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; | |
b8a9a11b TG |
149 | /* |
150 | * FIXME !!!!! | |
151 | * | |
152 | * This is just ass backwards. This maze has | |
153 | * unbalanced irq_enable/disable calls. So instead of | |
154 | * finding the root cause it works around the warning | |
155 | * in the irq_enable code by conditionally calling | |
156 | * into it. | |
157 | * | |
158 | * That's just wrong.The warning in the core code is | |
159 | * there to tell people to fix their assymetries in | |
160 | * their own code, not by abusing the core information | |
161 | * to avoid it. | |
162 | * | |
163 | * I so wish that the assymetry would be the other way | |
164 | * round and a few more irq_disable calls render that | |
165 | * shit unusable forever. | |
166 | * | |
167 | * tglx | |
168 | */ | |
57310c3c | 169 | if (irqd_irq_disabled(irq_get_irq_data(dev->irq))) |
91150af3 | 170 | enable_irq(dev->irq); |
57310c3c | 171 | } |
8535ef05 MM |
172 | } |
173 | ||
d2b0f6f7 GS |
174 | static bool eeh_dev_removed(struct eeh_dev *edev) |
175 | { | |
176 | /* EEH device removed ? */ | |
177 | if (!edev || (edev->mode & EEH_DEV_REMOVED)) | |
178 | return true; | |
179 | ||
180 | return false; | |
181 | } | |
182 | ||
cb5b5624 | 183 | /** |
29f8bf1b | 184 | * eeh_report_error - Report pci error to each device driver |
9b3c76f0 | 185 | * @data: eeh device |
29f8bf1b | 186 | * @userdata: return value |
a84f273c GS |
187 | * |
188 | * Report an EEH error to each device driver, collect up and | |
189 | * merge the device driver responses. Cumulative response | |
cb5b5624 | 190 | * passed back in "userdata". |
77bd7415 | 191 | */ |
9b3c76f0 | 192 | static void *eeh_report_error(void *data, void *userdata) |
77bd7415 | 193 | { |
9b3c76f0 GS |
194 | struct eeh_dev *edev = (struct eeh_dev *)data; |
195 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
18eb3b39 | 196 | enum pci_ers_result rc, *res = userdata; |
feadf7c0 | 197 | struct pci_driver *driver; |
77bd7415 | 198 | |
d2b0f6f7 GS |
199 | if (!dev || eeh_dev_removed(edev)) |
200 | return NULL; | |
77bd7415 LV |
201 | dev->error_state = pci_channel_io_frozen; |
202 | ||
feadf7c0 GS |
203 | driver = eeh_pcid_get(dev); |
204 | if (!driver) return NULL; | |
77bd7415 | 205 | |
8535ef05 MM |
206 | eeh_disable_irq(dev); |
207 | ||
6a1ca373 | 208 | if (!driver->err_handler || |
feadf7c0 GS |
209 | !driver->err_handler->error_detected) { |
210 | eeh_pcid_put(dev); | |
9b3c76f0 | 211 | return NULL; |
feadf7c0 | 212 | } |
77bd7415 | 213 | |
29f8bf1b | 214 | rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen); |
2a50f144 LV |
215 | |
216 | /* A driver that needs a reset trumps all others */ | |
217 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
18eb3b39 | 218 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 219 | |
feadf7c0 | 220 | eeh_pcid_put(dev); |
9b3c76f0 | 221 | return NULL; |
6a1ca373 LV |
222 | } |
223 | ||
224 | /** | |
29f8bf1b | 225 | * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled |
9b3c76f0 | 226 | * @data: eeh device |
29f8bf1b | 227 | * @userdata: return value |
6a1ca373 | 228 | * |
638799b3 LV |
229 | * Tells each device driver that IO ports, MMIO and config space I/O |
230 | * are now enabled. Collects up and merges the device driver responses. | |
231 | * Cumulative response passed back in "userdata". | |
6a1ca373 | 232 | */ |
9b3c76f0 | 233 | static void *eeh_report_mmio_enabled(void *data, void *userdata) |
6a1ca373 | 234 | { |
9b3c76f0 GS |
235 | struct eeh_dev *edev = (struct eeh_dev *)data; |
