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
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.
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
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.
23 * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com>
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>
31 #include <asm/eeh_event.h>
32 #include <asm/ppc-pci.h>
33 #include <asm/pci-bridge.h>
38 * eeh_pcid_name - Retrieve name of PCI device driver
41 * This routine is used to retrieve the name of PCI device driver
44 static inline const char *eeh_pcid_name(struct pci_dev *pdev)
46 if (pdev && pdev->dev.driver)
47 return pdev->dev.driver->name;
52 * eeh_pcid_get - Get the PCI device driver
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.
60 static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev)
62 if (!pdev || !pdev->driver)
65 if (!try_module_get(pdev->driver->driver.owner))
72 * eeh_pcid_put - Dereference on the PCI device driver
75 * The function is called to do dereference on the PCI device
76 * driver of the indicated PCI device.
78 static inline void eeh_pcid_put(struct pci_dev *pdev)
80 if (!pdev || !pdev->driver)
83 module_put(pdev->driver->driver.owner);
87 * eeh_disable_irq - Disable interrupt for the recovering device
90 * This routine must be called when reporting temporary or permanent
91 * error to the particular PCI device to disable interrupt of that
92 * device. If the device has enabled MSI or MSI-X interrupt, we needn't
93 * do real work because EEH should freeze DMA transfers for those PCI
94 * devices encountering EEH errors, which includes MSI or MSI-X.
96 static void eeh_disable_irq(struct pci_dev *dev)
98 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
100 /* Don't disable MSI and MSI-X interrupts. They are
101 * effectively disabled by the DMA Stopped state
102 * when an EEH error occurs.
104 if (dev->msi_enabled || dev->msix_enabled)
107 if (!irq_has_action(dev->irq))
110 edev->mode |= EEH_DEV_IRQ_DISABLED;
111 disable_irq_nosync(dev->irq);
115 * eeh_enable_irq - Enable interrupt for the recovering device
118 * This routine must be called to enable interrupt while failed
119 * device could be resumed.
121 static void eeh_enable_irq(struct pci_dev *dev)
123 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
125 if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
126 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
130 * This is just ass backwards. This maze has
131 * unbalanced irq_enable/disable calls. So instead of
132 * finding the root cause it works around the warning
133 * in the irq_enable code by conditionally calling
136 * That's just wrong.The warning in the core code is
137 * there to tell people to fix their assymetries in
138 * their own code, not by abusing the core information
141 * I so wish that the assymetry would be the other way
142 * round and a few more irq_disable calls render that
143 * shit unusable forever.
147 if (irqd_irq_disabled(irq_get_irq_data(dev->irq)))
148 enable_irq(dev->irq);
152 static bool eeh_dev_removed(struct eeh_dev *edev)
154 /* EEH device removed ? */
155 if (!edev || (edev->mode & EEH_DEV_REMOVED))
161 static void *eeh_dev_save_state(void *data, void *userdata)
163 struct eeh_dev *edev = data;
164 struct pci_dev *pdev;
170 * We cannot access the config space on some adapters.
171 * Otherwise, it will cause fenced PHB. We don't save
172 * the content in their config space and will restore
173 * from the initial config space saved when the EEH
176 if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED))
179 pdev = eeh_dev_to_pci_dev(edev);
183 pci_save_state(pdev);
188 * eeh_report_error - Report pci error to each device driver
190 * @userdata: return value
192 * Report an EEH error to each device driver, collect up and
193 * merge the device driver responses. Cumulative response
194 * passed back in "userdata".
196 static void *eeh_report_error(void *data, void *userdata)
198 struct eeh_dev *edev = (struct eeh_dev *)data;
199 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
200 enum pci_ers_result rc, *res = userdata;
201 struct pci_driver *driver;
203 if (!dev || eeh_dev_removed(edev))
205 dev->error_state = pci_channel_io_frozen;
207 driver = eeh_pcid_get(dev);
208 if (!driver) return NULL;
210 eeh_disable_irq(dev);
212 if (!driver->err_handler ||
213 !driver->err_handler->error_detected) {
218 rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
220 /* A driver that needs a reset trumps all others */
221 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
222 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
229 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
231 * @userdata: return value
233 * Tells each device driver that IO ports, MMIO and config space I/O
234 * are now enabled. Collects up and merges the device driver responses.
235 * Cumulative response passed back in "userdata".
