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 */
eeh_pcid_name(struct pci_dev * pdev)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 */
eeh_pcid_get(struct pci_dev * pdev)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 */
eeh_pcid_put(struct pci_dev * pdev)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 #if 0
87 static void print_device_node_tree(struct pci_dn *pdn, int dent)
88 {
89 int i;
90 struct device_node *pc;
91
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;
104 }
105 }
106 #endif
107
108 /**
109 * eeh_disable_irq - Disable interrupt for the recovering device
110 * @dev: PCI device
111 *
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.
117 */
eeh_disable_irq(struct pci_dev * dev)118 static void eeh_disable_irq(struct pci_dev *dev)
119 {
120 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
121
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.
125 */
126 if (dev->msi_enabled || dev->msix_enabled)
127 return;
128
129 if (!irq_has_action(dev->irq))
130 return;
131
132 edev->mode |= EEH_DEV_IRQ_DISABLED;
133 disable_irq_nosync(dev->irq);
134 }
135
136 /**
137 * eeh_enable_irq - Enable interrupt for the recovering device
138 * @dev: PCI device
139 *
140 * This routine must be called to enable interrupt while failed
141 * device could be resumed.
142 */
eeh_enable_irq(struct pci_dev * dev)143 static void eeh_enable_irq(struct pci_dev *dev)
144 {
145 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
146
147 if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
148 edev->mode &= ~EEH_DEV_IRQ_DISABLED;
149 enable_irq(dev->irq);
150 }
151 }
152
153 /**
154 * eeh_report_error - Report pci error to each device driver
155 * @data: eeh device
156 * @userdata: return value
157 *
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".
161 */
eeh_report_error(void * data,void * userdata)162 static void *eeh_report_error(void *data, void *userdata)
163 {
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;
168
169 /* We might not have the associated PCI device,
170 * then we should continue for next one.
171 */
172 if (!dev) return NULL;
173 dev->error_state = pci_channel_io_frozen;
174
175 driver = eeh_pcid_get(dev);
176 if (!driver) return NULL;
177
178 eeh_disable_irq(dev);
179
180 if (!driver->err_handler ||
181 !driver->err_handler->error_detected) {
182 eeh_pcid_put(dev);
183 return NULL;
184 }
185
186 rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
187
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;
191
192 eeh_pcid_put(dev);
193 return NULL;
194 }
195
196 /**
197 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
198 * @data: eeh device
199 * @userdata: return value
200 *
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".
204 */
eeh_report_mmio_enabled(void * data,void * userdata)205 static void *eeh_report_mmio_enabled(void *data, void *userdata)
206 {
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;
211
212 driver = eeh_pcid_get(dev);
213 if (!driver) return NULL;
214
215 if (!driver->err_handler ||
216 !driver->err_handler->mmio_enabled) {
217 eeh_pcid_put(dev);
218 return NULL;
219 }
220
221 rc = driver->err_handler->mmio_enabled(dev);
222
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;
226
227 eeh_pcid_put(dev);
228 return NULL;
229 }
230
231 /**
232 * eeh_report_reset - Tell device that slot has been reset
233 * @data: eeh device
234 * @userdata: return value
235 *
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.
240 */
eeh_report_reset(void * data,void * userdata)241 static void *eeh_report_reset(void *data, void *userdata)
242 {
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;
247
248 if (!dev) return NULL;
249 dev->error_state = pci_channel_io_normal;
250
251 driver = eeh_pcid_get(dev);
252 if (!driver) return NULL;
253
254 eeh_enable_irq(dev);
255
256 if (!driver->err_handler ||
257 !driver->err_handler->slot_reset) {
258 eeh_pcid_put(dev);
259 return NULL;
260 }
261
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;
267
268 eeh_pcid_put(dev);
269 return NULL;
270 }
271
272 /**
273 * eeh_report_resume - Tell device to resume normal operations
274 * @data: eeh device
275 * @userdata: return value
276 *
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.
