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 | ||
67086e32 WY |
37 | struct eeh_rmv_data { |
38 | struct list_head edev_list; | |
39 | int removed; | |
40 | }; | |
41 | ||
29f8bf1b GS |
42 | /** |
43 | * eeh_pcid_name - Retrieve name of PCI device driver | |
44 | * @pdev: PCI device | |
45 | * | |
46 | * This routine is used to retrieve the name of PCI device driver | |
47 | * if that's valid. | |
48 | */ | |
40a7cd92 | 49 | static inline const char *eeh_pcid_name(struct pci_dev *pdev) |
77bd7415 | 50 | { |
273d2803 | 51 | if (pdev && pdev->dev.driver) |
77bd7415 LV |
52 | return pdev->dev.driver->name; |
53 | return ""; | |
54 | } | |
55 | ||
feadf7c0 GS |
56 | /** |
57 | * eeh_pcid_get - Get the PCI device driver | |
58 | * @pdev: PCI device | |
59 | * | |
60 | * The function is used to retrieve the PCI device driver for | |
61 | * the indicated PCI device. Besides, we will increase the reference | |
62 | * of the PCI device driver to prevent that being unloaded on | |
63 | * the fly. Otherwise, kernel crash would be seen. | |
64 | */ | |
65 | static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev) | |
66 | { | |
67 | if (!pdev || !pdev->driver) | |
68 | return NULL; | |
69 | ||
70 | if (!try_module_get(pdev->driver->driver.owner)) | |
71 | return NULL; | |
72 | ||
73 | return pdev->driver; | |
74 | } | |
75 | ||
76 | /** | |
77 | * eeh_pcid_put - Dereference on the PCI device driver | |
78 | * @pdev: PCI device | |
79 | * | |
80 | * The function is called to do dereference on the PCI device | |
81 | * driver of the indicated PCI device. | |
82 | */ | |
83 | static inline void eeh_pcid_put(struct pci_dev *pdev) | |
84 | { | |
85 | if (!pdev || !pdev->driver) | |
86 | return; | |
87 | ||
88 | module_put(pdev->driver->driver.owner); | |
89 | } | |
90 | ||
8535ef05 | 91 | /** |
29f8bf1b GS |
92 | * eeh_disable_irq - Disable interrupt for the recovering device |
93 | * @dev: PCI device | |
94 | * | |
95 | * This routine must be called when reporting temporary or permanent | |
96 | * error to the particular PCI device to disable interrupt of that | |
97 | * device. If the device has enabled MSI or MSI-X interrupt, we needn't | |
98 | * do real work because EEH should freeze DMA transfers for those PCI | |
99 | * devices encountering EEH errors, which includes MSI or MSI-X. | |
8535ef05 MM |
100 | */ |
101 | static void eeh_disable_irq(struct pci_dev *dev) | |
102 | { | |
40a7cd92 | 103 | struct eeh_dev *edev = pci_dev_to_eeh_dev(dev); |
8535ef05 MM |
104 | |
105 | /* Don't disable MSI and MSI-X interrupts. They are | |
106 | * effectively disabled by the DMA Stopped state | |
107 | * when an EEH error occurs. | |
29f8bf1b | 108 | */ |
8535ef05 MM |
109 | if (dev->msi_enabled || dev->msix_enabled) |
110 | return; | |
111 | ||
59e3f837 | 112 | if (!irq_has_action(dev->irq)) |
8535ef05 MM |
113 | return; |
114 | ||
dbbceee1 | 115 | edev->mode |= EEH_DEV_IRQ_DISABLED; |
8535ef05 MM |
116 | disable_irq_nosync(dev->irq); |
117 | } | |
118 | ||
119 | /** | |
29f8bf1b GS |
120 | * eeh_enable_irq - Enable interrupt for the recovering device |
121 | * @dev: PCI device | |
122 | * | |
123 | * This routine must be called to enable interrupt while failed | |
124 | * device could be resumed. | |
8535ef05 MM |
125 | */ |
126 | static void eeh_enable_irq(struct pci_dev *dev) | |
127 | { | |
40a7cd92 | 128 | struct eeh_dev *edev = pci_dev_to_eeh_dev(dev); |
8535ef05 | 129 | |
dbbceee1 GS |
130 | if ((edev->mode) & EEH_DEV_IRQ_DISABLED) { |
131 | edev->mode &= ~EEH_DEV_IRQ_DISABLED; | |
b8a9a11b TG |
132 | /* |
133 | * FIXME !!!!! | |
134 | * | |
135 | * This is just ass backwards. This maze has | |
136 | * unbalanced irq_enable/disable calls. So instead of | |
137 | * finding the root cause it works around the warning | |
138 | * in the irq_enable code by conditionally calling | |
139 | * into it. | |
140 | * | |
141 | * That's just wrong.The warning in the core code is | |
142 | * there to tell people to fix their assymetries in | |
143 | * their own code, not by abusing the core information | |
144 | * to avoid it. | |
145 | * | |
146 | * I so wish that the assymetry would be the other way | |
147 | * round and a few more irq_disable calls render that | |
148 | * shit unusable forever. | |
149 | * | |
150 | * tglx | |
151 | */ | |
57310c3c | 152 | if (irqd_irq_disabled(irq_get_irq_data(dev->irq))) |
91150af3 | 153 | enable_irq(dev->irq); |
57310c3c | 154 | } |
8535ef05 MM |
155 | } |
156 | ||
d2b0f6f7 GS |
157 | static bool eeh_dev_removed(struct eeh_dev *edev) |
158 | { | |
159 | /* EEH device removed ? */ | |
160 | if (!edev || (edev->mode & EEH_DEV_REMOVED)) | |
161 | return true; | |
162 | ||
163 | return false; | |
164 | } | |
165 | ||
5cfb20b9 GS |
166 | static void *eeh_dev_save_state(void *data, void *userdata) |
167 | { | |
168 | struct eeh_dev *edev = data; | |
169 | struct pci_dev *pdev; | |
170 | ||
171 | if (!edev) | |
172 | return NULL; | |
