Commit | Line | Data |
---|---|---|
77bd7415 LV |
1 | /* |
2 | * PCI Error Recovery Driver for RPA-compliant PPC64 platform. | |
3c8c90ab LV |
3 | * Copyright IBM Corp. 2004 2005 |
4 | * Copyright Linas Vepstas <linas@linas.org> 2004, 2005 | |
77bd7415 LV |
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 | * | |
3c8c90ab | 23 | * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com> |
77bd7415 LV |
24 | */ |
25 | #include <linux/delay.h> | |
77bd7415 | 26 | #include <linux/interrupt.h> |
ac325acd | 27 | #include <linux/irq.h> |
feadf7c0 | 28 | #include <linux/module.h> |
77bd7415 LV |
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 | ||
29f8bf1b GS |
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 | */ | |
40a7cd92 | 44 | static inline const char *eeh_pcid_name(struct pci_dev *pdev) |
77bd7415 | 45 | { |
273d2803 | 46 | if (pdev && pdev->dev.driver) |
77bd7415 LV |
47 | return pdev->dev.driver->name; |
48 | return ""; | |
49 | } | |
50 | ||
feadf7c0 GS |
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 | ||
dcfcfe75 | 86 | #if 0 |
8e01520c | 87 | static void print_device_node_tree(struct pci_dn *pdn, int dent) |
77bd7415 LV |
88 | { |
89 | int i; | |
8e01520c AM |
90 | struct device_node *pc; |
91 | ||
92 | if (!pdn) | |
93 | return; | |
94 | for (i = 0; i < dent; i++) | |
77bd7415 LV |
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; | |
8e01520c | 100 | pc = pdn->node->child; |
77bd7415 LV |
101 | while (pc) { |
102 | print_device_node_tree(PCI_DN(pc), dent); | |
103 | pc = pc->sibling; | |
104 | } | |
105 | } | |
106 | #endif | |
107 | ||
8535ef05 | 108 | /** |
29f8bf1b GS |
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. | |
8535ef05 MM |
117 | */ |
118 | static void eeh_disable_irq(struct pci_dev *dev) | |
119 | { | |
40a7cd92 | 120 | struct eeh_dev *edev = pci_dev_to_eeh_dev(dev); |
8535ef05 MM |
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. | |
29f8bf1b | 125 | */ |
8535ef05 MM |
126 | if (dev->msi_enabled || dev->msix_enabled) |
127 | return; | |
128 | ||
59e3f837 | 129 | if (!irq_has_action(dev->irq)) |
8535ef05 MM |
130 | return; |
131 | ||
dbbceee1 | 132 | edev->mode |= EEH_DEV_IRQ_DISABLED; |
8535ef05 MM |
133 | disable_irq_nosync(dev->irq); |
134 | } | |
135 | ||
136 | /** | |
29f8bf1b GS |
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. | |
8535ef05 MM |
142 | */ |
143 | static void eeh_enable_irq(struct pci_dev *dev) | |
144 | { | |
40a7cd92 | 145 | struct eeh_dev *edev = pci_dev_to_eeh_dev(dev); |
8535ef05 | 146 | |
dbbceee1 GS |
147 | if ((edev->mode) & EEH_DEV_IRQ_DISABLED) { |
148 | edev->mode &= ~EEH_DEV_IRQ_DISABLED; | |
8535ef05 MM |
149 | enable_irq(dev->irq); |
150 | } | |
151 | } | |
152 | ||
cb5b5624 | 153 | /** |
29f8bf1b | 154 | * eeh_report_error - Report pci error to each device driver |
9b3c76f0 | 155 | * @data: eeh device |
29f8bf1b | 156 | * @userdata: return value |
a84f273c GS |
157 | * |
158 | * Report an EEH error to each device driver, collect up and | |
159 | * merge the device driver responses. Cumulative response | |
cb5b5624 | 160 | * passed back in "userdata". |
77bd7415 | 161 | */ |