236 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 237 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 238 | struct pci_driver *driver; |
6a1ca373 | 239 | |
d2b0f6f7 GS |
240 | if (!dev || eeh_dev_removed(edev)) |
241 | return NULL; | |
242 | ||
feadf7c0 GS |
243 | driver = eeh_pcid_get(dev); |
244 | if (!driver) return NULL; | |
9b3c76f0 | 245 | |
feadf7c0 | 246 | if (!driver->err_handler || |
f26c7a03 GS |
247 | !driver->err_handler->mmio_enabled || |
248 | (edev->mode & EEH_DEV_NO_HANDLER)) { | |
feadf7c0 | 249 | eeh_pcid_put(dev); |
9b3c76f0 | 250 | return NULL; |
feadf7c0 | 251 | } |
6a1ca373 | 252 | |
29f8bf1b | 253 | rc = driver->err_handler->mmio_enabled(dev); |
2a50f144 LV |
254 | |
255 | /* A driver that needs a reset trumps all others */ | |
256 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
6a1ca373 | 257 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 258 | |
feadf7c0 | 259 | eeh_pcid_put(dev); |
9b3c76f0 | 260 | return NULL; |
77bd7415 LV |
261 | } |
262 | ||
cb5b5624 | 263 | /** |
29f8bf1b | 264 | * eeh_report_reset - Tell device that slot has been reset |
9b3c76f0 | 265 | * @data: eeh device |
29f8bf1b GS |
266 | * @userdata: return value |
267 | * | |
268 | * This routine must be called while EEH tries to reset particular | |
269 | * PCI device so that the associated PCI device driver could take | |
270 | * some actions, usually to save data the driver needs so that the | |
271 | * driver can work again while the device is recovered. | |
77bd7415 | 272 | */ |
9b3c76f0 | 273 | static void *eeh_report_reset(void *data, void *userdata) |
77bd7415 | 274 | { |
9b3c76f0 GS |
275 | struct eeh_dev *edev = (struct eeh_dev *)data; |
276 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 277 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 278 | struct pci_driver *driver; |
77bd7415 | 279 | |
d2b0f6f7 GS |
280 | if (!dev || eeh_dev_removed(edev)) |
281 | return NULL; | |
c58dc575 MM |
282 | dev->error_state = pci_channel_io_normal; |
283 | ||
feadf7c0 GS |
284 | driver = eeh_pcid_get(dev); |
285 | if (!driver) return NULL; | |
286 | ||
8535ef05 MM |
287 | eeh_enable_irq(dev); |
288 | ||
6a1ca373 | 289 | if (!driver->err_handler || |
f26c7a03 GS |
290 | !driver->err_handler->slot_reset || |
291 | (edev->mode & EEH_DEV_NO_HANDLER)) { | |
feadf7c0 | 292 | eeh_pcid_put(dev); |
9b3c76f0 | 293 | return NULL; |
feadf7c0 | 294 | } |
77bd7415 | 295 | |
6a1ca373 | 296 | rc = driver->err_handler->slot_reset(dev); |
5794dbcb LV |
297 | if ((*res == PCI_ERS_RESULT_NONE) || |
298 | (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; | |
6a1ca373 LV |
299 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
300 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
70298c6e | 301 | |
feadf7c0 | 302 | eeh_pcid_put(dev); |
9b3c76f0 | 303 | return NULL; |
77bd7415 LV |
304 | } |
305 | ||
cb5b5624 | 306 | /** |
29f8bf1b | 307 | * eeh_report_resume - Tell device to resume normal operations |
9b3c76f0 | 308 | * @data: eeh device |
29f8bf1b GS |
309 | * @userdata: return value |
310 | * | |
311 | * This routine must be called to notify the device driver that it | |
312 | * could resume so that the device driver can do some initialization | |
313 | * to make the recovered device work again. | |
cb5b5624 | 314 | */ |
9b3c76f0 | 315 | static void *eeh_report_resume(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 | ||
d2b0f6f7 GS |
321 | if (!dev || eeh_dev_removed(edev)) |
322 | return NULL; | |
77bd7415 LV |
323 | dev->error_state = pci_channel_io_normal; |
324 | ||
feadf7c0 GS |
325 | driver = eeh_pcid_get(dev); |
326 | if (!driver) return NULL; | |
