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 | ||
feadf7c0 GS |
42 | /** |
43 | * eeh_pcid_get - Get the PCI device driver | |
44 | * @pdev: PCI device | |
45 | * | |
46 | * The function is used to retrieve the PCI device driver for | |
47 | * the indicated PCI device. Besides, we will increase the reference | |
48 | * of the PCI device driver to prevent that being unloaded on | |
49 | * the fly. Otherwise, kernel crash would be seen. | |
50 | */ | |
51 | static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev) | |
52 | { | |
53 | if (!pdev || !pdev->driver) | |
54 | return NULL; | |
55 | ||
56 | if (!try_module_get(pdev->driver->driver.owner)) | |
57 | return NULL; | |
58 | ||
59 | return pdev->driver; | |
60 | } | |
61 | ||
62 | /** | |
63 | * eeh_pcid_put - Dereference on the PCI device driver | |
64 | * @pdev: PCI device | |
65 | * | |
66 | * The function is called to do dereference on the PCI device | |
67 | * driver of the indicated PCI device. | |
68 | */ | |
69 | static inline void eeh_pcid_put(struct pci_dev *pdev) | |
70 | { | |
71 | if (!pdev || !pdev->driver) | |
72 | return; | |
73 | ||
74 | module_put(pdev->driver->driver.owner); | |
75 | } | |
76 | ||
8535ef05 | 77 | /** |
29f8bf1b GS |
78 | * eeh_disable_irq - Disable interrupt for the recovering device |
79 | * @dev: PCI device | |
80 | * | |
81 | * This routine must be called when reporting temporary or permanent | |
82 | * error to the particular PCI device to disable interrupt of that | |
83 | * device. If the device has enabled MSI or MSI-X interrupt, we needn't | |
84 | * do real work because EEH should freeze DMA transfers for those PCI | |
85 | * devices encountering EEH errors, which includes MSI or MSI-X. | |
8535ef05 MM |
86 | */ |
87 | static void eeh_disable_irq(struct pci_dev *dev) | |
88 | { | |
40a7cd92 | 89 | struct eeh_dev *edev = pci_dev_to_eeh_dev(dev); |
8535ef05 MM |
90 | |
91 | /* Don't disable MSI and MSI-X interrupts. They are | |
92 | * effectively disabled by the DMA Stopped state | |
93 | * when an EEH error occurs. | |
29f8bf1b | 94 | */ |
8535ef05 MM |
95 | if (dev->msi_enabled || dev->msix_enabled) |
96 | return; | |
97 | ||
59e3f837 | 98 | if (!irq_has_action(dev->irq)) |
8535ef05 MM |
99 | return; |
100 | ||
dbbceee1 | 101 | edev->mode |= EEH_DEV_IRQ_DISABLED; |
8535ef05 MM |
102 | disable_irq_nosync(dev->irq); |
103 | } | |
104 | ||
105 | /** | |
29f8bf1b GS |
106 | * eeh_enable_irq - Enable interrupt for the recovering device |
107 | * @dev: PCI device | |
108 | * | |
109 | * This routine must be called to enable interrupt while failed | |
110 | * device could be resumed. | |
8535ef05 MM |
111 | */ |
112 | static void eeh_enable_irq(struct pci_dev *dev) | |
113 | { | |
40a7cd92 | 114 | struct eeh_dev *edev = pci_dev_to_eeh_dev(dev); |
8535ef05 | 115 | |
dbbceee1 GS |
116 | if ((edev->mode) & EEH_DEV_IRQ_DISABLED) { |
117 | edev->mode &= ~EEH_DEV_IRQ_DISABLED; | |
b8a9a11b TG |
118 | /* |
119 | * FIXME !!!!! | |
120 | * | |
121 | * This is just ass backwards. This maze has | |
122 | * unbalanced irq_enable/disable calls. So instead of | |
123 | * finding the root cause it works around the warning | |
124 | * in the irq_enable code by conditionally calling | |
125 | * into it. | |
126 | * | |
127 | * That's just wrong.The warning in the core code is | |
027dfac6 | 128 | * there to tell people to fix their asymmetries in |
b8a9a11b TG |
129 | * their own code, not by abusing the core information |
130 | * to avoid it. | |
131 | * | |
132 | * I so wish that the assymetry would be the other way | |
133 | * round and a few more irq_disable calls render that | |
134 | * shit unusable forever. | |
135 | * | |
136 | * tglx | |
137 | */ | |
57310c3c | 138 | if (irqd_irq_disabled(irq_get_irq_data(dev->irq))) |
91150af3 | 139 | enable_irq(dev->irq); |
57310c3c | 140 | } |
8535ef05 MM |
141 | } |
142 | ||
d2b0f6f7 GS |
143 | static bool eeh_dev_removed(struct eeh_dev *edev) |
144 | { | |
145 | /* EEH device removed ? */ | |
146 | if (!edev || (edev->mode & EEH_DEV_REMOVED)) | |
147 | return true; | |
148 | ||
149 | return false; | |
150 | } | |
151 | ||
5cfb20b9 GS |
152 | static void *eeh_dev_save_state(void *data, void *userdata) |
153 | { | |
154 | struct eeh_dev *edev = data; | |
155 | struct pci_dev *pdev; | |
156 | ||
157 | if (!edev) | |
158 | return NULL; | |
159 | ||
5a0cdbfd GS |
160 | /* |
161 | * We cannot access the config space on some adapters. | |
162 | * Otherwise, it will cause fenced PHB. We don't save | |
163 | * the content in their config space and will restore | |
164 | * from the initial config space saved when the EEH | |
165 | * device is created. | |
166 | */ | |
167 | if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED)) | |
168 | return NULL; | |
169 | ||
5cfb20b9 GS |
170 | pdev = eeh_dev_to_pci_dev(edev); |
171 | if (!pdev) | |
172 | return NULL; | |
173 | ||
174 | pci_save_state(pdev); | |
175 | return NULL; | |
176 | } | |
177 | ||
cb5b5624 | 178 | /** |
29f8bf1b | 179 | * eeh_report_error - Report pci error to each device driver |
