~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/powerpc/kernel/eeh_driver.c

Version: ~ [ linux-5.3-rc5 ] ~ [ linux-5.2.9 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.67 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.139 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.72 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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
  5  *
  6  * All rights reserved.
  7  *
  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.
 12  *
 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.
 18  *
 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.
 22  *
 23  * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com>
 24  */
 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>
 36 
 37 /**
 38  * eeh_pcid_name - Retrieve name of PCI device driver
 39  * @pdev: PCI device
 40  *
 41  * This routine is used to retrieve the name of PCI device driver
 42  * if that's valid.
 43  */
 44 static inline const char *eeh_pcid_name(struct pci_dev *pdev)
 45 {
 46         if (pdev && pdev->dev.driver)
 47                 return pdev->dev.driver->name;
 48         return "";
 49 }
 50 
 51 /**
 52  * eeh_pcid_get - Get the PCI device driver
 53  * @pdev: PCI device
 54  *
 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.
 59  */
 60 static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev)
 61 {
 62         if (!pdev || !pdev->driver)
 63                 return NULL;
 64 
 65         if (!try_module_get(pdev->driver->driver.owner))
 66                 return NULL;
 67 
 68         return pdev->driver;
 69 }
 70 
 71 /**
 72  * eeh_pcid_put - Dereference on the PCI device driver
 73  * @pdev: PCI device
 74  *
 75  * The function is called to do dereference on the PCI device
 76  * driver of the indicated PCI device.
 77  */
 78 static inline void eeh_pcid_put(struct pci_dev *pdev)
 79 {
 80         if (!pdev || !pdev->driver)
 81                 return;
 82 
 83         module_put(pdev->driver->driver.owner);
 84 }
 85 
 86 /**
 87  * eeh_disable_irq - Disable interrupt for the recovering device
 88  * @dev: PCI device
 89  *
 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.
 95  */
 96 static void eeh_disable_irq(struct pci_dev *dev)
 97 {
 98         struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
 99 
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.
103          */
104         if (dev->msi_enabled || dev->msix_enabled)
105                 return;
106 
107         if (!irq_has_action(dev->irq))
108                 return;
109 
110         edev->mode |= EEH_DEV_IRQ_DISABLED;
111         disable_irq_nosync(dev->irq);
112 }
113 
114 /**
115  * eeh_enable_irq - Enable interrupt for the recovering device
116  * @dev: PCI device
117  *
118  * This routine must be called to enable interrupt while failed
119  * device could be resumed.
120  */
121 static void eeh_enable_irq(struct pci_dev *dev)
122 {
123         struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
124 
125         if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
126                 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
127                 /*
128                  * FIXME !!!!!
129                  *
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
134                  * into it.
135                  *
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
139                  * to avoid it.
140                  *
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.
144                  *
145                  *      tglx
146                  */
147                 if (irqd_irq_disabled(irq_get_irq_data(dev->irq)))
148                         enable_irq(dev->irq);
149         }
150 }
151 
152 static bool eeh_dev_removed(struct eeh_dev *edev)
153 {
154         /* EEH device removed ? */
155         if (!edev || (edev->mode & EEH_DEV_REMOVED))
156                 return true;
157 
158         return false;
159 }
160 
161 static void *eeh_dev_save_state(void *data, void *userdata)
162 {
163         struct eeh_dev *edev = data;
164         struct pci_dev *pdev;
165 
166         if (!edev)
167                 return NULL;
168 
169         pdev = eeh_dev_to_pci_dev(edev);
170         if (!pdev)
171                 return NULL;
172 
173         pci_save_state(pdev);
174         return NULL;
175 }
176 
177 /**
178  * eeh_report_error - Report pci error to each device driver
179  * @data: eeh device
180  * @userdata: return value
181  *
182  * Report an EEH error to each device driver, collect up and
183  * merge the device driver responses. Cumulative response
184  * passed back in "userdata".