d0e70341 | 327 | |
8535ef05 MM |
328 | eeh_enable_irq(dev); |
329 | ||
d0e70341 | 330 | if (!driver->err_handler || |
f26c7a03 GS |
331 | !driver->err_handler->resume || |
332 | (edev->mode & EEH_DEV_NO_HANDLER)) { | |
333 | edev->mode &= ~EEH_DEV_NO_HANDLER; | |
feadf7c0 | 334 | eeh_pcid_put(dev); |
9b3c76f0 | 335 | return NULL; |
feadf7c0 | 336 | } |
77bd7415 LV |
337 | |
338 | driver->err_handler->resume(dev); | |
70298c6e | 339 | |
feadf7c0 | 340 | eeh_pcid_put(dev); |
9b3c76f0 | 341 | return NULL; |
77bd7415 LV |
342 | } |
343 | ||
cb5b5624 | 344 | /** |
29f8bf1b | 345 | * eeh_report_failure - Tell device driver that device is dead. |
9b3c76f0 | 346 | * @data: eeh device |
29f8bf1b | 347 | * @userdata: return value |
cb5b5624 LV |
348 | * |
349 | * This informs the device driver that the device is permanently | |
350 | * dead, and that no further recovery attempts will be made on it. | |
351 | */ | |
9b3c76f0 | 352 | static void *eeh_report_failure(void *data, void *userdata) |
77bd7415 | 353 | { |
9b3c76f0 GS |
354 | struct eeh_dev *edev = (struct eeh_dev *)data; |
355 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
356 | struct pci_driver *driver; | |
357 | ||
d2b0f6f7 GS |
358 | if (!dev || eeh_dev_removed(edev)) |
359 | return NULL; | |
77bd7415 LV |
360 | dev->error_state = pci_channel_io_perm_failure; |
361 | ||
feadf7c0 GS |
362 | driver = eeh_pcid_get(dev); |
363 | if (!driver) return NULL; | |
77bd7415 | 364 | |
8535ef05 MM |
365 | eeh_disable_irq(dev); |
366 | ||
367 | if (!driver->err_handler || | |
feadf7c0 GS |
368 | !driver->err_handler->error_detected) { |
369 | eeh_pcid_put(dev); | |
9b3c76f0 | 370 | return NULL; |
feadf7c0 | 371 | } |
8535ef05 | 372 | |
77bd7415 | 373 | driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); |
70298c6e | 374 | |
feadf7c0 | 375 | eeh_pcid_put(dev); |
9b3c76f0 | 376 | return NULL; |
77bd7415 LV |
377 | } |
378 | ||
f5c57710 GS |
379 | static void *eeh_rmv_device(void *data, void *userdata) |
380 | { | |
381 | struct pci_driver *driver; | |
382 | struct eeh_dev *edev = (struct eeh_dev *)data; | |
383 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
384 | int *removed = (int *)userdata; | |
385 | ||
386 | /* | |
387 | * Actually, we should remove the PCI bridges as well. | |
388 | * However, that's lots of complexity to do that, | |
389 | * particularly some of devices under the bridge might | |
390 | * support EEH. So we just care about PCI devices for | |
391 | * simplicity here. | |
392 | */ | |
393 | if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) | |
394 | return NULL; | |
8cc6b6cd | 395 | |
d2b0f6f7 GS |
396 | /* |
397 | * We rely on count-based pcibios_release_device() to | |
398 | * detach permanently offlined PEs. Unfortunately, that's | |
399 | * not reliable enough. We might have the permanently | |
400 | * offlined PEs attached, but we needn't take care of | |
401 | * them and their child devices. | |
402 | */ | |
403 | if (eeh_dev_removed(edev)) | |
404 | return NULL; | |
405 | ||
f5c57710 | 406 | driver = eeh_pcid_get(dev); |
8cc6b6cd TLSC |
407 | if (driver) { |
408 | eeh_pcid_put(dev); | |
409 | if (driver->err_handler) | |
410 | return NULL; | |
411 | } | |
f5c57710 GS |
412 | |
413 | /* Remove it from PCI subsystem */ | |
414 | pr_debug("EEH: Removing %s without EEH sensitive driver\n", | |
415 | pci_name(dev)); | |
416 | edev->bus = dev->bus; | |
417 | edev->mode |= EEH_DEV_DISCONNECTED; | |
418 | (*removed)++; | |
419 | ||
1c2042c8 | 420 | pci_lock_rescan_remove(); |
f5c57710 | 421 | pci_stop_and_remove_bus_device(dev); |