173 | ||
5a0cdbfd GS |
174 | /* |
175 | * We cannot access the config space on some adapters. | |
176 | * Otherwise, it will cause fenced PHB. We don't save | |
177 | * the content in their config space and will restore | |
178 | * from the initial config space saved when the EEH | |
179 | * device is created. | |
180 | */ | |
181 | if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED)) | |
182 | return NULL; | |
183 | ||
5cfb20b9 GS |
184 | pdev = eeh_dev_to_pci_dev(edev); |
185 | if (!pdev) | |
186 | return NULL; | |
187 | ||
188 | pci_save_state(pdev); | |
189 | return NULL; | |
190 | } | |
191 | ||
cb5b5624 | 192 | /** |
29f8bf1b | 193 | * eeh_report_error - Report pci error to each device driver |
9b3c76f0 | 194 | * @data: eeh device |
29f8bf1b | 195 | * @userdata: return value |
a84f273c GS |
196 | * |
197 | * Report an EEH error to each device driver, collect up and | |
198 | * merge the device driver responses. Cumulative response | |
cb5b5624 | 199 | * passed back in "userdata". |
77bd7415 | 200 | */ |
9b3c76f0 | 201 | static void *eeh_report_error(void *data, void *userdata) |
77bd7415 | 202 | { |
9b3c76f0 GS |
203 | struct eeh_dev *edev = (struct eeh_dev *)data; |
204 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
18eb3b39 | 205 | enum pci_ers_result rc, *res = userdata; |
feadf7c0 | 206 | struct pci_driver *driver; |
77bd7415 | 207 | |
2311cca5 | 208 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 | 209 | return NULL; |
77bd7415 LV |
210 | dev->error_state = pci_channel_io_frozen; |
211 | ||
feadf7c0 GS |
212 | driver = eeh_pcid_get(dev); |
213 | if (!driver) return NULL; | |
77bd7415 | 214 | |
8535ef05 MM |
215 | eeh_disable_irq(dev); |
216 | ||
6a1ca373 | 217 | if (!driver->err_handler || |
feadf7c0 GS |
218 | !driver->err_handler->error_detected) { |
219 | eeh_pcid_put(dev); | |
9b3c76f0 | 220 | return NULL; |
feadf7c0 | 221 | } |
77bd7415 | 222 | |
29f8bf1b | 223 | rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen); |
2a50f144 LV |
224 | |
225 | /* A driver that needs a reset trumps all others */ | |
226 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
18eb3b39 | 227 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 228 | |
67086e32 | 229 | edev->in_error = true; |
feadf7c0 | 230 | eeh_pcid_put(dev); |
9b3c76f0 | 231 | return NULL; |
6a1ca373 LV |
232 | } |
233 | ||
234 | /** | |
29f8bf1b | 235 | * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled |
9b3c76f0 | 236 | * @data: eeh device |
29f8bf1b | 237 | * @userdata: return value |
6a1ca373 | 238 | * |
638799b3 LV |
239 | * Tells each device driver that IO ports, MMIO and config space I/O |
240 | * are now enabled. Collects up and merges the device driver responses. | |
241 | * Cumulative response passed back in "userdata". | |
6a1ca373 | 242 | */ |
9b3c76f0 | 243 | static void *eeh_report_mmio_enabled(void *data, void *userdata) |
6a1ca373 | 244 | { |
9b3c76f0 GS |
245 | struct eeh_dev *edev = (struct eeh_dev *)data; |
246 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 247 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 248 | struct pci_driver *driver; |
6a1ca373 | 249 | |
2311cca5 | 250 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 GS |
251 | return NULL; |
252 | ||
feadf7c0 GS |
253 | driver = eeh_pcid_get(dev); |
254 | if (!driver) return NULL; | |
9b3c76f0 | 255 | |
feadf7c0 | 256 | if (!driver->err_handler || |
f26c7a03 GS |
257 | !driver->err_handler->mmio_enabled || |
258 | (edev->mode & EEH_DEV_NO_HANDLER)) { | |
feadf7c0 | 259 | eeh_pcid_put(dev); |
9b3c76f0 | 260 | return NULL; |
feadf7c0 | 261 | } |
6a1ca373 | 262 | |
29f8bf1b | 263 | rc = driver->err_handler->mmio_enabled(dev); |
2a50f144 LV |
264 | |
265 | /* A driver that needs a reset trumps all others */ | |
266 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
6a1ca373 | 267 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 268 | |
feadf7c0 | 269 | eeh_pcid_put(dev); |
9b3c76f0 | 270 | return NULL; |
77bd7415 LV |
271 | } |
272 | ||
cb5b5624 | 273 | /** |
29f8bf1b | 274 | * eeh_report_reset - Tell device that slot has been reset |
9b3c76f0 | 275 | * @data: eeh device |
29f8bf1b GS |
276 | * @userdata: return value |
277 | * | |
278 | * This routine must be called while EEH tries to reset particular | |
279 | * PCI device so that the associated PCI device driver could take | |
280 | * some actions, usually to save data the driver needs so that the | |
281 | * driver can work again while the device is recovered. | |
77bd7415 | 282 | */ |
9b3c76f0 | 283 | static void *eeh_report_reset(void *data, void *userdata) |
77bd7415 | 284 | { |
9b3c76f0 GS |
285 | struct eeh_dev *edev = (struct eeh_dev *)data; |
286 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 287 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 288 | struct pci_driver *driver; |
77bd7415 | 289 | |
2311cca5 | 290 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 | 291 | return NULL; |