237 static void *eeh_report_mmio_enabled(void *data, void *userdata)
239 struct eeh_dev *edev = (struct eeh_dev *)data;
240 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
241 enum pci_ers_result rc, *res = userdata;
242 struct pci_driver *driver;
244 if (!dev || eeh_dev_removed(edev))
247 driver = eeh_pcid_get(dev);
248 if (!driver) return NULL;
250 if (!driver->err_handler ||
251 !driver->err_handler->mmio_enabled ||
252 (edev->mode & EEH_DEV_NO_HANDLER)) {
257 rc = driver->err_handler->mmio_enabled(dev);
259 /* A driver that needs a reset trumps all others */
260 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
261 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
268 * eeh_report_reset - Tell device that slot has been reset
270 * @userdata: return value
272 * This routine must be called while EEH tries to reset particular
273 * PCI device so that the associated PCI device driver could take
274 * some actions, usually to save data the driver needs so that the
275 * driver can work again while the device is recovered.
277 static void *eeh_report_reset(void *data, void *userdata)
279 struct eeh_dev *edev = (struct eeh_dev *)data;
280 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
281 enum pci_ers_result rc, *res = userdata;
282 struct pci_driver *driver;
284 if (!dev || eeh_dev_removed(edev))
286 dev->error_state = pci_channel_io_normal;
288 driver = eeh_pcid_get(dev);
289 if (!driver) return NULL;
293 if (!driver->err_handler ||
294 !driver->err_handler->slot_reset ||
295 (edev->mode & EEH_DEV_NO_HANDLER)) {
300 rc = driver->err_handler->slot_reset(dev);
301 if ((*res == PCI_ERS_RESULT_NONE) ||
302 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
303 if (*res == PCI_ERS_RESULT_DISCONNECT &&
304 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
310 static void *eeh_dev_restore_state(void *data, void *userdata)
312 struct eeh_dev *edev = data;
313 struct pci_dev *pdev;
319 * The content in the config space isn't saved because
320 * the blocked config space on some adapters. We have
321 * to restore the initial saved config space when the
322 * EEH device is created.
324 if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED)) {
325 if (list_is_last(&edev->list, &edev->pe->edevs))
326 eeh_pe_restore_bars(edev->pe);
331 pdev = eeh_dev_to_pci_dev(edev);
335 pci_restore_state(pdev);
340 * eeh_report_resume - Tell device to resume normal operations
342 * @userdata: return value
344 * This routine must be called to notify the device driver that it
345 * could resume so that the device driver can do some initialization
346 * to make the recovered device work again.
348 static void *eeh_report_resume(void *data, void *userdata)
350 struct eeh_dev *edev = (struct eeh_dev *)data;
351 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
352 struct pci_driver *driver;
354 if (!dev || eeh_dev_removed(edev))
356 dev->error_state = pci_channel_io_normal;
358 driver = eeh_pcid_get(dev);
359 if (!driver) return NULL;
363 if (!driver->err_handler ||
364 !driver->err_handler->resume ||
365 (edev->mode & EEH_DEV_NO_HANDLER)) {
366 edev->mode &= ~EEH_DEV_NO_HANDLER;
371 driver->err_handler->resume(dev);
378 * eeh_report_failure - Tell device driver that device is dead.
380 * @userdata: return value
382 * This informs the device driver that the device is permanently
383 * dead, and that no further recovery attempts will be made on it.
385 static void *eeh_report_failure(void *data, void *userdata)
387 struct eeh_dev *edev = (struct eeh_dev *)data;
388 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
389 struct pci_driver *driver;
391 if (!dev || eeh_dev_removed(edev))
393 dev->error_state = pci_channel_io_perm_failure;
395 driver = eeh_pcid_get(dev);
396 if (!driver) return NULL;
398 eeh_disable_irq(dev);
400 if (!driver->err_handler ||
401 !driver->err_handler->error_detected) {
406 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
412 static void *eeh_rmv_device(void *data, void *userdata)
414 struct pci_driver *driver;
415 struct eeh_dev *edev = (struct eeh_dev *)data;
416 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
417 int *removed = (int *)userdata;
420 * Actually, we should remove the PCI bridges as well.
421 * However, that's lots of complexity to do that,
422 * particularly some of devices under the bridge might
423 * support EEH. So we just care about PCI devices for
426 if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
430 * We rely on count-based pcibios_release_device() to
431 * detach permanently offlined PEs. Unfortunately, that's
432 * not reliable enough. We might have the permanently
433 * offlined PEs attached, but we needn't take care of
434 * them and their child devices.
436 if (eeh_dev_removed(edev))
439 driver = eeh_pcid_get(dev);
442 if (driver->err_handler &&
443 driver->err_handler->error_detected &&
444 driver->err_handler->slot_reset)
448 /* Remove it from PCI subsystem */
449 pr_debug("EEH: Removing %s without EEH sensitive driver\n",
451 edev->bus = dev->bus;
452 edev->mode |= EEH_DEV_DISCONNECTED;
455 pci_lock_rescan_remove();
456 pci_stop_and_remove_bus_device(dev);
457 pci_unlock_rescan_remove();
462 static void *eeh_pe_detach_dev(void *data, void *userdata)
464 struct eeh_pe *pe = (struct eeh_pe *)data;
465 struct eeh_dev *edev, *tmp;
467 eeh_pe_for_each_dev(pe, edev, tmp) {
468 if (!(edev->mode & EEH_DEV_DISCONNECTED))
471 edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
472 eeh_rmv_from_parent_pe(edev);
479 * Explicitly clear PE's frozen state for PowerNV where
480 * we have frozen PE until BAR restore is completed. It's
481 * harmless to clear it for pSeries. To be consistent with
482 * PE reset (for 3 times), we try to clear the frozen state
483 * for 3 times as well.
485 static void *__eeh_clear_pe_frozen_state(void *data, void *flag)
487 struct eeh_pe *pe = (struct eeh_pe *)data;
488 bool clear_sw_state = *(bool *)flag;
491 for (i = 0; rc && i < 3; i++)
492 rc = eeh_unfreeze_pe(pe, clear_sw_state);
494 /* Stop immediately on any errors */
496 pr_warn("%s: Failure %d unfreezing PHB#%x-PE#%x\n",
497 __func__, rc, pe->phb->global_number, pe->addr);
504 static int eeh_clear_pe_frozen_state(struct eeh_pe *pe,
509 rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, &clear_sw_state);
511 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
513 return rc ? -EIO : 0;
516 int eeh_pe_reset_and_recover(struct eeh_pe *pe)
520 /* Bail if the PE is being recovered */
521 if (pe->state & EEH_PE_RECOVERING)
524 /* Put the PE into recovery mode */
525 eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
528 eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL);
531 ret = eeh_reset_pe(pe);
533 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
537 /* Unfreeze the PE */
538 ret = eeh_clear_pe_frozen_state(pe, true);
540 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
544 /* Notify completion of reset */
545 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
547 /* Restore device state */
548 eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL);
551 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
553 /* Clear recovery mode */
554 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
560 * eeh_reset_device - Perform actual reset of a pci slot
562 * @bus: PCI bus corresponding to the isolcated slot
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.
568 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
570 struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
571 struct timeval tstamp;
572 int cnt, rc, removed = 0;
574 /* pcibios will clear the counter; save the value */
575 cnt = pe->freeze_count;
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().
584 eeh_pe_state_mark(pe, EEH_PE_KEEP);
586 eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
587 pci_lock_rescan_remove();
588 pcibios_remove_pci_devices(bus);
589 pci_unlock_rescan_remove();
590 } else if (frozen_bus) {
591 eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
595 * Reset the pci controller. (Asserts RST#; resets config space).
596 * Reconfigure bridges and devices. Don't try to bring the system
597 * up if the reset failed for some reason.
599 * During the reset, it's very dangerous to have uncontrolled PCI
600 * config accesses. So we prefer to block them. However, controlled
601 * PCI config accesses initiated from EEH itself are allowed.
603 rc = eeh_reset_pe(pe);
607 pci_lock_rescan_remove();
610 eeh_ops->configure_bridge(pe);
611 eeh_pe_restore_bars(pe);
613 /* Clear frozen state */
614 rc = eeh_clear_pe_frozen_state(pe, false);
616 pci_unlock_rescan_remove();
620 /* Give the system 5 seconds to finish running the user-space
621 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
622 * this is a hack, but if we don't do this, and try to bring
623 * the device up before the scripts have taken it down,
624 * potentially weird things happen.
627 pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
631 * The EEH device is still connected with its parent
632 * PE. We should disconnect it so the binding can be
633 * rebuilt when adding PCI devices.