1c2042c8 | 422 | pci_unlock_rescan_remove(); |
f5c57710 GS |
423 | |
424 | return NULL; | |
425 | } | |
426 | ||
427 | static void *eeh_pe_detach_dev(void *data, void *userdata) | |
428 | { | |
429 | struct eeh_pe *pe = (struct eeh_pe *)data; | |
430 | struct eeh_dev *edev, *tmp; | |
431 | ||
432 | eeh_pe_for_each_dev(pe, edev, tmp) { | |
433 | if (!(edev->mode & EEH_DEV_DISCONNECTED)) | |
434 | continue; | |
435 | ||
436 | edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED); | |
437 | eeh_rmv_from_parent_pe(edev); | |
438 | } | |
439 | ||
440 | return NULL; | |
441 | } | |
442 | ||
78954700 GS |
443 | /* |
444 | * Explicitly clear PE's frozen state for PowerNV where | |
445 | * we have frozen PE until BAR restore is completed. It's | |
446 | * harmless to clear it for pSeries. To be consistent with | |
447 | * PE reset (for 3 times), we try to clear the frozen state | |
448 | * for 3 times as well. | |
449 | */ | |
2c665992 | 450 | static void *__eeh_clear_pe_frozen_state(void *data, void *flag) |
78954700 | 451 | { |
2c665992 | 452 | struct eeh_pe *pe = (struct eeh_pe *)data; |
c9dd0143 | 453 | int i, rc = 1; |
78954700 | 454 | |
c9dd0143 GS |
455 | for (i = 0; rc && i < 3; i++) |
456 | rc = eeh_unfreeze_pe(pe, false); | |
78954700 | 457 | |
c9dd0143 | 458 | /* Stop immediately on any errors */ |
2c665992 | 459 | if (rc) { |
c9dd0143 GS |
460 | pr_warn("%s: Failure %d unfreezing PHB#%x-PE#%x\n", |
461 | __func__, rc, pe->phb->global_number, pe->addr); | |
2c665992 GS |
462 | return (void *)pe; |
463 | } | |
464 | ||
465 | return NULL; | |
466 | } | |
467 | ||
468 | static int eeh_clear_pe_frozen_state(struct eeh_pe *pe) | |
469 | { | |
470 | void *rc; | |
471 | ||
472 | rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, NULL); | |
473 | if (!rc) | |
78954700 GS |
474 | eeh_pe_state_clear(pe, EEH_PE_ISOLATED); |
475 | ||
2c665992 | 476 | return rc ? -EIO : 0; |
78954700 GS |
477 | } |
478 | ||
77bd7415 | 479 | /** |
29f8bf1b | 480 | * eeh_reset_device - Perform actual reset of a pci slot |
9b3c76f0 | 481 | * @pe: EEH PE |
29f8bf1b | 482 | * @bus: PCI bus corresponding to the isolcated slot |
77bd7415 | 483 | * |
29f8bf1b GS |
484 | * This routine must be called to do reset on the indicated PE. |
485 | * During the reset, udev might be invoked because those affected | |
486 | * PCI devices will be removed and then added. | |
77bd7415 | 487 | */ |
9b3c76f0 | 488 | static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus) |
77bd7415 | 489 | { |
f5c57710 | 490 | struct pci_bus *frozen_bus = eeh_pe_bus_get(pe); |
5a71978e | 491 | struct timeval tstamp; |
f5c57710 | 492 | int cnt, rc, removed = 0; |
42405456 LV |
493 | |
494 | /* pcibios will clear the counter; save the value */ | |
9b3c76f0 | 495 | cnt = pe->freeze_count; |
5a71978e | 496 | tstamp = pe->tstamp; |
42405456 | 497 | |
20ee6a97 GS |
498 | /* |
499 | * We don't remove the corresponding PE instances because | |
500 | * we need the information afterwords. The attached EEH | |
501 | * devices are expected to be attached soon when calling | |
502 | * into pcibios_add_pci_devices(). | |
503 | */ | |
f5c57710 | 504 | eeh_pe_state_mark(pe, EEH_PE_KEEP); |
1c2042c8 RW |
505 | if (bus) { |
506 | pci_lock_rescan_remove(); | |
807a827d | 507 | pcibios_remove_pci_devices(bus); |
1c2042c8 RW |
508 | pci_unlock_rescan_remove(); |
509 | } else if (frozen_bus) { | |
f5c57710 | 510 | eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed); |
1c2042c8 | 511 | } |
77bd7415 | 512 | |
d0914f50 GS |
513 | /* |
514 | * Reset the pci controller. (Asserts RST#; resets config space). | |