9b3c76f0 | 162 | static void *eeh_report_error(void *data, void *userdata) |
77bd7415 | 163 | { |
9b3c76f0 GS |
164 | struct eeh_dev *edev = (struct eeh_dev *)data; |
165 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
18eb3b39 | 166 | enum pci_ers_result rc, *res = userdata; |
feadf7c0 | 167 | struct pci_driver *driver; |
77bd7415 | 168 | |
9b3c76f0 GS |
169 | /* We might not have the associated PCI device, |
170 | * then we should continue for next one. | |
171 | */ | |
172 | if (!dev) return NULL; | |
77bd7415 LV |
173 | dev->error_state = pci_channel_io_frozen; |
174 | ||
feadf7c0 GS |
175 | driver = eeh_pcid_get(dev); |
176 | if (!driver) return NULL; | |
77bd7415 | 177 | |
8535ef05 MM |
178 | eeh_disable_irq(dev); |
179 | ||
6a1ca373 | 180 | if (!driver->err_handler || |
feadf7c0 GS |
181 | !driver->err_handler->error_detected) { |
182 | eeh_pcid_put(dev); | |
9b3c76f0 | 183 | return NULL; |
feadf7c0 | 184 | } |
77bd7415 | 185 | |
29f8bf1b | 186 | rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen); |
2a50f144 LV |
187 | |
188 | /* A driver that needs a reset trumps all others */ | |
189 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
18eb3b39 | 190 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 191 | |
feadf7c0 | 192 | eeh_pcid_put(dev); |
9b3c76f0 | 193 | return NULL; |
6a1ca373 LV |
194 | } |
195 | ||
196 | /** | |
29f8bf1b | 197 | * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled |
9b3c76f0 | 198 | * @data: eeh device |
29f8bf1b | 199 | * @userdata: return value |
6a1ca373 | 200 | * |
638799b3 LV |
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". | |
6a1ca373 | 204 | */ |
9b3c76f0 | 205 | static void *eeh_report_mmio_enabled(void *data, void *userdata) |
6a1ca373 | 206 | { |
9b3c76f0 GS |
207 | struct eeh_dev *edev = (struct eeh_dev *)data; |
208 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 209 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 210 | struct pci_driver *driver; |
6a1ca373 | 211 | |
feadf7c0 GS |
212 | driver = eeh_pcid_get(dev); |
213 | if (!driver) return NULL; | |
9b3c76f0 | 214 | |
feadf7c0 GS |
215 | if (!driver->err_handler || |
216 | !driver->err_handler->mmio_enabled) { | |
217 | eeh_pcid_put(dev); | |
9b3c76f0 | 218 | return NULL; |
feadf7c0 | 219 | } |
6a1ca373 | 220 | |
29f8bf1b | 221 | rc = driver->err_handler->mmio_enabled(dev); |
2a50f144 LV |
222 | |
223 | /* A driver that needs a reset trumps all others */ | |
224 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
6a1ca373 | 225 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 226 | |
feadf7c0 | 227 | eeh_pcid_put(dev); |
9b3c76f0 | 228 | return NULL; |
77bd7415 LV |
229 | } |
230 | ||
cb5b5624 | 231 | /** |
29f8bf1b | 232 | * eeh_report_reset - Tell device that slot has been reset |
9b3c76f0 | 233 | * @data: eeh device |
29f8bf1b GS |
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. | |
77bd7415 | 240 | */ |
9b3c76f0 | 241 | static void *eeh_report_reset(void *data, void *userdata) |
77bd7415 | 242 | { |
9b3c76f0 GS |
243 | struct eeh_dev *edev = (struct eeh_dev *)data; |
244 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 245 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 246 | struct pci_driver *driver; |