9b3c76f0 | 180 | * @data: eeh device |
29f8bf1b | 181 | * @userdata: return value |
a84f273c GS |
182 | * |
183 | * Report an EEH error to each device driver, collect up and | |
184 | * merge the device driver responses. Cumulative response | |
cb5b5624 | 185 | * passed back in "userdata". |
77bd7415 | 186 | */ |
9b3c76f0 | 187 | static void *eeh_report_error(void *data, void *userdata) |
77bd7415 | 188 | { |
9b3c76f0 GS |
189 | struct eeh_dev *edev = (struct eeh_dev *)data; |
190 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
18eb3b39 | 191 | enum pci_ers_result rc, *res = userdata; |
feadf7c0 | 192 | struct pci_driver *driver; |
77bd7415 | 193 | |
2311cca5 | 194 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 | 195 | return NULL; |
f0295e04 MN |
196 | |
197 | device_lock(&dev->dev); | |
77bd7415 LV |
198 | dev->error_state = pci_channel_io_frozen; |
199 | ||
feadf7c0 | 200 | driver = eeh_pcid_get(dev); |
f0295e04 | 201 | if (!driver) goto out_no_dev; |
77bd7415 | 202 | |
8535ef05 MM |
203 | eeh_disable_irq(dev); |
204 | ||
6a1ca373 | 205 | if (!driver->err_handler || |
f0295e04 MN |
206 | !driver->err_handler->error_detected) |
207 | goto out; | |
77bd7415 | 208 | |
29f8bf1b | 209 | rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen); |
2a50f144 LV |
210 | |
211 | /* A driver that needs a reset trumps all others */ | |
212 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
18eb3b39 | 213 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 214 | |
67086e32 | 215 | edev->in_error = true; |
856e1eb9 | 216 | pci_uevent_ers(dev, PCI_ERS_RESULT_NONE); |
f0295e04 MN |
217 | |
218 | out: | |
219 | eeh_pcid_put(dev); | |
220 | out_no_dev: | |
221 | device_unlock(&dev->dev); | |
9b3c76f0 | 222 | return NULL; |
6a1ca373 LV |
223 | } |
224 | ||
225 | /** | |
29f8bf1b | 226 | * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled |
9b3c76f0 | 227 | * @data: eeh device |
29f8bf1b | 228 | * @userdata: return value |
6a1ca373 | 229 | * |
638799b3 LV |
230 | * Tells each device driver that IO ports, MMIO and config space I/O |
231 | * are now enabled. Collects up and merges the device driver responses. | |
232 | * Cumulative response passed back in "userdata". | |
6a1ca373 | 233 | */ |
9b3c76f0 | 234 | static void *eeh_report_mmio_enabled(void *data, void *userdata) |
6a1ca373 | 235 | { |
9b3c76f0 GS |
236 | struct eeh_dev *edev = (struct eeh_dev *)data; |
237 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 238 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 239 | struct pci_driver *driver; |
6a1ca373 | 240 | |
2311cca5 | 241 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 GS |
242 | return NULL; |
243 | ||
f0295e04 | 244 | device_lock(&dev->dev); |
feadf7c0 | 245 | driver = eeh_pcid_get(dev); |
f0295e04 | 246 | if (!driver) goto out_no_dev; |
9b3c76f0 | 247 | |
feadf7c0 | 248 | if (!driver->err_handler || |
f26c7a03 | 249 | !driver->err_handler->mmio_enabled || |
f0295e04 MN |
250 | (edev->mode & EEH_DEV_NO_HANDLER)) |
251 | goto out; | |
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 | |
f0295e04 | 259 | out: |
feadf7c0 | 260 | eeh_pcid_put(dev); |
f0295e04 MN |
261 | out_no_dev: |
262 | device_unlock(&dev->dev); | |
9b3c76f0 | 263 | return NULL; |
77bd7415 LV |
264 | } |
265 | ||
cb5b5624 | 266 | /** |
29f8bf1b | 267 | * eeh_report_reset - Tell device that slot has been reset |
9b3c76f0 | 268 | * @data: eeh device |
29f8bf1b GS |
269 | * @userdata: return value |
270 | * | |
271 | * This routine must be called while EEH tries to reset particular | |
272 | * PCI device so that the associated PCI device driver could take | |
273 | * some actions, usually to save data the driver needs so that the | |
274 | * driver can work again while the device is recovered. | |
77bd7415 | 275 | */ |
9b3c76f0 | 276 | static void *eeh_report_reset(void *data, void *userdata) |
77bd7415 | 277 | { |
9b3c76f0 GS |
278 | struct eeh_dev *edev = (struct eeh_dev *)data; |
279 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 280 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 281 | struct pci_driver *driver; |
77bd7415 | 282 | |
2311cca5 | 283 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 | 284 | return NULL; |
f0295e04 MN |
285 | |
286 | device_lock(&dev->dev); | |
c58dc575 MM |
287 | dev->error_state = pci_channel_io_normal; |
288 | ||
feadf7c0 | 289 | driver = eeh_pcid_get(dev); |
f0295e04 | 290 | if (!driver) goto out_no_dev; |
feadf7c0 | 291 | |
8535ef05 MM |
292 | eeh_enable_irq(dev); |
293 | ||
6a1ca373 | 294 | if (!driver->err_handler || |
f26c7a03 | 295 | !driver->err_handler->slot_reset || |
67086e32 | 296 | (edev->mode & EEH_DEV_NO_HANDLER) || |
f0295e04 MN |
297 | (!edev->in_error)) |
298 | goto out; | |
77bd7415 | 299 | |
6a1ca373 | 300 | rc = driver->err_handler->slot_reset(dev); |
5794dbcb LV |
301 | if ((*res == PCI_ERS_RESULT_NONE) || |
302 | (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; | |
6a1ca373 LV |
303 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