280 */
eeh_report_resume(void * data,void * userdata)281 static void *eeh_report_resume(void *data, void *userdata)
282 {
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;
286
287 if (!dev) return NULL;
288 dev->error_state = pci_channel_io_normal;
289
290 driver = eeh_pcid_get(dev);
291 if (!driver) return NULL;
292
293 eeh_enable_irq(dev);
294
295 if (!driver->err_handler ||
296 !driver->err_handler->resume) {
297 eeh_pcid_put(dev);
298 return NULL;
299 }
300
301 driver->err_handler->resume(dev);
302
303 eeh_pcid_put(dev);
304 return NULL;
305 }
306
307 /**
308 * eeh_report_failure - Tell device driver that device is dead.
309 * @data: eeh device
310 * @userdata: return value
311 *
312 * This informs the device driver that the device is permanently
313 * dead, and that no further recovery attempts will be made on it.
314 */
eeh_report_failure(void * data,void * userdata)315 static void *eeh_report_failure(void *data, void *userdata)
316 {
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;
320
321 if (!dev) return NULL;
322 dev->error_state = pci_channel_io_perm_failure;
323
324 driver = eeh_pcid_get(dev);
325 if (!driver) return NULL;
326
327 eeh_disable_irq(dev);
328
329 if (!driver->err_handler ||
330 !driver->err_handler->error_detected) {
331 eeh_pcid_put(dev);
332 return NULL;
333 }
334
335 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
336
337 eeh_pcid_put(dev);
338 return NULL;
339 }
340
341 /**
342 * eeh_reset_device - Perform actual reset of a pci slot
343 * @pe: EEH PE
344 * @bus: PCI bus corresponding to the isolcated slot
345 *
346 * This routine must be called to do reset on the indicated PE.
347 * During the reset, udev might be invoked because those affected
348 * PCI devices will be removed and then added.
349 */
eeh_reset_device(struct eeh_pe * pe,struct pci_bus * bus)350 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
351 {
352 int cnt, rc;
353
354 /* pcibios will clear the counter; save the value */
355 cnt = pe->freeze_count;
356
357 /*
358 * We don't remove the corresponding PE instances because
359 * we need the information afterwords. The attached EEH
360 * devices are expected to be attached soon when calling
361 * into pcibios_add_pci_devices().
362 */
363 if (bus)
364 __pcibios_remove_pci_devices(bus, 0);
365
366 /* Reset the pci controller. (Asserts RST#; resets config space).
367 * Reconfigure bridges and devices. Don't try to bring the system
368 * up if the reset failed for some reason.
369 */
370 rc = eeh_reset_pe(pe);
371 if (rc)
372 return rc;
373
374 /* Restore PE */
375 eeh_ops->configure_bridge(pe);
376 eeh_pe_restore_bars(pe);
377
378 /* Give the system 5 seconds to finish running the user-space
379 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
380 * this is a hack, but if we don't do this, and try to bring
381 * the device up before the scripts have taken it down,
382 * potentially weird things happen.
383 */
384 if (bus) {
385 ssleep(5);
386 pcibios_add_pci_devices(bus);
387 }
388 pe->freeze_count = cnt;
389
390 return 0;
391 }
392
393 /* The longest amount of time to wait for a pci device
394 * to come back on line, in seconds.
395 */
396 #define MAX_WAIT_FOR_RECOVERY 150
397
398 /**
399 * eeh_handle_event - Reset a PCI device after hard lockup.
400 * @pe: EEH PE
401 *
402 * While PHB detects address or data parity errors on particular PCI
403 * slot, the associated PE will be frozen. Besides, DMA's occurring
404 * to wild addresses (which usually happen due to bugs in device
405 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
406 * #PERR or other misc PCI-related errors also can trigger EEH errors.
407 *
408 * Recovery process consists of unplugging the device driver (which
409 * generated hotplug events to userspace), then issuing a PCI #RST to
410 * the device, then reconfiguring the PCI config space for all bridges
411 * & devices under this slot, and then finally restarting the device
412 * drivers (which cause a second set of hotplug events to go out to
413 * userspace).