77bd7415 | 247 | |
feadf7c0 | 248 | if (!dev) return NULL; |
c58dc575 MM |
249 | dev->error_state = pci_channel_io_normal; |
250 | ||
feadf7c0 GS |
251 | driver = eeh_pcid_get(dev); |
252 | if (!driver) return NULL; | |
253 | ||
8535ef05 MM |
254 | eeh_enable_irq(dev); |
255 | ||
6a1ca373 | 256 | if (!driver->err_handler || |
feadf7c0 GS |
257 | !driver->err_handler->slot_reset) { |
258 | eeh_pcid_put(dev); | |
9b3c76f0 | 259 | return NULL; |
feadf7c0 | 260 | } |
77bd7415 | 261 | |
6a1ca373 | 262 | rc = driver->err_handler->slot_reset(dev); |
5794dbcb LV |
263 | if ((*res == PCI_ERS_RESULT_NONE) || |
264 | (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; | |
6a1ca373 LV |
265 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
266 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
70298c6e | 267 | |
feadf7c0 | 268 | eeh_pcid_put(dev); |
9b3c76f0 | 269 | return NULL; |
77bd7415 LV |
270 | } |
271 | ||
cb5b5624 | 272 | /** |
29f8bf1b | 273 | * eeh_report_resume - Tell device to resume normal operations |
9b3c76f0 | 274 | * @data: eeh device |
29f8bf1b GS |
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. | |
cb5b5624 | 280 | */ |
9b3c76f0 | 281 | static void *eeh_report_resume(void *data, void *userdata) |
77bd7415 | 282 | { |
9b3c76f0 GS |
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; | |
77bd7415 LV |
288 | dev->error_state = pci_channel_io_normal; |
289 | ||
feadf7c0 GS |
290 | driver = eeh_pcid_get(dev); |
291 | if (!driver) return NULL; | |
d0e70341 | 292 | |
8535ef05 MM |
293 | eeh_enable_irq(dev); |
294 | ||
d0e70341 | 295 | if (!driver->err_handler || |
feadf7c0 GS |
296 | !driver->err_handler->resume) { |
297 | eeh_pcid_put(dev); | |
9b3c76f0 | 298 | return NULL; |
feadf7c0 | 299 | } |
77bd7415 LV |
300 | |
301 | driver->err_handler->resume(dev); | |
70298c6e | 302 | |
feadf7c0 | 303 | eeh_pcid_put(dev); |
9b3c76f0 | 304 | return NULL; |
77bd7415 LV |
305 | } |
306 | ||
cb5b5624 | 307 | /** |
29f8bf1b | 308 | * eeh_report_failure - Tell device driver that device is dead. |
9b3c76f0 | 309 | * @data: eeh device |
29f8bf1b | 310 | * @userdata: return value |
cb5b5624 LV |
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 | */ | |
9b3c76f0 | 315 | static void *eeh_report_failure(void *data, void *userdata) |
77bd7415 | 316 | { |
9b3c76f0 GS |
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; | |
77bd7415 LV |
322 | dev->error_state = pci_channel_io_perm_failure; |
323 | ||
feadf7c0 GS |
324 | driver = eeh_pcid_get(dev); |
325 | if (!driver) return NULL; | |
77bd7415 | 326 | |
8535ef05 MM |
327 | eeh_disable_irq(dev); |
328 | ||
329 | if (!driver->err_handler || | |
feadf7c0 GS |
330 | !driver->err_handler->error_detected) { |
331 | eeh_pcid_put(dev); | |
9b3c76f0 | 332 | return NULL; |
feadf7c0 | 333 | } |
8535ef05 | 334 | |
77bd7415 | 335 | driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); |
70298c6e | 336 | |
feadf7c0 | 337 | eeh_pcid_put(dev); |
9b3c76f0 | 338 | return NULL; |
77bd7415 LV |
339 | } |
340 | ||
f5c57710 GS |
341 | static void *eeh_rmv_device(void *data, void *userdata) |
342 | { | |