185  */
186 static void *eeh_report_error(void *data, void *userdata)
187 {
188         struct eeh_dev *edev = (struct eeh_dev *)data;
189         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
190         enum pci_ers_result rc, *res = userdata;
191         struct pci_driver *driver;
192 
193         if (!dev || eeh_dev_removed(edev))
194                 return NULL;
195         dev->error_state = pci_channel_io_frozen;
196 
197         driver = eeh_pcid_get(dev);
198         if (!driver) return NULL;
199 
200         eeh_disable_irq(dev);
201 
202         if (!driver->err_handler ||
203             !driver->err_handler->error_detected) {
204                 eeh_pcid_put(dev);
205                 return NULL;
206         }
207 
208         rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
209 
210         /* A driver that needs a reset trumps all others */
211         if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
212         if (*res == PCI_ERS_RESULT_NONE) *res = rc;
213 
214         eeh_pcid_put(dev);
215         return NULL;
216 }
217 
218 /**
219  * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
220  * @data: eeh device
221  * @userdata: return value
222  *
223  * Tells each device driver that IO ports, MMIO and config space I/O
224  * are now enabled. Collects up and merges the device driver responses.
225  * Cumulative response passed back in "userdata".
226  */
227 static void *eeh_report_mmio_enabled(void *data, void *userdata)
228 {
229         struct eeh_dev *edev = (struct eeh_dev *)data;
230         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
231         enum pci_ers_result rc, *res = userdata;
232         struct pci_driver *driver;
233 
234         if (!dev || eeh_dev_removed(edev))
235                 return NULL;
236 
237         driver = eeh_pcid_get(dev);
238         if (!driver) return NULL;
239 
240         if (!driver->err_handler ||
241             !driver->err_handler->mmio_enabled ||
242             (edev->mode & EEH_DEV_NO_HANDLER)) {
243                 eeh_pcid_put(dev);
244                 return NULL;
245         }
246 
247         rc = driver->err_handler->mmio_enabled(dev);
248 
249         /* A driver that needs a reset trumps all others */
250         if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
251         if (*res == PCI_ERS_RESULT_NONE) *res = rc;
252 
253         eeh_pcid_put(dev);
254         return NULL;
255 }
256 
257 /**
258  * eeh_report_reset - Tell device that slot has been reset
259  * @data: eeh device
260  * @userdata: return value
261  *
262  * This routine must be called while EEH tries to reset particular
263  * PCI device so that the associated PCI device driver could take
264  * some actions, usually to save data the driver needs so that the
265  * driver can work again while the device is recovered.
266  */
267 static void *eeh_report_reset(void *data, void *userdata)
268 {
269         struct eeh_dev *edev = (struct eeh_dev *)data;
270         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
271         enum pci_ers_result rc, *res = userdata;
272         struct pci_driver *driver;
273 
274         if (!dev || eeh_dev_removed(edev))
275                 return NULL;
276         dev->error_state = pci_channel_io_normal;
277 
278         driver = eeh_pcid_get(dev);
279         if (!driver) return NULL;
280 
281         eeh_enable_irq(dev);
282 
283         if (!driver->err_handler ||
284             !driver->err_handler->slot_reset ||
285             (edev->mode & EEH_DEV_NO_HANDLER)) {
286                 eeh_pcid_put(dev);
287                 return NULL;
288         }
289 
290         rc = driver->err_handler->slot_reset(dev);
291         if ((*res == PCI_ERS_RESULT_NONE) ||
292             (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
293         if (*res == PCI_ERS_RESULT_DISCONNECT &&
294              rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
295 
296         eeh_pcid_put(dev);
297         return NULL;
298 }
299 
300 static void *eeh_dev_restore_state(void *data, void *userdata)
301 {
302         struct eeh_dev *edev = data;
303         struct pci_dev *pdev;
304 
305         if (!edev)
306                 return NULL;
307 
308         pdev = eeh_dev_to_pci_dev(edev);
309         if (!pdev)
310                 return NULL;
311 
312         pci_restore_state(pdev);
313         return NULL;
314 }
315 
316 /**
317  * eeh_report_resume - Tell device to resume normal operations
318  * @data: eeh device
319  * @userdata: return value
320  *
321  * This routine must be called to notify the device driver that it
322  * could resume so that the device driver can do some initialization
323  * to make the recovered device work again.