304 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
70298c6e | 305 | |
f0295e04 | 306 | out: |
feadf7c0 | 307 | eeh_pcid_put(dev); |
f0295e04 MN |
308 | out_no_dev: |
309 | device_unlock(&dev->dev); | |
9b3c76f0 | 310 | return NULL; |
77bd7415 LV |
311 | } |
312 | ||
5cfb20b9 GS |
313 | static void *eeh_dev_restore_state(void *data, void *userdata) |
314 | { | |
315 | struct eeh_dev *edev = data; | |
316 | struct pci_dev *pdev; | |
317 | ||
318 | if (!edev) | |
319 | return NULL; | |
320 | ||
5a0cdbfd GS |
321 | /* |
322 | * The content in the config space isn't saved because | |
323 | * the blocked config space on some adapters. We have | |
324 | * to restore the initial saved config space when the | |
325 | * EEH device is created. | |
326 | */ | |
327 | if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED)) { | |
328 | if (list_is_last(&edev->list, &edev->pe->edevs)) | |
329 | eeh_pe_restore_bars(edev->pe); | |
330 | ||
331 | return NULL; | |
332 | } | |
333 | ||
5cfb20b9 GS |
334 | pdev = eeh_dev_to_pci_dev(edev); |
335 | if (!pdev) | |
336 | return NULL; | |
337 | ||
338 | pci_restore_state(pdev); | |
339 | return NULL; | |
340 | } | |
341 | ||
cb5b5624 | 342 | /** |
29f8bf1b | 343 | * eeh_report_resume - Tell device to resume normal operations |
9b3c76f0 | 344 | * @data: eeh device |
29f8bf1b GS |
345 | * @userdata: return value |
346 | * | |
347 | * This routine must be called to notify the device driver that it | |
348 | * could resume so that the device driver can do some initialization | |
349 | * to make the recovered device work again. | |
cb5b5624 | 350 | */ |
9b3c76f0 | 351 | static void *eeh_report_resume(void *data, void *userdata) |
77bd7415 | 352 | { |
9b3c76f0 GS |
353 | struct eeh_dev *edev = (struct eeh_dev *)data; |
354 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
67086e32 | 355 | bool was_in_error; |
9b3c76f0 GS |
356 | struct pci_driver *driver; |
357 | ||
2311cca5 | 358 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 | 359 | return NULL; |
f0295e04 MN |
360 | |
361 | device_lock(&dev->dev); | |
77bd7415 LV |
362 | dev->error_state = pci_channel_io_normal; |
363 | ||
feadf7c0 | 364 | driver = eeh_pcid_get(dev); |
f0295e04 | 365 | if (!driver) goto out_no_dev; |
d0e70341 | 366 | |
67086e32 WY |
367 | was_in_error = edev->in_error; |
368 | edev->in_error = false; | |
8535ef05 MM |
369 | eeh_enable_irq(dev); |
370 | ||
d0e70341 | 371 | if (!driver->err_handler || |
f26c7a03 | 372 | !driver->err_handler->resume || |
67086e32 | 373 | (edev->mode & EEH_DEV_NO_HANDLER) || !was_in_error) { |
f26c7a03 | 374 | edev->mode &= ~EEH_DEV_NO_HANDLER; |
f0295e04 | 375 | goto out; |
feadf7c0 | 376 | } |
77bd7415 LV |
377 | |
378 | driver->err_handler->resume(dev); | |
70298c6e | 379 | |
856e1eb9 | 380 | pci_uevent_ers(dev, PCI_ERS_RESULT_RECOVERED); |
f0295e04 MN |
381 | out: |
382 | eeh_pcid_put(dev); | |
856e1eb9 | 383 | #ifdef CONFIG_PCI_IOV |
521ca5a9 JA |
384 | if (eeh_ops->notify_resume && eeh_dev_to_pdn(edev)) |
385 | eeh_ops->notify_resume(eeh_dev_to_pdn(edev)); | |
856e1eb9 | 386 | #endif |
f0295e04 MN |
387 | out_no_dev: |
388 | device_unlock(&dev->dev); | |
9b3c76f0 | 389 | return NULL; |
77bd7415 LV |
390 | } |
391 | ||
cb5b5624 | 392 | /** |
29f8bf1b | 393 | * eeh_report_failure - Tell device driver that device is dead. |
9b3c76f0 | 394 | * @data: eeh device |
29f8bf1b | 395 | * @userdata: return value |
cb5b5624 LV |
396 | * |
397 | * This informs the device driver that the device is permanently | |
398 | * dead, and that no further recovery attempts will be made on it. | |
399 | */ | |
9b3c76f0 | 400 | static void *eeh_report_failure(void *data, void *userdata) |
77bd7415 | 401 | { |
9b3c76f0 GS |
402 | struct eeh_dev *edev = (struct eeh_dev *)data; |
403 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
404 | struct pci_driver *driver; | |
405 | ||
2311cca5 | 406 | if (!dev || eeh_dev_removed(edev) || eeh_pe_passed(edev->pe)) |
d2b0f6f7 | 407 | return NULL; |
f0295e04 MN |
408 | |
409 | device_lock(&dev->dev); | |
77bd7415 LV |
410 | dev->error_state = pci_channel_io_perm_failure; |
411 | ||
feadf7c0 | 412 | driver = eeh_pcid_get(dev); |
f0295e04 | 413 | if (!driver) goto out_no_dev; |
77bd7415 | 414 | |
8535ef05 MM |
415 | eeh_disable_irq(dev); |
416 | ||
417 | if (!driver->err_handler || | |
f0295e04 MN |
418 | !driver->err_handler->error_detected) |
419 | goto out; | |
8535ef05 | 420 | |
77bd7415 | 421 | driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); |
70298c6e | 422 | |
856e1eb9 | 423 | pci_uevent_ers(dev, PCI_ERS_RESULT_DISCONNECT); |
f0295e04 MN |
424 | out: |
425 | eeh_pcid_put(dev); | |
426 | out_no_dev: | |
427 | device_unlock(&dev->dev); | |
9b3c76f0 | 428 | return NULL; |
77bd7415 LV |
429 | } |
430 | ||
67086e32 WY |
431 | static void *eeh_add_virt_device(void *data, void *userdata) |
432 | { | |
433 | struct pci_driver *driver; | |
434 | struct eeh_dev *edev = (struct eeh_dev *)data; | |
435 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