b6495c0c | 515 | * Reconfigure bridges and devices. Don't try to bring the system |
29f8bf1b | 516 | * up if the reset failed for some reason. |
d0914f50 GS |
517 | * |
518 | * During the reset, it's very dangerous to have uncontrolled PCI | |
519 | * config accesses. So we prefer to block them. However, controlled | |
520 | * PCI config accesses initiated from EEH itself are allowed. | |
29f8bf1b | 521 | */ |
d0914f50 | 522 | eeh_pe_state_mark(pe, EEH_PE_RESET); |
9b3c76f0 | 523 | rc = eeh_reset_pe(pe); |
d0914f50 GS |
524 | if (rc) { |
525 | eeh_pe_state_clear(pe, EEH_PE_RESET); | |
b6495c0c | 526 | return rc; |
d0914f50 | 527 | } |
77bd7415 | 528 | |
1c2042c8 RW |
529 | pci_lock_rescan_remove(); |
530 | ||
9b3c76f0 GS |
531 | /* Restore PE */ |
532 | eeh_ops->configure_bridge(pe); | |
533 | eeh_pe_restore_bars(pe); | |
d0914f50 | 534 | eeh_pe_state_clear(pe, EEH_PE_RESET); |
77bd7415 | 535 | |
78954700 GS |
536 | /* Clear frozen state */ |
537 | rc = eeh_clear_pe_frozen_state(pe); | |
538 | if (rc) | |
539 | return rc; | |
540 | ||
77bd7415 | 541 | /* Give the system 5 seconds to finish running the user-space |
a84f273c GS |
542 | * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, |
543 | * this is a hack, but if we don't do this, and try to bring | |
544 | * the device up before the scripts have taken it down, | |
77bd7415 LV |
545 | * potentially weird things happen. |
546 | */ | |
547 | if (bus) { | |
f5c57710 | 548 | pr_info("EEH: Sleep 5s ahead of complete hotplug\n"); |
29f8bf1b | 549 | ssleep(5); |
f5c57710 GS |
550 | |
551 | /* | |
552 | * The EEH device is still connected with its parent | |
553 | * PE. We should disconnect it so the binding can be | |
554 | * rebuilt when adding PCI devices. | |
555 | */ | |
556 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); | |
77bd7415 | 557 | pcibios_add_pci_devices(bus); |
f5c57710 GS |
558 | } else if (frozen_bus && removed) { |
559 | pr_info("EEH: Sleep 5s ahead of partial hotplug\n"); | |
560 | ssleep(5); | |
561 | ||
562 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); | |
563 | pcibios_add_pci_devices(frozen_bus); | |
77bd7415 | 564 | } |
f5c57710 | 565 | eeh_pe_state_clear(pe, EEH_PE_KEEP); |
5a71978e GS |
566 | |
567 | pe->tstamp = tstamp; | |
9b3c76f0 | 568 | pe->freeze_count = cnt; |
b6495c0c | 569 | |
1c2042c8 | 570 | pci_unlock_rescan_remove(); |
b6495c0c | 571 | return 0; |
77bd7415 LV |
572 | } |
573 | ||
574 | /* The longest amount of time to wait for a pci device | |
575 | * to come back on line, in seconds. | |
576 | */ | |
fb48dc22 | 577 | #define MAX_WAIT_FOR_RECOVERY 300 |
77bd7415 | 578 | |
8a6b1bc7 | 579 | static void eeh_handle_normal_event(struct eeh_pe *pe) |
77bd7415 | 580 | { |
77bd7415 | 581 | struct pci_bus *frozen_bus; |
b6495c0c | 582 | int rc = 0; |
18eb3b39 | 583 | enum pci_ers_result result = PCI_ERS_RESULT_NONE; |
77bd7415 | 584 | |
9b3c76f0 | 585 | frozen_bus = eeh_pe_bus_get(pe); |
77bd7415 | 586 | if (!frozen_bus) { |
9b3c76f0 GS |
587 | pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n", |
588 | __func__, pe->phb->global_number, pe->addr); | |
589 | return; | |
77bd7415 LV |
590 | } |
591 | ||
5a71978e | 592 | eeh_pe_update_time_stamp(pe); |
9b3c76f0 GS |
593 | pe->freeze_count++; |
594 | if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES) | |
8df83028 | 595 | goto excess_failures; |
0dae2743 | 596 | pr_warn("EEH: This PCI device has failed %d times in the last hour\n", |
9b3c76f0 | 597 | pe->freeze_count); |
77bd7415 LV |
598 | |
599 | /* Walk the various device drivers attached to this slot through | |
