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); |
91150af3 | 146 | struct irq_desc *desc; |
8535ef05 | 147 | |
dbbceee1 GS |
148 | if ((edev->mode) & EEH_DEV_IRQ_DISABLED) { |
149 | edev->mode &= ~EEH_DEV_IRQ_DISABLED; | |
91150af3 GS |
150 | |
151 | desc = irq_to_desc(dev->irq); | |
152 | if (desc && desc->depth > 0) | |
153 | enable_irq(dev->irq); | |
8535ef05 MM |
154 | } |
155 | } | |
156 | ||
cb5b5624 | 157 | /** |
29f8bf1b | 158 | * eeh_report_error - Report pci error to each device driver |
9b3c76f0 | 159 | * @data: eeh device |
29f8bf1b | 160 | * @userdata: return value |
a84f273c GS |
161 | * |
162 | * Report an EEH error to each device driver, collect up and | |
163 | * merge the device driver responses. Cumulative response | |
cb5b5624 | 164 | * passed back in "userdata". |
77bd7415 | 165 | */ |
9b3c76f0 | 166 | static void *eeh_report_error(void *data, void *userdata) |
77bd7415 | 167 | { |
9b3c76f0 GS |
168 | struct eeh_dev *edev = (struct eeh_dev *)data; |
169 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
18eb3b39 | 170 | enum pci_ers_result rc, *res = userdata; |
feadf7c0 | 171 | struct pci_driver *driver; |
77bd7415 | 172 | |
9b3c76f0 GS |
173 | /* We might not have the associated PCI device, |
174 | * then we should continue for next one. | |
175 | */ | |
176 | if (!dev) return NULL; | |
77bd7415 LV |
177 | dev->error_state = pci_channel_io_frozen; |
178 | ||
feadf7c0 GS |
179 | driver = eeh_pcid_get(dev); |
180 | if (!driver) return NULL; | |
77bd7415 | 181 | |
8535ef05 MM |
182 | eeh_disable_irq(dev); |
183 | ||
6a1ca373 | 184 | if (!driver->err_handler || |
feadf7c0 GS |
185 | !driver->err_handler->error_detected) { |
186 | eeh_pcid_put(dev); | |
9b3c76f0 | 187 | return NULL; |
feadf7c0 | 188 | } |
77bd7415 | 189 | |
29f8bf1b | 190 | rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen); |
2a50f144 LV |
191 | |
192 | /* A driver that needs a reset trumps all others */ | |
193 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
18eb3b39 | 194 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 195 | |
feadf7c0 | 196 | eeh_pcid_put(dev); |
9b3c76f0 | 197 | return NULL; |
6a1ca373 LV |
198 | } |
199 | ||
200 | /** | |
29f8bf1b | 201 | * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled |
9b3c76f0 | 202 | * @data: eeh device |
29f8bf1b | 203 | * @userdata: return value |
6a1ca373 | 204 | * |
638799b3 LV |
205 | * Tells each device driver that IO ports, MMIO and config space I/O |
206 | * are now enabled. Collects up and merges the device driver responses. | |
207 | * Cumulative response passed back in "userdata". | |
6a1ca373 | 208 | */ |
9b3c76f0 | 209 | static void *eeh_report_mmio_enabled(void *data, void *userdata) |
6a1ca373 | 210 | { |
9b3c76f0 GS |
211 | struct eeh_dev *edev = (struct eeh_dev *)data; |
212 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 213 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 214 | struct pci_driver *driver; |
6a1ca373 | 215 | |
feadf7c0 GS |
216 | driver = eeh_pcid_get(dev); |
217 | if (!driver) return NULL; | |
9b3c76f0 | 218 | |
feadf7c0 GS |
219 | if (!driver->err_handler || |
220 | !driver->err_handler->mmio_enabled) { | |
221 | eeh_pcid_put(dev); | |
9b3c76f0 | 222 | return NULL; |
feadf7c0 | 223 | } |
6a1ca373 | 224 | |
29f8bf1b | 225 | rc = driver->err_handler->mmio_enabled(dev); |
2a50f144 LV |
226 | |
227 | /* A driver that needs a reset trumps all others */ | |
228 | if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
6a1ca373 | 229 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
70298c6e | 230 | |
feadf7c0 | 231 | eeh_pcid_put(dev); |
9b3c76f0 | 232 | return NULL; |
77bd7415 LV |
233 | } |
234 | ||
cb5b5624 | 235 | /** |
29f8bf1b | 236 | * eeh_report_reset - Tell device that slot has been reset |
9b3c76f0 | 237 | * @data: eeh device |
29f8bf1b GS |
238 | * @userdata: return value |
239 | * | |
240 | * This routine must be called while EEH tries to reset particular | |
241 | * PCI device so that the associated PCI device driver could take | |
242 | * some actions, usually to save data the driver needs so that the | |
243 | * driver can work again while the device is recovered. | |
77bd7415 | 244 | */ |
9b3c76f0 | 245 | static void *eeh_report_reset(void *data, void *userdata) |
77bd7415 | 246 | { |
9b3c76f0 GS |
247 | struct eeh_dev *edev = (struct eeh_dev *)data; |
248 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
6a1ca373 | 249 | enum pci_ers_result rc, *res = userdata; |
9b3c76f0 | 250 | struct pci_driver *driver; |
77bd7415 | 251 | |
feadf7c0 | 252 | if (!dev) return NULL; |
c58dc575 MM |