c58dc575 MM |
292 | dev->error_state = pci_channel_io_normal; |
293 | ||
feadf7c0 GS |
294 | driver = eeh_pcid_get(dev); |
295 | if (!driver) return NULL; | |
296 | ||
8535ef05 MM |
297 | eeh_enable_irq(dev); |
298 | ||
6a1ca373 | 299 | if (!driver->err_handler || |
f26c7a03 | 300 | !driver->err_handler->slot_reset || |
67086e32 WY |
301 | (edev->mode & EEH_DEV_NO_HANDLER) || |
302 | (!edev->in_error)) { | |
feadf7c0 | 303 | eeh_pcid_put(dev); |
9b3c76f0 | 304 | return NULL; |
feadf7c0 | 305 | } |
77bd7415 | 306 | |
6a1ca373 | 307 | rc = driver->err_handler->slot_reset(dev); |
5794dbcb LV |
308 | if ((*res == PCI_ERS_RESULT_NONE) || |
309 | (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; | |
6a1ca373 LV |
310 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
311 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
70298c6e | 312 | |
feadf7c0 | 313 | eeh_pcid_put(dev); |
9b3c76f0 | 314 | return NULL; |
77bd7415 LV |
315 | } |
316 | ||
5cfb20b9 GS |
317 | static void *eeh_dev_restore_state(void *data, void *userdata) |
318 | { | |
319 | struct eeh_dev *edev = data; | |
320 | struct pci_dev *pdev; | |
321 | ||
322 | if (!edev) | |
323 | return NULL; | |
324 | ||
5a0cdbfd GS |
325 | /* |
326 | * The content in the config space isn't saved because | |
327 | * the blocked config space on some adapters. We have | |
328 | * to restore the initial saved config space when the | |
329 | * EEH device is created. | |
330 | */ | |
331 | if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED)) { | |
332 | if (list_is_last(&edev->list, &edev->pe->edevs)) | |
333 | eeh_pe_restore_bars(edev->pe); | |
334 | ||
335 | return NULL; | |
336 | } | |
337 | ||
5cfb20b9 GS |
338 | pdev = eeh_dev_to_pci_dev(edev); |
339 | if (!pdev) | |
340 | return NULL; | |
341 | ||
342 | pci_restore_state(pdev); | |
343 | return NULL; | |
344 | } | |
345 | ||
cb5b5624 | 346 | /** |
29f8bf1b | 347 | * eeh_report_resume - Tell device to resume normal operations |
9b3c76f0 | 348 | * @data: eeh device |
29f8bf1b GS |
349 | * @userdata: return value |
350 | * | |
351 | * This routine must be called to notify the device driver that it | |
352 | * could resume so that the device driver can do some initialization | |
353 | * to make the recovered device work again. | |
cb5b5624 | 354 | */ |
9b3c76f0 | 355 | static void *eeh_report_resume(void *data, void *userdata) |
77bd7415 | 356 | { |
9b3c76f0 GS |
357 | struct eeh_dev *edev = (struct eeh_dev *)data; |
358 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
67086e32 | 359 | bool was_in_error; |
9b3c76f0 GS |
360 | struct pci_driver *driver; |
361 | ||
2311cca5 | 362 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 | 363 | return NULL; |
77bd7415 LV |
364 | dev->error_state = pci_channel_io_normal; |
365 | ||
feadf7c0 GS |
366 | driver = eeh_pcid_get(dev); |
367 | if (!driver) return NULL; | |
d0e70341 | 368 | |
67086e32 WY |
369 | was_in_error = edev->in_error; |
370 | edev->in_error = false; | |
8535ef05 MM |
371 | eeh_enable_irq(dev); |
372 | ||
d0e70341 | 373 | if (!driver->err_handler || |
f26c7a03 | 374 | !driver->err_handler->resume || |
67086e32 | 375 | (edev->mode & EEH_DEV_NO_HANDLER) || !was_in_error) { |
f26c7a03 | 376 | edev->mode &= ~EEH_DEV_NO_HANDLER; |
feadf7c0 | 377 | eeh_pcid_put(dev); |
9b3c76f0 | 378 | return NULL; |
feadf7c0 | 379 | } |
77bd7415 LV |
380 | |
381 | driver->err_handler->resume(dev); | |
70298c6e | 382 | |
feadf7c0 | 383 | eeh_pcid_put(dev); |
9b3c76f0 | 384 | return NULL; |
77bd7415 LV |
385 | } |
386 | ||
cb5b5624 | 387 | /** |
29f8bf1b | 388 | * eeh_report_failure - Tell device driver that device is dead. |
9b3c76f0 | 389 | * @data: eeh device |
29f8bf1b | 390 | * @userdata: return value |
cb5b5624 LV |
391 | * |
392 | * This informs the device driver that the device is permanently | |
393 | * dead, and that no further recovery attempts will be made on it. | |
394 | */ | |
9b3c76f0 | 395 | static void *eeh_report_failure(void *data, void *userdata) |
77bd7415 | 396 | { |
9b3c76f0 GS |
397 | struct eeh_dev *edev = (struct eeh_dev *)data; |
398 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
399 | struct pci_driver *driver; | |
400 | ||
2311cca5 | 401 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 | 402 | return NULL; |
77bd7415 LV |
403 | dev->error_state = pci_channel_io_perm_failure; |
404 | ||
feadf7c0 GS |
405 | driver = eeh_pcid_get(dev); |
406 | if (!driver) return NULL; | |
77bd7415 | 407 | |
8535ef05 MM |
408 | eeh_disable_irq(dev); |
409 | ||
410 | if (!driver->err_handler || | |
feadf7c0 GS |
411 | !driver->err_handler->error_detected) { |
412 | eeh_pcid_put(dev); | |
9b3c76f0 | 413 | return NULL; |
feadf7c0 | 414 | } |
8535ef05 | 415 | |
77bd7415 | 416 | driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); |
70298c6e | 417 | |
feadf7c0 | 418 | eeh_pcid_put(dev); |
9b3c76f0 | 419 | return NULL; |
77bd7415 LV |
420 | } |
421 | ||
67086e32 WY |