600 | * a reset sequence, giving each an opportunity to do what it needs | |
601 | * to accomplish the reset. Each child gets a report of the | |
602 | * status ... if any child can't handle the reset, then the entire | |
603 | * slot is dlpar removed and added. | |
604 | */ | |
56ca4fde | 605 | pr_info("EEH: Notify device drivers to shutdown\n"); |
9b3c76f0 | 606 | eeh_pe_dev_traverse(pe, eeh_report_error, &result); |
77bd7415 | 607 | |
5f1a7c81 | 608 | /* Get the current PCI slot state. This can take a long time, |
29f8bf1b GS |
609 | * sometimes over 3 seconds for certain systems. |
610 | */ | |
9b3c76f0 | 611 | rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000); |
eb594a47 | 612 | if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) { |
0dae2743 | 613 | pr_warn("EEH: Permanent failure\n"); |
5f1a7c81 LV |
614 | goto hard_fail; |
615 | } | |
616 | ||
ede8ca26 LV |
617 | /* Since rtas may enable MMIO when posting the error log, |
618 | * don't post the error log until after all dev drivers | |
17213c3b LV |
619 | * have been informed. |
620 | */ | |
56ca4fde | 621 | pr_info("EEH: Collect temporary log\n"); |
9b3c76f0 | 622 | eeh_slot_error_detail(pe, EEH_LOG_TEMP); |
ede8ca26 | 623 | |
77bd7415 LV |
624 | /* If all device drivers were EEH-unaware, then shut |
625 | * down all of the device drivers, and hope they | |
626 | * go down willingly, without panicing the system. | |
627 | */ | |
18eb3b39 | 628 | if (result == PCI_ERS_RESULT_NONE) { |
56ca4fde | 629 | pr_info("EEH: Reset with hotplug activity\n"); |
9b3c76f0 | 630 | rc = eeh_reset_device(pe, frozen_bus); |
e0f90b64 | 631 | if (rc) { |
0dae2743 GS |
632 | pr_warn("%s: Unable to reset, err=%d\n", |
633 | __func__, rc); | |
b6495c0c | 634 | goto hard_fail; |
e0f90b64 | 635 | } |
77bd7415 LV |
636 | } |
637 | ||
6a1ca373 LV |
638 | /* If all devices reported they can proceed, then re-enable MMIO */ |
639 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { | |
56ca4fde | 640 | pr_info("EEH: Enable I/O for affected devices\n"); |
9b3c76f0 | 641 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO); |
6a1ca373 | 642 | |
fa1be476 LV |
643 | if (rc < 0) |
644 | goto hard_fail; | |
6a1ca373 LV |
645 | if (rc) { |
646 | result = PCI_ERS_RESULT_NEED_RESET; | |
647 | } else { | |
56ca4fde | 648 | pr_info("EEH: Notify device drivers to resume I/O\n"); |
9b3c76f0 | 649 | eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result); |
6a1ca373 | 650 | } |
77bd7415 LV |
651 | } |
652 | ||
6a1ca373 | 653 | /* If all devices reported they can proceed, then re-enable DMA */ |
18eb3b39 | 654 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { |
56ca4fde | 655 | pr_info("EEH: Enabled DMA for affected devices\n"); |
9b3c76f0 | 656 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA); |
6a1ca373 | 657 | |
fa1be476 LV |
658 | if (rc < 0) |
659 | goto hard_fail; | |
35845a78 | 660 | if (rc) { |
6a1ca373 | 661 | result = PCI_ERS_RESULT_NEED_RESET; |
35845a78 GS |
662 | } else { |
663 | /* | |
664 | * We didn't do PE reset for the case. The PE | |
665 | * is still in frozen state. Clear it before | |
666 | * resuming the PE. | |
667 | */ | |
668 | eeh_pe_state_clear(pe, EEH_PE_ISOLATED); | |
d0e70341 | 669 | result = PCI_ERS_RESULT_RECOVERED; |
35845a78 | 670 | } |
6a1ca373 LV |
671 | } |
672 | ||
673 | /* If any device has a hard failure, then shut off everything. */ | |
e0f90b64 | 674 | if (result == PCI_ERS_RESULT_DISCONNECT) { |
0dae2743 | 675 | pr_warn("EEH: Device driver gave up\n"); |
6a1ca373 | 676 | goto hard_fail; |
e0f90b64 | 677 | } |
6a1ca373 LV |
678 | |