414 */
eeh_handle_event(struct eeh_pe * pe)415 void eeh_handle_event(struct eeh_pe *pe)
416 {
417 struct pci_bus *frozen_bus;
418 int rc = 0;
419 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
420
421 frozen_bus = eeh_pe_bus_get(pe);
422 if (!frozen_bus) {
423 pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
424 __func__, pe->phb->global_number, pe->addr);
425 return;
426 }
427
428 pe->freeze_count++;
429 if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES)
430 goto excess_failures;
431 pr_warning("EEH: This PCI device has failed %d times in the last hour\n",
432 pe->freeze_count);
433
434 /* Walk the various device drivers attached to this slot through
435 * a reset sequence, giving each an opportunity to do what it needs
436 * to accomplish the reset. Each child gets a report of the
437 * status ... if any child can't handle the reset, then the entire
438 * slot is dlpar removed and added.
439 */
440 eeh_pe_dev_traverse(pe, eeh_report_error, &result);
441
442 /* Get the current PCI slot state. This can take a long time,
443 * sometimes over 3 seconds for certain systems.
444 */
445 rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
446 if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
447 printk(KERN_WARNING "EEH: Permanent failure\n");
448 goto hard_fail;
449 }
450
451 /* Since rtas may enable MMIO when posting the error log,
452 * don't post the error log until after all dev drivers
453 * have been informed.
454 */
455 eeh_slot_error_detail(pe, EEH_LOG_TEMP);
456
457 /* If all device drivers were EEH-unaware, then shut
458 * down all of the device drivers, and hope they
459 * go down willingly, without panicing the system.
460 */
461 if (result == PCI_ERS_RESULT_NONE) {
462 rc = eeh_reset_device(pe, frozen_bus);
463 if (rc) {
464 printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc);
465 goto hard_fail;
466 }
467 }
468
469 /* If all devices reported they can proceed, then re-enable MMIO */
470 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
471 rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
472
473 if (rc < 0)
474 goto hard_fail;
475 if (rc) {
476 result = PCI_ERS_RESULT_NEED_RESET;
477 } else {
478 result = PCI_ERS_RESULT_NONE;
479 eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
480 }
481 }
482
483 /* If all devices reported they can proceed, then re-enable DMA */
484 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
485 rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
486
487 if (rc < 0)
488 goto hard_fail;
489 if (rc)
490 result = PCI_ERS_RESULT_NEED_RESET;
491 else
492 result = PCI_ERS_RESULT_RECOVERED;
493 }
494
495 /* If any device has a hard failure, then shut off everything. */
496 if (result == PCI_ERS_RESULT_DISCONNECT) {
497 printk(KERN_WARNING "EEH: Device driver gave up\n");
498 goto hard_fail;
499 }
500
501 /* If any device called out for a reset, then reset the slot */
502 if (result == PCI_ERS_RESULT_NEED_RESET) {
503 rc = eeh_reset_device(pe, NULL);
504 if (rc) {
505 printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc);
506 goto hard_fail;
507 }
508 result = PCI_ERS_RESULT_NONE;
509 eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
510 }
511
512 /* All devices should claim they have recovered by now. */
513 if ((result != PCI_ERS_RESULT_RECOVERED) &&
514 (result != PCI_ERS_RESULT_NONE)) {
515 printk(KERN_WARNING "EEH: Not recovered\n");
516 goto hard_fail;
517 }
518
519 /* Tell all device drivers that they can resume operations */
520 eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
521
522 return;
523
524 excess_failures:
525 /*
526 * About 90% of all real-life EEH failures in the field
527 * are due to poorly seated PCI cards. Only 10% or so are
528 * due to actual, failed cards.
529 */
530 pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
531 "last hour and has been permanently disabled.\n"
532 "Please try reseating or replacing it.\n",
533 pe->phb->global_number, pe->addr,
534 pe->freeze_count);
535 goto perm_error;
536
537 hard_fail:
538 pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
539 "Please try reseating or replacing it\n",
540 pe->phb->global_number, pe->addr);
541
542 perm_error:
543 eeh_slot_error_detail(pe, EEH_LOG_PERM);
544
545 /* Notify all devices that they're about to go down. */
546 eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
547
548 /* Shut down the device drivers for good. */
549 if (frozen_bus)
550 pcibios_remove_pci_devices(frozen_bus);
551 }
552
553