422 | static void *eeh_add_virt_device(void *data, void *userdata) |
423 | { | |
424 | struct pci_driver *driver; | |
425 | struct eeh_dev *edev = (struct eeh_dev *)data; | |
426 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
427 | struct pci_dn *pdn = eeh_dev_to_pdn(edev); | |
428 | ||
429 | if (!(edev->physfn)) { | |
430 | pr_warn("%s: EEH dev %04x:%02x:%02x.%01x not for VF\n", | |
431 | __func__, edev->phb->global_number, pdn->busno, | |
432 | PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn)); | |
433 | return NULL; | |
434 | } | |
435 | ||
436 | driver = eeh_pcid_get(dev); | |
437 | if (driver) { | |
438 | eeh_pcid_put(dev); | |
439 | if (driver->err_handler) | |
440 | return NULL; | |
441 | } | |
442 | ||
443 | #ifdef CONFIG_PPC_POWERNV | |
444 | pci_iov_add_virtfn(edev->physfn, pdn->vf_index, 0); | |
445 | #endif | |
446 | return NULL; | |
447 | } | |
448 | ||
f5c57710 GS |
449 | static void *eeh_rmv_device(void *data, void *userdata) |
450 | { | |
451 | struct pci_driver *driver; | |
452 | struct eeh_dev *edev = (struct eeh_dev *)data; | |
453 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
67086e32 WY |
454 | struct eeh_rmv_data *rmv_data = (struct eeh_rmv_data *)userdata; |
455 | int *removed = rmv_data ? &rmv_data->removed : NULL; | |
f5c57710 GS |
456 | |
457 | /* | |
458 | * Actually, we should remove the PCI bridges as well. | |
459 | * However, that's lots of complexity to do that, | |
460 | * particularly some of devices under the bridge might | |
461 | * support EEH. So we just care about PCI devices for | |
462 | * simplicity here. | |
463 | */ | |
93de6901 | 464 | if (!dev || (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)) |
f5c57710 | 465 | return NULL; |
8cc6b6cd | 466 | |
d2b0f6f7 GS |
467 | /* |
468 | * We rely on count-based pcibios_release_device() to | |
469 | * detach permanently offlined PEs. Unfortunately, that's | |
470 | * not reliable enough. We might have the permanently | |
471 | * offlined PEs attached, but we needn't take care of | |
472 | * them and their child devices. | |
473 | */ | |
474 | if (eeh_dev_removed(edev)) | |
475 | return NULL; | |
476 | ||
f5c57710 | 477 | driver = eeh_pcid_get(dev); |
8cc6b6cd TLSC |
478 | if (driver) { |
479 | eeh_pcid_put(dev); | |
3fa7bf72 GS |
480 | if (removed && |
481 | eeh_pe_passed(edev->pe)) | |
482 | return NULL; | |
67086e32 WY |
483 | if (removed && |
484 | driver->err_handler && | |
f2da4ccf | 485 | driver->err_handler->error_detected && |
f6bf0fa1 | 486 | driver->err_handler->slot_reset) |
8cc6b6cd TLSC |
487 | return NULL; |
488 | } | |
f5c57710 GS |
489 | |
490 | /* Remove it from PCI subsystem */ | |
491 | pr_debug("EEH: Removing %s without EEH sensitive driver\n", | |
492 | pci_name(dev)); | |
493 | edev->bus = dev->bus; | |
494 | edev->mode |= EEH_DEV_DISCONNECTED; | |
67086e32 WY |
495 | if (removed) |
496 | (*removed)++; | |
f5c57710 | 497 | |
67086e32 WY |
498 | if (edev->physfn) { |
499 | #ifdef CONFIG_PPC_POWERNV | |
500 | struct pci_dn *pdn = eeh_dev_to_pdn(edev); | |
501 | ||
502 | pci_iov_remove_virtfn(edev->physfn, pdn->vf_index, 0); | |
503 | edev->pdev = NULL; | |
504 | ||
505 | /* | |
506 | * We have to set the VF PE number to invalid one, which is | |
507 | * required to plug the VF successfully. | |
508 | */ | |
509 | pdn->pe_number = IODA_INVALID_PE; | |
510 | #endif | |
511 | if (rmv_data) | |
512 | list_add(&edev->rmv_list, &rmv_data->edev_list); | |
513 | } else { | |
514 | pci_lock_rescan_remove(); | |
515 | pci_stop_and_remove_bus_device(dev); | |
516 | pci_unlock_rescan_remove(); | |
517 | } | |
f5c57710 GS |
518 | |
519 | return NULL; | |
520 | } | |
521 | ||
522 | static void *eeh_pe_detach_dev(void *data, void *userdata) | |
523 | { | |
524 | struct eeh_pe *pe = (struct eeh_pe *)data; | |
525 | struct eeh_dev *edev, *tmp; | |
526 | ||
527 | eeh_pe_for_each_dev(pe, edev, tmp) { | |
528 | if (!(edev->mode & EEH_DEV_DISCONNECTED)) | |
529 | continue; | |
530 | ||
531 | edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED); | |
532 | eeh_rmv_from_parent_pe(edev); | |
533 | } | |
534 | ||
535 | return NULL; | |
536 | } | |
537 | ||
78954700 GS |
538 | /* |
539 | * Explicitly clear PE's frozen state for PowerNV where | |
540 | * we have frozen PE until BAR restore is completed. It's | |
541 | * harmless to clear it for pSeries. To be consistent with | |
542 | * PE reset (for 3 times), we try to clear the frozen state | |
543 | * for 3 times as well. | |
544 | */ | |
2c665992 | 545 | static void *__eeh_clear_pe_frozen_state(void *data, void *flag) |
78954700 | 546 | { |
2c665992 | 547 | struct eeh_pe *pe = (struct eeh_pe *)data; |
5cfb20b9 | 548 | bool *clear_sw_state = flag; |
c9dd0143 | 549 | int i, rc = 1; |
78954700 | 550 | |
c9dd0143 | 551 | for (i = 0; rc && i < 3; i++) |
5cfb20b9 | 552 | rc = eeh_unfreeze_pe(pe, clear_sw_state); |
78954700 | 553 | |
c9dd0143 | 554 | /* Stop immediately on any errors */ |
2c665992 | 555 | if (rc) { |
