 741                          * is still in frozen state. Clear it before
 742                          * resuming the PE.
 743                          */
 744                         eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
 745                         result = PCI_ERS_RESULT_RECOVERED;
 746                 }
 747         }
 748
 749         /* If any device has a hard failure, then shut off everything. */
 750         if (result == PCI_ERS_RESULT_DISCONNECT) {
 751                 pr_warn("EEH: Device driver gave up\n");
 752                 goto hard_fail;
 753         }
 754
 755         /* If any device called out for a reset, then reset the slot */
 756         if (result == PCI_ERS_RESULT_NEED_RESET) {
 757                 pr_info("EEH: Reset without hotplug activity\n");
 758                 rc = eeh_reset_device(pe, NULL);
 759                 if (rc) {
 760                         pr_warn("%s: Cannot reset, err=%d\n",
 761                                 __func__, rc);
 762                         goto hard_fail;
 763                 }
 764
 765                 pr_info("EEH: Notify device drivers "
 766                         "the completion of reset\n");
 767                 result = PCI_ERS_RESULT_NONE;
 768                 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
 769         }
 770
 771         /* All devices should claim they have recovered by now. */
 772         if ((result != PCI_ERS_RESULT_RECOVERED) &&
 773             (result != PCI_ERS_RESULT_NONE)) {
 774                 pr_warn("EEH: Not recovered\n");
 775                 goto hard_fail;
 776         }
 777
 778         /* Tell all device drivers that they can resume operations */
 779         pr_info("EEH: Notify device driver to resume\n");
 780         eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
 781
 782         return;
 783
 784 excess_failures:
 785         /*
 786          * About 90% of all real-life EEH failures in the field
 787          * are due to poorly seated PCI cards. Only 10% or so are
 788          * due to actual, failed cards.
 789          */
 790         pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
 791                "last hour and has been permanently disabled.\n"
 792                "Please try reseating or replacing it.\n",
 793                 pe->phb->global_number, pe->addr,
 794                 pe->freeze_count);
 795         goto perm_error;
 796
 797 hard_fail:
 798         pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
 799                "Please try reseating or replacing it\n",
 800                 pe->phb->global_number, pe->addr);
 801
 802 perm_error:
 803         eeh_slot_error_detail(pe, EEH_LOG_PERM);
 804
 805         /* Notify all devices that they're about to go down. */
 806         eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
 807
 808         /* Mark the PE to be removed permanently */
 809         eeh_pe_state_mark(pe, EEH_PE_REMOVED);
 810
 811         /*
 812          * Shut down the device drivers for good. We mark
 813          * all removed devices correctly to avoid access
 814          * the their PCI config any more.
 815          */
 816         if (frozen_bus) {
 817                 eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
 818
 819                 pci_lock_rescan_remove();
 820                 pcibios_remove_pci_devices(frozen_bus);
 821                 pci_unlock_rescan_remove();
 822         }
 823 }
 824
 825 static void eeh_handle_special_event(void)
 826 {
 827         struct eeh_pe *pe, *phb_pe;
 828         struct pci_bus *bus;
 829         struct pci_controller *hose;
 830         unsigned long flags;
 831         int rc;
 832
 833
 834         do {
 835                 rc = eeh_ops->next_error(&pe);
 836
 837                 switch (rc) {
 838                 case EEH_NEXT_ERR_DEAD_IOC:
 839                         /* Mark all PHBs in dead state */
 840                         eeh_serialize_lock(&flags);
 841
 842                         /* Purge all events */
 843                         eeh_remove_event(NULL, true);
 844
 845                         list_for_each_entry(hose, &hose_list, list_node) {
 846                                 phb_pe = eeh_phb_pe_get(hose);
 847                                 if (!phb_pe) continue;
 848
 849                                 eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
 850                         }
 851
 852                         eeh_serialize_unlock(flags);
 853
 854                         break;
 855                 case EEH_NEXT_ERR_FROZEN_PE:
 856                 case EEH_NEXT_ERR_FENCED_PHB:
 857                 case EEH_NEXT_ERR_DEAD_PHB:
 858                         /* Mark the PE in fenced state */
 859                         eeh_serialize_lock(&flags);
 860
 861                         /* Purge all events of the PHB */
 862                         eeh_remove_event(pe, true);
 863
 864                         if (rc == EEH_NEXT_ERR_DEAD_PHB)
 865                                 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
 866                         else
 867                                 eeh_pe_state_mark(pe,
 868                                         EEH_PE_ISOLATED | EEH_PE_RECOVERING);
 869
 870                         eeh_serialize_unlock(flags);
 871
 872                         break;
 873                 case EEH_NEXT_ERR_NONE:
 874                         return;
 875                 default:
 876                         pr_warn("%s: Invalid value %d from next_error()\n",
 877                                 __func__, rc);
 878                         return;
 879                 }
 880
 881                 /*
 882                  * For fenced PHB and frozen PE, it's handled as normal
 883                  * event. We have to remove the affected PHBs for dead
 884                  * PHB and IOC
 885                  */
 886                 if (rc == EEH_NEXT_ERR_FROZEN_PE ||
 887                     rc == EEH_NEXT_ERR_FENCED_PHB) {
 888                         eeh_handle_normal_event(pe);
 889                         eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
 890                 } else {
 891                         pci_lock_rescan_remove();
 892                         list_for_each_entry(hose, &hose_list, list_node) {
 893                                 phb_pe = eeh_phb_pe_get(hose);
 894                                 if (!phb_pe ||
 895                                     !(phb_pe->state & EEH_PE_ISOLATED) ||
 896                                     (phb_pe->state & EEH_PE_RECOVERING))
 897                                         continue;
 898
 899                                 /* Notify all devices to be down */
 900                                 bus = eeh_pe_bus_get(phb_pe);
 901                                 eeh_pe_dev_traverse(pe,
 902                                         eeh_report_failure, NULL);
 903                                 pcibios_remove_pci_devices(bus);
 904                         }
 905                         pci_unlock_rescan_remove();
 906                 }
 907
 908                 /*
 909                  * If we have detected dead IOC, we needn't proceed
 910                  * any more since all PHBs would have been removed
 911                  */
 912                 if (rc == EEH_NEXT_ERR_DEAD_IOC)
 913                         break;
 914         } while (rc != EEH_NEXT_ERR_NONE);
 915 }
 916
 917 /**
 918  * eeh_handle_event - Reset a PCI device after hard lockup.
 919  * @pe: EEH PE
 920  *
 921  * While PHB detects address or data parity errors on particular PCI
 922  * slot, the associated PE will be frozen. Besides, DMA's occurring
 923  * to wild addresses (which usually happen due to bugs in device
 924  * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
 925  * #PERR or other misc PCI-related errors also can trigger EEH errors.
 926  *
 927  * Recovery process consists of unplugging the device driver (which
 928  * generated hotplug events to userspace), then issuing a PCI #RST to
 929  * the device, then reconfiguring the PCI config space for all bridges
 930  * & devices under this slot, and then finally restarting the device
 931  * drivers (which cause a second set of hotplug events to go out to
 932  * userspace).
 933  */
 934 void eeh_handle_event(struct eeh_pe *pe)
 935 {
 936         if (pe)
 937                 eeh_handle_normal_event(pe);
 938         else
 939                 eeh_handle_special_event();
 940 }