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; | |
347 | ||
348 | /* | |
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 | |
353 | * simplicity here. | |
354 | */ | |
355 | if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) | |
356 | return NULL; | |
357 | driver = eeh_pcid_get(dev); | |
358 | if (driver && driver->err_handler) | |
359 | return NULL; | |
360 | ||
361 | /* Remove it from PCI subsystem */ | |
362 | pr_debug("EEH: Removing %s without EEH sensitive driver\n", | |
363 | pci_name(dev)); | |
364 | edev->bus = dev->bus; | |
365 | edev->mode |= EEH_DEV_DISCONNECTED; | |
366 | (*removed)++; | |
367 | ||
368 | pci_stop_and_remove_bus_device(dev); | |
369 | ||
370 | return NULL; | |
371 | } | |
372 | ||
373 | static void *eeh_pe_detach_dev(void *data, void *userdata) | |
374 | { | |
375 | struct eeh_pe *pe = (struct eeh_pe *)data; | |
376 | struct eeh_dev *edev, *tmp; | |
377 | ||
378 | eeh_pe_for_each_dev(pe, edev, tmp) { | |
379 | if (!(edev->mode & EEH_DEV_DISCONNECTED)) | |
380 | continue; | |
381 | ||
382 | edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED); | |
383 | eeh_rmv_from_parent_pe(edev); | |
384 | } | |
385 | ||
386 | return NULL; | |
387 | } | |
388 | ||
77bd7415 | 389 | /** |
29f8bf1b | 390 | * eeh_reset_device - Perform actual reset of a pci slot |
9b3c76f0 | 391 | * @pe: EEH PE |
29f8bf1b | 392 | * @bus: PCI bus corresponding to the isolcated slot |
77bd7415 | 393 | * |
29f8bf1b GS |
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. | |
77bd7415 | 397 | */ |
9b3c76f0 | 398 | static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus) |
77bd7415 | 399 | { |
f5c57710 | 400 | struct pci_bus *frozen_bus = eeh_pe_bus_get(pe); |
5a71978e | 401 | struct timeval tstamp; |
f5c57710 | 402 | int cnt, rc, removed = 0; |
42405456 LV |
403 | |
404 | /* pcibios will clear the counter; save the value */ | |
9b3c76f0 | 405 | cnt = pe->freeze_count; |
5a71978e | 406 | tstamp = pe->tstamp; |
42405456 | 407 | |
20ee6a97 GS |
408 | /* |
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(). | |
413 | */ | |
f5c57710 GS |
414 | eeh_pe_state_mark(pe, EEH_PE_KEEP); |
415 | if (bus) | |
807a827d | 416 | pcibios_remove_pci_devices(bus); |
f5c57710 GS |
417 | else if (frozen_bus) |
418 | eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed); | |
77bd7415 LV |
419 | |
420 | /* Reset the pci controller. (Asserts RST#; resets config space). | |
b6495c0c | 421 | * Reconfigure bridges and devices. Don't try to bring the system |
29f8bf1b GS |
422 | * up if the reset failed for some reason. |
423 | */ | |
9b3c76f0 | 424 | rc = eeh_reset_pe(pe); |
b6495c0c LV |
425 | if (rc) |
426 | return rc; | |
77bd7415 | 427 | |
9b3c76f0 GS |
428 | /* Restore PE */ |
429 | eeh_ops->configure_bridge(pe); | |
430 | eeh_pe_restore_bars(pe); | |
77bd7415 LV |
431 | |
432 | /* Give the system 5 seconds to finish running the user-space | |
a84f273c GS |
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, | |
77bd7415 LV |
436 | * potentially weird things happen. |
437 | */ | |
438 | if (bus) { | |
f5c57710 | 439 | pr_info("EEH: Sleep 5s ahead of complete hotplug\n"); |
29f8bf1b | 440 | ssleep(5); |
f5c57710 GS |