c9dd0143 GS |
556 | pr_warn("%s: Failure %d unfreezing PHB#%x-PE#%x\n", |
557 | __func__, rc, pe->phb->global_number, pe->addr); | |
2c665992 GS |
558 | return (void *)pe; |
559 | } | |
560 | ||
561 | return NULL; | |
562 | } | |
563 | ||
5cfb20b9 GS |
564 | static int eeh_clear_pe_frozen_state(struct eeh_pe *pe, |
565 | bool clear_sw_state) | |
2c665992 GS |
566 | { |
567 | void *rc; | |
568 | ||
5cfb20b9 | 569 | rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, &clear_sw_state); |
2c665992 | 570 | if (!rc) |
78954700 GS |
571 | eeh_pe_state_clear(pe, EEH_PE_ISOLATED); |
572 | ||
2c665992 | 573 | return rc ? -EIO : 0; |
78954700 GS |
574 | } |
575 | ||
5cfb20b9 GS |
576 | int eeh_pe_reset_and_recover(struct eeh_pe *pe) |
577 | { | |
2efc771f | 578 | int ret; |
5cfb20b9 GS |
579 | |
580 | /* Bail if the PE is being recovered */ | |
581 | if (pe->state & EEH_PE_RECOVERING) | |
582 | return 0; | |
583 | ||
584 | /* Put the PE into recovery mode */ | |
585 | eeh_pe_state_mark(pe, EEH_PE_RECOVERING); | |
586 | ||
587 | /* Save states */ | |
588 | eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL); | |
589 | ||
5cfb20b9 | 590 | /* Issue reset */ |
5cfb20b9 GS |
591 | ret = eeh_reset_pe(pe); |
592 | if (ret) { | |
28bf36f9 | 593 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); |
5cfb20b9 GS |
594 | return ret; |
595 | } | |
5cfb20b9 GS |
596 | |
597 | /* Unfreeze the PE */ | |
598 | ret = eeh_clear_pe_frozen_state(pe, true); | |
599 | if (ret) { | |
600 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); | |
601 | return ret; | |
602 | } | |
603 | ||
5cfb20b9 GS |
604 | /* Restore device state */ |
605 | eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL); | |
606 | ||
5cfb20b9 GS |
607 | /* Clear recovery mode */ |
608 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); | |
609 | ||
610 | return 0; | |
611 | } | |
612 | ||
77bd7415 | 613 | /** |
29f8bf1b | 614 | * eeh_reset_device - Perform actual reset of a pci slot |
9b3c76f0 | 615 | * @pe: EEH PE |
29f8bf1b | 616 | * @bus: PCI bus corresponding to the isolcated slot |
77bd7415 | 617 | * |
29f8bf1b GS |
618 | * This routine must be called to do reset on the indicated PE. |
619 | * During the reset, udev might be invoked because those affected | |
620 | * PCI devices will be removed and then added. | |
77bd7415 | 621 | */ |
67086e32 WY |
622 | static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus, |
623 | struct eeh_rmv_data *rmv_data) | |
77bd7415 | 624 | { |
f5c57710 | 625 | struct pci_bus *frozen_bus = eeh_pe_bus_get(pe); |
5a71978e | 626 | struct timeval tstamp; |
67086e32 WY |
627 | int cnt, rc; |
628 | struct eeh_dev *edev; | |
42405456 LV |
629 | |
630 | /* pcibios will clear the counter; save the value */ | |
9b3c76f0 | 631 | cnt = pe->freeze_count; |
5a71978e | 632 | tstamp = pe->tstamp; |
42405456 | 633 | |
20ee6a97 GS |
634 | /* |
635 | * We don't remove the corresponding PE instances because | |
636 | * we need the information afterwords. The attached EEH | |
637 | * devices are expected to be attached soon when calling | |
bd251b89 | 638 | * into pci_hp_add_devices(). |
20ee6a97 | 639 | */ |
f5c57710 | 640 | eeh_pe_state_mark(pe, EEH_PE_KEEP); |
1c2042c8 | 641 | if (bus) { |
67086e32 WY |
642 | if (pe->type & EEH_PE_VF) { |
643 | eeh_pe_dev_traverse(pe, eeh_rmv_device, NULL); | |
644 | } else { | |
67086e32 | 645 | pci_lock_rescan_remove(); |
bd251b89 | 646 | pci_hp_remove_devices(bus); |
67086e32 WY |
647 | pci_unlock_rescan_remove(); |
648 | } | |
1c2042c8 | 649 | } else if (frozen_bus) { |
cca0e542 | 650 | eeh_pe_dev_traverse(pe, eeh_rmv_device, rmv_data); |
1c2042c8 | 651 | } |
77bd7415 | 652 | |
d0914f50 GS |
653 | /* |
654 | * Reset the pci controller. (Asserts RST#; resets config space). | |
b6495c0c | 655 | * Reconfigure bridges and devices. Don't try to bring the system |
29f8bf1b | 656 | * up if the reset failed for some reason. |
d0914f50 GS |
657 | * |
658 | * During the reset, it's very dangerous to have uncontrolled PCI | |
659 | * config accesses. So we prefer to block them. However, controlled | |
660 | * PCI config accesses initiated from EEH itself are allowed. | |
29f8bf1b | 661 | */ |
9b3c76f0 | 662 | rc = eeh_reset_pe(pe); |
28bf36f9 | 663 | if (rc) |
b6495c0c | 664 | return rc; |
77bd7415 | 665 | |
1c2042c8 RW |
666 | pci_lock_rescan_remove(); |
667 | ||
9b3c76f0 GS |
668 | /* Restore PE */ |
669 | eeh_ops->configure_bridge(pe); | |
670 | eeh_pe_restore_bars(pe); | |
77bd7415 | 671 | |
dc9c41bd AD |
672 | /* Clear frozen state */ |
673 | rc = eeh_clear_pe_frozen_state(pe, false); | |
674 | if (rc) | |
675 | return rc; | |
78954700 | 676 | |
77bd7415 | 677 | /* Give the system 5 seconds to finish running the user-space |
a84f273c GS |
678 | * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, |
679 | * this is a hack, but if we don't do this, and try to bring | |
680 | * the device up before the scripts have taken it down, | |
