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);
147 if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
148 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
149 enable_irq(dev->irq);
154 * eeh_report_error - Report pci error to each device driver
156 * @userdata: return value
158 * Report an EEH error to each device driver, collect up and
159 * merge the device driver responses. Cumulative response
160 * passed back in "userdata".
162 static void *eeh_report_error(void *data, void *userdata)
164 struct eeh_dev *edev = (struct eeh_dev *)data;
165 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
166 enum pci_ers_result rc, *res = userdata;
167 struct pci_driver *driver;
169 /* We might not have the associated PCI device,
170 * then we should continue for next one.
172 if (!dev) return NULL;
173 dev->error_state = pci_channel_io_frozen;
175 driver = eeh_pcid_get(dev);
176 if (!driver) return NULL;
178 eeh_disable_irq(dev);
180 if (!driver->err_handler ||
181 !driver->err_handler->error_detected) {
186 rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
188 /* A driver that needs a reset trumps all others */
189 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
190 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
197 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
199 * @userdata: return value
201 * Tells each device driver that IO ports, MMIO and config space I/O
202 * are now enabled. Collects up and merges the device driver responses.
203 * Cumulative response passed back in "userdata".
205 static void *eeh_report_mmio_enabled(void *data, void *userdata)
207 struct eeh_dev *edev = (struct eeh_dev *)data;
208 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
209 enum pci_ers_result rc, *res = userdata;
210 struct pci_driver *driver;
212 driver = eeh_pcid_get(dev);
213 if (!driver) return NULL;
215 if (!driver->err_handler ||
216 !driver->err_handler->mmio_enabled) {
221 rc = driver->err_handler->mmio_enabled(dev);
223 /* A driver that needs a reset trumps all others */
224 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
225 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
232 * eeh_report_reset - Tell device that slot has been reset
234 * @userdata: return value
236 * This routine must be called while EEH tries to reset particular
237 * PCI device so that the associated PCI device driver could take
238 * some actions, usually to save data the driver needs so that the
239 * driver can work again while the device is recovered.
241 static void *eeh_report_reset(void *data, void *userdata)
243 struct eeh_dev *edev = (struct eeh_dev *)data;
244 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
245 enum pci_ers_result rc, *res = userdata;
246 struct pci_driver *driver;
248 if (!dev) return NULL;
249 dev->error_state = pci_channel_io_normal;
251 driver = eeh_pcid_get(dev);
252 if (!driver) return NULL;
256 if (!driver->err_handler ||
257 !driver->err_handler->slot_reset) {
262 rc = driver->err_handler->slot_reset(dev);
263 if ((*res == PCI_ERS_RESULT_NONE) ||
264 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
265 if (*res == PCI_ERS_RESULT_DISCONNECT &&
266 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
273 * eeh_report_resume - Tell device to resume normal operations
275 * @userdata: return value
277 * This routine must be called to notify the device driver that it
278 * could resume so that the device driver can do some initialization
279 * to make the recovered device work again.
281 static void *eeh_report_resume(void *data, void *userdata)
283 struct eeh_dev *edev = (struct eeh_dev *)data;
284 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
285 struct pci_driver *driver;
287 if (!dev) return NULL;
288 dev->error_state = pci_channel_io_normal;
290 driver = eeh_pcid_get(dev);
291 if (!driver) return NULL;
295 if (!driver->err_handler ||
296 !driver->err_handler->resume) {
301 driver->err_handler->resume(dev);
308 * eeh_report_failure - Tell device driver that device is dead.
310 * @userdata: return value
312 * This informs the device driver that the device is permanently
313 * dead, and that no further recovery attempts will be made on it.
315 static void *eeh_report_failure(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) return NULL;
322 dev->error_state = pci_channel_io_perm_failure;
324 driver = eeh_pcid_get(dev);
325 if (!driver) return NULL;
327 eeh_disable_irq(dev);
329 if (!driver->err_handler ||
330 !driver->err_handler->error_detected) {
335 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
341 static void *eeh_rmv_device(void *data, void *userdata)
343 struct pci_driver *driver;
344 struct eeh_dev *edev = (struct eeh_dev *)data;
345 struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
346 int *removed = (int *)userdata;
349 * Actually, we should remove the PCI bridges as well.
