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 static void print_device_node_tree(struct pci_dn *pdn, int dent)
90 struct device_node *pc;
94 for (i = 0; i < dent; i++)
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);
100 pc = pdn->node->child;
102 print_device_node_tree(PCI_DN(pc), dent);
109 * eeh_disable_irq - Disable interrupt for the recovering 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)
129 if (!irq_has_action(dev->irq))
132 edev->mode |= EEH_DEV_IRQ_DISABLED;
133 disable_irq_nosync(dev->irq);
137 * eeh_enable_irq - Enable interrupt for the recovering 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);
146 struct irq_desc *desc;
148 if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
149 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
151 desc = irq_to_desc(dev->irq);
152 if (desc && desc->depth > 0)
153 enable_irq(dev->irq);
158 * eeh_report_error - Report pci error to each device driver
160 * @userdata: return value
162 * Report an EEH error to each device driver, collect up and
163 * merge the device driver responses. Cumulative response
164 * passed back in "userdata".
166 static void *eeh_report_error(void *data, void *userdata)
168 struct eeh_dev *edev = (struct eeh_dev *)data;
169 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
170 enum pci_ers_result rc, *res = userdata;
171 struct pci_driver *driver;
173 /* We might not have the associated PCI device,
174 * then we should continue for next one.
176 if (!dev) return NULL;
177 dev->error_state = pci_channel_io_frozen;
179 driver = eeh_pcid_get(dev);
180 if (!driver) return NULL;
182 eeh_disable_irq(dev);
184 if (!driver->err_handler ||
185 !driver->err_handler->error_detected) {
190 rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
192 /* A driver that needs a reset trumps all others */
193 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
194 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
201 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
203 * @userdata: return value
205 * Tells each device driver that IO ports, MMIO and config space I/O
206 * are now enabled. Collects up and merges the device driver responses.
207 * Cumulative response passed back in "userdata".
209 static void *eeh_report_mmio_enabled(void *data, void *userdata)
211 struct eeh_dev *edev = (struct eeh_dev *)data;
212 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
213 enum pci_ers_result rc, *res = userdata;
214 struct pci_driver *driver;
216 driver = eeh_pcid_get(dev);
217 if (!driver) return NULL;
219 if (!driver->err_handler ||
220 !driver->err_handler->mmio_enabled ||
221 (edev->mode & EEH_DEV_NO_HANDLER)) {
226 rc = driver->err_handler->mmio_enabled(dev);
228 /* A driver that needs a reset trumps all others */
229 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
230 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
237 * eeh_report_reset - Tell device that slot has been reset
239 * @userdata: return value
241 * This routine must be called while EEH tries to reset particular
242 * PCI device so that the associated PCI device driver could take
243 * some actions, usually to save data the driver needs so that the
244 * driver can work again while the device is recovered.
246 static void *eeh_report_reset(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) return NULL;
254 dev->error_state = pci_channel_io_normal;
256 driver = eeh_pcid_get(dev);
257 if (!driver) return NULL;
261 if (!driver->err_handler ||
262 !driver->err_handler->slot_reset ||
263 (edev->mode & EEH_DEV_NO_HANDLER)) {
268 rc = driver->err_handler->slot_reset(dev);
269 if ((*res == PCI_ERS_RESULT_NONE) ||
270 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
271 if (*res == PCI_ERS_RESULT_DISCONNECT &&
272 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
279 * eeh_report_resume - Tell device to resume normal operations
281 * @userdata: return value
283 * This routine must be called to notify the device driver that it
284 * could resume so that the device driver can do some initialization
285 * to make the recovered device work again.
287 static void *eeh_report_resume(void *data, void *userdata)
289 struct eeh_dev *edev = (struct eeh_dev *)data;
290 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
291 struct pci_driver *driver;
293 if (!dev) return NULL;
294 dev->error_state = pci_channel_io_normal;
296 driver = eeh_pcid_get(dev);
297 if (!driver) return NULL;
301 if (!driver->err_handler ||
302 !driver->err_handler->resume ||
303 (edev->mode & EEH_DEV_NO_HANDLER)) {
304 edev->mode &= ~EEH_DEV_NO_HANDLER;
309 driver->err_handler->resume(dev);
316 * eeh_report_failure - Tell device driver that device is dead.