679 | /* If any device called out for a reset, then reset the slot */ | |
680 | if (result == PCI_ERS_RESULT_NEED_RESET) { | |
56ca4fde | 681 | pr_info("EEH: Reset without hotplug activity\n"); |
9b3c76f0 | 682 | rc = eeh_reset_device(pe, NULL); |
e0f90b64 | 683 | if (rc) { |
0dae2743 GS |
684 | pr_warn("%s: Cannot reset, err=%d\n", |
685 | __func__, rc); | |
b6495c0c | 686 | goto hard_fail; |
e0f90b64 | 687 | } |
56ca4fde GS |
688 | |
689 | pr_info("EEH: Notify device drivers " | |
690 | "the completion of reset\n"); | |
6a1ca373 | 691 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 692 | eeh_pe_dev_traverse(pe, eeh_report_reset, &result); |
77bd7415 LV |
693 | } |
694 | ||
6a1ca373 | 695 | /* All devices should claim they have recovered by now. */ |
90fdd613 LV |
696 | if ((result != PCI_ERS_RESULT_RECOVERED) && |
697 | (result != PCI_ERS_RESULT_NONE)) { | |
0dae2743 | 698 | pr_warn("EEH: Not recovered\n"); |
6a1ca373 | 699 | goto hard_fail; |
e0f90b64 | 700 | } |
6a1ca373 | 701 | |
77bd7415 | 702 | /* Tell all device drivers that they can resume operations */ |
56ca4fde | 703 | pr_info("EEH: Notify device driver to resume\n"); |
9b3c76f0 | 704 | eeh_pe_dev_traverse(pe, eeh_report_resume, NULL); |
b6495c0c | 705 | |
9b3c76f0 | 706 | return; |
a84f273c | 707 | |
8df83028 | 708 | excess_failures: |
b6495c0c LV |
709 | /* |
710 | * About 90% of all real-life EEH failures in the field | |
711 | * are due to poorly seated PCI cards. Only 10% or so are | |
712 | * due to actual, failed cards. | |
713 | */ | |
9b3c76f0 GS |
714 | pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n" |
715 | "last hour and has been permanently disabled.\n" | |
716 | "Please try reseating or replacing it.\n", | |
717 | pe->phb->global_number, pe->addr, | |
718 | pe->freeze_count); | |
8df83028 LV |
719 | goto perm_error; |
720 | ||
721 | hard_fail: | |
9b3c76f0 GS |
722 | pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n" |
723 | "Please try reseating or replacing it\n", | |
724 | pe->phb->global_number, pe->addr); | |
b6495c0c | 725 | |
8df83028 | 726 | perm_error: |
9b3c76f0 | 727 | eeh_slot_error_detail(pe, EEH_LOG_PERM); |
b6495c0c LV |
728 | |
729 | /* Notify all devices that they're about to go down. */ | |
9b3c76f0 | 730 | eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); |
b6495c0c | 731 | |
d2b0f6f7 GS |
732 | /* Mark the PE to be removed permanently */ |
733 | pe->freeze_count = EEH_MAX_ALLOWED_FREEZES + 1; | |
734 | ||
735 | /* | |
736 | * Shut down the device drivers for good. We mark | |
737 | * all removed devices correctly to avoid access | |
738 | * the their PCI config any more. | |
739 | */ | |
1c2042c8 | 740 | if (frozen_bus) { |
d2b0f6f7 GS |
741 | eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED); |
742 | ||
1c2042c8 | 743 | pci_lock_rescan_remove(); |
9b3c76f0 | 744 | pcibios_remove_pci_devices(frozen_bus); |
1c2042c8 RW |
745 | pci_unlock_rescan_remove(); |
746 | } | |
77bd7415 | 747 | } |
8a6b1bc7 GS |
748 | |
749 | static void eeh_handle_special_event(void) | |
750 | { | |
751 | struct eeh_pe *pe, *phb_pe; | |
752 | struct pci_bus *bus; | |
7e4e7867 | 753 | struct pci_controller *hose; |
8a6b1bc7 | 754 | unsigned long flags; |
7e4e7867 | 755 | int rc; |
8a6b1bc7 | 756 | |
8a6b1bc7 | 757 | |
7e4e7867 GS |
758 | do { |
759 | rc = eeh_ops->next_error(&pe); | |
760 | ||
761 | switch (rc) { | |
762 | case EEH_NEXT_ERR_DEAD_IOC: | |
763 | /* Mark all PHBs in dead state */ | |
764 | eeh_serialize_lock(&flags); | |
765 | ||
766 | /* Purge all events */ | |
5c7a35e3 | 767 | eeh_remove_event(NULL, true); |
7e4e7867 GS |
768 | |