350 * However, that's lots of complexity to do that,
351 * particularly some of devices under the bridge might
352 * support EEH. So we just care about PCI devices for
355 if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
357 driver = eeh_pcid_get(dev);
358 if (driver && driver->err_handler)
361 /* Remove it from PCI subsystem */
362 pr_debug("EEH: Removing %s without EEH sensitive driver\n",
364 edev->bus = dev->bus;
365 edev->mode |= EEH_DEV_DISCONNECTED;
368 pci_stop_and_remove_bus_device(dev);
373 static void *eeh_pe_detach_dev(void *data, void *userdata)
375 struct eeh_pe *pe = (struct eeh_pe *)data;
376 struct eeh_dev *edev, *tmp;
378 eeh_pe_for_each_dev(pe, edev, tmp) {
379 if (!(edev->mode & EEH_DEV_DISCONNECTED))
382 edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
383 eeh_rmv_from_parent_pe(edev);
390 * eeh_reset_device - Perform actual reset of a pci slot
392 * @bus: PCI bus corresponding to the isolcated slot
394 * This routine must be called to do reset on the indicated PE.
395 * During the reset, udev might be invoked because those affected
396 * PCI devices will be removed and then added.
398 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
400 struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
401 struct timeval tstamp;
402 int cnt, rc, removed = 0;
404 /* pcibios will clear the counter; save the value */
405 cnt = pe->freeze_count;
409 * We don't remove the corresponding PE instances because
410 * we need the information afterwords. The attached EEH
411 * devices are expected to be attached soon when calling
412 * into pcibios_add_pci_devices().
414 eeh_pe_state_mark(pe, EEH_PE_KEEP);
416 pcibios_remove_pci_devices(bus);
418 eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
420 /* Reset the pci controller. (Asserts RST#; resets config space).
421 * Reconfigure bridges and devices. Don't try to bring the system
422 * up if the reset failed for some reason.
424 rc = eeh_reset_pe(pe);
429 eeh_ops->configure_bridge(pe);
430 eeh_pe_restore_bars(pe);
432 /* Give the system 5 seconds to finish running the user-space
433 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
434 * this is a hack, but if we don't do this, and try to bring
435 * the device up before the scripts have taken it down,
436 * potentially weird things happen.
439 pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
443 * The EEH device is still connected with its parent
444 * PE. We should disconnect it so the binding can be
445 * rebuilt when adding PCI devices.
447 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
448 pcibios_add_pci_devices(bus);
449 } else if (frozen_bus && removed) {
450 pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
453 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
454 pcibios_add_pci_devices(frozen_bus);
456 eeh_pe_state_clear(pe, EEH_PE_KEEP);
459 pe->freeze_count = cnt;
464 /* The longest amount of time to wait for a pci device
465 * to come back on line, in seconds.
467 #define MAX_WAIT_FOR_RECOVERY 150
469 static void eeh_handle_normal_event(struct eeh_pe *pe)
471 struct pci_bus *frozen_bus;
473 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
475 frozen_bus = eeh_pe_bus_get(pe);
477 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
478 __func__, pe->phb->global_number, pe->addr);
482 eeh_pe_update_time_stamp(pe);
484 if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES)
485 goto excess_failures;
486 pr_warning("EEH: This PCI device has failed %d times in the last hour\n",
489 /* Walk the various device drivers attached to this slot through
490 * a reset sequence, giving each an opportunity to do what it needs
491 * to accomplish the reset. Each child gets a report of the
492 * status ... if any child can't handle the reset, then the entire
493 * slot is dlpar removed and added.
495 pr_info("EEH: Notify device drivers to shutdown\n");
496 eeh_pe_dev_traverse(pe, eeh_report_error, &result);
498 /* Get the current PCI slot state. This can take a long time,
499 * sometimes over 3 seconds for certain systems.
501 rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
502 if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
503 pr_warning("EEH: Permanent failure\n");
507 /* Since rtas may enable MMIO when posting the error log,
508 * don't post the error log until after all dev drivers
509 * have been informed.
511 pr_info("EEH: Collect temporary log\n");
512 eeh_slot_error_detail(pe, EEH_LOG_TEMP);
514 /* If all device drivers were EEH-unaware, then shut
515 * down all of the device drivers, and hope they
516 * go down willingly, without panicing the system.