324  */
325 static void *eeh_report_resume(void *data, void *userdata)
326 {
327         struct eeh_dev *edev = (struct eeh_dev *)data;
328         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
329         struct pci_driver *driver;
330 
331         if (!dev || eeh_dev_removed(edev))
332                 return NULL;
333         dev->error_state = pci_channel_io_normal;
334 
335         driver = eeh_pcid_get(dev);
336         if (!driver) return NULL;
337 
338         eeh_enable_irq(dev);
339 
340         if (!driver->err_handler ||
341             !driver->err_handler->resume ||
342             (edev->mode & EEH_DEV_NO_HANDLER)) {
343                 edev->mode &= ~EEH_DEV_NO_HANDLER;
344                 eeh_pcid_put(dev);
345                 return NULL;
346         }
347 
348         driver->err_handler->resume(dev);
349 
350         eeh_pcid_put(dev);
351         return NULL;
352 }
353 
354 /**
355  * eeh_report_failure - Tell device driver that device is dead.
356  * @data: eeh device
357  * @userdata: return value
358  *
359  * This informs the device driver that the device is permanently
360  * dead, and that no further recovery attempts will be made on it.
361  */
362 static void *eeh_report_failure(void *data, void *userdata)
363 {
364         struct eeh_dev *edev = (struct eeh_dev *)data;
365         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
366         struct pci_driver *driver;
367 
368         if (!dev || eeh_dev_removed(edev))
369                 return NULL;
370         dev->error_state = pci_channel_io_perm_failure;
371 
372         driver = eeh_pcid_get(dev);
373         if (!driver) return NULL;
374 
375         eeh_disable_irq(dev);
376 
377         if (!driver->err_handler ||
378             !driver->err_handler->error_detected) {
379                 eeh_pcid_put(dev);
380                 return NULL;
381         }
382 
383         driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
384 
385         eeh_pcid_put(dev);
386         return NULL;
387 }
388 
389 static void *eeh_rmv_device(void *data, void *userdata)
390 {
391         struct pci_driver *driver;
392         struct eeh_dev *edev = (struct eeh_dev *)data;
393         struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
394         int *removed = (int *)userdata;
395 
396         /*
397          * Actually, we should remove the PCI bridges as well.
398          * However, that's lots of complexity to do that,
399          * particularly some of devices under the bridge might
400          * support EEH. So we just care about PCI devices for
401          * simplicity here.
402          */
403         if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
404                 return NULL;
405 
406         /*
407          * We rely on count-based pcibios_release_device() to
408          * detach permanently offlined PEs. Unfortunately, that's
409          * not reliable enough. We might have the permanently
410          * offlined PEs attached, but we needn't take care of
411          * them and their child devices.
412          */
413         if (eeh_dev_removed(edev))
414                 return NULL;
415 
416         driver = eeh_pcid_get(dev);
417         if (driver) {
418                 eeh_pcid_put(dev);
419                 if (driver->err_handler)
420                         return NULL;
421         }
422 
423         /* Remove it from PCI subsystem */
424         pr_debug("EEH: Removing %s without EEH sensitive driver\n",
425                  pci_name(dev));
426         edev->bus = dev->bus;
427         edev->mode |= EEH_DEV_DISCONNECTED;
428         (*removed)++;
429 
430         pci_lock_rescan_remove();
431         pci_stop_and_remove_bus_device(dev);
432         pci_unlock_rescan_remove();
433 
434         return NULL;
435 }
436 
437 static void *eeh_pe_detach_dev(void *data, void *userdata)
438 {
439         struct eeh_pe *pe = (struct eeh_pe *)data;
440         struct eeh_dev *edev, *tmp;
441 
442         eeh_pe_for_each_dev(pe, edev, tmp) {
443                 if (!(edev->mode & EEH_DEV_DISCONNECTED))
444                         continue;
445 
446                 edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
447                 eeh_rmv_from_parent_pe(edev);
448         }
449 
450         return NULL;
451 }
452 
453 /*
454  * Explicitly clear PE's frozen state for PowerNV where
455  * we have frozen PE until BAR restore is completed. It's
456  * harmless to clear it for pSeries. To be consistent with
457  * PE reset (for 3 times), we try to clear the frozen state
458  * for 3 times as well.