77bd7415 LV |
681 | * potentially weird things happen. |
682 | */ | |
683 | if (bus) { | |
f5c57710 | 684 | pr_info("EEH: Sleep 5s ahead of complete hotplug\n"); |
29f8bf1b | 685 | ssleep(5); |
f5c57710 GS |
686 | |
687 | /* | |
688 | * The EEH device is still connected with its parent | |
689 | * PE. We should disconnect it so the binding can be | |
690 | * rebuilt when adding PCI devices. | |
691 | */ | |
67086e32 | 692 | edev = list_first_entry(&pe->edevs, struct eeh_dev, list); |
f5c57710 | 693 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); |
a3aa256b | 694 | if (pe->type & EEH_PE_VF) { |
67086e32 | 695 | eeh_add_virt_device(edev, NULL); |
a3aa256b GS |
696 | } else { |
697 | eeh_pe_state_clear(pe, EEH_PE_PRI_BUS); | |
bd251b89 | 698 | pci_hp_add_devices(bus); |
a3aa256b | 699 | } |
67086e32 | 700 | } else if (frozen_bus && rmv_data->removed) { |
f5c57710 GS |
701 | pr_info("EEH: Sleep 5s ahead of partial hotplug\n"); |
702 | ssleep(5); | |
703 | ||
67086e32 | 704 | edev = list_first_entry(&pe->edevs, struct eeh_dev, list); |
f5c57710 | 705 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); |
67086e32 WY |
706 | if (pe->type & EEH_PE_VF) |
707 | eeh_add_virt_device(edev, NULL); | |
708 | else | |
bd251b89 | 709 | pci_hp_add_devices(frozen_bus); |
77bd7415 | 710 | } |
f5c57710 | 711 | eeh_pe_state_clear(pe, EEH_PE_KEEP); |
5a71978e GS |
712 | |
713 | pe->tstamp = tstamp; | |
9b3c76f0 | 714 | pe->freeze_count = cnt; |
b6495c0c | 715 | |
1c2042c8 | 716 | pci_unlock_rescan_remove(); |
b6495c0c | 717 | return 0; |
77bd7415 LV |
718 | } |
719 | ||
720 | /* The longest amount of time to wait for a pci device | |
721 | * to come back on line, in seconds. | |
722 | */ | |
fb48dc22 | 723 | #define MAX_WAIT_FOR_RECOVERY 300 |
77bd7415 | 724 | |
8a6b1bc7 | 725 | static void eeh_handle_normal_event(struct eeh_pe *pe) |
77bd7415 | 726 | { |
77bd7415 | 727 | struct pci_bus *frozen_bus; |
67086e32 | 728 | struct eeh_dev *edev, *tmp; |
b6495c0c | 729 | int rc = 0; |
18eb3b39 | 730 | enum pci_ers_result result = PCI_ERS_RESULT_NONE; |
67086e32 | 731 | struct eeh_rmv_data rmv_data = {LIST_HEAD_INIT(rmv_data.edev_list), 0}; |
77bd7415 | 732 | |
9b3c76f0 | 733 | frozen_bus = eeh_pe_bus_get(pe); |
77bd7415 | 734 | if (!frozen_bus) { |
9b3c76f0 GS |
735 | pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n", |
736 | __func__, pe->phb->global_number, pe->addr); | |
737 | return; | |
77bd7415 LV |
738 | } |
739 | ||
5a71978e | 740 | eeh_pe_update_time_stamp(pe); |
9b3c76f0 | 741 | pe->freeze_count++; |
1b28f170 | 742 | if (pe->freeze_count > eeh_max_freezes) |
8df83028 | 743 | goto excess_failures; |
0dae2743 | 744 | pr_warn("EEH: This PCI device has failed %d times in the last hour\n", |
9b3c76f0 | 745 | pe->freeze_count); |
77bd7415 LV |
746 | |
747 | /* Walk the various device drivers attached to this slot through | |
748 | * a reset sequence, giving each an opportunity to do what it needs | |
749 | * to accomplish the reset. Each child gets a report of the | |
750 | * status ... if any child can't handle the reset, then the entire | |
751 | * slot is dlpar removed and added. | |
8234fced GS |
752 | * |
753 | * When the PHB is fenced, we have to issue a reset to recover from | |
754 | * the error. Override the result if necessary to have partially | |
755 | * hotplug for this case. | |
77bd7415 | 756 | */ |
56ca4fde | 757 | pr_info("EEH: Notify device drivers to shutdown\n"); |
9b3c76f0 | 758 | eeh_pe_dev_traverse(pe, eeh_report_error, &result); |
8234fced GS |
759 | if ((pe->type & EEH_PE_PHB) && |
760 | result != PCI_ERS_RESULT_NONE && | |
761 | result != PCI_ERS_RESULT_NEED_RESET) | |
762 | result = PCI_ERS_RESULT_NEED_RESET; | |
77bd7415 | 763 | |
5f1a7c81 | 764 | /* Get the current PCI slot state. This can take a long time, |
2ac3990c | 765 | * sometimes over 300 seconds for certain systems. |
29f8bf1b | 766 | */ |
9b3c76f0 | 767 | rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000); |
eb594a47 | 768 | if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) { |
0dae2743 | 769 | pr_warn("EEH: Permanent failure\n"); |
5f1a7c81 LV |
770 | goto hard_fail; |
771 | } | |
772 | ||
ede8ca26 LV |
773 | /* Since rtas may enable MMIO when posting the error log, |
774 | * don't post the error log until after all dev drivers | |
17213c3b LV |
775 | * have been informed. |
776 | */ | |
56ca4fde | 777 | pr_info("EEH: Collect temporary log\n"); |
9b3c76f0 | 778 | eeh_slot_error_detail(pe, EEH_LOG_TEMP); |
ede8ca26 | 779 | |
77bd7415 LV |
780 | /* If all device drivers were EEH-unaware, then shut |
781 | * down all of the device drivers, and hope they | |
782 | * go down willingly, without panicing the system. | |
783 | */ | |
18eb3b39 | 784 | if (result == PCI_ERS_RESULT_NONE) { |
56ca4fde | 785 | pr_info("EEH: Reset with hotplug activity\n"); |
67086e32 | 786 | rc = eeh_reset_device(pe, frozen_bus, NULL); |
e0f90b64 | 787 | if (rc) { |
0dae2743 GS |
