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 | */ | |
169 | if (irqd_irq_disabled(irq_get_irq_data(dev->irq)) | |
91150af3 | 170 | enable_irq(dev->irq); |
8535ef05 MM |
171 | } |
172 | ||
cb5b5624 | 173 | /** |
29f8bf1b | 174 | * eeh_report_error - Report pci error to each device driver |
9b3c76f0 | 175 | * @data: eeh device |
29f8bf1b | 176 | * @userdata: return value |
a84f273c GS |
177 | * |
178 | * Report an EEH error to each device driver, collect up and | |
179 | * merge the device driver responses. Cumulative response | |
cb5b5624 | 180 | * passed back in "userdata". |
77bd7415 | 181 | */ |
9b3c76f0 | 182 | static void *eeh_report_error(void *data, void *userdata) |
77bd7415 | 183 | { |
9b3c76f0 GS |
184 | struct eeh_dev *edev = (struct eeh_dev *)data; |
185 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
18eb3b39 | 186 | enum pci_ers_result rc, *res = userdata; |
feadf7c0 | 187 | struct pci_driver *driver; |
77bd7415 | 188 | |
9b3c76f0 GS |
189 | /* We might not have the associated PCI device, |
190 | * then we should continue for next one. | |
191 | */ | |
192 | if (!dev) return NULL; | |
77bd7415 LV |
193 | dev->error_state = pci_channel_io_frozen; |
194 | ||
feadf7c0 GS |
195 | driver = eeh_pcid_get(dev); |
196 | if (!driver) return NULL; | |
77bd7415 | 197 | |
8535ef05 MM |
198 | eeh_disable_irq(dev); |
199 | ||
6a1ca373 | 200 | if (!driver->err_handler || |
feadf7c0 GS |
201 | !driver->err_handler->error_detected) { |
202 | eeh_pcid_put(dev); | |
9b3c76f0 | 203 | return NULL; |
feadf7c0 | 204 | } |
77bd7415 | 205 | |
29f8bf1b | 206 | rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen); |
2a50f144 LV |
207 | |
208 | /* A driver that needs a reset trumps all others */ | |
209 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
18eb3b39 | 210 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 211 | |
feadf7c0 | 212 | eeh_pcid_put(dev); |
9b3c76f0 | 213 | return NULL; |
6a1ca373 LV |
214 | } |
215 | ||
216 | /** | |
29f8bf1b | 217 | * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled |
9b3c76f0 | 218 | * @data: eeh device |
29f8bf1b | 219 | * @userdata: return value |
6a1ca373 | 220 | * |
638799b3 LV |
221 | * Tells each device driver that IO ports, MMIO and config space I/O |
222 | * are now enabled. Collects up and merges the device driver responses. | |
223 | * Cumulative response passed back in "userdata". | |
6a1ca373 | 224 | */ |
9b3c76f0 | 225 | static void *eeh_report_mmio_enabled(void *data, void *userdata) |
6a1ca373 | 226 | { |
9b3c76f0 GS |
227 | struct eeh_dev *edev = (struct eeh_dev *)data; |
228 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 229 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 230 | struct pci_driver *driver; |
6a1ca373 | 231 | |
feadf7c0 GS |
232 | driver = eeh_pcid_get(dev); |
233 | if (!driver) return NULL; | |
9b3c76f0 | 234 | |
feadf7c0 | 235 | if (!driver->err_handler || |
f26c7a03 GS |
236 | !driver->err_handler->mmio_enabled || |
237 | (edev->mode & EEH_DEV_NO_HANDLER)) { | |
feadf7c0 | 238 | eeh_pcid_put(dev); |
9b3c76f0 | 239 | return NULL; |
feadf7c0 | 240 | } |
6a1ca373 | 241 | |
29f8bf1b | 242 | rc = driver->err_handler->mmio_enabled(dev); |
2a50f144 LV |
243 | |
244 | /* A driver that needs a reset trumps all others */ | |
245 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
6a1ca373 | 246 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 247 | |
feadf7c0 | 248 | eeh_pcid_put(dev); |
9b3c76f0 | 249 | return NULL; |
77bd7415 LV |
250 | } |
251 | ||