459  */
460 static void *__eeh_clear_pe_frozen_state(void *data, void *flag)
461 {
462         struct eeh_pe *pe = (struct eeh_pe *)data;
463         bool *clear_sw_state = flag;
464         int i, rc = 1;
465 
466         for (i = 0; rc && i < 3; i++)
467                 rc = eeh_unfreeze_pe(pe, clear_sw_state);
468 
469         /* Stop immediately on any errors */
470         if (rc) {
471                 pr_warn("%s: Failure %d unfreezing PHB#%x-PE#%x\n",
472                         __func__, rc, pe->phb->global_number, pe->addr);
473                 return (void *)pe;
474         }
475 
476         return NULL;
477 }
478 
479 static int eeh_clear_pe_frozen_state(struct eeh_pe *pe,
480                                      bool clear_sw_state)
481 {
482         void *rc;
483 
484         rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, &clear_sw_state);
485         if (!rc)
486                 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
487 
488         return rc ? -EIO : 0;
489 }
490 
491 int eeh_pe_reset_and_recover(struct eeh_pe *pe)
492 {
493         int result, ret;
494 
495         /* Bail if the PE is being recovered */
496         if (pe->state & EEH_PE_RECOVERING)
497                 return 0;
498 
499         /* Put the PE into recovery mode */
500         eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
501 
502         /* Save states */
503         eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL);
504 
505         /* Report error */
506         eeh_pe_dev_traverse(pe, eeh_report_error, &result);
507 
508         /* Issue reset */
509         ret = eeh_reset_pe(pe);
510         if (ret) {
511                 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
512                 return ret;
513         }
514 
515         /* Unfreeze the PE */
516         ret = eeh_clear_pe_frozen_state(pe, true);
517         if (ret) {
518                 eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
519                 return ret;
520         }
521 
522         /* Notify completion of reset */
523         eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
524 
525         /* Restore device state */
526         eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL);
527 
528         /* Resume */
529         eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
530 
531         /* Clear recovery mode */
532         eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
533 
534         return 0;
535 }
536 
537 /**
538  * eeh_reset_device - Perform actual reset of a pci slot
539  * @pe: EEH PE
540  * @bus: PCI bus corresponding to the isolcated slot
541  *
542  * This routine must be called to do reset on the indicated PE.
543  * During the reset, udev might be invoked because those affected
544  * PCI devices will be removed and then added.
545  */
546 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
547 {
548         struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
549         struct timeval tstamp;
550         int cnt, rc, removed = 0;
551 
552         /* pcibios will clear the counter; save the value */
553         cnt = pe->freeze_count;
554         tstamp = pe->tstamp;
555 
556         /*
557          * We don't remove the corresponding PE instances because
558          * we need the information afterwords. The attached EEH
559          * devices are expected to be attached soon when calling
560          * into pcibios_add_pci_devices().
561          */
562         eeh_pe_state_mark(pe, EEH_PE_KEEP);
563         if (bus) {
564                 pci_lock_rescan_remove();
565                 pcibios_remove_pci_devices(bus);
566                 pci_unlock_rescan_remove();
567         } else if (frozen_bus) {
568                 eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
569         }
570 
571         /*
572          * Reset the pci controller. (Asserts RST#; resets config space).
573          * Reconfigure bridges and devices. Don't try to bring the system
574          * up if the reset failed for some reason.