436 | struct pci_dn *pdn = eeh_dev_to_pdn(edev); | |
437 | ||
438 | if (!(edev->physfn)) { | |
439 | pr_warn("%s: EEH dev %04x:%02x:%02x.%01x not for VF\n", | |
69672bd7 | 440 | __func__, pdn->phb->global_number, pdn->busno, |
67086e32 WY |
441 | PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn)); |
442 | return NULL; | |
443 | } | |
444 | ||
445 | driver = eeh_pcid_get(dev); | |
446 | if (driver) { | |
46d4be41 SB |
447 | if (driver->err_handler) { |
448 | eeh_pcid_put(dev); | |
67086e32 | 449 | return NULL; |
46d4be41 SB |
450 | } |
451 | eeh_pcid_put(dev); | |
67086e32 WY |
452 | } |
453 | ||
988fc3ba | 454 | #ifdef CONFIG_PCI_IOV |
753f6124 | 455 | pci_iov_add_virtfn(edev->physfn, pdn->vf_index); |
67086e32 WY |
456 | #endif |
457 | return NULL; | |
458 | } | |
459 | ||
f5c57710 GS |
460 | static void *eeh_rmv_device(void *data, void *userdata) |
461 | { | |
462 | struct pci_driver *driver; | |
463 | struct eeh_dev *edev = (struct eeh_dev *)data; | |
464 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
67086e32 WY |
465 | struct eeh_rmv_data *rmv_data = (struct eeh_rmv_data *)userdata; |
466 | int *removed = rmv_data ? &rmv_data->removed : NULL; | |
f5c57710 GS |
467 | |
468 | /* | |
469 | * Actually, we should remove the PCI bridges as well. | |
470 | * However, that's lots of complexity to do that, | |
471 | * particularly some of devices under the bridge might | |
472 | * support EEH. So we just care about PCI devices for | |
473 | * simplicity here. | |
474 | */ | |
93de6901 | 475 | if (!dev || (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)) |
f5c57710 | 476 | return NULL; |
8cc6b6cd | 477 | |
d2b0f6f7 GS |
478 | /* |
479 | * We rely on count-based pcibios_release_device() to | |
480 | * detach permanently offlined PEs. Unfortunately, that's | |
481 | * not reliable enough. We might have the permanently | |
482 | * offlined PEs attached, but we needn't take care of | |
483 | * them and their child devices. | |
484 | */ | |
485 | if (eeh_dev_removed(edev)) | |
486 | return NULL; | |
487 | ||
46d4be41 SB |
488 | if (removed) { |
489 | if (eeh_pe_passed(edev->pe)) | |
8cc6b6cd | 490 | return NULL; |
46d4be41 SB |
491 | driver = eeh_pcid_get(dev); |
492 | if (driver) { | |
493 | if (driver->err_handler && | |
494 | driver->err_handler->error_detected && | |
495 | driver->err_handler->slot_reset) { | |
496 | eeh_pcid_put(dev); | |
497 | return NULL; | |
498 | } | |
499 | eeh_pcid_put(dev); | |
500 | } | |
8cc6b6cd | 501 | } |
f5c57710 GS |
502 | |
503 | /* Remove it from PCI subsystem */ | |
504 | pr_debug("EEH: Removing %s without EEH sensitive driver\n", | |
505 | pci_name(dev)); | |
506 | edev->bus = dev->bus; | |
507 | edev->mode |= EEH_DEV_DISCONNECTED; | |
67086e32 WY |
508 | if (removed) |
509 | (*removed)++; | |
f5c57710 | 510 | |
67086e32 | 511 | if (edev->physfn) { |
988fc3ba | 512 | #ifdef CONFIG_PCI_IOV |
67086e32 WY |
513 | struct pci_dn *pdn = eeh_dev_to_pdn(edev); |
514 | ||
753f6124 | 515 | pci_iov_remove_virtfn(edev->physfn, pdn->vf_index); |
67086e32 WY |
516 | edev->pdev = NULL; |
517 | ||
518 | /* | |
519 | * We have to set the VF PE number to invalid one, which is | |
520 | * required to plug the VF successfully. | |
521 | */ | |
522 | pdn->pe_number = IODA_INVALID_PE; | |
523 | #endif | |
524 | if (rmv_data) | |
525 | list_add(&edev->rmv_list, &rmv_data->edev_list); | |
526 | } else { | |
527 | pci_lock_rescan_remove(); | |
528 | pci_stop_and_remove_bus_device(dev); | |
529 | pci_unlock_rescan_remove(); | |
530 | } | |
f5c57710 GS |
531 | |
532 | return NULL; | |
533 | } | |
534 | ||
535 | static void *eeh_pe_detach_dev(void *data, void *userdata) | |
536 | { | |
537 | struct eeh_pe *pe = (struct eeh_pe *)data; | |
538 | struct eeh_dev *edev, *tmp; | |
539 | ||
540 | eeh_pe_for_each_dev(pe, edev, tmp) { | |
541 | if (!(edev->mode & EEH_DEV_DISCONNECTED)) | |
542 | continue; | |
543 | ||
544 | edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED); | |
545 | eeh_rmv_from_parent_pe(edev); | |
546 | } | |
547 | ||
548 | return NULL; | |
549 | } | |
550 | ||
78954700 GS |
551 | /* |
552 | * Explicitly clear PE's frozen state for PowerNV where | |
553 | * we have frozen PE until BAR restore is completed. It's | |
554 | * harmless to clear it for pSeries. To be consistent with | |
555 | * PE reset (for 3 times), we try to clear the frozen state | |
556 | * for 3 times as well. | |
557 | */ | |
2c665992 | 558 | static void *__eeh_clear_pe_frozen_state(void *data, void *flag) |
78954700 | 559 | { |
2c665992 | 560 | struct eeh_pe *pe = (struct eeh_pe *)data; |
f05fea5b | 561 | bool clear_sw_state = *(bool *)flag; |
c9dd0143 | 562 | int i, rc = 1; |
78954700 | 563 | |
c9dd0143 | 564 | for (i = 0; rc && i < 3; i++) |
5cfb20b9 | 565 | rc = eeh_unfreeze_pe(pe, clear_sw_state); |
78954700 | 566 | |
c9dd0143 | 567 | /* Stop immediately on any errors */ |
2c665992 | 568 | if (rc) { |
c9dd0143 GS |
569 | pr_warn("%s: Failure %d unfreezing PHB#%x-PE#%x\n", |
570 | __func__, rc, pe->phb->global_number, pe->addr); | |