441 | |
442 | /* | |
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. | |
446 | */ | |
447 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); | |
77bd7415 | 448 | pcibios_add_pci_devices(bus); |
f5c57710 GS |
449 | } else if (frozen_bus && removed) { |
450 | pr_info("EEH: Sleep 5s ahead of partial hotplug\n"); | |
451 | ssleep(5); | |
452 | ||
453 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); | |
454 | pcibios_add_pci_devices(frozen_bus); | |
77bd7415 | 455 | } |
f5c57710 | 456 | eeh_pe_state_clear(pe, EEH_PE_KEEP); |
5a71978e GS |
457 | |
458 | pe->tstamp = tstamp; | |
9b3c76f0 | 459 | pe->freeze_count = cnt; |
b6495c0c LV |
460 | |
461 | return 0; | |
77bd7415 LV |
462 | } |
463 | ||
464 | /* The longest amount of time to wait for a pci device | |
465 | * to come back on line, in seconds. | |
466 | */ | |
fa1be476 | 467 | #define MAX_WAIT_FOR_RECOVERY 150 |
77bd7415 | 468 | |
8a6b1bc7 | 469 | static void eeh_handle_normal_event(struct eeh_pe *pe) |
77bd7415 | 470 | { |
77bd7415 | 471 | struct pci_bus *frozen_bus; |
b6495c0c | 472 | int rc = 0; |
18eb3b39 | 473 | enum pci_ers_result result = PCI_ERS_RESULT_NONE; |
77bd7415 | 474 | |
9b3c76f0 | 475 | frozen_bus = eeh_pe_bus_get(pe); |
77bd7415 | 476 | if (!frozen_bus) { |
9b3c76f0 GS |
477 | pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n", |
478 | __func__, pe->phb->global_number, pe->addr); | |
479 | return; | |
77bd7415 LV |
480 | } |
481 | ||
5a71978e | 482 | eeh_pe_update_time_stamp(pe); |
9b3c76f0 GS |
483 | pe->freeze_count++; |
484 | if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES) | |
8df83028 | 485 | goto excess_failures; |
9b3c76f0 GS |
486 | pr_warning("EEH: This PCI device has failed %d times in the last hour\n", |
487 | pe->freeze_count); | |
77bd7415 LV |
488 | |
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. | |
494 | */ | |
56ca4fde | 495 | pr_info("EEH: Notify device drivers to shutdown\n"); |
9b3c76f0 | 496 | eeh_pe_dev_traverse(pe, eeh_report_error, &result); |
77bd7415 | 497 | |
5f1a7c81 | 498 | /* Get the current PCI slot state. This can take a long time, |
29f8bf1b GS |
499 | * sometimes over 3 seconds for certain systems. |
500 | */ | |
9b3c76f0 | 501 | rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000); |
eb594a47 | 502 | if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) { |
56ca4fde | 503 | pr_warning("EEH: Permanent failure\n"); |
5f1a7c81 LV |
504 | goto hard_fail; |
505 | } | |
506 | ||
ede8ca26 LV |
507 | /* Since rtas may enable MMIO when posting the error log, |
508 | * don't post the error log until after all dev drivers | |
17213c3b LV |
509 | * have been informed. |
510 | */ | |
56ca4fde | 511 | pr_info("EEH: Collect temporary log\n"); |
9b3c76f0 | 512 | eeh_slot_error_detail(pe, EEH_LOG_TEMP); |
ede8ca26 | 513 | |
77bd7415 LV |
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. | |
517 | */ | |
18eb3b39 | 518 | if (result == PCI_ERS_RESULT_NONE) { |
56ca4fde | 519 | pr_info("EEH: Reset with hotplug activity\n"); |
9b3c76f0 | 520 | rc = eeh_reset_device(pe, frozen_bus); |
e0f90b64 | 521 | if (rc) { |
56ca4fde GS |
522 | pr_warning("%s: Unable to reset, err=%d\n", |