575          *
576          * During the reset, it's very dangerous to have uncontrolled PCI
577          * config accesses. So we prefer to block them. However, controlled
578          * PCI config accesses initiated from EEH itself are allowed.
579          */
580         rc = eeh_reset_pe(pe);
581         if (rc)
582                 return rc;
583 
584         pci_lock_rescan_remove();
585 
586         /* Restore PE */
587         eeh_ops->configure_bridge(pe);
588         eeh_pe_restore_bars(pe);
589 
590         /* Clear frozen state */
591         rc = eeh_clear_pe_frozen_state(pe, false);
592         if (rc)
593                 return rc;
594 
595         /* Give the system 5 seconds to finish running the user-space
596          * hotplug shutdown scripts, e.g. ifdown for ethernet.  Yes,
597          * this is a hack, but if we don't do this, and try to bring
598          * the device up before the scripts have taken it down,
599          * potentially weird things happen.
600          */
601         if (bus) {
602                 pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
603                 ssleep(5);
604 
605                 /*
606                  * The EEH device is still connected with its parent
607                  * PE. We should disconnect it so the binding can be
608                  * rebuilt when adding PCI devices.
609                  */
610                 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
611                 pcibios_add_pci_devices(bus);
612         } else if (frozen_bus && removed) {
613                 pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
614                 ssleep(5);
615 
616                 eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
617                 pcibios_add_pci_devices(frozen_bus);
618         }
619         eeh_pe_state_clear(pe, EEH_PE_KEEP);
620 
621         pe->tstamp = tstamp;
622         pe->freeze_count = cnt;
623 
624         pci_unlock_rescan_remove();
625         return 0;
626 }
627 
628 /* The longest amount of time to wait for a pci device
629  * to come back on line, in seconds.
630  */
631 #define MAX_WAIT_FOR_RECOVERY 300
632 
633 static void eeh_handle_normal_event(struct eeh_pe *pe)
634 {
635         struct pci_bus *frozen_bus;
636         int rc = 0;
637         enum pci_ers_result result = PCI_ERS_RESULT_NONE;
638 
639         frozen_bus = eeh_pe_bus_get(pe);
640         if (!frozen_bus) {
641                 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
642                         __func__, pe->phb->global_number, pe->addr);
643                 return;
644         }
645 
646         eeh_pe_update_time_stamp(pe);
647         pe->freeze_count++;
648         if (pe->freeze_count > eeh_max_freezes)
649                 goto excess_failures;
650         pr_warn("EEH: This PCI device has failed %d times in the last hour\n",
651                 pe->freeze_count);
652 
653         /* Walk the various device drivers attached to this slot through
654          * a reset sequence, giving each an opportunity to do what it needs
655          * to accomplish the reset.  Each child gets a report of the
656          * status ... if any child can't handle the reset, then the entire
657          * slot is dlpar removed and added.
658          */
659         pr_info("EEH: Notify device drivers to shutdown\n");
660         eeh_pe_dev_traverse(pe, eeh_report_error, &result);
661 
662         /* Get the current PCI slot state. This can take a long time,
663          * sometimes over 300 seconds for certain systems.
664          */
665         rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
666         if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
667                 pr_warn("EEH: Permanent failure\n");
668                 goto hard_fail;
669         }
670 
671         /* Since rtas may enable MMIO when posting the error log,
672          * don't post the error log until after all dev drivers
673          * have been informed.
674          */
675         pr_info("EEH: Collect temporary log\n");
676         eeh_slot_error_detail(pe, EEH_LOG_TEMP);
677 
678         /* If all device drivers were EEH-unaware, then shut
679          * down all of the device drivers, and hope they
680          * go down willingly, without panicing the system.