518 if (result == PCI_ERS_RESULT_NONE) {
519 pr_info("EEH: Reset with hotplug activity\n");
520 rc = eeh_reset_device(pe, frozen_bus);
522 pr_warning("%s: Unable to reset, err=%d\n",
528 /* If all devices reported they can proceed, then re-enable MMIO */
529 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
530 pr_info("EEH: Enable I/O for affected devices\n");
531 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
536 result = PCI_ERS_RESULT_NEED_RESET;
538 pr_info("EEH: Notify device drivers to resume I/O\n");
539 result = PCI_ERS_RESULT_NONE;
540 eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
544 /* If all devices reported they can proceed, then re-enable DMA */
545 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
546 pr_info("EEH: Enabled DMA for affected devices\n");
547 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
552 result = PCI_ERS_RESULT_NEED_RESET;
554 result = PCI_ERS_RESULT_RECOVERED;
557 /* If any device has a hard failure, then shut off everything. */
558 if (result == PCI_ERS_RESULT_DISCONNECT) {
559 pr_warning("EEH: Device driver gave up\n");
563 /* If any device called out for a reset, then reset the slot */
564 if (result == PCI_ERS_RESULT_NEED_RESET) {
565 pr_info("EEH: Reset without hotplug activity\n");
566 rc = eeh_reset_device(pe, NULL);
568 pr_warning("%s: Cannot reset, err=%d\n",
573 pr_info("EEH: Notify device drivers "
574 "the completion of reset\n");
575 result = PCI_ERS_RESULT_NONE;
576 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
579 /* All devices should claim they have recovered by now. */
580 if ((result != PCI_ERS_RESULT_RECOVERED) &&
581 (result != PCI_ERS_RESULT_NONE)) {
582 pr_warning("EEH: Not recovered\n");
586 /* Tell all device drivers that they can resume operations */
587 pr_info("EEH: Notify device driver to resume\n");
588 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
594 * About 90% of all real-life EEH failures in the field
595 * are due to poorly seated PCI cards. Only 10% or so are
596 * due to actual, failed cards.
598 pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
599 "last hour and has been permanently disabled.\n"
600 "Please try reseating or replacing it.\n",
601 pe->phb->global_number, pe->addr,
606 pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
607 "Please try reseating or replacing it\n",
608 pe->phb->global_number, pe->addr);
611 eeh_slot_error_detail(pe, EEH_LOG_PERM);
613 /* Notify all devices that they're about to go down. */
614 eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
616 /* Shut down the device drivers for good. */
618 pcibios_remove_pci_devices(frozen_bus);
621 static void eeh_handle_special_event(void)
623 struct eeh_pe *pe, *phb_pe;
625 struct pci_controller *hose, *tmp;
630 * The return value from next_error() has been classified as follows.
631 * It might be good to enumerate them. However, next_error() is only
632 * supported by PowerNV platform for now. So it would be fine to use
635 * 4 - Dead IOC 3 - Dead PHB
636 * 2 - Fenced PHB 1 - Frozen PE
640 rc = eeh_ops->next_error(&pe);
646 /* Mark all PHBs in dead state */
647 eeh_serialize_lock(&flags);
648 list_for_each_entry_safe(hose, tmp,
649 &hose_list, list_node) {
650 phb_pe = eeh_phb_pe_get(hose);
651 if (!phb_pe) continue;
653 eeh_pe_state_mark(phb_pe,
654 EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
656 eeh_serialize_unlock(flags);
658 /* Purge all events */
659 eeh_remove_event(NULL);
664 /* Mark the PE in fenced state */
665 eeh_serialize_lock(&flags);
667 eeh_pe_state_mark(pe,
668 EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
670 eeh_pe_state_mark(pe,
671 EEH_PE_ISOLATED | EEH_PE_RECOVERING);
672 eeh_serialize_unlock(flags);
674 /* Purge all events of the PHB */
675 eeh_remove_event(pe);
678 pr_err("%s: Invalid value %d from next_error()\n",
684 * For fenced PHB and frozen PE, it's handled as normal
685 * event. We have to remove the affected PHBs for dead
688 if (rc == 2 || rc == 1)
689 eeh_handle_normal_event(pe);
691 list_for_each_entry_safe(hose, tmp,
692 &hose_list, list_node) {
693 phb_pe = eeh_phb_pe_get(hose);
694 if (!phb_pe || !(phb_pe->state & EEH_PE_PHB_DEAD))
697 bus = eeh_pe_bus_get(phb_pe);
698 /* Notify all devices that they're about to go down. */
699 eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
700 pcibios_remove_pci_devices(bus);
706 * eeh_handle_event - Reset a PCI device after hard lockup.
709 * While PHB detects address or data parity errors on particular PCI
710 * slot, the associated PE will be frozen. Besides, DMA's occurring
711 * to wild addresses (which usually happen due to bugs in device
712 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
713 * #PERR or other misc PCI-related errors also can trigger EEH errors.
715 * Recovery process consists of unplugging the device driver (which
716 * generated hotplug events to userspace), then issuing a PCI #RST to
717 * the device, then reconfiguring the PCI config space for all bridges
718 * & devices under this slot, and then finally restarting the device
719 * drivers (which cause a second set of hotplug events to go out to
722 void eeh_handle_event(struct eeh_pe *pe)
725 eeh_handle_normal_event(pe);
727 eeh_handle_special_event();