769 | list_for_each_entry(hose, &hose_list, list_node) { | |
770 | phb_pe = eeh_phb_pe_get(hose); | |
771 | if (!phb_pe) continue; | |
772 | ||
9e049375 | 773 | eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED); |
7e4e7867 GS |
774 | } |
775 | ||
776 | eeh_serialize_unlock(flags); | |
777 | ||
778 | break; | |
779 | case EEH_NEXT_ERR_FROZEN_PE: | |
780 | case EEH_NEXT_ERR_FENCED_PHB: | |
781 | case EEH_NEXT_ERR_DEAD_PHB: | |
782 | /* Mark the PE in fenced state */ | |
783 | eeh_serialize_lock(&flags); | |
784 | ||
785 | /* Purge all events of the PHB */ | |
5c7a35e3 | 786 | eeh_remove_event(pe, true); |
7e4e7867 GS |
787 | |
788 | if (rc == EEH_NEXT_ERR_DEAD_PHB) | |
9e049375 | 789 | eeh_pe_state_mark(pe, EEH_PE_ISOLATED); |
7e4e7867 GS |
790 | else |
791 | eeh_pe_state_mark(pe, | |
792 | EEH_PE_ISOLATED | EEH_PE_RECOVERING); | |
793 | ||
794 | eeh_serialize_unlock(flags); | |
795 | ||
796 | break; | |
797 | case EEH_NEXT_ERR_NONE: | |
798 | return; | |
799 | default: | |
800 | pr_warn("%s: Invalid value %d from next_error()\n", | |
801 | __func__, rc); | |
802 | return; | |
8a6b1bc7 | 803 | } |
8a6b1bc7 | 804 | |
7e4e7867 GS |
805 | /* |
806 | * For fenced PHB and frozen PE, it's handled as normal | |
807 | * event. We have to remove the affected PHBs for dead | |
808 | * PHB and IOC | |
809 | */ | |
810 | if (rc == EEH_NEXT_ERR_FROZEN_PE || | |
811 | rc == EEH_NEXT_ERR_FENCED_PHB) { | |
812 | eeh_handle_normal_event(pe); | |
9e049375 | 813 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); |
7e4e7867 | 814 | } else { |
1b17366d | 815 | pci_lock_rescan_remove(); |
7e4e7867 GS |
816 | list_for_each_entry(hose, &hose_list, list_node) { |
817 | phb_pe = eeh_phb_pe_get(hose); | |
818 | if (!phb_pe || | |
9e049375 GS |
819 | !(phb_pe->state & EEH_PE_ISOLATED) || |
820 | (phb_pe->state & EEH_PE_RECOVERING)) | |
7e4e7867 GS |
821 | continue; |
822 | ||
823 | /* Notify all devices to be down */ | |
824 | bus = eeh_pe_bus_get(phb_pe); | |
825 | eeh_pe_dev_traverse(pe, | |
826 | eeh_report_failure, NULL); | |
827 | pcibios_remove_pci_devices(bus); | |
828 | } | |
1b17366d | 829 | pci_unlock_rescan_remove(); |
8a6b1bc7 | 830 | } |
7e4e7867 GS |
831 | |
832 | /* | |
833 | * If we have detected dead IOC, we needn't proceed | |
834 | * any more since all PHBs would have been removed | |
835 | */ | |
836 | if (rc == EEH_NEXT_ERR_DEAD_IOC) | |
837 | break; | |
838 | } while (rc != EEH_NEXT_ERR_NONE); | |
8a6b1bc7 GS |
839 | } |
840 | ||
841 | /** | |
842 | * eeh_handle_event - Reset a PCI device after hard lockup. | |
843 | * @pe: EEH PE | |
844 | * | |
845 | * While PHB detects address or data parity errors on particular PCI | |
846 | * slot, the associated PE will be frozen. Besides, DMA's occurring | |
847 | * to wild addresses (which usually happen due to bugs in device | |
848 | * drivers or in PCI adapter firmware) can cause EEH error. #SERR, | |
849 | * #PERR or other misc PCI-related errors also can trigger EEH errors. | |
850 | * | |
851 | * Recovery process consists of unplugging the device driver (which | |
852 | * generated hotplug events to userspace), then issuing a PCI #RST to | |
853 | * the device, then reconfiguring the PCI config space for all bridges | |
854 | * & devices under this slot, and then finally restarting the device | |
855 | * drivers (which cause a second set of hotplug events to go out to | |
856 | * userspace). | |
857 | */ | |
858 | void eeh_handle_event(struct eeh_pe *pe) | |
859 | { | |
860 | if (pe) | |
861 | eeh_handle_normal_event(pe); | |
862 | else | |
863 | eeh_handle_special_event(); | |
864 | } |