681          */
682         if (result == PCI_ERS_RESULT_NONE) {
683                 pr_info("EEH: Reset with hotplug activity\n");
684                 rc = eeh_reset_device(pe, frozen_bus);
685                 if (rc) {
686                         pr_warn("%s: Unable to reset, err=%d\n",
687                                 __func__, rc);
688                         goto hard_fail;
689                 }
690         }
691 
692         /* If all devices reported they can proceed, then re-enable MMIO */
693         if (result == PCI_ERS_RESULT_CAN_RECOVER) {
694                 pr_info("EEH: Enable I/O for affected devices\n");
695                 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
696 
697                 if (rc < 0)
698                         goto hard_fail;
699                 if (rc) {
700                         result = PCI_ERS_RESULT_NEED_RESET;
701                 } else {
702                         pr_info("EEH: Notify device drivers to resume I/O\n");
703                         eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
704                 }
705         }
706 
707         /* If all devices reported they can proceed, then re-enable DMA */
708         if (result == PCI_ERS_RESULT_CAN_RECOVER) {
709                 pr_info("EEH: Enabled DMA for affected devices\n");
710                 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
711 
712                 if (rc < 0)
713                         goto hard_fail;
714                 if (rc) {
715                         result = PCI_ERS_RESULT_NEED_RESET;
716                 } else {
717                         /*
718                          * We didn't do PE reset for the case. The PE
719                          * is still in frozen state. Clear it before
720                          * resuming the PE.
721                          */
722                         eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
723                         result = PCI_ERS_RESULT_RECOVERED;
724                 }
725         }
726 
727         /* If any device has a hard failure, then shut off everything. */
728         if (result == PCI_ERS_RESULT_DISCONNECT) {
729                 pr_warn("EEH: Device driver gave up\n");
730                 goto hard_fail;
731         }
732 
733         /* If any device called out for a reset, then reset the slot */
734         if (result == PCI_ERS_RESULT_NEED_RESET) {
735                 pr_info("EEH: Reset without hotplug activity\n");
736                 rc = eeh_reset_device(pe, NULL);
737                 if (rc) {
738                         pr_warn("%s: Cannot reset, err=%d\n",
739                                 __func__, rc);
740                         goto hard_fail;
741                 }
742 
743                 pr_info("EEH: Notify device drivers "
744                         "the completion of reset\n");
745                 result = PCI_ERS_RESULT_NONE;
746                 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
747         }
748 
749         /* All devices should claim they have recovered by now. */
750         if ((result != PCI_ERS_RESULT_RECOVERED) &&
751             (result != PCI_ERS_RESULT_NONE)) {
752                 pr_warn("EEH: Not recovered\n");
753                 goto hard_fail;
754         }
755 
756         /* Tell all device drivers that they can resume operations */
757         pr_info("EEH: Notify device driver to resume\n");
758         eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
759 
760         return;
761 
762 excess_failures:
763         /*
764          * About 90% of all real-life EEH failures in the field
765          * are due to poorly seated PCI cards. Only 10% or so are
766          * due to actual, failed cards.
767          */
768         pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
769                "last hour and has been permanently disabled.\n"
770                "Please try reseating or replacing it.\n",
771                 pe->phb->global_number, pe->addr,
772                 pe->freeze_count);
773         goto perm_error;
774 
775 hard_fail:
776         pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
777                "Please try reseating or replacing it\n",
778                 pe->phb->global_number, pe->addr);
779 
780 perm_error:
781         eeh_slot_error_detail(pe, EEH_LOG_PERM);
782 
783         /* Notify all devices that they're about to go down. */
784         eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
785 
786         /* Mark the PE to be removed permanently */
787         eeh_pe_state_mark(pe, EEH_PE_REMOVED);
788 
789         /*
790          * Shut down the device drivers for good. We mark
791          * all removed devices correctly to avoid access
792          * the their PCI config any more.