318 * @userdata: return value
320 * This informs the device driver that the device is permanently
321 * dead, and that no further recovery attempts will be made on it.
323 static void *eeh_report_failure(void *data, void *userdata)
325 struct eeh_dev *edev = (struct eeh_dev *)data;
326 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
327 struct pci_driver *driver;
329 if (!dev) return NULL;
330 dev->error_state = pci_channel_io_perm_failure;
332 driver = eeh_pcid_get(dev);
333 if (!driver) return NULL;
335 eeh_disable_irq(dev);
337 if (!driver->err_handler ||
338 !driver->err_handler->error_detected) {
343 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
349 static void *eeh_rmv_device(void *data, void *userdata)
351 struct pci_driver *driver;
352 struct eeh_dev *edev = (struct eeh_dev *)data;
353 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
354 int *removed = (int *)userdata;
357 * Actually, we should remove the PCI bridges as well.
358 * However, that's lots of complexity to do that,
359 * particularly some of devices under the bridge might
360 * support EEH. So we just care about PCI devices for
363 if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
366 driver = eeh_pcid_get(dev);
369 if (driver->err_handler)
373 /* Remove it from PCI subsystem */
374 pr_debug("EEH: Removing %s without EEH sensitive driver\n",
376 edev->bus = dev->bus;
377 edev->mode |= EEH_DEV_DISCONNECTED;
380 pci_lock_rescan_remove();
381 pci_stop_and_remove_bus_device(dev);
382 pci_unlock_rescan_remove();
387 static void *eeh_pe_detach_dev(void *data, void *userdata)
389 struct eeh_pe *pe = (struct eeh_pe *)data;
390 struct eeh_dev *edev, *tmp;
392 eeh_pe_for_each_dev(pe, edev, tmp) {
393 if (!(edev->mode & EEH_DEV_DISCONNECTED))
396 edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
397 eeh_rmv_from_parent_pe(edev);
404 * eeh_reset_device - Perform actual reset of a pci slot
406 * @bus: PCI bus corresponding to the isolcated slot
408 * This routine must be called to do reset on the indicated PE.
409 * During the reset, udev might be invoked because those affected
410 * PCI devices will be removed and then added.
412 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
414 struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
415 struct timeval tstamp;
416 int cnt, rc, removed = 0;
418 /* pcibios will clear the counter; save the value */
419 cnt = pe->freeze_count;
423 * We don't remove the corresponding PE instances because
424 * we need the information afterwords. The attached EEH
425 * devices are expected to be attached soon when calling
426 * into pcibios_add_pci_devices().
428 eeh_pe_state_mark(pe, EEH_PE_KEEP);
430 pci_lock_rescan_remove();
431 pcibios_remove_pci_devices(bus);
432 pci_unlock_rescan_remove();
433 } else if (frozen_bus) {
434 eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
437 /* Reset the pci controller. (Asserts RST#; resets config space).
438 * Reconfigure bridges and devices. Don't try to bring the system
439 * up if the reset failed for some reason.
441 rc = eeh_reset_pe(pe);
445 pci_lock_rescan_remove();
448 eeh_ops->configure_bridge(pe);
449 eeh_pe_restore_bars(pe);
451 /* Give the system 5 seconds to finish running the user-space
452 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
453 * this is a hack, but if we don't do this, and try to bring
454 * the device up before the scripts have taken it down,
455 * potentially weird things happen.
458 pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
462 * The EEH device is still connected with its parent
463 * PE. We should disconnect it so the binding can be
464 * rebuilt when adding PCI devices.
466 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
467 pcibios_add_pci_devices(bus);
468 } else if (frozen_bus && removed) {
469 pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
472 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
473 pcibios_add_pci_devices(frozen_bus);
475 eeh_pe_state_clear(pe, EEH_PE_KEEP);
478 pe->freeze_count = cnt;
480 pci_unlock_rescan_remove();
484 /* The longest amount of time to wait for a pci device
485 * to come back on line, in seconds.
487 #define MAX_WAIT_FOR_RECOVERY 300
489 static void eeh_handle_normal_event(struct eeh_pe *pe)
491 struct pci_bus *frozen_bus;
493 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
495 frozen_bus = eeh_pe_bus_get(pe);
497 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
498 __func__, pe->phb->global_number, pe->addr);
502 eeh_pe_update_time_stamp(pe);
504 if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES)
505 goto excess_failures;
506 pr_warning("EEH: This PCI device has failed %d times in the last hour\n",
509 /* Walk the various device drivers attached to this slot through
510 * a reset sequence, giving each an opportunity to do what it needs
511 * to accomplish the reset. Each child gets a report of the
512 * status ... if any child can't handle the reset, then the entire
513 * slot is dlpar removed and added.
515 pr_info("EEH: Notify device drivers to shutdown\n");
516 eeh_pe_dev_traverse(pe, eeh_report_error, &result);
518 /* Get the current PCI slot state. This can take a long time,
519 * sometimes over 3 seconds for certain systems.
521 rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
522 if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
523 pr_warning("EEH: Permanent failure\n");
527 /* Since rtas may enable MMIO when posting the error log,
528 * don't post the error log until after all dev drivers
529 * have been informed.
531 pr_info("EEH: Collect temporary log\n");
532 eeh_slot_error_detail(pe, EEH_LOG_TEMP);
534 /* If all device drivers were EEH-unaware, then shut
535 * down all of the device drivers, and hope they
536 * go down willingly, without panicing the system.
538 if (result == PCI_ERS_RESULT_NONE) {
539 pr_info("EEH: Reset with hotplug activity\n");
540 rc = eeh_reset_device(pe, frozen_bus);
542 pr_warning("%s: Unable to reset, err=%d\n",
548 /* If all devices reported they can proceed, then re-enable MMIO */
549 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
550 pr_info("EEH: Enable I/O for affected devices\n");
551 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
556 result = PCI_ERS_RESULT_NEED_RESET;
558 pr_info("EEH: Notify device drivers to resume I/O\n");
559 result = PCI_ERS_RESULT_NONE;
560 eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
564 /* If all devices reported they can proceed, then re-enable DMA */
565 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
566 pr_info("EEH: Enabled DMA for affected devices\n");
567 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
572 result = PCI_ERS_RESULT_NEED_RESET;
574 result = PCI_ERS_RESULT_RECOVERED;
577 /* If any device has a hard failure, then shut off everything. */
578 if (result == PCI_ERS_RESULT_DISCONNECT) {
579 pr_warning("EEH: Device driver gave up\n");
583 /* If any device called out for a reset, then reset the slot */
584 if (result == PCI_ERS_RESULT_NEED_RESET) {
585 pr_info("EEH: Reset without hotplug activity\n");
586 rc = eeh_reset_device(pe, NULL);
588 pr_warning("%s: Cannot reset, err=%d\n",
593 pr_info("EEH: Notify device drivers "
594 "the completion of reset\n");
595 result = PCI_ERS_RESULT_NONE;
596 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
599 /* All devices should claim they have recovered by now. */
600 if ((result != PCI_ERS_RESULT_RECOVERED) &&
601 (result != PCI_ERS_RESULT_NONE)) {
602 pr_warning("EEH: Not recovered\n");
606 /* Tell all device drivers that they can resume operations */
607 pr_info("EEH: Notify device driver to resume\n");
608 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
614 * About 90% of all real-life EEH failures in the field
615 * are due to poorly seated PCI cards. Only 10% or so are
616 * due to actual, failed cards.