2c665992 GS |
571 | return (void *)pe; |
572 | } | |
573 | ||
574 | return NULL; | |
575 | } | |
576 | ||
5cfb20b9 GS |
577 | static int eeh_clear_pe_frozen_state(struct eeh_pe *pe, |
578 | bool clear_sw_state) | |
2c665992 GS |
579 | { |
580 | void *rc; | |
581 | ||
5cfb20b9 | 582 | rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, &clear_sw_state); |
2c665992 | 583 | if (!rc) |
78954700 GS |
584 | eeh_pe_state_clear(pe, EEH_PE_ISOLATED); |
585 | ||
2c665992 | 586 | return rc ? -EIO : 0; |
78954700 GS |
587 | } |
588 | ||
5cfb20b9 GS |
589 | int eeh_pe_reset_and_recover(struct eeh_pe *pe) |
590 | { | |
2efc771f | 591 | int ret; |
5cfb20b9 GS |
592 | |
593 | /* Bail if the PE is being recovered */ | |
594 | if (pe->state & EEH_PE_RECOVERING) | |
595 | return 0; | |
596 | ||
597 | /* Put the PE into recovery mode */ | |
598 | eeh_pe_state_mark(pe, EEH_PE_RECOVERING); | |
599 | ||
600 | /* Save states */ | |
601 | eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL); | |
602 | ||
5cfb20b9 | 603 | /* Issue reset */ |
6654c936 | 604 | ret = eeh_pe_reset_full(pe); |
5cfb20b9 | 605 | if (ret) { |
28bf36f9 | 606 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); |
5cfb20b9 GS |
607 | return ret; |
608 | } | |
5cfb20b9 GS |
609 | |
610 | /* Unfreeze the PE */ | |
611 | ret = eeh_clear_pe_frozen_state(pe, true); | |
612 | if (ret) { | |
613 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); | |
614 | return ret; | |
615 | } | |
616 | ||
5cfb20b9 GS |
617 | /* Restore device state */ |
618 | eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL); | |
619 | ||
5cfb20b9 GS |
620 | /* Clear recovery mode */ |
621 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); | |
622 | ||
623 | return 0; | |
624 | } | |
625 | ||
77bd7415 | 626 | /** |
29f8bf1b | 627 | * eeh_reset_device - Perform actual reset of a pci slot |
5fd13460 | 628 | * @driver_eeh_aware: Does the device's driver provide EEH support? |
9b3c76f0 | 629 | * @pe: EEH PE |
29f8bf1b | 630 | * @bus: PCI bus corresponding to the isolcated slot |
5fd13460 | 631 | * @rmv_data: Optional, list to record removed devices |
77bd7415 | 632 | * |
29f8bf1b GS |
633 | * This routine must be called to do reset on the indicated PE. |
634 | * During the reset, udev might be invoked because those affected | |
635 | * PCI devices will be removed and then added. | |
77bd7415 | 636 | */ |
67086e32 | 637 | static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus, |
5fd13460 SB |
638 | struct eeh_rmv_data *rmv_data, |
639 | bool driver_eeh_aware) | |
77bd7415 | 640 | { |
edfd17ff | 641 | time64_t tstamp; |
67086e32 WY |
642 | int cnt, rc; |
643 | struct eeh_dev *edev; | |
42405456 LV |
644 | |
645 | /* pcibios will clear the counter; save the value */ | |
9b3c76f0 | 646 | cnt = pe->freeze_count; |
5a71978e | 647 | tstamp = pe->tstamp; |
42405456 | 648 | |
20ee6a97 GS |
649 | /* |
650 | * We don't remove the corresponding PE instances because | |
651 | * we need the information afterwords. The attached EEH | |
652 | * devices are expected to be attached soon when calling | |
bd251b89 | 653 | * into pci_hp_add_devices(). |
20ee6a97 | 654 | */ |
f5c57710 | 655 | eeh_pe_state_mark(pe, EEH_PE_KEEP); |
54048cf8 | 656 | if (driver_eeh_aware || (pe->type & EEH_PE_VF)) { |
cca0e542 | 657 | eeh_pe_dev_traverse(pe, eeh_rmv_device, rmv_data); |
54048cf8 SB |
658 | } else { |
659 | pci_lock_rescan_remove(); | |
660 | pci_hp_remove_devices(bus); | |
661 | pci_unlock_rescan_remove(); | |
1c2042c8 | 662 | } |
77bd7415 | 663 | |
d0914f50 GS |
664 | /* |
665 | * Reset the pci controller. (Asserts RST#; resets config space). | |
b6495c0c | 666 | * Reconfigure bridges and devices. Don't try to bring the system |
29f8bf1b | 667 | * up if the reset failed for some reason. |
d0914f50 GS |
668 | * |
669 | * During the reset, it's very dangerous to have uncontrolled PCI | |
670 | * config accesses. So we prefer to block them. However, controlled | |
671 | * PCI config accesses initiated from EEH itself are allowed. | |
29f8bf1b | 672 | */ |
6654c936 | 673 | rc = eeh_pe_reset_full(pe); |
28bf36f9 | 674 | if (rc) |
b6495c0c | 675 | return rc; |
77bd7415 | 676 | |
1c2042c8 RW |
677 | pci_lock_rescan_remove(); |
678 | ||
9b3c76f0 GS |
679 | /* Restore PE */ |
680 | eeh_ops->configure_bridge(pe); | |
681 | eeh_pe_restore_bars(pe); | |
77bd7415 | 682 | |
dc9c41bd AD |
683 | /* Clear frozen state */ |
684 | rc = eeh_clear_pe_frozen_state(pe, false); | |
409bf7f8 AD |
685 | if (rc) { |
686 | pci_unlock_rescan_remove(); | |
dc9c41bd | 687 | return rc; |
409bf7f8 | 688 | } |
78954700 | 689 | |
77bd7415 | 690 | /* Give the system 5 seconds to finish running the user-space |
a84f273c GS |
691 | * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, |
692 | * this is a hack, but if we don't do this, and try to bring | |
693 | * the device up before the scripts have taken it down, | |
77bd7415 LV |
694 | * potentially weird things happen. |
695 | */ | |
54048cf8 SB |
696 | if (!driver_eeh_aware || rmv_data->removed) { |