788 | pr_warn("%s: Unable to reset, err=%d\n", |
789 | __func__, rc); | |
b6495c0c | 790 | goto hard_fail; |
e0f90b64 | 791 | } |
77bd7415 LV |
792 | } |
793 | ||
6a1ca373 LV |
794 | /* If all devices reported they can proceed, then re-enable MMIO */ |
795 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { | |
56ca4fde | 796 | pr_info("EEH: Enable I/O for affected devices\n"); |
9b3c76f0 | 797 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO); |
6a1ca373 | 798 | |
fa1be476 LV |
799 | if (rc < 0) |
800 | goto hard_fail; | |
6a1ca373 LV |
801 | if (rc) { |
802 | result = PCI_ERS_RESULT_NEED_RESET; | |
803 | } else { | |
56ca4fde | 804 | pr_info("EEH: Notify device drivers to resume I/O\n"); |
9b3c76f0 | 805 | eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result); |
6a1ca373 | 806 | } |
77bd7415 LV |
807 | } |
808 | ||
6a1ca373 | 809 | /* If all devices reported they can proceed, then re-enable DMA */ |
18eb3b39 | 810 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { |
56ca4fde | 811 | pr_info("EEH: Enabled DMA for affected devices\n"); |
9b3c76f0 | 812 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA); |
6a1ca373 | 813 | |
fa1be476 LV |
814 | if (rc < 0) |
815 | goto hard_fail; | |
35845a78 | 816 | if (rc) { |
6a1ca373 | 817 | result = PCI_ERS_RESULT_NEED_RESET; |
35845a78 GS |
818 | } else { |
819 | /* | |
820 | * We didn't do PE reset for the case. The PE | |
821 | * is still in frozen state. Clear it before | |
822 | * resuming the PE. | |
823 | */ | |
824 | eeh_pe_state_clear(pe, EEH_PE_ISOLATED); | |
d0e70341 | 825 | result = PCI_ERS_RESULT_RECOVERED; |
35845a78 | 826 | } |
6a1ca373 LV |
827 | } |
828 | ||
829 | /* If any device has a hard failure, then shut off everything. */ | |
e0f90b64 | 830 | if (result == PCI_ERS_RESULT_DISCONNECT) { |
0dae2743 | 831 | pr_warn("EEH: Device driver gave up\n"); |
6a1ca373 | 832 | goto hard_fail; |
e0f90b64 | 833 | } |
6a1ca373 LV |
834 | |
835 | /* If any device called out for a reset, then reset the slot */ | |
836 | if (result == PCI_ERS_RESULT_NEED_RESET) { | |
56ca4fde | 837 | pr_info("EEH: Reset without hotplug activity\n"); |
67086e32 | 838 | rc = eeh_reset_device(pe, NULL, &rmv_data); |
e0f90b64 | 839 | if (rc) { |
0dae2743 GS |
840 | pr_warn("%s: Cannot reset, err=%d\n", |
841 | __func__, rc); | |
b6495c0c | 842 | goto hard_fail; |
e0f90b64 | 843 | } |
56ca4fde GS |
844 | |
845 | pr_info("EEH: Notify device drivers " | |
846 | "the completion of reset\n"); | |
6a1ca373 | 847 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 848 | eeh_pe_dev_traverse(pe, eeh_report_reset, &result); |
77bd7415 LV |
849 | } |
850 | ||
6a1ca373 | 851 | /* All devices should claim they have recovered by now. */ |
90fdd613 LV |
852 | if ((result != PCI_ERS_RESULT_RECOVERED) && |
853 | (result != PCI_ERS_RESULT_NONE)) { | |
0dae2743 | 854 | pr_warn("EEH: Not recovered\n"); |
6a1ca373 | 855 | goto hard_fail; |
e0f90b64 | 856 | } |
6a1ca373 | 857 | |
67086e32 WY |
858 | /* |
859 | * For those hot removed VFs, we should add back them after PF get | |
860 | * recovered properly. | |
861 | */ | |
862 | list_for_each_entry_safe(edev, tmp, &rmv_data.edev_list, rmv_list) { | |
863 | eeh_add_virt_device(edev, NULL); | |
864 | list_del(&edev->rmv_list); | |
865 | } | |
866 | ||
77bd7415 | 867 | /* Tell all device drivers that they can resume operations */ |
56ca4fde | 868 | pr_info("EEH: Notify device driver to resume\n"); |
9b3c76f0 | 869 | eeh_pe_dev_traverse(pe, eeh_report_resume, NULL); |
b6495c0c | 870 | |
9b3c76f0 | 871 | return; |
a84f273c | 872 | |
8df83028 | 873 | excess_failures: |
b6495c0c LV |
874 | /* |
875 | * About 90% of all real-life EEH failures in the field | |
876 | * are due to poorly seated PCI cards. Only 10% or so are | |
877 | * due to actual, failed cards. | |
878 | */ | |
9b3c76f0 GS |
879 | pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n" |
880 | "last hour and has been permanently disabled.\n" | |
881 | "Please try reseating or replacing it.\n", | |
882 | pe->phb->global_number, pe->addr, | |
883 | pe->freeze_count); | |
8df83028 LV |
884 | goto perm_error; |
885 | ||
886 | hard_fail: | |
9b3c76f0 GS |
887 | pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n" |
888 | "Please try reseating or replacing it\n", | |
889 | pe->phb->global_number, pe->addr); | |
b6495c0c | 890 | |
8df83028 | 891 | perm_error: |
9b3c76f0 | 892 | eeh_slot_error_detail(pe, EEH_LOG_PERM); |
b6495c0c LV |
893 | |
894 | /* Notify all devices that they're about to go down. */ | |
9b3c76f0 | 895 | eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); |
b6495c0c | 896 | |
d2b0f6f7 | 897 | /* Mark the PE to be removed permanently */ |
432227e9 | 898 | eeh_pe_state_mark(pe, EEH_PE_REMOVED); |
d2b0f6f7 GS |
899 | |
900 | /* | |
901 | * Shut down the device drivers for good. We mark | |
902 | * all removed devices correctly to avoid access | |
903 | * the their PCI config any more. | |
904 | */ | |