618 pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
619 "last hour and has been permanently disabled.\n"
620 "Please try reseating or replacing it.\n",
621 pe->phb->global_number, pe->addr,
626 pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
627 "Please try reseating or replacing it\n",
628 pe->phb->global_number, pe->addr);
631 eeh_slot_error_detail(pe, EEH_LOG_PERM);
633 /* Notify all devices that they're about to go down. */
634 eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
636 /* Shut down the device drivers for good. */
638 pci_lock_rescan_remove();
639 pcibios_remove_pci_devices(frozen_bus);
640 pci_unlock_rescan_remove();
644 static void eeh_handle_special_event(void)
646 struct eeh_pe *pe, *phb_pe;
648 struct pci_controller *hose;
654 rc = eeh_ops->next_error(&pe);
657 case EEH_NEXT_ERR_DEAD_IOC:
658 /* Mark all PHBs in dead state */
659 eeh_serialize_lock(&flags);
661 /* Purge all events */
662 eeh_remove_event(NULL);
664 list_for_each_entry(hose, &hose_list, list_node) {
665 phb_pe = eeh_phb_pe_get(hose);
666 if (!phb_pe) continue;
668 eeh_pe_state_mark(phb_pe,
669 EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
672 eeh_serialize_unlock(flags);
675 case EEH_NEXT_ERR_FROZEN_PE:
676 case EEH_NEXT_ERR_FENCED_PHB:
677 case EEH_NEXT_ERR_DEAD_PHB:
678 /* Mark the PE in fenced state */
679 eeh_serialize_lock(&flags);
681 /* Purge all events of the PHB */
682 eeh_remove_event(pe);
684 if (rc == EEH_NEXT_ERR_DEAD_PHB)
685 eeh_pe_state_mark(pe,
686 EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
688 eeh_pe_state_mark(pe,
689 EEH_PE_ISOLATED | EEH_PE_RECOVERING);
691 eeh_serialize_unlock(flags);
694 case EEH_NEXT_ERR_NONE:
697 pr_warn("%s: Invalid value %d from next_error()\n",
703 * For fenced PHB and frozen PE, it's handled as normal
704 * event. We have to remove the affected PHBs for dead
707 if (rc == EEH_NEXT_ERR_FROZEN_PE ||
708 rc == EEH_NEXT_ERR_FENCED_PHB) {
709 eeh_handle_normal_event(pe);
711 pci_lock_rescan_remove();
712 list_for_each_entry(hose, &hose_list, list_node) {
713 phb_pe = eeh_phb_pe_get(hose);
715 !(phb_pe->state & EEH_PE_PHB_DEAD))
718 /* Notify all devices to be down */
719 bus = eeh_pe_bus_get(phb_pe);
720 eeh_pe_dev_traverse(pe,
721 eeh_report_failure, NULL);
722 pcibios_remove_pci_devices(bus);
724 pci_unlock_rescan_remove();
728 * If we have detected dead IOC, we needn't proceed
729 * any more since all PHBs would have been removed
731 if (rc == EEH_NEXT_ERR_DEAD_IOC)
733 } while (rc != EEH_NEXT_ERR_NONE);
737 * eeh_handle_event - Reset a PCI device after hard lockup.
740 * While PHB detects address or data parity errors on particular PCI
741 * slot, the associated PE will be frozen. Besides, DMA's occurring
742 * to wild addresses (which usually happen due to bugs in device
743 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
744 * #PERR or other misc PCI-related errors also can trigger EEH errors.
746 * Recovery process consists of unplugging the device driver (which
747 * generated hotplug events to userspace), then issuing a PCI #RST to
748 * the device, then reconfiguring the PCI config space for all bridges
749 * & devices under this slot, and then finally restarting the device
750 * drivers (which cause a second set of hotplug events to go out to
753 void eeh_handle_event(struct eeh_pe *pe)
756 eeh_handle_normal_event(pe);
758 eeh_handle_special_event();