cb5b5624 | 252 | /** |
29f8bf1b | 253 | * eeh_report_reset - Tell device that slot has been reset |
9b3c76f0 | 254 | * @data: eeh device |
29f8bf1b GS |
255 | * @userdata: return value |
256 | * | |
257 | * This routine must be called while EEH tries to reset particular | |
258 | * PCI device so that the associated PCI device driver could take | |
259 | * some actions, usually to save data the driver needs so that the | |
260 | * driver can work again while the device is recovered. | |
77bd7415 | 261 | */ |
9b3c76f0 | 262 | static void *eeh_report_reset(void *data, void *userdata) |
77bd7415 | 263 | { |
9b3c76f0 GS |
264 | struct eeh_dev *edev = (struct eeh_dev *)data; |
265 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 266 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 267 | struct pci_driver *driver; |
77bd7415 | 268 | |
feadf7c0 | 269 | if (!dev) return NULL; |
c58dc575 MM |
270 | dev->error_state = pci_channel_io_normal; |
271 | ||
feadf7c0 GS |
272 | driver = eeh_pcid_get(dev); |
273 | if (!driver) return NULL; | |
274 | ||
8535ef05 MM |
275 | eeh_enable_irq(dev); |
276 | ||
6a1ca373 | 277 | if (!driver->err_handler || |
f26c7a03 GS |
278 | !driver->err_handler->slot_reset || |
279 | (edev->mode & EEH_DEV_NO_HANDLER)) { | |
feadf7c0 | 280 | eeh_pcid_put(dev); |
9b3c76f0 | 281 | return NULL; |
feadf7c0 | 282 | } |
77bd7415 | 283 | |
6a1ca373 | 284 | rc = driver->err_handler->slot_reset(dev); |
5794dbcb LV |
285 | if ((*res == PCI_ERS_RESULT_NONE) || |
286 | (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; | |
6a1ca373 LV |
287 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
288 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
70298c6e | 289 | |
feadf7c0 | 290 | eeh_pcid_put(dev); |
9b3c76f0 | 291 | return NULL; |
77bd7415 LV |
292 | } |
293 | ||
cb5b5624 | 294 | /** |
29f8bf1b | 295 | * eeh_report_resume - Tell device to resume normal operations |
9b3c76f0 | 296 | * @data: eeh device |
29f8bf1b GS |
297 | * @userdata: return value |
298 | * | |
299 | * This routine must be called to notify the device driver that it | |
300 | * could resume so that the device driver can do some initialization | |
301 | * to make the recovered device work again. | |
cb5b5624 | 302 | */ |
9b3c76f0 | 303 | static void *eeh_report_resume(void *data, void *userdata) |
77bd7415 | 304 | { |
9b3c76f0 GS |
305 | struct eeh_dev *edev = (struct eeh_dev *)data; |
306 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
307 | struct pci_driver *driver; | |
308 | ||
309 | if (!dev) return NULL; | |
77bd7415 LV |
310 | dev->error_state = pci_channel_io_normal; |
311 | ||
feadf7c0 GS |
312 | driver = eeh_pcid_get(dev); |
313 | if (!driver) return NULL; | |
d0e70341 | 314 | |
8535ef05 MM |
315 | eeh_enable_irq(dev); |
316 | ||
d0e70341 | 317 | if (!driver->err_handler || |
f26c7a03 GS |
318 | !driver->err_handler->resume || |
319 | (edev->mode & EEH_DEV_NO_HANDLER)) { | |
320 | edev->mode &= ~EEH_DEV_NO_HANDLER; | |
feadf7c0 | 321 | eeh_pcid_put(dev); |
9b3c76f0 | 322 | return NULL; |
feadf7c0 | 323 | } |
77bd7415 LV |
324 | |
325 | driver->err_handler->resume(dev); | |
70298c6e | 326 | |
feadf7c0 | 327 | eeh_pcid_put(dev); |
9b3c76f0 | 328 | return NULL; |
77bd7415 LV |
329 | } |
330 | ||
cb5b5624 | 331 | /** |
29f8bf1b | 332 | * eeh_report_failure - Tell device driver that device is dead. |
9b3c76f0 | 333 | * @data: eeh device |
29f8bf1b | 334 | * @userdata: return value |
cb5b5624 LV |
335 | * |
336 | * This informs the device driver that the device is permanently | |