523 | __func__, rc); | |
b6495c0c | 524 | goto hard_fail; |
e0f90b64 | 525 | } |
77bd7415 LV |
526 | } |
527 | ||
6a1ca373 LV |
528 | /* If all devices reported they can proceed, then re-enable MMIO */ |
529 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { | |
56ca4fde | 530 | pr_info("EEH: Enable I/O for affected devices\n"); |
9b3c76f0 | 531 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO); |
6a1ca373 | 532 | |
fa1be476 LV |
533 | if (rc < 0) |
534 | goto hard_fail; | |
6a1ca373 LV |
535 | if (rc) { |
536 | result = PCI_ERS_RESULT_NEED_RESET; | |
537 | } else { | |
56ca4fde | 538 | pr_info("EEH: Notify device drivers to resume I/O\n"); |
6a1ca373 | 539 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 540 | eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result); |
6a1ca373 | 541 | } |
77bd7415 LV |
542 | } |
543 | ||
6a1ca373 | 544 | /* If all devices reported they can proceed, then re-enable DMA */ |
18eb3b39 | 545 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { |
56ca4fde | 546 | pr_info("EEH: Enabled DMA for affected devices\n"); |
9b3c76f0 | 547 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA); |
6a1ca373 | 548 | |
fa1be476 LV |
549 | if (rc < 0) |
550 | goto hard_fail; | |
6a1ca373 LV |
551 | if (rc) |
552 | result = PCI_ERS_RESULT_NEED_RESET; | |
d0e70341 LV |
553 | else |
554 | result = PCI_ERS_RESULT_RECOVERED; | |
6a1ca373 LV |
555 | } |
556 | ||
557 | /* If any device has a hard failure, then shut off everything. */ | |
e0f90b64 | 558 | if (result == PCI_ERS_RESULT_DISCONNECT) { |
56ca4fde | 559 | pr_warning("EEH: Device driver gave up\n"); |
6a1ca373 | 560 | goto hard_fail; |
e0f90b64 | 561 | } |
6a1ca373 LV |
562 | |
563 | /* If any device called out for a reset, then reset the slot */ | |
564 | if (result == PCI_ERS_RESULT_NEED_RESET) { | |
56ca4fde | 565 | pr_info("EEH: Reset without hotplug activity\n"); |
9b3c76f0 | 566 | rc = eeh_reset_device(pe, NULL); |
e0f90b64 | 567 | if (rc) { |
56ca4fde GS |
568 | pr_warning("%s: Cannot reset, err=%d\n", |
569 | __func__, rc); | |
b6495c0c | 570 | goto hard_fail; |
e0f90b64 | 571 | } |
56ca4fde GS |
572 | |
573 | pr_info("EEH: Notify device drivers " | |
574 | "the completion of reset\n"); | |
6a1ca373 | 575 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 576 | eeh_pe_dev_traverse(pe, eeh_report_reset, &result); |
77bd7415 LV |
577 | } |
578 | ||
6a1ca373 | 579 | /* All devices should claim they have recovered by now. */ |
90fdd613 LV |
580 | if ((result != PCI_ERS_RESULT_RECOVERED) && |
581 | (result != PCI_ERS_RESULT_NONE)) { | |
56ca4fde | 582 | pr_warning("EEH: Not recovered\n"); |
6a1ca373 | 583 | goto hard_fail; |
e0f90b64 | 584 | } |
6a1ca373 | 585 | |
77bd7415 | 586 | /* Tell all device drivers that they can resume operations */ |
56ca4fde | 587 | pr_info("EEH: Notify device driver to resume\n"); |
9b3c76f0 | 588 | eeh_pe_dev_traverse(pe, eeh_report_resume, NULL); |
b6495c0c | 589 | |
9b3c76f0 | 590 | return; |
a84f273c | 591 | |
8df83028 | 592 | excess_failures: |
b6495c0c LV |
593 | /* |
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. | |
597 | */ | |
9b3c76f0 GS |
