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 struct list_head edev_list;
43 * eeh_pcid_name - Retrieve name of PCI device driver
46 * This routine is used to retrieve the name of PCI device driver
49 static inline const char *eeh_pcid_name(struct pci_dev *pdev)
51 if (pdev && pdev->dev.driver)
52 return pdev->dev.driver->name;
57 * eeh_pcid_get - Get the PCI device driver
60 * The function is used to retrieve the PCI device driver for
61 * the indicated PCI device. Besides, we will increase the reference
62 * of the PCI device driver to prevent that being unloaded on
63 * the fly. Otherwise, kernel crash would be seen.
65 static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev)
67 if (!pdev || !pdev->driver)
70 if (!try_module_get(pdev->driver->driver.owner))
77 * eeh_pcid_put - Dereference on the PCI device driver
80 * The function is called to do dereference on the PCI device
81 * driver of the indicated PCI device.
83 static inline void eeh_pcid_put(struct pci_dev *pdev)
85 if (!pdev || !pdev->driver)
88 module_put(pdev->driver->driver.owner);
92 * eeh_disable_irq - Disable interrupt for the recovering device
95 * This routine must be called when reporting temporary or permanent
96 * error to the particular PCI device to disable interrupt of that
97 * device. If the device has enabled MSI or MSI-X interrupt, we needn't
98 * do real work because EEH should freeze DMA transfers for those PCI
99 * devices encountering EEH errors, which includes MSI or MSI-X.
101 static void eeh_disable_irq(struct pci_dev *dev)
103 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
105 /* Don't disable MSI and MSI-X interrupts. They are
106 * effectively disabled by the DMA Stopped state
107 * when an EEH error occurs.
109 if (dev->msi_enabled || dev->msix_enabled)
112 if (!irq_has_action(dev->irq))
115 edev->mode |= EEH_DEV_IRQ_DISABLED;
116 disable_irq_nosync(dev->irq);
120 * eeh_enable_irq - Enable interrupt for the recovering device
123 * This routine must be called to enable interrupt while failed
124 * device could be resumed.
126 static void eeh_enable_irq(struct pci_dev *dev)
128 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
130 if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
131 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
135 * This is just ass backwards. This maze has
136 * unbalanced irq_enable/disable calls. So instead of
137 * finding the root cause it works around the warning
138 * in the irq_enable code by conditionally calling
141 * That's just wrong.The warning in the core code is
142 * there to tell people to fix their asymmetries in
143 * their own code, not by abusing the core information
146 * I so wish that the assymetry would be the other way
147 * round and a few more irq_disable calls render that
148 * shit unusable forever.
152 if (irqd_irq_disabled(irq_get_irq_data(dev->irq)))
153 enable_irq(dev->irq);
157 static bool eeh_dev_removed(struct eeh_dev *edev)
159 /* EEH device removed ? */
160 if (!edev || (edev->mode & EEH_DEV_REMOVED))
166 static void *eeh_dev_save_state(void *data, void *userdata)
168 struct eeh_dev *edev = data;
169 struct pci_dev *pdev;
175 * We cannot access the config space on some adapters.
176 * Otherwise, it will cause fenced PHB. We don't save
177 * the content in their config space and will restore
178 * from the initial config space saved when the EEH
181 if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED))
184 pdev = eeh_dev_to_pci_dev(edev);
188 pci_save_state(pdev);
193 * eeh_report_error - Report pci error to each device driver
195 * @userdata: return value
197 * Report an EEH error to each device driver, collect up and
198 * merge the device driver responses. Cumulative response
199 * passed back in "userdata".
201 static void *eeh_report_error(void *data, void *userdata)
203 struct eeh_dev *edev = (struct eeh_dev *)data;
204 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
205 enum pci_ers_result rc, *res = userdata;
206 struct pci_driver *driver;
208 if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe))
211 device_lock(&dev->dev);
212 dev->error_state = pci_channel_io_frozen;
214 driver = eeh_pcid_get(dev);
215 if (!driver) goto out_no_dev;
217 eeh_disable_irq(dev);
219 if (!driver->err_handler ||
220 !driver->err_handler->error_detected)
223 rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
225 /* A driver that needs a reset trumps all others */
226 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
227 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
229 edev->in_error = true;
233 device_unlock(&dev->dev);
238 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
240 * @userdata: return value
242 * Tells each device driver that IO ports, MMIO and config space I/O
243 * are now enabled. Collects up and merges the device driver responses.
244 * Cumulative response passed back in "userdata".
246 static void *eeh_report_mmio_enabled(void *data, void *userdata)
248 struct eeh_dev *edev = (struct eeh_dev *)data;
249 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
250 enum pci_ers_result rc, *res = userdata;
251 struct pci_driver *driver;
253 if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe))
256 device_lock(&dev->dev);
257 driver = eeh_pcid_get(dev);
258 if (!driver) goto out_no_dev;
260 if (!driver->err_handler ||
261 !driver->err_handler->mmio_enabled ||
262 (edev->mode & EEH_DEV_NO_HANDLER))
265 rc = driver->err_handler->mmio_enabled(dev);
267 /* A driver that needs a reset trumps all others */
268 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
269 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
274 device_unlock(&dev->dev);
279 * eeh_report_reset - Tell device that slot has been reset
281 * @userdata: return value
283 * This routine must be called while EEH tries to reset particular
284 * PCI device so that the associated PCI device driver could take
285 * some actions, usually to save data the driver needs so that the
286 * driver can work again while the device is recovered.
288 static void *eeh_report_reset(void *data, void *userdata)
290 struct eeh_dev *edev = (struct eeh_dev *)data;
291 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
292 enum pci_ers_result rc, *res = userdata;
293 struct pci_driver *driver;
295 if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe))
298 device_lock(&dev->dev);
299 dev->error_state = pci_channel_io_normal;
301 driver = eeh_pcid_get(dev);
302 if (!driver) goto out_no_dev;
306 if (!driver->err_handler ||
307 !driver->err_handler->slot_reset ||
308 (edev->mode & EEH_DEV_NO_HANDLER) ||
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;
321 device_unlock(&dev->dev);
325 static void *eeh_dev_restore_state(void *data, void *userdata)
327 struct eeh_dev *edev = data;
328 struct pci_dev *pdev;
334 * The content in the config space isn't saved because
335 * the blocked config space on some adapters. We have
336 * to restore the initial saved config space when the
337 * EEH device is created.
339 if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED)) {
340 if (list_is_last(&edev->list, &edev->pe->edevs))
341 eeh_pe_restore_bars(edev->pe);
346 pdev = eeh_dev_to_pci_dev(edev);
350 pci_restore_state(pdev);
355 * eeh_report_resume - Tell device to resume normal operations
357 * @userdata: return value
359 * This routine must be called to notify the device driver that it
360 * could resume so that the device driver can do some initialization
361 * to make the recovered device work again.
363 static void *eeh_report_resume(void *data, void *userdata)
365 struct eeh_dev *edev = (struct eeh_dev *)data;
366 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
368 struct pci_driver *driver;
370 if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe))
373 device_lock(&dev->dev);
374 dev->error_state = pci_channel_io_normal;
376 driver = eeh_pcid_get(dev);
377 if (!driver) goto out_no_dev;
379 was_in_error = edev->in_error;
380 edev->in_error = false;
383 if (!driver->err_handler ||
384 !driver->err_handler->resume ||
385 (edev->mode & EEH_DEV_NO_HANDLER) || !was_in_error) {
386 edev->mode &= ~EEH_DEV_NO_HANDLER;
390 driver->err_handler->resume(dev);
395 device_unlock(&dev->dev);
400 * eeh_report_failure - Tell device driver that device is dead.
402 * @userdata: return value
404 * This informs the device driver that the device is permanently
405 * dead, and that no further recovery attempts will be made on it.
407 static void *eeh_report_failure(void *data, void *userdata)
409 struct eeh_dev *edev = (struct eeh_dev *)data;
410 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
411 struct pci_driver *driver;
413 if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe))
416 device_lock(&dev->dev);
417 dev->error_state = pci_channel_io_perm_failure;
419 driver = eeh_pcid_get(dev);
420 if (!driver) goto out_no_dev;
422 eeh_disable_irq(dev);
424 if (!driver->err_handler ||
425 !driver->err_handler->error_detected)
428 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
433 device_unlock(&dev->dev);
437 static void *eeh_add_virt_device(void *data, void *userdata)
439 struct pci_driver *driver;
440 struct eeh_dev *edev = (struct eeh_dev *)data;
441 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
442 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
444 if (!(edev->physfn)) {
445 pr_warn("%s: EEH dev %04x:%02x:%02x.%01x not for VF\n",
446 __func__, edev->phb->global_number, pdn->busno,
447 PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
451 driver = eeh_pcid_get(dev);
453 if (driver->err_handler) {
460 #ifdef CONFIG_PPC_POWERNV
461 pci_iov_add_virtfn(edev->physfn, pdn->vf_index, 0);
466 static void *eeh_rmv_device(void *data, void *userdata)
468 struct pci_driver *driver;
469 struct eeh_dev *edev = (struct eeh_dev *)data;
470 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
471 struct eeh_rmv_data *rmv_data = (struct eeh_rmv_data *)userdata;
472 int *removed = rmv_data ? &rmv_data->removed : NULL;
475 * Actually, we should remove the PCI bridges as well.
476 * However, that's lots of complexity to do that,
477 * particularly some of devices under the bridge might
478 * support EEH. So we just care about PCI devices for
481 if (!dev || (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE))
485 * We rely on count-based pcibios_release_device() to
486 * detach permanently offlined PEs. Unfortunately, that's
487 * not reliable enough. We might have the permanently
488 * offlined PEs attached, but we needn't take care of
489 * them and their child devices.
491 if (eeh_dev_removed(edev))
495 if (eeh_pe_passed(edev->pe))
497 driver = eeh_pcid_get(dev);
499 if (driver->err_handler &&
500 driver->err_handler->error_detected &&
501 driver->err_handler->slot_reset) {
509 /* Remove it from PCI subsystem */
510 pr_debug("EEH: Removing %s without EEH sensitive driver\n",
512 edev->bus = dev->bus;
513 edev->mode |= EEH_DEV_DISCONNECTED;
518 #ifdef CONFIG_PPC_POWERNV
519 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
521 pci_iov_remove_virtfn(edev->physfn, pdn->vf_index, 0);
525 list_add(&edev->rmv_list, &rmv_data->edev_list);
527 pci_lock_rescan_remove();
528 pci_stop_and_remove_bus_device(dev);
529 pci_unlock_rescan_remove();
535 static void *eeh_pe_detach_dev(void *data, void *userdata)
537 struct eeh_pe *pe = (struct eeh_pe *)data;
538 struct eeh_dev *edev, *tmp;
540 eeh_pe_for_each_dev(pe, edev, tmp) {
541 if (!(edev->mode & EEH_DEV_DISCONNECTED))
544 edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
545 eeh_rmv_from_parent_pe(edev);
552 * Explicitly clear PE's frozen state for PowerNV where
553 * we have frozen PE until BAR restore is completed. It's
554 * harmless to clear it for pSeries. To be consistent with
555 * PE reset (for 3 times), we try to clear the frozen state
556 * for 3 times as well.
558 static void *__eeh_clear_pe_frozen_state(void *data, void *flag)
560 struct eeh_pe *pe = (struct eeh_pe *)data;
561 bool clear_sw_state = *(bool *)flag;
564 for (i = 0; rc && i < 3; i++)
565 rc = eeh_unfreeze_pe(pe, clear_sw_state);
567 /* Stop immediately on any errors */
569 pr_warn("%s: Failure %d unfreezing PHB#%x-PE#%x\n",
570 __func__, rc, pe->phb->global_number, pe->addr);
577 static int eeh_clear_pe_frozen_state(struct eeh_pe *pe,
582 rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, &clear_sw_state);
584 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
586 return rc ? -EIO : 0;
589 int eeh_pe_reset_and_recover(struct eeh_pe *pe)
593 /* Bail if the PE is being recovered */
594 if (pe->state & EEH_PE_RECOVERING)
597 /* Put the PE into recovery mode */
598 eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
601 eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL);
604 ret = eeh_reset_pe(pe);
606 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
610 /* Unfreeze the PE */
611 ret = eeh_clear_pe_frozen_state(pe, true);
613 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
617 /* Restore device state */
618 eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL);
620 /* Clear recovery mode */
621 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
627 * eeh_reset_device - Perform actual reset of a pci slot
629 * @bus: PCI bus corresponding to the isolcated slot
631 * This routine must be called to do reset on the indicated PE.
632 * During the reset, udev might be invoked because those affected
633 * PCI devices will be removed and then added.
635 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus,
636 struct eeh_rmv_data *rmv_data)
638 struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
639 struct timeval tstamp;
641 struct eeh_dev *edev;
643 /* pcibios will clear the counter; save the value */
644 cnt = pe->freeze_count;
648 * We don't remove the corresponding PE instances because
649 * we need the information afterwords. The attached EEH
650 * devices are expected to be attached soon when calling
651 * into pci_hp_add_devices().
653 eeh_pe_state_mark(pe, EEH_PE_KEEP);
655 if (pe->type & EEH_PE_VF) {
656 eeh_pe_dev_traverse(pe, eeh_rmv_device, NULL);
658 pci_lock_rescan_remove();
659 pci_hp_remove_devices(bus);
660 pci_unlock_rescan_remove();
662 } else if (frozen_bus) {
663 eeh_pe_dev_traverse(pe, eeh_rmv_device, rmv_data);
667 * Reset the pci controller. (Asserts RST#; resets config space).
668 * Reconfigure bridges and devices. Don't try to bring the system
669 * up if the reset failed for some reason.
671 * During the reset, it's very dangerous to have uncontrolled PCI
672 * config accesses. So we prefer to block them. However, controlled
673 * PCI config accesses initiated from EEH itself are allowed.
675 rc = eeh_reset_pe(pe);
679 pci_lock_rescan_remove();
682 eeh_ops->configure_bridge(pe);
683 eeh_pe_restore_bars(pe);
685 /* Clear frozen state */
686 rc = eeh_clear_pe_frozen_state(pe, false);
688 pci_unlock_rescan_remove();
692 /* Give the system 5 seconds to finish running the user-space
693 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
694 * this is a hack, but if we don't do this, and try to bring
695 * the device up before the scripts have taken it down,
696 * potentially weird things happen.
699 pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
703 * The EEH device is still connected with its parent
704 * PE. We should disconnect it so the binding can be
705 * rebuilt when adding PCI devices.
707 edev = list_first_entry(&pe->edevs, struct eeh_dev, list);
708 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
709 if (pe->type & EEH_PE_VF) {
710 eeh_add_virt_device(edev, NULL);
712 eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
713 pci_hp_add_devices(bus);
715 } else if (frozen_bus && rmv_data->removed) {
716 pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
719 edev = list_first_entry(&pe->edevs, struct eeh_dev, list);
720 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
721 if (pe->type & EEH_PE_VF)
722 eeh_add_virt_device(edev, NULL);
724 pci_hp_add_devices(frozen_bus);
726 eeh_pe_state_clear(pe, EEH_PE_KEEP);
729 pe->freeze_count = cnt;
731 pci_unlock_rescan_remove();
735 /* The longest amount of time to wait for a pci device
736 * to come back on line, in seconds.
738 #define MAX_WAIT_FOR_RECOVERY 300
740 static bool eeh_handle_normal_event(struct eeh_pe *pe)
742 struct pci_bus *frozen_bus;
743 struct eeh_dev *edev, *tmp;
745 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
746 struct eeh_rmv_data rmv_data = {LIST_HEAD_INIT(rmv_data.edev_list), 0};
748 frozen_bus = eeh_pe_bus_get(pe);
750 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
751 __func__, pe->phb->global_number, pe->addr);
755 eeh_pe_update_time_stamp(pe);
757 if (pe->freeze_count > eeh_max_freezes)
758 goto excess_failures;
759 pr_warn("EEH: This PCI device has failed %d times in the last hour\n",
762 /* Walk the various device drivers attached to this slot through
763 * a reset sequence, giving each an opportunity to do what it needs
764 * to accomplish the reset. Each child gets a report of the
765 * status ... if any child can't handle the reset, then the entire
766 * slot is dlpar removed and added.
768 * When the PHB is fenced, we have to issue a reset to recover from
769 * the error. Override the result if necessary to have partially
770 * hotplug for this case.
772 pr_info("EEH: Notify device drivers to shutdown\n");
773 eeh_pe_dev_traverse(pe, eeh_report_error, &result);
774 if ((pe->type & EEH_PE_PHB) &&
775 result != PCI_ERS_RESULT_NONE &&
776 result != PCI_ERS_RESULT_NEED_RESET)
777 result = PCI_ERS_RESULT_NEED_RESET;
779 /* Get the current PCI slot state. This can take a long time,
780 * sometimes over 300 seconds for certain systems.
782 rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
783 if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
784 pr_warn("EEH: Permanent failure\n");
788 /* Since rtas may enable MMIO when posting the error log,
789 * don't post the error log until after all dev drivers
790 * have been informed.
792 pr_info("EEH: Collect temporary log\n");
793 eeh_slot_error_detail(pe, EEH_LOG_TEMP);
795 /* If all device drivers were EEH-unaware, then shut
796 * down all of the device drivers, and hope they
797 * go down willingly, without panicing the system.
799 if (result == PCI_ERS_RESULT_NONE) {
800 pr_info("EEH: Reset with hotplug activity\n");
801 rc = eeh_reset_device(pe, frozen_bus, NULL);
803 pr_warn("%s: Unable to reset, err=%d\n",
809 /* If all devices reported they can proceed, then re-enable MMIO */
810 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
811 pr_info("EEH: Enable I/O for affected devices\n");
812 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
817 result = PCI_ERS_RESULT_NEED_RESET;
819 pr_info("EEH: Notify device drivers to resume I/O\n");
820 eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
824 /* If all devices reported they can proceed, then re-enable DMA */
825 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
826 pr_info("EEH: Enabled DMA for affected devices\n");
827 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
832 result = PCI_ERS_RESULT_NEED_RESET;
835 * We didn't do PE reset for the case. The PE
836 * is still in frozen state. Clear it before
839 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
840 result = PCI_ERS_RESULT_RECOVERED;
844 /* If any device has a hard failure, then shut off everything. */
845 if (result == PCI_ERS_RESULT_DISCONNECT) {
846 pr_warn("EEH: Device driver gave up\n");
850 /* If any device called out for a reset, then reset the slot */
851 if (result == PCI_ERS_RESULT_NEED_RESET) {
852 pr_info("EEH: Reset without hotplug activity\n");
853 rc = eeh_reset_device(pe, NULL, &rmv_data);
855 pr_warn("%s: Cannot reset, err=%d\n",
860 pr_info("EEH: Notify device drivers "
861 "the completion of reset\n");
862 result = PCI_ERS_RESULT_NONE;
863 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
866 /* All devices should claim they have recovered by now. */
867 if ((result != PCI_ERS_RESULT_RECOVERED) &&
868 (result != PCI_ERS_RESULT_NONE)) {
869 pr_warn("EEH: Not recovered\n");
874 * For those hot removed VFs, we should add back them after PF get
875 * recovered properly.
877 list_for_each_entry_safe(edev, tmp, &rmv_data.edev_list, rmv_list) {
878 eeh_add_virt_device(edev, NULL);
879 list_del(&edev->rmv_list);
882 /* Tell all device drivers that they can resume operations */
883 pr_info("EEH: Notify device driver to resume\n");
884 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
890 * About 90% of all real-life EEH failures in the field
891 * are due to poorly seated PCI cards. Only 10% or so are
892 * due to actual, failed cards.
894 pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
895 "last hour and has been permanently disabled.\n"
896 "Please try reseating or replacing it.\n",
897 pe->phb->global_number, pe->addr,
902 pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
903 "Please try reseating or replacing it\n",
904 pe->phb->global_number, pe->addr);
907 eeh_slot_error_detail(pe, EEH_LOG_PERM);
909 /* Notify all devices that they're about to go down. */
910 eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
912 /* Mark the PE to be removed permanently */
913 eeh_pe_state_mark(pe, EEH_PE_REMOVED);
916 * Shut down the device drivers for good. We mark
917 * all removed devices correctly to avoid access
918 * the their PCI config any more.
921 if (pe->type & EEH_PE_VF) {
922 eeh_pe_dev_traverse(pe, eeh_rmv_device, NULL);
923 eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
925 eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
926 eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
928 pci_lock_rescan_remove();
929 pci_hp_remove_devices(frozen_bus);
930 pci_unlock_rescan_remove();
932 /* The passed PE should no longer be used */
939 static void eeh_handle_special_event(void)
941 struct eeh_pe *pe, *phb_pe;
943 struct pci_controller *hose;
949 rc = eeh_ops->next_error(&pe);
952 case EEH_NEXT_ERR_DEAD_IOC:
953 /* Mark all PHBs in dead state */
954 eeh_serialize_lock(&flags);
956 /* Purge all events */
957 eeh_remove_event(NULL, true);
959 list_for_each_entry(hose, &hose_list, list_node) {
960 phb_pe = eeh_phb_pe_get(hose);
961 if (!phb_pe) continue;
963 eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
966 eeh_serialize_unlock(flags);
969 case EEH_NEXT_ERR_FROZEN_PE:
970 case EEH_NEXT_ERR_FENCED_PHB:
971 case EEH_NEXT_ERR_DEAD_PHB:
972 /* Mark the PE in fenced state */
973 eeh_serialize_lock(&flags);
975 /* Purge all events of the PHB */
976 eeh_remove_event(pe, true);
978 if (rc == EEH_NEXT_ERR_DEAD_PHB)
979 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
981 eeh_pe_state_mark(pe,
982 EEH_PE_ISOLATED | EEH_PE_RECOVERING);
984 eeh_serialize_unlock(flags);
987 case EEH_NEXT_ERR_NONE:
990 pr_warn("%s: Invalid value %d from next_error()\n",
996 * For fenced PHB and frozen PE, it's handled as normal
997 * event. We have to remove the affected PHBs for dead
1000 if (rc == EEH_NEXT_ERR_FROZEN_PE ||
1001 rc == EEH_NEXT_ERR_FENCED_PHB) {
1003 * eeh_handle_normal_event() can make the PE stale if it
1004 * determines that the PE cannot possibly be recovered.
1005 * Don't modify the PE state if that's the case.
1007 if (eeh_handle_normal_event(pe))
1010 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
1012 pci_lock_rescan_remove();
1013 list_for_each_entry(hose, &hose_list, list_node) {
1014 phb_pe = eeh_phb_pe_get(hose);
1016 !(phb_pe->state & EEH_PE_ISOLATED) ||
1017 (phb_pe->state & EEH_PE_RECOVERING))
1020 /* Notify all devices to be down */
1021 eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
1022 eeh_pe_dev_traverse(pe,
1023 eeh_report_failure, NULL);
1024 bus = eeh_pe_bus_get(phb_pe);
1026 pr_err("%s: Cannot find PCI bus for "
1029 pe->phb->global_number,
1033 pci_hp_remove_devices(bus);
1035 pci_unlock_rescan_remove();
1039 * If we have detected dead IOC, we needn't proceed
1040 * any more since all PHBs would have been removed
1042 if (rc == EEH_NEXT_ERR_DEAD_IOC)
1044 } while (rc != EEH_NEXT_ERR_NONE);
1048 * eeh_handle_event - Reset a PCI device after hard lockup.
1051 * While PHB detects address or data parity errors on particular PCI
1052 * slot, the associated PE will be frozen. Besides, DMA's occurring
1053 * to wild addresses (which usually happen due to bugs in device
1054 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
1055 * #PERR or other misc PCI-related errors also can trigger EEH errors.
1057 * Recovery process consists of unplugging the device driver (which
1058 * generated hotplug events to userspace), then issuing a PCI #RST to
1059 * the device, then reconfiguring the PCI config space for all bridges
1060 * & devices under this slot, and then finally restarting the device
1061 * drivers (which cause a second set of hotplug events to go out to
1064 void eeh_handle_event(struct eeh_pe *pe)
1067 eeh_handle_normal_event(pe);
1069 eeh_handle_special_event();