635 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
636 pcibios_add_pci_devices(bus);
637 } else if (frozen_bus && removed) {
638 pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
641 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
642 pcibios_add_pci_devices(frozen_bus);
644 eeh_pe_state_clear(pe, EEH_PE_KEEP);
647 pe->freeze_count = cnt;
649 pci_unlock_rescan_remove();
653 /* The longest amount of time to wait for a pci device
654 * to come back on line, in seconds.
656 #define MAX_WAIT_FOR_RECOVERY 300
658 static bool eeh_handle_normal_event(struct eeh_pe *pe)
660 struct pci_bus *frozen_bus;
662 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
664 frozen_bus = eeh_pe_bus_get(pe);
666 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
667 __func__, pe->phb->global_number, pe->addr);
671 eeh_pe_update_time_stamp(pe);
673 if (pe->freeze_count > eeh_max_freezes)
674 goto excess_failures;
675 pr_warn("EEH: This PCI device has failed %d times in the last hour\n",
678 /* Walk the various device drivers attached to this slot through
679 * a reset sequence, giving each an opportunity to do what it needs
680 * to accomplish the reset. Each child gets a report of the
681 * status ... if any child can't handle the reset, then the entire
682 * slot is dlpar removed and added.
684 * When the PHB is fenced, we have to issue a reset to recover from
685 * the error. Override the result if necessary to have partially
686 * hotplug for this case.
688 pr_info("EEH: Notify device drivers to shutdown\n");
689 eeh_pe_dev_traverse(pe, eeh_report_error, &result);
690 if ((pe->type & EEH_PE_PHB) &&
691 result != PCI_ERS_RESULT_NONE &&
692 result != PCI_ERS_RESULT_NEED_RESET)
693 result = PCI_ERS_RESULT_NEED_RESET;
695 /* Get the current PCI slot state. This can take a long time,
696 * sometimes over 300 seconds for certain systems.
698 rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
699 if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
700 pr_warn("EEH: Permanent failure\n");
704 /* Since rtas may enable MMIO when posting the error log,
705 * don't post the error log until after all dev drivers
706 * have been informed.
708 pr_info("EEH: Collect temporary log\n");
709 eeh_slot_error_detail(pe, EEH_LOG_TEMP);
711 /* If all device drivers were EEH-unaware, then shut
712 * down all of the device drivers, and hope they
713 * go down willingly, without panicing the system.
715 if (result == PCI_ERS_RESULT_NONE) {
716 pr_info("EEH: Reset with hotplug activity\n");
717 rc = eeh_reset_device(pe, frozen_bus);
719 pr_warn("%s: Unable to reset, err=%d\n",
725 /* If all devices reported they can proceed, then re-enable MMIO */
726 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
727 pr_info("EEH: Enable I/O for affected devices\n");
728 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
733 result = PCI_ERS_RESULT_NEED_RESET;
735 pr_info("EEH: Notify device drivers to resume I/O\n");
736 eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
740 /* If all devices reported they can proceed, then re-enable DMA */
741 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
742 pr_info("EEH: Enabled DMA for affected devices\n");
743 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
748 result = PCI_ERS_RESULT_NEED_RESET;
751 * We didn't do PE reset for the case. The PE
752 * is still in frozen state. Clear it before
755 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
756 result = PCI_ERS_RESULT_RECOVERED;
760 /* If any device has a hard failure, then shut off everything. */
761 if (result == PCI_ERS_RESULT_DISCONNECT) {
762 pr_warn("EEH: Device driver gave up\n");
766 /* If any device called out for a reset, then reset the slot */
767 if (result == PCI_ERS_RESULT_NEED_RESET) {
768 pr_info("EEH: Reset without hotplug activity\n");
769 rc = eeh_reset_device(pe, NULL);
771 pr_warn("%s: Cannot reset, err=%d\n",
776 pr_info("EEH: Notify device drivers "
777 "the completion of reset\n");
778 result = PCI_ERS_RESULT_NONE;
779 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
782 /* All devices should claim they have recovered by now. */
783 if ((result != PCI_ERS_RESULT_RECOVERED) &&
784 (result != PCI_ERS_RESULT_NONE)) {
785 pr_warn("EEH: Not recovered\n");
789 /* Tell all device drivers that they can resume operations */
790 pr_info("EEH: Notify device driver to resume\n");
791 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
797 * About 90% of all real-life EEH failures in the field
798 * are due to poorly seated PCI cards. Only 10% or so are
799 * due to actual, failed cards.