697 | pr_info("EEH: Sleep 5s ahead of %s hotplug\n", | |
698 | (driver_eeh_aware ? "partial" : "complete")); | |
29f8bf1b | 699 | ssleep(5); |
f5c57710 GS |
700 | |
701 | /* | |
702 | * The EEH device is still connected with its parent | |
703 | * PE. We should disconnect it so the binding can be | |
704 | * rebuilt when adding PCI devices. | |
705 | */ | |
67086e32 | 706 | edev = list_first_entry(&pe->edevs, struct eeh_dev, list); |
f5c57710 | 707 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); |
a3aa256b | 708 | if (pe->type & EEH_PE_VF) { |
67086e32 | 709 | eeh_add_virt_device(edev, NULL); |
a3aa256b | 710 | } else { |
54048cf8 SB |
711 | if (!driver_eeh_aware) |
712 | eeh_pe_state_clear(pe, EEH_PE_PRI_BUS); | |
bd251b89 | 713 | pci_hp_add_devices(bus); |
a3aa256b | 714 | } |
77bd7415 | 715 | } |
f5c57710 | 716 | eeh_pe_state_clear(pe, EEH_PE_KEEP); |
5a71978e GS |
717 | |
718 | pe->tstamp = tstamp; | |
9b3c76f0 | 719 | pe->freeze_count = cnt; |
b6495c0c | 720 | |
1c2042c8 | 721 | pci_unlock_rescan_remove(); |
b6495c0c | 722 | return 0; |
77bd7415 LV |
723 | } |
724 | ||
725 | /* The longest amount of time to wait for a pci device | |
726 | * to come back on line, in seconds. | |
727 | */ | |
fb48dc22 | 728 | #define MAX_WAIT_FOR_RECOVERY 300 |
77bd7415 | 729 | |
c0b64978 RC |
730 | /** |
731 | * eeh_handle_normal_event - Handle EEH events on a specific PE | |
37fd8125 SB |
732 | * @pe: EEH PE - which should not be used after we return, as it may |
733 | * have been invalidated. | |
c0b64978 RC |
734 | * |
735 | * Attempts to recover the given PE. If recovery fails or the PE has failed | |
736 | * too many times, remove the PE. | |
737 | * | |
68701780 SB |
738 | * While PHB detects address or data parity errors on particular PCI |
739 | * slot, the associated PE will be frozen. Besides, DMA's occurring | |
740 | * to wild addresses (which usually happen due to bugs in device | |
741 | * drivers or in PCI adapter firmware) can cause EEH error. #SERR, | |
742 | * #PERR or other misc PCI-related errors also can trigger EEH errors. | |
743 | * | |
744 | * Recovery process consists of unplugging the device driver (which | |
745 | * generated hotplug events to userspace), then issuing a PCI #RST to | |
746 | * the device, then reconfiguring the PCI config space for all bridges | |
747 | * & devices under this slot, and then finally restarting the device | |
748 | * drivers (which cause a second set of hotplug events to go out to | |
749 | * userspace). | |
c0b64978 | 750 | */ |
37fd8125 | 751 | void eeh_handle_normal_event(struct eeh_pe *pe) |
77bd7415 | 752 | { |
cd95f804 | 753 | struct pci_bus *bus; |
67086e32 | 754 | struct eeh_dev *edev, *tmp; |
b6495c0c | 755 | int rc = 0; |
18eb3b39 | 756 | enum pci_ers_result result = PCI_ERS_RESULT_NONE; |
67086e32 | 757 | struct eeh_rmv_data rmv_data = {LIST_HEAD_INIT(rmv_data.edev_list), 0}; |
77bd7415 | 758 | |
cd95f804 SB |
759 | bus = eeh_pe_bus_get(pe); |
760 | if (!bus) { | |
1f52f176 | 761 | pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n", |
9b3c76f0 | 762 | __func__, pe->phb->global_number, pe->addr); |
37fd8125 | 763 | return; |
77bd7415 LV |
764 | } |
765 | ||
37fd8125 SB |
766 | eeh_pe_state_mark(pe, EEH_PE_RECOVERING); |
767 | ||
5a71978e | 768 | eeh_pe_update_time_stamp(pe); |
9b3c76f0 | 769 | pe->freeze_count++; |
c0b64978 | 770 | if (pe->freeze_count > eeh_max_freezes) { |
796b9f5b | 771 | pr_err("EEH: PHB#%x-PE#%x has failed %d times in the last hour and has been permanently disabled.\n", |
c0b64978 RC |
772 | pe->phb->global_number, pe->addr, |
773 | pe->freeze_count); | |
774 | goto hard_fail; | |
775 | } | |
796b9f5b SB |
776 | pr_warn("EEH: This PCI device has failed %d times in the last hour and will be permanently disabled after %d failures.\n", |
777 | pe->freeze_count, eeh_max_freezes); | |
77bd7415 LV |
778 | |
779 | /* Walk the various device drivers attached to this slot through | |
780 | * a reset sequence, giving each an opportunity to do what it needs | |
781 | * to accomplish the reset. Each child gets a report of the | |
782 | * status ... if any child can't handle the reset, then the entire | |
783 | * slot is dlpar removed and added. | |
8234fced GS |
784 | * |
785 | * When the PHB is fenced, we have to issue a reset to recover from | |
786 | * the error. Override the result if necessary to have partially | |
787 | * hotplug for this case. | |
77bd7415 | 788 | */ |
56ca4fde | 789 | pr_info("EEH: Notify device drivers to shutdown\n"); |
9b3c76f0 | 790 | eeh_pe_dev_traverse(pe, eeh_report_error, &result); |
8234fced GS |
791 | if ((pe->type & EEH_PE_PHB) && |
792 | result != PCI_ERS_RESULT_NONE && | |
793 | result != PCI_ERS_RESULT_NEED_RESET) | |
794 | result = PCI_ERS_RESULT_NEED_RESET; | |
77bd7415 | 795 | |
5f1a7c81 | 796 | /* Get the current PCI slot state. This can take a long time, |
2ac3990c | 797 | * sometimes over 300 seconds for certain systems. |
29f8bf1b | 798 | */ |
9b3c76f0 | 799 | rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000); |
eb594a47 | 800 | if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) { |
0dae2743 | 801 | pr_warn("EEH: Permanent failure\n"); |
5f1a7c81 LV |
802 | goto hard_fail; |
803 | } | |
804 | ||
ede8ca26 LV |
805 | /* Since rtas may enable MMIO when posting the error log, |
806 | * don't post the error log until after all dev drivers | |
17213c3b LV |
807 | * have been informed. |
808 | */ | |
56ca4fde | 809 | pr_info("EEH: Collect temporary log\n"); |
9b3c76f0 | 810 | eeh_slot_error_detail(pe, EEH_LOG_TEMP); |
ede8ca26 | 811 | |
77bd7415 LV |
812 | /* If all device drivers were EEH-unaware, then shut |
813 | * down all of the device drivers, and hope they | |
814 | * go down willingly, without panicing the system. | |
815 | */ | |
18eb3b39 | 816 | if (result == PCI_ERS_RESULT_NONE) { |
56ca4fde | 817 | pr_info("EEH: Reset with hotplug activity\n"); |
5fd13460 | 818 | rc = eeh_reset_device(pe, bus, NULL, false); |
e0f90b64 | 819 | if (rc) { |
0dae2743 GS |
820 | pr_warn("%s: Unable to reset, err=%d\n", |
821 | __func__, rc); | |
b6495c0c | 822 | goto hard_fail; |
e0f90b64 | 823 | } |
77bd7415 LV |
824 | } |
825 | ||
6a1ca373 LV |
826 | /* If all devices reported they can proceed, then re-enable MMIO */ |
827 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { | |
56ca4fde | 828 | pr_info("EEH: Enable I/O for affected devices\n"); |
9b3c76f0 | 829 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO); |
6a1ca373 | 830 | |
fa1be476 LV |
831 | if (rc < 0) |
832 | goto hard_fail; | |
6a1ca373 LV |
833 | if (rc) { |
834 | result = PCI_ERS_RESULT_NEED_RESET; | |
835 | } else { | |
56ca4fde | 836 | pr_info("EEH: Notify device drivers to resume I/O\n"); |
9b3c76f0 | 837 | eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result); |
6a1ca373 | 838 | } |
77bd7415 LV |
839 | } |
840 | ||
6a1ca373 | 841 | /* If all devices reported they can proceed, then re-enable DMA */ |
18eb3b39 | 842 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { |
56ca4fde | 843 | pr_info("EEH: Enabled DMA for affected devices\n"); |
9b3c76f0 | 844 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA); |
6a1ca373 | 845 | |
fa1be476 LV |
846 | if (rc < 0) |
847 | goto hard_fail; | |
35845a78 | 848 | if (rc) { |
6a1ca373 | 849 | result = PCI_ERS_RESULT_NEED_RESET; |
35845a78 GS |
850 | } else { |
851 | /* | |
852 | * We didn't do PE reset for the case. The PE | |
853 | * is still in frozen state. Clear it before | |
854 | * resuming the PE. | |
855 | */ | |
856 | eeh_pe_state_clear(pe, EEH_PE_ISOLATED); | |
d0e70341 | 857 | result = PCI_ERS_RESULT_RECOVERED; |
35845a78 | 858 | } |
6a1ca373 LV |
859 | } |
860 | ||
861 | /* If any device has a hard failure, then shut off everything. */ | |
e0f90b64 | 862 | if (result == PCI_ERS_RESULT_DISCONNECT) { |
0dae2743 | 863 | pr_warn("EEH: Device driver gave up\n"); |
6a1ca373 | 864 | goto hard_fail; |
e0f90b64 | 865 | } |
6a1ca373 LV |
866 | |
867 | /* If any device called out for a reset, then reset the slot */ | |
868 | if (result == PCI_ERS_RESULT_NEED_RESET) { | |
56ca4fde | 869 | pr_info("EEH: Reset without hotplug activity\n"); |
5fd13460 | 870 | rc = eeh_reset_device(pe, bus, &rmv_data, true); |
e0f90b64 | 871 | if (rc) { |
0dae2743 GS |
872 | pr_warn("%s: Cannot reset, err=%d\n", |
873 | __func__, rc); | |
b6495c0c | 874 | goto hard_fail; |
e0f90b64 | 875 | } |
56ca4fde GS |
876 | |
877 | pr_info("EEH: Notify device drivers " | |
878 | "the completion of reset\n"); | |
6a1ca373 | 879 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 880 | eeh_pe_dev_traverse(pe, eeh_report_reset, &result); |
77bd7415 LV |
881 | } |
882 | ||
6a1ca373 | 883 | /* All devices should claim they have recovered by now. */ |
90fdd613 LV |
884 | if ((result != PCI_ERS_RESULT_RECOVERED) && |
885 | (result != PCI_ERS_RESULT_NONE)) { | |
0dae2743 | 886 | pr_warn("EEH: Not recovered\n"); |
6a1ca373 | 887 | goto hard_fail; |
e0f90b64 | 888 | } |
6a1ca373 | 889 | |
67086e32 WY |
890 | /* |
891 | * For those hot removed VFs, we should add back them after PF get | |
892 | * recovered properly. | |
893 | */ | |
894 | list_for_each_entry_safe(edev, tmp, &rmv_data.edev_list, rmv_list) { | |
895 | eeh_add_virt_device(edev, NULL); | |
896 | list_del(&edev->rmv_list); | |
897 | } | |
898 | ||
77bd7415 | 899 | /* Tell all device drivers that they can resume operations */ |
56ca4fde | 900 | pr_info("EEH: Notify device driver to resume\n"); |
9b3c76f0 | 901 | eeh_pe_dev_traverse(pe, eeh_report_resume, NULL); |
b6495c0c | 902 | |
796b9f5b | 903 | pr_info("EEH: Recovery successful.\n"); |
37fd8125 | 904 | goto final; |
a84f273c | 905 | |
c0b64978 | 906 | hard_fail: |
b6495c0c LV |
907 | /* |
908 | * About 90% of all real-life EEH failures in the field | |
909 | * are due to poorly seated PCI cards. Only 10% or so are | |
910 | * due to actual, failed cards. | |
911 | */ | |
1f52f176 | 912 | pr_err("EEH: Unable to recover from failure from PHB#%x-PE#%x.\n" |
9b3c76f0 GS |
913 | "Please try reseating or replacing it\n", |
914 | pe->phb->global_number, pe->addr); | |
b6495c0c | 915 | |
9b3c76f0 | 916 | eeh_slot_error_detail(pe, EEH_LOG_PERM); |
b6495c0c LV |
917 | |
918 | /* Notify all devices that they're about to go down. */ | |
9b3c76f0 | 919 | eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); |
b6495c0c | 920 | |
d2b0f6f7 | 921 | /* Mark the PE to be removed permanently */ |
432227e9 | 922 | eeh_pe_state_mark(pe, EEH_PE_REMOVED); |
d2b0f6f7 GS |
923 | |
924 | /* | |
925 | * Shut down the device drivers for good. We mark | |
926 | * all removed devices correctly to avoid access | |
927 | * the their PCI config any more. | |
928 | */ | |
5b86ac9e SB |
929 | if (pe->type & EEH_PE_VF) { |
930 | eeh_pe_dev_traverse(pe, eeh_rmv_device, NULL); | |
931 | eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED); | |
932 | } else { | |
933 | eeh_pe_state_clear(pe, EEH_PE_PRI_BUS); | |
934 | eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED); | |
d2b0f6f7 | 935 | |
5b86ac9e | 936 | pci_lock_rescan_remove(); |
cd95f804 | 937 | pci_hp_remove_devices(bus); |
5b86ac9e SB |
938 | pci_unlock_rescan_remove(); |
939 | /* The passed PE should no longer be used */ | |
940 | return; | |
1c2042c8 | 941 | } |
37fd8125 SB |
942 | final: |
943 | eeh_pe_state_clear(pe, EEH_PE_RECOVERING); | |
77bd7415 | 944 | } |
8a6b1bc7 | 945 | |
c0b64978 RC |
946 | /** |
947 | * eeh_handle_special_event - Handle EEH events without a specific failing PE | |
948 | * | |
949 | * Called when an EEH event is detected but can't be narrowed down to a | |
950 | * specific PE. Iterates through possible failures and handles them as | |
951 | * necessary. | |
952 | */ | |
68701780 | 953 | void eeh_handle_special_event(void) |
8a6b1bc7 GS |
954 | { |
955 | struct eeh_pe *pe, *phb_pe; | |
956 | struct pci_bus *bus; | |
7e4e7867 | 957 | struct pci_controller *hose; |
8a6b1bc7 | 958 | unsigned long flags; |
7e4e7867 | 959 | int rc; |
8a6b1bc7 | 960 | |
8a6b1bc7 | 961 | |
7e4e7867 GS |
962 | do { |
963 | rc = eeh_ops->next_error(&pe); | |
964 | ||
965 | switch (rc) { | |
966 | case EEH_NEXT_ERR_DEAD_IOC: | |
967 | /* Mark all PHBs in dead state */ | |
968 | eeh_serialize_lock(&flags); | |
969 | ||
970 | /* Purge all events */ | |
5c7a35e3 | 971 | eeh_remove_event(NULL, true); |
7e4e7867 GS |
972 | |
973 | list_for_each_entry(hose, &hose_list, list_node) { | |
974 | phb_pe = eeh_phb_pe_get(hose); | |
975 | if (!phb_pe) continue; | |
976 | ||
9e049375 | 977 | eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED); |
7e4e7867 GS |
978 | } |
979 | ||
980 | eeh_serialize_unlock(flags); | |
981 | ||
982 | break; | |
983 | case EEH_NEXT_ERR_FROZEN_PE: | |
984 | case EEH_NEXT_ERR_FENCED_PHB: | |
985 | case EEH_NEXT_ERR_DEAD_PHB: | |
986 | /* Mark the PE in fenced state */ | |
987 | eeh_serialize_lock(&flags); | |
988 | ||
989 | /* Purge all events of the PHB */ | |
5c7a35e3 | 990 | eeh_remove_event(pe, true); |
7e4e7867 GS |
991 | |
992 | if (rc == EEH_NEXT_ERR_DEAD_PHB) | |
9e049375 | 993 | eeh_pe_state_mark(pe, EEH_PE_ISOLATED); |
7e4e7867 GS |
994 | else |
995 | eeh_pe_state_mark(pe, | |
996 | EEH_PE_ISOLATED | EEH_PE_RECOVERING); | |
997 | ||
998 | eeh_serialize_unlock(flags); | |
999 | ||
1000 | break; | |
1001 | case EEH_NEXT_ERR_NONE: | |
1002 | return; | |
1003 | default: | |
1004 | pr_warn("%s: Invalid value %d from next_error()\n", | |
1005 | __func__, rc); | |
1006 | return; | |
8a6b1bc7 | 1007 | } |
8a6b1bc7 | 1008 | |
7e4e7867 GS |
1009 | /* |
1010 | * For fenced PHB and frozen PE, it's handled as normal | |
1011 | * event. We have to remove the affected PHBs for dead | |
1012 | * PHB and IOC | |
1013 | */ | |
1014 | if (rc == EEH_NEXT_ERR_FROZEN_PE || | |
1015 | rc == EEH_NEXT_ERR_FENCED_PHB) { | |
37fd8125 | 1016 | eeh_handle_normal_event(pe); |
7e4e7867 | 1017 | } else { |
1b17366d | 1018 | pci_lock_rescan_remove(); |
7e4e7867 GS |
1019 | list_for_each_entry(hose, &hose_list, list_node) { |
1020 | phb_pe = eeh_phb_pe_get(hose); | |
1021 | if (!phb_pe || | |
9e049375 GS |
1022 | !(phb_pe->state & EEH_PE_ISOLATED) || |
1023 | (phb_pe->state & EEH_PE_RECOVERING)) | |
7e4e7867 GS |
1024 | continue; |
1025 | ||
1026 | /* Notify all devices to be down */ | |
05ba75f8 | 1027 | eeh_pe_state_clear(pe, EEH_PE_PRI_BUS); |
af2e3a00 RC |
1028 | eeh_pe_dev_traverse(pe, |
1029 | eeh_report_failure, NULL); | |
7e4e7867 | 1030 | bus = eeh_pe_bus_get(phb_pe); |
04fec21c RC |
1031 | if (!bus) { |
1032 | pr_err("%s: Cannot find PCI bus for " | |
1f52f176 | 1033 | "PHB#%x-PE#%x\n", |
04fec21c RC |
1034 | __func__, |
1035 | pe->phb->global_number, | |
1036 | pe->addr); | |
1037 | break; | |
1038 | } | |
bd251b89 | 1039 | pci_hp_remove_devices(bus); |
7e4e7867 | 1040 | } |
1b17366d | 1041 | pci_unlock_rescan_remove(); |
8a6b1bc7 | 1042 | } |
7e4e7867 GS |
1043 | |
1044 | /* | |
1045 | * If we have detected dead IOC, we needn't proceed | |
1046 | * any more since all PHBs would have been removed | |
1047 | */ | |
1048 | if (rc == EEH_NEXT_ERR_DEAD_IOC) | |
1049 | break; | |
1050 | } while (rc != EEH_NEXT_ERR_NONE); | |
8a6b1bc7 | 1051 | } |