337 | * dead, and that no further recovery attempts will be made on it. | |
338 | */ | |
9b3c76f0 | 339 | static void *eeh_report_failure(void *data, void *userdata) |
77bd7415 | 340 | { |
9b3c76f0 GS |
341 | struct eeh_dev *edev = (struct eeh_dev *)data; |
342 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
343 | struct pci_driver *driver; | |
344 | ||
345 | if (!dev) return NULL; | |
77bd7415 LV |
346 | dev->error_state = pci_channel_io_perm_failure; |
347 | ||
feadf7c0 GS |
348 | driver = eeh_pcid_get(dev); |
349 | if (!driver) return NULL; | |
77bd7415 | 350 | |
8535ef05 MM |
351 | eeh_disable_irq(dev); |
352 | ||
353 | if (!driver->err_handler || | |
feadf7c0 GS |
354 | !driver->err_handler->error_detected) { |
355 | eeh_pcid_put(dev); | |
9b3c76f0 | 356 | return NULL; |
feadf7c0 | 357 | } |
8535ef05 | 358 | |
77bd7415 | 359 | driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); |
70298c6e | 360 | |
feadf7c0 | 361 | eeh_pcid_put(dev); |
9b3c76f0 | 362 | return NULL; |
77bd7415 LV |
363 | } |
364 | ||
f5c57710 GS |
365 | static void *eeh_rmv_device(void *data, void *userdata) |
366 | { | |
367 | struct pci_driver *driver; | |
368 | struct eeh_dev *edev = (struct eeh_dev *)data; | |
369 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
370 | int *removed = (int *)userdata; | |
371 | ||
372 | /* | |
373 | * Actually, we should remove the PCI bridges as well. | |
374 | * However, that's lots of complexity to do that, | |
375 | * particularly some of devices under the bridge might | |
376 | * support EEH. So we just care about PCI devices for | |
377 | * simplicity here. | |
378 | */ | |
379 | if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) | |
380 | return NULL; | |
8cc6b6cd | 381 | |
f5c57710 | 382 | driver = eeh_pcid_get(dev); |
8cc6b6cd TLSC |
383 | if (driver) { |
384 | eeh_pcid_put(dev); | |
385 | if (driver->err_handler) | |
386 | return NULL; | |
387 | } | |
f5c57710 GS |
388 | |
389 | /* Remove it from PCI subsystem */ | |
390 | pr_debug("EEH: Removing %s without EEH sensitive driver\n", | |
391 | pci_name(dev)); | |
392 | edev->bus = dev->bus; | |
393 | edev->mode |= EEH_DEV_DISCONNECTED; | |
394 | (*removed)++; | |
395 | ||
1c2042c8 | 396 | pci_lock_rescan_remove(); |
f5c57710 | 397 | pci_stop_and_remove_bus_device(dev); |
1c2042c8 | 398 | pci_unlock_rescan_remove(); |
f5c57710 GS |
399 | |
400 | return NULL; | |
401 | } | |
402 | ||
403 | static void *eeh_pe_detach_dev(void *data, void *userdata) | |
404 | { | |
405 | struct eeh_pe *pe = (struct eeh_pe *)data; | |
406 | struct eeh_dev *edev, *tmp; | |
407 | ||
408 | eeh_pe_for_each_dev(pe, edev, tmp) { | |
409 | if (!(edev->mode & EEH_DEV_DISCONNECTED)) | |
410 | continue; | |
411 | ||
412 | edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED); | |
413 | eeh_rmv_from_parent_pe(edev); | |
414 | } | |
415 | ||
416 | return NULL; | |
417 | } | |
418 | ||
77bd7415 | 419 | /** |
29f8bf1b | 420 | * eeh_reset_device - Perform actual reset of a pci slot |
9b3c76f0 | 421 | * @pe: EEH PE |
29f8bf1b | 422 | * @bus: PCI bus corresponding to the isolcated slot |
77bd7415 | 423 | * |
29f8bf1b GS |
424 | * This routine must be called to do reset on the indicated PE. |
425 | * During the reset, udev might be invoked because those affected | |
426 | * PCI devices will be removed and then added. | |
77bd7415 | 427 | */ |
9b3c76f0 | 428 | static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus) |
77bd7415 | 429 | { |
f5c57710 | 430 | struct pci_bus *frozen_bus = eeh_pe_bus_get(pe); |
5a71978e | 431 | struct timeval tstamp; |
f5c57710 | 432 | int cnt, rc, removed = 0; |
42405456 LV |
433 | |
434 | /* pcibios will clear the counter; save the value */ | |
9b3c76f0 | 435 | cnt = pe->freeze_count; |
5a71978e | 436 | tstamp = pe->tstamp; |
42405456 | 437 | |
20ee6a97 GS |
438 | /* |
439 | * We don't remove the corresponding PE instances because | |
440 | * we need the information afterwords. The attached EEH | |
441 | * devices are expected to be attached soon when calling | |
442 | * into pcibios_add_pci_devices(). | |
443 | */ | |
f5c57710 | 444 | eeh_pe_state_mark(pe, EEH_PE_KEEP); |
1c2042c8 RW |
445 | if (bus) { |
446 | pci_lock_rescan_remove(); | |
807a827d | 447 | pcibios_remove_pci_devices(bus); |
1c2042c8 RW |
448 | pci_unlock_rescan_remove(); |
449 | } else if (frozen_bus) { | |
f5c57710 | 450 | eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed); |
1c2042c8 | 451 | } |
77bd7415 LV |
452 | |
453 | /* Reset the pci controller. (Asserts RST#; resets config space). | |
b6495c0c | 454 | * Reconfigure bridges and devices. Don't try to bring the system |
29f8bf1b GS |
455 | * up if the reset failed for some reason. |
456 | */ | |
9b3c76f0 | 457 | rc = eeh_reset_pe(pe); |
b6495c0c LV |
458 | if (rc) |
459 | return rc; | |
77bd7415 | 460 | |
1c2042c8 RW |
461 | pci_lock_rescan_remove(); |
462 | ||
9b3c76f0 GS |
463 | /* Restore PE */ |
464 | eeh_ops->configure_bridge(pe); | |
465 | eeh_pe_restore_bars(pe); | |
77bd7415 LV |
466 | |
467 | /* Give the system 5 seconds to finish running the user-space | |
a84f273c GS |
468 | * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, |
469 | * this is a hack, but if we don't do this, and try to bring | |
470 | * the device up before the scripts have taken it down, | |
77bd7415 LV |
471 | * potentially weird things happen. |
472 | */ | |
473 | if (bus) { | |
f5c57710 | 474 | pr_info("EEH: Sleep 5s ahead of complete hotplug\n"); |
29f8bf1b | 475 | ssleep(5); |
f5c57710 GS |
476 | |
477 | /* | |
478 | * The EEH device is still connected with its parent | |
479 | * PE. We should disconnect it so the binding can be | |
480 | * rebuilt when adding PCI devices. | |
481 | */ | |
482 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); | |
77bd7415 | 483 | pcibios_add_pci_devices(bus); |
f5c57710 GS |
484 | } else if (frozen_bus && removed) { |
485 | pr_info("EEH: Sleep 5s ahead of partial hotplug\n"); | |
486 | ssleep(5); | |
487 | ||
488 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); | |
489 | pcibios_add_pci_devices(frozen_bus); | |
77bd7415 | 490 | } |
f5c57710 | 491 | eeh_pe_state_clear(pe, EEH_PE_KEEP); |
5a71978e GS |
492 | |
493 | pe->tstamp = tstamp; | |
9b3c76f0 | 494 | pe->freeze_count = cnt; |
b6495c0c | 495 | |
1c2042c8 | 496 | pci_unlock_rescan_remove(); |
b6495c0c | 497 | return 0; |
77bd7415 LV |
498 | } |
499 | ||
500 | /* The longest amount of time to wait for a pci device | |
501 | * to come back on line, in seconds. | |
502 | */ | |
fb48dc22 | 503 | #define MAX_WAIT_FOR_RECOVERY 300 |
77bd7415 | 504 | |
8a6b1bc7 | 505 | static void eeh_handle_normal_event(struct eeh_pe *pe) |
77bd7415 | 506 | { |
77bd7415 | 507 | struct pci_bus *frozen_bus; |
b6495c0c | 508 | int rc = 0; |
18eb3b39 | 509 | enum pci_ers_result result = PCI_ERS_RESULT_NONE; |
77bd7415 | 510 | |
9b3c76f0 | 511 | frozen_bus = eeh_pe_bus_get(pe); |
77bd7415 | 512 | if (!frozen_bus) { |
9b3c76f0 GS |
513 | pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n", |
514 | __func__, pe->phb->global_number, pe->addr); | |
515 | return; | |
77bd7415 LV |
516 | } |
517 | ||
5a71978e | 518 | eeh_pe_update_time_stamp(pe); |
9b3c76f0 GS |
519 | pe->freeze_count++; |
520 | if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES) | |
8df83028 | 521 | goto excess_failures; |
9b3c76f0 GS |
522 | pr_warning("EEH: This PCI device has failed %d times in the last hour\n", |
523 | pe->freeze_count); | |
77bd7415 LV |
524 | |
525 | /* Walk the various device drivers attached to this slot through | |
526 | * a reset sequence, giving each an opportunity to do what it needs | |
527 | * to accomplish the reset. Each child gets a report of the | |
528 | * status ... if any child can't handle the reset, then the entire | |
529 | * slot is dlpar removed and added. | |
530 | */ | |
56ca4fde | 531 | pr_info("EEH: Notify device drivers to shutdown\n"); |
9b3c76f0 | 532 | eeh_pe_dev_traverse(pe, eeh_report_error, &result); |
77bd7415 | 533 | |
5f1a7c81 | 534 | /* Get the current PCI slot state. This can take a long time, |
29f8bf1b GS |
535 | * sometimes over 3 seconds for certain systems. |
536 | */ | |
9b3c76f0 | 537 | rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000); |
eb594a47 | 538 | if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) { |
56ca4fde | 539 | pr_warning("EEH: Permanent failure\n"); |
5f1a7c81 LV |
540 | goto hard_fail; |
541 | } | |
542 | ||
ede8ca26 LV |
543 | /* Since rtas may enable MMIO when posting the error log, |
544 | * don't post the error log until after all dev drivers | |
17213c3b LV |
545 | * have been informed. |
546 | */ | |
56ca4fde | 547 | pr_info("EEH: Collect temporary log\n"); |
9b3c76f0 | 548 | eeh_slot_error_detail(pe, EEH_LOG_TEMP); |
ede8ca26 | 549 | |
77bd7415 LV |
550 | /* If all device drivers were EEH-unaware, then shut |
551 | * down all of the device drivers, and hope they | |
552 | * go down willingly, without panicing the system. | |
553 | */ | |
18eb3b39 | 554 | if (result == PCI_ERS_RESULT_NONE) { |
56ca4fde | 555 | pr_info("EEH: Reset with hotplug activity\n"); |
9b3c76f0 | 556 | rc = eeh_reset_device(pe, frozen_bus); |
e0f90b64 | 557 | if (rc) { |
56ca4fde GS |
558 | pr_warning("%s: Unable to reset, err=%d\n", |
559 | __func__, rc); | |
b6495c0c | 560 | goto hard_fail; |
e0f90b64 | 561 | } |
77bd7415 LV |
562 | } |
563 | ||
6a1ca373 LV |
564 | /* If all devices reported they can proceed, then re-enable MMIO */ |
565 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { | |
56ca4fde | 566 | pr_info("EEH: Enable I/O for affected devices\n"); |
9b3c76f0 | 567 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO); |
6a1ca373 | 568 | |
fa1be476 LV |
569 | if (rc < 0) |
570 | goto hard_fail; | |
6a1ca373 LV |
571 | if (rc) { |
572 | result = PCI_ERS_RESULT_NEED_RESET; | |
573 | } else { | |
56ca4fde | 574 | pr_info("EEH: Notify device drivers to resume I/O\n"); |
6a1ca373 | 575 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 576 | eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result); |
6a1ca373 | 577 | } |
77bd7415 LV |
578 | } |
579 | ||
6a1ca373 | 580 | /* If all devices reported they can proceed, then re-enable DMA */ |
18eb3b39 | 581 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { |
56ca4fde | 582 | pr_info("EEH: Enabled DMA for affected devices\n"); |
9b3c76f0 | 583 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA); |
6a1ca373 | 584 | |
fa1be476 LV |
585 | if (rc < 0) |
586 | goto hard_fail; | |
6a1ca373 LV |
587 | if (rc) |
588 | result = PCI_ERS_RESULT_NEED_RESET; | |
d0e70341 LV |
589 | else |
590 | result = PCI_ERS_RESULT_RECOVERED; | |
6a1ca373 LV |
591 | } |
592 | ||
593 | /* If any device has a hard failure, then shut off everything. */ | |
e0f90b64 | 594 | if (result == PCI_ERS_RESULT_DISCONNECT) { |
56ca4fde | 595 | pr_warning("EEH: Device driver gave up\n"); |
6a1ca373 | 596 | goto hard_fail; |
e0f90b64 | 597 | } |
6a1ca373 LV |
598 | |
599 | /* If any device called out for a reset, then reset the slot */ | |
600 | if (result == PCI_ERS_RESULT_NEED_RESET) { | |
56ca4fde | 601 | pr_info("EEH: Reset without hotplug activity\n"); |
9b3c76f0 | 602 | rc = eeh_reset_device(pe, NULL); |
e0f90b64 | 603 | if (rc) { |
56ca4fde GS |
604 | pr_warning("%s: Cannot reset, err=%d\n", |
605 | __func__, rc); | |
b6495c0c | 606 | goto hard_fail; |
e0f90b64 | 607 | } |
56ca4fde GS |
608 | |
609 | pr_info("EEH: Notify device drivers " | |
610 | "the completion of reset\n"); | |
6a1ca373 | 611 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 612 | eeh_pe_dev_traverse(pe, eeh_report_reset, &result); |
77bd7415 LV |
613 | } |
614 | ||
6a1ca373 | 615 | /* All devices should claim they have recovered by now. */ |
90fdd613 LV |
616 | if ((result != PCI_ERS_RESULT_RECOVERED) && |
617 | (result != PCI_ERS_RESULT_NONE)) { | |
56ca4fde | 618 | pr_warning("EEH: Not recovered\n"); |
6a1ca373 | 619 | goto hard_fail; |
e0f90b64 | 620 | } |
6a1ca373 | 621 | |
77bd7415 | 622 | /* Tell all device drivers that they can resume operations */ |
56ca4fde | 623 | pr_info("EEH: Notify device driver to resume\n"); |
9b3c76f0 | 624 | eeh_pe_dev_traverse(pe, eeh_report_resume, NULL); |
b6495c0c | 625 | |
9b3c76f0 | 626 | return; |
a84f273c | 627 | |
8df83028 | 628 | excess_failures: |
b6495c0c LV |
629 | /* |
630 | * About 90% of all real-life EEH failures in the field | |
631 | * are due to poorly seated PCI cards. Only 10% or so are | |
632 | * due to actual, failed cards. | |
633 | */ | |
9b3c76f0 GS |
634 | pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n" |
635 | "last hour and has been permanently disabled.\n" | |
636 | "Please try reseating or replacing it.\n", | |
637 | pe->phb->global_number, pe->addr, | |
638 | pe->freeze_count); | |
8df83028 LV |
639 | goto perm_error; |
640 | ||
641 | hard_fail: | |
9b3c76f0 GS |
642 | pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n" |
643 | "Please try reseating or replacing it\n", | |
644 | pe->phb->global_number, pe->addr); | |
b6495c0c | 645 | |
8df83028 | 646 | perm_error: |
9b3c76f0 | 647 | eeh_slot_error_detail(pe, EEH_LOG_PERM); |
b6495c0c LV |
648 | |
649 | /* Notify all devices that they're about to go down. */ | |
9b3c76f0 | 650 | eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); |
b6495c0c LV |
651 | |
652 | /* Shut down the device drivers for good. */ | |
1c2042c8 RW |
653 | if (frozen_bus) { |
654 | pci_lock_rescan_remove(); | |
9b3c76f0 | 655 | pcibios_remove_pci_devices(frozen_bus); |
1c2042c8 RW |
656 | pci_unlock_rescan_remove(); |
657 | } | |
77bd7415 | 658 | } |
8a6b1bc7 GS |
659 | |
660 | static void eeh_handle_special_event(void) | |
661 | { | |
662 | struct eeh_pe *pe, *phb_pe; | |
663 | struct pci_bus *bus; | |
7e4e7867 | 664 | struct pci_controller *hose; |
8a6b1bc7 | 665 | unsigned long flags; |
7e4e7867 | 666 | int rc; |
8a6b1bc7 | 667 | |
8a6b1bc7 | 668 | |
7e4e7867 GS |
669 | do { |
670 | rc = eeh_ops->next_error(&pe); | |
671 | ||
672 | switch (rc) { | |
673 | case EEH_NEXT_ERR_DEAD_IOC: | |
674 | /* Mark all PHBs in dead state */ | |
675 | eeh_serialize_lock(&flags); | |
676 | ||
677 | /* Purge all events */ | |
678 | eeh_remove_event(NULL); | |
679 | ||
680 | list_for_each_entry(hose, &hose_list, list_node) { | |
681 | phb_pe = eeh_phb_pe_get(hose); | |
682 | if (!phb_pe) continue; | |
683 | ||
684 | eeh_pe_state_mark(phb_pe, | |
685 | EEH_PE_ISOLATED | EEH_PE_PHB_DEAD); | |
686 | } | |
687 | ||
688 | eeh_serialize_unlock(flags); | |
689 | ||
690 | break; | |
691 | case EEH_NEXT_ERR_FROZEN_PE: | |
692 | case EEH_NEXT_ERR_FENCED_PHB: | |
693 | case EEH_NEXT_ERR_DEAD_PHB: | |
694 | /* Mark the PE in fenced state */ | |
695 | eeh_serialize_lock(&flags); | |
696 | ||
697 | /* Purge all events of the PHB */ | |
698 | eeh_remove_event(pe); | |
699 | ||
700 | if (rc == EEH_NEXT_ERR_DEAD_PHB) | |
701 | eeh_pe_state_mark(pe, | |
702 | EEH_PE_ISOLATED | EEH_PE_PHB_DEAD); | |
703 | else | |
704 | eeh_pe_state_mark(pe, | |
705 | EEH_PE_ISOLATED | EEH_PE_RECOVERING); | |
706 | ||
707 | eeh_serialize_unlock(flags); | |
708 | ||
709 | break; | |
710 | case EEH_NEXT_ERR_NONE: | |
711 | return; | |
712 | default: | |
713 | pr_warn("%s: Invalid value %d from next_error()\n", | |
714 | __func__, rc); | |
715 | return; | |
8a6b1bc7 | 716 | } |
8a6b1bc7 | 717 | |
7e4e7867 GS |
718 | /* |
719 | * For fenced PHB and frozen PE, it's handled as normal | |
720 | * event. We have to remove the affected PHBs for dead | |
721 | * PHB and IOC | |
722 | */ | |
723 | if (rc == EEH_NEXT_ERR_FROZEN_PE || | |
724 | rc == EEH_NEXT_ERR_FENCED_PHB) { | |
725 | eeh_handle_normal_event(pe); | |
726 | } else { | |
1b17366d | 727 | pci_lock_rescan_remove(); |
7e4e7867 GS |
728 | list_for_each_entry(hose, &hose_list, list_node) { |
729 | phb_pe = eeh_phb_pe_get(hose); | |
730 | if (!phb_pe || | |
731 | !(phb_pe->state & EEH_PE_PHB_DEAD)) | |
732 | continue; | |
733 | ||
734 | /* Notify all devices to be down */ | |
735 | bus = eeh_pe_bus_get(phb_pe); | |
736 | eeh_pe_dev_traverse(pe, | |
737 | eeh_report_failure, NULL); | |
738 | pcibios_remove_pci_devices(bus); | |
739 | } | |
1b17366d | 740 | pci_unlock_rescan_remove(); |
8a6b1bc7 | 741 | } |
7e4e7867 GS |
742 | |
743 | /* | |
744 | * If we have detected dead IOC, we needn't proceed | |
745 | * any more since all PHBs would have been removed | |
746 | */ | |
747 | if (rc == EEH_NEXT_ERR_DEAD_IOC) | |
748 | break; | |
749 | } while (rc != EEH_NEXT_ERR_NONE); | |
8a6b1bc7 GS |
750 | } |
751 | ||
752 | /** | |
753 | * eeh_handle_event - Reset a PCI device after hard lockup. | |
754 | * @pe: EEH PE | |
755 | * | |
756 | * While PHB detects address or data parity errors on particular PCI | |
757 | * slot, the associated PE will be frozen. Besides, DMA's occurring | |
758 | * to wild addresses (which usually happen due to bugs in device | |
759 | * drivers or in PCI adapter firmware) can cause EEH error. #SERR, | |
760 | * #PERR or other misc PCI-related errors also can trigger EEH errors. | |
761 | * | |
762 | * Recovery process consists of unplugging the device driver (which | |
763 | * generated hotplug events to userspace), then issuing a PCI #RST to | |
764 | * the device, then reconfiguring the PCI config space for all bridges | |
765 | * & devices under this slot, and then finally restarting the device | |
766 | * drivers (which cause a second set of hotplug events to go out to | |
767 | * userspace). | |
768 | */ | |
769 | void eeh_handle_event(struct eeh_pe *pe) | |
770 | { | |
771 | if (pe) | |
772 | eeh_handle_normal_event(pe); | |
773 | else | |
774 | eeh_handle_special_event(); | |
775 | } |