801 pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
802 "last hour and has been permanently disabled.\n"
803 "Please try reseating or replacing it.\n",
804 pe->phb->global_number, pe->addr,
809 pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
810 "Please try reseating or replacing it\n",
811 pe->phb->global_number, pe->addr);
814 eeh_slot_error_detail(pe, EEH_LOG_PERM);
816 /* Notify all devices that they're about to go down. */
817 eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
819 /* Mark the PE to be removed permanently */
820 eeh_pe_state_mark(pe, EEH_PE_REMOVED);
823 * Shut down the device drivers for good. We mark
824 * all removed devices correctly to avoid access
825 * the their PCI config any more.
828 eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
829 eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
831 pci_lock_rescan_remove();
832 pcibios_remove_pci_devices(frozen_bus);
833 pci_unlock_rescan_remove();
835 /* The passed PE should no longer be used */
841 static void eeh_handle_special_event(void)
843 struct eeh_pe *pe, *phb_pe;
845 struct pci_controller *hose;
851 rc = eeh_ops->next_error(&pe);
854 case EEH_NEXT_ERR_DEAD_IOC:
855 /* Mark all PHBs in dead state */
856 eeh_serialize_lock(&flags);
858 /* Purge all events */
859 eeh_remove_event(NULL, true);
861 list_for_each_entry(hose, &hose_list, list_node) {
862 phb_pe = eeh_phb_pe_get(hose);
863 if (!phb_pe) continue;
865 eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
868 eeh_serialize_unlock(flags);
871 case EEH_NEXT_ERR_FROZEN_PE:
872 case EEH_NEXT_ERR_FENCED_PHB:
873 case EEH_NEXT_ERR_DEAD_PHB:
874 /* Mark the PE in fenced state */
875 eeh_serialize_lock(&flags);
877 /* Purge all events of the PHB */
878 eeh_remove_event(pe, true);
880 if (rc == EEH_NEXT_ERR_DEAD_PHB)
881 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
883 eeh_pe_state_mark(pe,
884 EEH_PE_ISOLATED | EEH_PE_RECOVERING);
886 eeh_serialize_unlock(flags);
889 case EEH_NEXT_ERR_NONE:
892 pr_warn("%s: Invalid value %d from next_error()\n",
898 * For fenced PHB and frozen PE, it's handled as normal
899 * event. We have to remove the affected PHBs for dead
902 if (rc == EEH_NEXT_ERR_FROZEN_PE ||
903 rc == EEH_NEXT_ERR_FENCED_PHB) {
905 * eeh_handle_normal_event() can make the PE stale if it
906 * determines that the PE cannot possibly be recovered.
907 * Don't modify the PE state if that's the case.
909 if (eeh_handle_normal_event(pe))
912 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
914 pci_lock_rescan_remove();
915 list_for_each_entry(hose, &hose_list, list_node) {
916 phb_pe = eeh_phb_pe_get(hose);
918 !(phb_pe->state & EEH_PE_ISOLATED) ||
919 (phb_pe->state & EEH_PE_RECOVERING))
922 /* Notify all devices to be down */
923 eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
924 bus = eeh_pe_bus_get(phb_pe);
926 pr_err("%s: Cannot find PCI bus for "
929 pe->phb->global_number,
933 eeh_pe_dev_traverse(pe,
934 eeh_report_failure, NULL);
935 pcibios_remove_pci_devices(bus);
937 pci_unlock_rescan_remove();
941 * If we have detected dead IOC, we needn't proceed
942 * any more since all PHBs would have been removed
944 if (rc == EEH_NEXT_ERR_DEAD_IOC)
946 } while (rc != EEH_NEXT_ERR_NONE);
950 * eeh_handle_event - Reset a PCI device after hard lockup.
953 * While PHB detects address or data parity errors on particular PCI
954 * slot, the associated PE will be frozen. Besides, DMA's occurring
955 * to wild addresses (which usually happen due to bugs in device
956 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
957 * #PERR or other misc PCI-related errors also can trigger EEH errors.
959 * Recovery process consists of unplugging the device driver (which
960 * generated hotplug events to userspace), then issuing a PCI #RST to
961 * the device, then reconfiguring the PCI config space for all bridges
962 * & devices under this slot, and then finally restarting the device
963 * drivers (which cause a second set of hotplug events to go out to
966 void eeh_handle_event(struct eeh_pe *pe)
969 eeh_handle_normal_event(pe);
971 eeh_handle_special_event();