793          */
794         if (frozen_bus) {
795                 eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
796 
797                 pci_lock_rescan_remove();
798                 pcibios_remove_pci_devices(frozen_bus);
799                 pci_unlock_rescan_remove();
800         }
801 }
802 
803 static void eeh_handle_special_event(void)
804 {
805         struct eeh_pe *pe, *phb_pe;
806         struct pci_bus *bus;
807         struct pci_controller *hose;
808         unsigned long flags;
809         int rc;
810 
811 
812         do {
813                 rc = eeh_ops->next_error(&pe);
814 
815                 switch (rc) {
816                 case EEH_NEXT_ERR_DEAD_IOC:
817                         /* Mark all PHBs in dead state */
818                         eeh_serialize_lock(&flags);
819 
820                         /* Purge all events */
821                         eeh_remove_event(NULL, true);
822 
823                         list_for_each_entry(hose, &hose_list, list_node) {
824                                 phb_pe = eeh_phb_pe_get(hose);
825                                 if (!phb_pe) continue;
826 
827                                 eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
828                         }
829 
830                         eeh_serialize_unlock(flags);
831 
832                         break;
833                 case EEH_NEXT_ERR_FROZEN_PE:
834                 case EEH_NEXT_ERR_FENCED_PHB:
835                 case EEH_NEXT_ERR_DEAD_PHB:
836                         /* Mark the PE in fenced state */
837                         eeh_serialize_lock(&flags);
838 
839                         /* Purge all events of the PHB */
840                         eeh_remove_event(pe, true);
841 
842                         if (rc == EEH_NEXT_ERR_DEAD_PHB)
843                                 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
844                         else
845                                 eeh_pe_state_mark(pe,
846                                         EEH_PE_ISOLATED | EEH_PE_RECOVERING);
847 
848                         eeh_serialize_unlock(flags);
849 
850                         break;
851                 case EEH_NEXT_ERR_NONE:
852                         return;
853                 default:
854                         pr_warn("%s: Invalid value %d from next_error()\n",
855                                 __func__, rc);
856                         return;
857                 }
858 
859                 /*
860                  * For fenced PHB and frozen PE, it's handled as normal
861                  * event. We have to remove the affected PHBs for dead
862                  * PHB and IOC
863                  */
864                 if (rc == EEH_NEXT_ERR_FROZEN_PE ||
865                     rc == EEH_NEXT_ERR_FENCED_PHB) {
866                         eeh_handle_normal_event(pe);
867                         eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
868                 } else {
869                         pci_lock_rescan_remove();
870                         list_for_each_entry(hose, &hose_list, list_node) {
871                                 phb_pe = eeh_phb_pe_get(hose);
872                                 if (!phb_pe ||
873                                     !(phb_pe->state & EEH_PE_ISOLATED) ||
874                                     (phb_pe->state & EEH_PE_RECOVERING))
875                                         continue;
876 
877                                 /* Notify all devices to be down */
878                                 bus = eeh_pe_bus_get(phb_pe);
879                                 eeh_pe_dev_traverse(pe,
880                                         eeh_report_failure, NULL);
881                                 pcibios_remove_pci_devices(bus);
882                         }
883                         pci_unlock_rescan_remove();
884                 }
885 
886                 /*
887                  * If we have detected dead IOC, we needn't proceed
888                  * any more since all PHBs would have been removed
889                  */
890                 if (rc == EEH_NEXT_ERR_DEAD_IOC)
891                         break;
892         } while (rc != EEH_NEXT_ERR_NONE);
893 }
894 
895 /**
896  * eeh_handle_event - Reset a PCI device after hard lockup.
897  * @pe: EEH PE
898  *
899  * While PHB detects address or data parity errors on particular PCI
900  * slot, the associated PE will be frozen. Besides, DMA's occurring
901  * to wild addresses (which usually happen due to bugs in device
902  * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
903  * #PERR or other misc PCI-related errors also can trigger EEH errors.
904  *
905  * Recovery process consists of unplugging the device driver (which
906  * generated hotplug events to userspace), then issuing a PCI #RST to
907  * the device, then reconfiguring the PCI config space for all bridges
908  * & devices under this slot, and then finally restarting the device
909  * drivers (which cause a second set of hotplug events to go out to
910  * userspace).
911  */
912 void eeh_handle_event(struct eeh_pe *pe)
913 {
914         if (pe)
915                 eeh_handle_normal_event(pe);
916         else
917                 eeh_handle_special_event();
918 }
919 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp