Merge tag 'locks-v3.16-2' of git://git.samba.org/jlayton/linux
[linux/fpc-iii.git] / arch / powerpc / kernel / eeh_driver.c
blob420da61d4ce001d74ba64b675f8615b65f6e812a
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
6 * All rights reserved.
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
17 * details.
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>
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>
37 /**
38 * eeh_pcid_name - Retrieve name of PCI device driver
39 * @pdev: PCI device
41 * This routine is used to retrieve the name of PCI device driver
42 * if that's valid.
44 static inline const char *eeh_pcid_name(struct pci_dev *pdev)
46 if (pdev && pdev->dev.driver)
47 return pdev->dev.driver->name;
48 return "";
51 /**
52 * eeh_pcid_get - Get the PCI device driver
53 * @pdev: PCI device
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)
63 return NULL;
65 if (!try_module_get(pdev->driver->driver.owner))
66 return NULL;
68 return pdev->driver;
71 /**
72 * eeh_pcid_put - Dereference on the PCI device driver
73 * @pdev: PCI device
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)
81 return;
83 module_put(pdev->driver->driver.owner);
86 #if 0
87 static void print_device_node_tree(struct pci_dn *pdn, int dent)
89 int i;
90 struct device_node *pc;
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;
106 #endif
109 * eeh_disable_irq - Disable interrupt for the recovering device
110 * @dev: PCI device
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.
118 static void eeh_disable_irq(struct pci_dev *dev)
120 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
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.
126 if (dev->msi_enabled || dev->msix_enabled)
127 return;
129 if (!irq_has_action(dev->irq))
130 return;
132 edev->mode |= EEH_DEV_IRQ_DISABLED;
133 disable_irq_nosync(dev->irq);
137 * eeh_enable_irq - Enable interrupt for the recovering device
138 * @dev: PCI device
140 * This routine must be called to enable interrupt while failed
141 * device could be resumed.
143 static void eeh_enable_irq(struct pci_dev *dev)
145 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
147 if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
148 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
150 * FIXME !!!!!
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.
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.
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.
167 * tglx
169 if (irqd_irq_disabled(irq_get_irq_data(dev->irq)))
170 enable_irq(dev->irq);
174 static bool eeh_dev_removed(struct eeh_dev *edev)
176 /* EEH device removed ? */
177 if (!edev || (edev->mode & EEH_DEV_REMOVED))
178 return true;
180 return false;
184 * eeh_report_error - Report pci error to each device driver
185 * @data: eeh device
186 * @userdata: return value
188 * Report an EEH error to each device driver, collect up and
189 * merge the device driver responses. Cumulative response
190 * passed back in "userdata".
192 static void *eeh_report_error(void *data, void *userdata)
194 struct eeh_dev *edev = (struct eeh_dev *)data;
195 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
196 enum pci_ers_result rc, *res = userdata;
197 struct pci_driver *driver;
199 if (!dev || eeh_dev_removed(edev))
200 return NULL;
201 dev->error_state = pci_channel_io_frozen;
203 driver = eeh_pcid_get(dev);
204 if (!driver) return NULL;
206 eeh_disable_irq(dev);
208 if (!driver->err_handler ||
209 !driver->err_handler->error_detected) {
210 eeh_pcid_put(dev);
211 return NULL;
214 rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
216 /* A driver that needs a reset trumps all others */
217 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
218 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
220 eeh_pcid_put(dev);
221 return NULL;
225 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
226 * @data: eeh device
227 * @userdata: return value
229 * Tells each device driver that IO ports, MMIO and config space I/O
230 * are now enabled. Collects up and merges the device driver responses.
231 * Cumulative response passed back in "userdata".
233 static void *eeh_report_mmio_enabled(void *data, void *userdata)
235 struct eeh_dev *edev = (struct eeh_dev *)data;
236 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
237 enum pci_ers_result rc, *res = userdata;
238 struct pci_driver *driver;
240 if (!dev || eeh_dev_removed(edev))
241 return NULL;
243 driver = eeh_pcid_get(dev);
244 if (!driver) return NULL;
246 if (!driver->err_handler ||
247 !driver->err_handler->mmio_enabled ||
248 (edev->mode & EEH_DEV_NO_HANDLER)) {
249 eeh_pcid_put(dev);
250 return NULL;
253 rc = driver->err_handler->mmio_enabled(dev);
255 /* A driver that needs a reset trumps all others */
256 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
257 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
259 eeh_pcid_put(dev);
260 return NULL;
264 * eeh_report_reset - Tell device that slot has been reset
265 * @data: eeh device
266 * @userdata: return value
268 * This routine must be called while EEH tries to reset particular
269 * PCI device so that the associated PCI device driver could take
270 * some actions, usually to save data the driver needs so that the
271 * driver can work again while the device is recovered.
273 static void *eeh_report_reset(void *data, void *userdata)
275 struct eeh_dev *edev = (struct eeh_dev *)data;
276 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
277 enum pci_ers_result rc, *res = userdata;
278 struct pci_driver *driver;
280 if (!dev || eeh_dev_removed(edev))
281 return NULL;
282 dev->error_state = pci_channel_io_normal;
284 driver = eeh_pcid_get(dev);
285 if (!driver) return NULL;
287 eeh_enable_irq(dev);
289 if (!driver->err_handler ||
290 !driver->err_handler->slot_reset ||
291 (edev->mode & EEH_DEV_NO_HANDLER)) {
292 eeh_pcid_put(dev);
293 return NULL;
296 rc = driver->err_handler->slot_reset(dev);
297 if ((*res == PCI_ERS_RESULT_NONE) ||
298 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
299 if (*res == PCI_ERS_RESULT_DISCONNECT &&
300 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
302 eeh_pcid_put(dev);
303 return NULL;
307 * eeh_report_resume - Tell device to resume normal operations
308 * @data: eeh device
309 * @userdata: return value
311 * This routine must be called to notify the device driver that it
312 * could resume so that the device driver can do some initialization
313 * to make the recovered device work again.
315 static void *eeh_report_resume(void *data, void *userdata)
317 struct eeh_dev *edev = (struct eeh_dev *)data;
318 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
319 struct pci_driver *driver;
321 if (!dev || eeh_dev_removed(edev))
322 return NULL;
323 dev->error_state = pci_channel_io_normal;
325 driver = eeh_pcid_get(dev);
326 if (!driver) return NULL;
328 eeh_enable_irq(dev);
330 if (!driver->err_handler ||
331 !driver->err_handler->resume ||
332 (edev->mode & EEH_DEV_NO_HANDLER)) {
333 edev->mode &= ~EEH_DEV_NO_HANDLER;
334 eeh_pcid_put(dev);
335 return NULL;
338 driver->err_handler->resume(dev);
340 eeh_pcid_put(dev);
341 return NULL;
345 * eeh_report_failure - Tell device driver that device is dead.
346 * @data: eeh device
347 * @userdata: return value
349 * This informs the device driver that the device is permanently
350 * dead, and that no further recovery attempts will be made on it.
352 static void *eeh_report_failure(void *data, void *userdata)
354 struct eeh_dev *edev = (struct eeh_dev *)data;
355 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
356 struct pci_driver *driver;
358 if (!dev || eeh_dev_removed(edev))
359 return NULL;
360 dev->error_state = pci_channel_io_perm_failure;
362 driver = eeh_pcid_get(dev);
363 if (!driver) return NULL;
365 eeh_disable_irq(dev);
367 if (!driver->err_handler ||
368 !driver->err_handler->error_detected) {
369 eeh_pcid_put(dev);
370 return NULL;
373 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
375 eeh_pcid_put(dev);
376 return NULL;
379 static void *eeh_rmv_device(void *data, void *userdata)
381 struct pci_driver *driver;
382 struct eeh_dev *edev = (struct eeh_dev *)data;
383 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
384 int *removed = (int *)userdata;
387 * Actually, we should remove the PCI bridges as well.
388 * However, that's lots of complexity to do that,
389 * particularly some of devices under the bridge might
390 * support EEH. So we just care about PCI devices for
391 * simplicity here.
393 if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
394 return NULL;
397 * We rely on count-based pcibios_release_device() to
398 * detach permanently offlined PEs. Unfortunately, that's
399 * not reliable enough. We might have the permanently
400 * offlined PEs attached, but we needn't take care of
401 * them and their child devices.
403 if (eeh_dev_removed(edev))
404 return NULL;
406 driver = eeh_pcid_get(dev);
407 if (driver) {
408 eeh_pcid_put(dev);
409 if (driver->err_handler)
410 return NULL;
413 /* Remove it from PCI subsystem */
414 pr_debug("EEH: Removing %s without EEH sensitive driver\n",
415 pci_name(dev));
416 edev->bus = dev->bus;
417 edev->mode |= EEH_DEV_DISCONNECTED;
418 (*removed)++;
420 pci_lock_rescan_remove();
421 pci_stop_and_remove_bus_device(dev);
422 pci_unlock_rescan_remove();
424 return NULL;
427 static void *eeh_pe_detach_dev(void *data, void *userdata)
429 struct eeh_pe *pe = (struct eeh_pe *)data;
430 struct eeh_dev *edev, *tmp;
432 eeh_pe_for_each_dev(pe, edev, tmp) {
433 if (!(edev->mode & EEH_DEV_DISCONNECTED))
434 continue;
436 edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
437 eeh_rmv_from_parent_pe(edev);
440 return NULL;
444 * Explicitly clear PE's frozen state for PowerNV where
445 * we have frozen PE until BAR restore is completed. It's
446 * harmless to clear it for pSeries. To be consistent with
447 * PE reset (for 3 times), we try to clear the frozen state
448 * for 3 times as well.
450 static void *__eeh_clear_pe_frozen_state(void *data, void *flag)
452 struct eeh_pe *pe = (struct eeh_pe *)data;
453 int i, rc;
455 for (i = 0; i < 3; i++) {
456 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
457 if (rc)
458 continue;
459 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
460 if (!rc)
461 break;
464 /* The PE has been isolated, clear it */
465 if (rc) {
466 pr_warn("%s: Can't clear frozen PHB#%x-PE#%x (%d)\n",
467 __func__, pe->phb->global_number, pe->addr, rc);
468 return (void *)pe;
471 return NULL;
474 static int eeh_clear_pe_frozen_state(struct eeh_pe *pe)
476 void *rc;
478 rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, NULL);
479 if (!rc)
480 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
482 return rc ? -EIO : 0;
486 * eeh_reset_device - Perform actual reset of a pci slot
487 * @pe: EEH PE
488 * @bus: PCI bus corresponding to the isolcated slot
490 * This routine must be called to do reset on the indicated PE.
491 * During the reset, udev might be invoked because those affected
492 * PCI devices will be removed and then added.
494 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
496 struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
497 struct timeval tstamp;
498 int cnt, rc, removed = 0;
500 /* pcibios will clear the counter; save the value */
501 cnt = pe->freeze_count;
502 tstamp = pe->tstamp;
505 * We don't remove the corresponding PE instances because
506 * we need the information afterwords. The attached EEH
507 * devices are expected to be attached soon when calling
508 * into pcibios_add_pci_devices().
510 eeh_pe_state_mark(pe, EEH_PE_KEEP);
511 if (bus) {
512 pci_lock_rescan_remove();
513 pcibios_remove_pci_devices(bus);
514 pci_unlock_rescan_remove();
515 } else if (frozen_bus) {
516 eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
520 * Reset the pci controller. (Asserts RST#; resets config space).
521 * Reconfigure bridges and devices. Don't try to bring the system
522 * up if the reset failed for some reason.
524 * During the reset, it's very dangerous to have uncontrolled PCI
525 * config accesses. So we prefer to block them. However, controlled
526 * PCI config accesses initiated from EEH itself are allowed.
528 eeh_pe_state_mark(pe, EEH_PE_RESET);
529 rc = eeh_reset_pe(pe);
530 if (rc) {
531 eeh_pe_state_clear(pe, EEH_PE_RESET);
532 return rc;
535 pci_lock_rescan_remove();
537 /* Restore PE */
538 eeh_ops->configure_bridge(pe);
539 eeh_pe_restore_bars(pe);
540 eeh_pe_state_clear(pe, EEH_PE_RESET);
542 /* Clear frozen state */
543 rc = eeh_clear_pe_frozen_state(pe);
544 if (rc)
545 return rc;
547 /* Give the system 5 seconds to finish running the user-space
548 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
549 * this is a hack, but if we don't do this, and try to bring
550 * the device up before the scripts have taken it down,
551 * potentially weird things happen.
553 if (bus) {
554 pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
555 ssleep(5);
558 * The EEH device is still connected with its parent
559 * PE. We should disconnect it so the binding can be
560 * rebuilt when adding PCI devices.
562 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
563 pcibios_add_pci_devices(bus);
564 } else if (frozen_bus && removed) {
565 pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
566 ssleep(5);
568 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
569 pcibios_add_pci_devices(frozen_bus);
571 eeh_pe_state_clear(pe, EEH_PE_KEEP);
573 pe->tstamp = tstamp;
574 pe->freeze_count = cnt;
576 pci_unlock_rescan_remove();
577 return 0;
580 /* The longest amount of time to wait for a pci device
581 * to come back on line, in seconds.
583 #define MAX_WAIT_FOR_RECOVERY 300
585 static void eeh_handle_normal_event(struct eeh_pe *pe)
587 struct pci_bus *frozen_bus;
588 int rc = 0;
589 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
591 frozen_bus = eeh_pe_bus_get(pe);
592 if (!frozen_bus) {
593 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
594 __func__, pe->phb->global_number, pe->addr);
595 return;
598 eeh_pe_update_time_stamp(pe);
599 pe->freeze_count++;
600 if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES)
601 goto excess_failures;
602 pr_warning("EEH: This PCI device has failed %d times in the last hour\n",
603 pe->freeze_count);
605 /* Walk the various device drivers attached to this slot through
606 * a reset sequence, giving each an opportunity to do what it needs
607 * to accomplish the reset. Each child gets a report of the
608 * status ... if any child can't handle the reset, then the entire
609 * slot is dlpar removed and added.
611 pr_info("EEH: Notify device drivers to shutdown\n");
612 eeh_pe_dev_traverse(pe, eeh_report_error, &result);
614 /* Get the current PCI slot state. This can take a long time,
615 * sometimes over 3 seconds for certain systems.
617 rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
618 if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
619 pr_warning("EEH: Permanent failure\n");
620 goto hard_fail;
623 /* Since rtas may enable MMIO when posting the error log,
624 * don't post the error log until after all dev drivers
625 * have been informed.
627 pr_info("EEH: Collect temporary log\n");
628 eeh_slot_error_detail(pe, EEH_LOG_TEMP);
630 /* If all device drivers were EEH-unaware, then shut
631 * down all of the device drivers, and hope they
632 * go down willingly, without panicing the system.
634 if (result == PCI_ERS_RESULT_NONE) {
635 pr_info("EEH: Reset with hotplug activity\n");
636 rc = eeh_reset_device(pe, frozen_bus);
637 if (rc) {
638 pr_warning("%s: Unable to reset, err=%d\n",
639 __func__, rc);
640 goto hard_fail;
644 /* If all devices reported they can proceed, then re-enable MMIO */
645 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
646 pr_info("EEH: Enable I/O for affected devices\n");
647 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
649 if (rc < 0)
650 goto hard_fail;
651 if (rc) {
652 result = PCI_ERS_RESULT_NEED_RESET;
653 } else {
654 pr_info("EEH: Notify device drivers to resume I/O\n");
655 eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
659 /* If all devices reported they can proceed, then re-enable DMA */
660 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
661 pr_info("EEH: Enabled DMA for affected devices\n");
662 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
664 if (rc < 0)
665 goto hard_fail;
666 if (rc) {
667 result = PCI_ERS_RESULT_NEED_RESET;
668 } else {
670 * We didn't do PE reset for the case. The PE
671 * is still in frozen state. Clear it before
672 * resuming the PE.
674 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
675 result = PCI_ERS_RESULT_RECOVERED;
679 /* If any device has a hard failure, then shut off everything. */
680 if (result == PCI_ERS_RESULT_DISCONNECT) {
681 pr_warning("EEH: Device driver gave up\n");
682 goto hard_fail;
685 /* If any device called out for a reset, then reset the slot */
686 if (result == PCI_ERS_RESULT_NEED_RESET) {
687 pr_info("EEH: Reset without hotplug activity\n");
688 rc = eeh_reset_device(pe, NULL);
689 if (rc) {
690 pr_warning("%s: Cannot reset, err=%d\n",
691 __func__, rc);
692 goto hard_fail;
695 pr_info("EEH: Notify device drivers "
696 "the completion of reset\n");
697 result = PCI_ERS_RESULT_NONE;
698 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
701 /* All devices should claim they have recovered by now. */
702 if ((result != PCI_ERS_RESULT_RECOVERED) &&
703 (result != PCI_ERS_RESULT_NONE)) {
704 pr_warning("EEH: Not recovered\n");
705 goto hard_fail;
708 /* Tell all device drivers that they can resume operations */
709 pr_info("EEH: Notify device driver to resume\n");
710 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
712 return;
714 excess_failures:
716 * About 90% of all real-life EEH failures in the field
717 * are due to poorly seated PCI cards. Only 10% or so are
718 * due to actual, failed cards.
720 pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
721 "last hour and has been permanently disabled.\n"
722 "Please try reseating or replacing it.\n",
723 pe->phb->global_number, pe->addr,
724 pe->freeze_count);
725 goto perm_error;
727 hard_fail:
728 pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
729 "Please try reseating or replacing it\n",
730 pe->phb->global_number, pe->addr);
732 perm_error:
733 eeh_slot_error_detail(pe, EEH_LOG_PERM);
735 /* Notify all devices that they're about to go down. */
736 eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
738 /* Mark the PE to be removed permanently */
739 pe->freeze_count = EEH_MAX_ALLOWED_FREEZES + 1;
742 * Shut down the device drivers for good. We mark
743 * all removed devices correctly to avoid access
744 * the their PCI config any more.
746 if (frozen_bus) {
747 eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
749 pci_lock_rescan_remove();
750 pcibios_remove_pci_devices(frozen_bus);
751 pci_unlock_rescan_remove();
755 static void eeh_handle_special_event(void)
757 struct eeh_pe *pe, *phb_pe;
758 struct pci_bus *bus;
759 struct pci_controller *hose;
760 unsigned long flags;
761 int rc;
764 do {
765 rc = eeh_ops->next_error(&pe);
767 switch (rc) {
768 case EEH_NEXT_ERR_DEAD_IOC:
769 /* Mark all PHBs in dead state */
770 eeh_serialize_lock(&flags);
772 /* Purge all events */
773 eeh_remove_event(NULL, true);
775 list_for_each_entry(hose, &hose_list, list_node) {
776 phb_pe = eeh_phb_pe_get(hose);
777 if (!phb_pe) continue;
779 eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
782 eeh_serialize_unlock(flags);
784 break;
785 case EEH_NEXT_ERR_FROZEN_PE:
786 case EEH_NEXT_ERR_FENCED_PHB:
787 case EEH_NEXT_ERR_DEAD_PHB:
788 /* Mark the PE in fenced state */
789 eeh_serialize_lock(&flags);
791 /* Purge all events of the PHB */
792 eeh_remove_event(pe, true);
794 if (rc == EEH_NEXT_ERR_DEAD_PHB)
795 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
796 else
797 eeh_pe_state_mark(pe,
798 EEH_PE_ISOLATED | EEH_PE_RECOVERING);
800 eeh_serialize_unlock(flags);
802 break;
803 case EEH_NEXT_ERR_NONE:
804 return;
805 default:
806 pr_warn("%s: Invalid value %d from next_error()\n",
807 __func__, rc);
808 return;
812 * For fenced PHB and frozen PE, it's handled as normal
813 * event. We have to remove the affected PHBs for dead
814 * PHB and IOC
816 if (rc == EEH_NEXT_ERR_FROZEN_PE ||
817 rc == EEH_NEXT_ERR_FENCED_PHB) {
818 eeh_handle_normal_event(pe);
819 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
820 } else {
821 pci_lock_rescan_remove();
822 list_for_each_entry(hose, &hose_list, list_node) {
823 phb_pe = eeh_phb_pe_get(hose);
824 if (!phb_pe ||
825 !(phb_pe->state & EEH_PE_ISOLATED) ||
826 (phb_pe->state & EEH_PE_RECOVERING))
827 continue;
829 /* Notify all devices to be down */
830 bus = eeh_pe_bus_get(phb_pe);
831 eeh_pe_dev_traverse(pe,
832 eeh_report_failure, NULL);
833 pcibios_remove_pci_devices(bus);
835 pci_unlock_rescan_remove();
839 * If we have detected dead IOC, we needn't proceed
840 * any more since all PHBs would have been removed
842 if (rc == EEH_NEXT_ERR_DEAD_IOC)
843 break;
844 } while (rc != EEH_NEXT_ERR_NONE);
848 * eeh_handle_event - Reset a PCI device after hard lockup.
849 * @pe: EEH PE
851 * While PHB detects address or data parity errors on particular PCI
852 * slot, the associated PE will be frozen. Besides, DMA's occurring
853 * to wild addresses (which usually happen due to bugs in device
854 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
855 * #PERR or other misc PCI-related errors also can trigger EEH errors.
857 * Recovery process consists of unplugging the device driver (which
858 * generated hotplug events to userspace), then issuing a PCI #RST to
859 * the device, then reconfiguring the PCI config space for all bridges
860 * & devices under this slot, and then finally restarting the device
861 * drivers (which cause a second set of hotplug events to go out to
862 * userspace).
864 void eeh_handle_event(struct eeh_pe *pe)
866 if (pe)
867 eeh_handle_normal_event(pe);
868 else
869 eeh_handle_special_event();