1c2042c8 | 905 | if (frozen_bus) { |
67086e32 WY |
906 | if (pe->type & EEH_PE_VF) { |
907 | eeh_pe_dev_traverse(pe, eeh_rmv_device, NULL); | |
908 | eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED); | |
909 | } else { | |
910 | eeh_pe_state_clear(pe, EEH_PE_PRI_BUS); | |
911 | eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED); | |
d2b0f6f7 | 912 | |
67086e32 | 913 | pci_lock_rescan_remove(); |
bd251b89 | 914 | pci_hp_remove_devices(frozen_bus); |
67086e32 WY |
915 | pci_unlock_rescan_remove(); |
916 | } | |
1c2042c8 | 917 | } |
77bd7415 | 918 | } |
8a6b1bc7 GS |
919 | |
920 | static void eeh_handle_special_event(void) | |
921 | { | |
922 | struct eeh_pe *pe, *phb_pe; | |
923 | struct pci_bus *bus; | |
7e4e7867 | 924 | struct pci_controller *hose; |
8a6b1bc7 | 925 | unsigned long flags; |
7e4e7867 | 926 | int rc; |
8a6b1bc7 | 927 | |
8a6b1bc7 | 928 | |
7e4e7867 GS |
929 | do { |
930 | rc = eeh_ops->next_error(&pe); | |
931 | ||
932 | switch (rc) { | |
933 | case EEH_NEXT_ERR_DEAD_IOC: | |
934 | /* Mark all PHBs in dead state */ | |
935 | eeh_serialize_lock(&flags); | |
936 | ||
937 | /* Purge all events */ | |
5c7a35e3 | 938 | eeh_remove_event(NULL, true); |
7e4e7867 GS |
939 | |
940 | list_for_each_entry(hose, &hose_list, list_node) { | |
941 | phb_pe = eeh_phb_pe_get(hose); | |
942 | if (!phb_pe) continue; | |
943 | ||
9e049375 | 944 | eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED); |
7e4e7867 GS |
945 | } |
946 | ||
947 | eeh_serialize_unlock(flags); | |
948 | ||
949 | break; | |
950 | case EEH_NEXT_ERR_FROZEN_PE: | |
951 | case EEH_NEXT_ERR_FENCED_PHB: | |
952 | case EEH_NEXT_ERR_DEAD_PHB: | |
953 | /* Mark the PE in fenced state */ | |
954 | eeh_serialize_lock(&flags); | |
955 | ||
956 | /* Purge all events of the PHB */ | |
5c7a35e3 | 957 | eeh_remove_event(pe, true); |
7e4e7867 GS |
958 | |
959 | if (rc == EEH_NEXT_ERR_DEAD_PHB) | |
9e049375 | 960 | eeh_pe_state_mark(pe, EEH_PE_ISOLATED); |
7e4e7867 GS |
961 | else |
962 | eeh_pe_state_mark(pe, | |
963 | EEH_PE_ISOLATED | EEH_PE_RECOVERING); | |
964 | ||
965 | eeh_serialize_unlock(flags); | |
966 | ||
967 | break; | |
968 | case EEH_NEXT_ERR_NONE: | |
969 | return; | |
970 | default: | |
971 | pr_warn("%s: Invalid value %d from next_error()\n", | |
972 | __func__, rc); | |
973 | return; | |
8a6b1bc7 | 974 | } |
8a6b1bc7 | 975 | |
7e4e7867 GS |
976 | /* |
977 | * For fenced PHB and frozen PE, it's handled as normal | |
978 | * event. We have to remove the affected PHBs for dead | |
979 | * PHB and IOC | |
980 | */ | |
981 | if (rc == EEH_NEXT_ERR_FROZEN_PE || | |
982 | rc == EEH_NEXT_ERR_FENCED_PHB) { | |
983 | eeh_handle_normal_event(pe); | |
9e049375 | 984 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); |
7e4e7867 | 985 | } else { |
1b17366d | 986 | pci_lock_rescan_remove(); |
7e4e7867 GS |
987 | list_for_each_entry(hose, &hose_list, list_node) { |
988 | phb_pe = eeh_phb_pe_get(hose); | |
989 | if (!phb_pe || | |
9e049375 GS |
990 | !(phb_pe->state & EEH_PE_ISOLATED) || |
991 | (phb_pe->state & EEH_PE_RECOVERING)) | |
7e4e7867 GS |
992 | continue; |
993 | ||
994 | /* Notify all devices to be down */ | |
05ba75f8 | 995 | eeh_pe_state_clear(pe, EEH_PE_PRI_BUS); |
7e4e7867 GS |
996 | bus = eeh_pe_bus_get(phb_pe); |
997 | eeh_pe_dev_traverse(pe, | |
998 | eeh_report_failure, NULL); | |
bd251b89 | 999 | pci_hp_remove_devices(bus); |
7e4e7867 | 1000 | } |
1b17366d | 1001 | pci_unlock_rescan_remove(); |
8a6b1bc7 | 1002 | } |
7e4e7867 GS |
1003 | |
1004 | /* | |
1005 | * If we have detected dead IOC, we needn't proceed | |
1006 | * any more since all PHBs would have been removed | |
1007 | */ | |
1008 | if (rc == EEH_NEXT_ERR_DEAD_IOC) | |
1009 | break; | |
1010 | } while (rc != EEH_NEXT_ERR_NONE); | |
8a6b1bc7 GS |
1011 | } |
1012 | ||
1013 | /** | |
1014 | * eeh_handle_event - Reset a PCI device after hard lockup. | |
1015 | * @pe: EEH PE | |
1016 | * | |
1017 | * While PHB detects address or data parity errors on particular PCI | |
1018 | * slot, the associated PE will be frozen. Besides, DMA's occurring | |
1019 | * to wild addresses (which usually happen due to bugs in device | |
1020 | * drivers or in PCI adapter firmware) can cause EEH error. #SERR, | |
1021 | * #PERR or other misc PCI-related errors also can trigger EEH errors. | |
1022 | * | |
1023 | * Recovery process consists of unplugging the device driver (which | |
1024 | * generated hotplug events to userspace), then issuing a PCI #RST to | |
1025 | * the device, then reconfiguring the PCI config space for all bridges | |
1026 | * & devices under this slot, and then finally restarting the device | |
1027 | * drivers (which cause a second set of hotplug events to go out to | |
1028 | * userspace). | |
1029 | */ | |
1030 | void eeh_handle_event(struct eeh_pe *pe) | |
1031 | { | |
1032 | if (pe) | |
1033 | eeh_handle_normal_event(pe); | |
1034 | else | |
1035 | eeh_handle_special_event(); | |
1036 | } |