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, | |
602 | pe->freeze_count); | |
8df83028 LV |
603 | goto perm_error; |
604 | ||
605 | hard_fail: | |
9b3c76f0 GS |
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); | |
b6495c0c | 609 | |
8df83028 | 610 | perm_error: |
9b3c76f0 | 611 | eeh_slot_error_detail(pe, EEH_LOG_PERM); |
b6495c0c LV |
612 | |
613 | /* Notify all devices that they're about to go down. */ | |
9b3c76f0 | 614 | eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); |
b6495c0c LV |
615 | |
616 | /* Shut down the device drivers for good. */ | |
9b3c76f0 GS |
617 | if (frozen_bus) |
618 | pcibios_remove_pci_devices(frozen_bus); | |
77bd7415 | 619 | } |
8a6b1bc7 GS |
620 | |
621 | static void eeh_handle_special_event(void) | |
622 | { | |
623 | struct eeh_pe *pe, *phb_pe; | |
624 | struct pci_bus *bus; | |
625 | struct pci_controller *hose, *tmp; | |
626 | unsigned long flags; | |
627 | int rc = 0; | |
628 | ||
629 | /* | |
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 | |
633 | * integer directly: | |
634 | * | |
635 | * 4 - Dead IOC 3 - Dead PHB | |
636 | * 2 - Fenced PHB 1 - Frozen PE | |
637 | * 0 - No error found | |
638 | * | |
639 | */ | |
640 | rc = eeh_ops->next_error(&pe); | |
641 | if (rc <= 0) | |
642 | return; | |
643 | ||
644 | switch (rc) { | |
645 | case 4: | |
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; | |
652 | ||
653 | eeh_pe_state_mark(phb_pe, | |
654 | EEH_PE_ISOLATED | EEH_PE_PHB_DEAD); | |
655 | } | |
656 | eeh_serialize_unlock(flags); | |
657 | ||
658 | /* Purge all events */ | |
659 | eeh_remove_event(NULL); | |
660 | break; | |
661 | case 3: | |
662 | case 2: | |
663 | case 1: | |
664 | /* Mark the PE in fenced state */ | |
665 | eeh_serialize_lock(&flags); | |
666 | if (rc == 3) | |
667 | eeh_pe_state_mark(pe, | |
668 | EEH_PE_ISOLATED | EEH_PE_PHB_DEAD); | |
669 | else | |
670 | eeh_pe_state_mark(pe, | |
671 | EEH_PE_ISOLATED | EEH_PE_RECOVERING); | |
672 | eeh_serialize_unlock(flags); | |
673 | ||
674 | /* Purge all events of the PHB */ | |
675 | eeh_remove_event(pe); | |
676 | break; | |
677 | default: | |
678 | pr_err("%s: Invalid value %d from next_error()\n", | |
679 | __func__, rc); | |
680 | return; | |
681 | } | |
682 | ||
683 | /* | |
684 | * For fenced PHB and frozen PE, it's handled as normal | |
685 | * event. We have to remove the affected PHBs for dead | |
686 | * PHB and IOC | |
687 | */ | |
688 | if (rc == 2 || rc == 1) | |
689 | eeh_handle_normal_event(pe); | |
690 | else { | |
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)) | |
695 | continue; | |
696 | ||
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); | |
701 | } | |
702 | } | |
703 | } | |
704 | ||
705 | /** | |
706 | * eeh_handle_event - Reset a PCI device after hard lockup. | |
707 | * @pe: EEH PE | |
708 | * | |
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. | |
714 | * | |
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 | |
720 | * userspace). | |
721 | */ | |
722 | void eeh_handle_event(struct eeh_pe *pe) | |
723 | { | |
724 | if (pe) | |
725 | eeh_handle_normal_event(pe); | |
726 | else | |
727 | eeh_handle_special_event(); | |
728 | } |