253 | dev->error_state = pci_channel_io_normal; |
254 | ||
feadf7c0 GS |
255 | driver = eeh_pcid_get(dev); |
256 | if (!driver) return NULL; | |
257 | ||
8535ef05 MM |
258 | eeh_enable_irq(dev); |
259 | ||
6a1ca373 | 260 | if (!driver->err_handler || |
feadf7c0 GS |
261 | !driver->err_handler->slot_reset) { |
262 | eeh_pcid_put(dev); | |
9b3c76f0 | 263 | return NULL; |
feadf7c0 | 264 | } |
77bd7415 | 265 | |
6a1ca373 | 266 | rc = driver->err_handler->slot_reset(dev); |
5794dbcb LV |
267 | if ((*res == PCI_ERS_RESULT_NONE) || |
268 | (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; | |
6a1ca373 LV |
269 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
270 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
70298c6e | 271 | |
feadf7c0 | 272 | eeh_pcid_put(dev); |
9b3c76f0 | 273 | return NULL; |
77bd7415 LV |
274 | } |
275 | ||
cb5b5624 | 276 | /** |
29f8bf1b | 277 | * eeh_report_resume - Tell device to resume normal operations |
9b3c76f0 | 278 | * @data: eeh device |
29f8bf1b GS |
279 | * @userdata: return value |
280 | * | |
281 | * This routine must be called to notify the device driver that it | |
282 | * could resume so that the device driver can do some initialization | |
283 | * to make the recovered device work again. | |
cb5b5624 | 284 | */ |
9b3c76f0 | 285 | static void *eeh_report_resume(void *data, void *userdata) |
77bd7415 | 286 | { |
9b3c76f0 GS |
287 | struct eeh_dev *edev = (struct eeh_dev *)data; |
288 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
289 | struct pci_driver *driver; | |
290 | ||
291 | if (!dev) return NULL; | |
77bd7415 LV |
292 | dev->error_state = pci_channel_io_normal; |
293 | ||
feadf7c0 GS |
294 | driver = eeh_pcid_get(dev); |
295 | if (!driver) return NULL; | |
d0e70341 | 296 | |
8535ef05 MM |
297 | eeh_enable_irq(dev); |
298 | ||
d0e70341 | 299 | if (!driver->err_handler || |
feadf7c0 GS |
300 | !driver->err_handler->resume) { |
301 | eeh_pcid_put(dev); | |
9b3c76f0 | 302 | return NULL; |
feadf7c0 | 303 | } |
77bd7415 LV |
304 | |
305 | driver->err_handler->resume(dev); | |
70298c6e | 306 | |
feadf7c0 | 307 | eeh_pcid_put(dev); |
9b3c76f0 | 308 | return NULL; |
77bd7415 LV |
309 | } |
310 | ||
cb5b5624 | 311 | /** |
29f8bf1b | 312 | * eeh_report_failure - Tell device driver that device is dead. |
9b3c76f0 | 313 | * @data: eeh device |
29f8bf1b | 314 | * @userdata: return value |
cb5b5624 LV |
315 | * |
316 | * This informs the device driver that the device is permanently | |
317 | * dead, and that no further recovery attempts will be made on it. | |
318 | */ | |
9b3c76f0 | 319 | static void *eeh_report_failure(void *data, void *userdata) |
77bd7415 | 320 | { |
9b3c76f0 GS |
321 | struct eeh_dev *edev = (struct eeh_dev *)data; |
322 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
323 | struct pci_driver *driver; | |
324 | ||
325 | if (!dev) return NULL; | |
77bd7415 LV |
326 | dev->error_state = pci_channel_io_perm_failure; |
327 | ||
feadf7c0 GS |
328 | driver = eeh_pcid_get(dev); |
329 | if (!driver) return NULL; | |
77bd7415 | 330 | |
8535ef05 MM |
331 | eeh_disable_irq(dev); |
332 | ||
333 | if (!driver->err_handler || | |
feadf7c0 GS |
334 | !driver->err_handler->error_detected) { |
335 | eeh_pcid_put(dev); | |
9b3c76f0 | 336 | return NULL; |
feadf7c0 | 337 | } |
8535ef05 | 338 | |
77bd7415 | 339 | driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); |
70298c6e | 340 | |
feadf7c0 | 341 | eeh_pcid_put(dev); |
9b3c76f0 | 342 | return NULL; |
77bd7415 LV |
343 | } |
344 | ||
f5c57710 GS |
345 | static void *eeh_rmv_device(void *data, void *userdata) |
346 | { | |
347 | struct pci_driver *driver; | |
348 | struct eeh_dev *edev = (struct eeh_dev *)data; | |
349 | struct pci_dev *dev = eeh_dev_to_pci_dev(edev); | |
350 | int *removed = (int *)userdata; | |
351 | ||
352 | /* | |
353 | * Actually, we should remove the PCI bridges as well. | |
354 | * However, that's lots of complexity to do that, | |
355 | * particularly some of devices under the bridge might | |
356 | * support EEH. So we just care about PCI devices for | |
357 | * simplicity here. | |
358 | */ | |
359 | if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) | |
360 | return NULL; | |
361 | driver = eeh_pcid_get(dev); | |
362 | if (driver && driver->err_handler) | |
363 | return NULL; | |
364 | ||
365 | /* Remove it from PCI subsystem */ | |
366 | pr_debug("EEH: Removing %s without EEH sensitive driver\n", | |
367 | pci_name(dev)); | |
368 | edev->bus = dev->bus; | |
369 | edev->mode |= EEH_DEV_DISCONNECTED; | |
370 | (*removed)++; | |
371 | ||
372 | pci_stop_and_remove_bus_device(dev); | |
373 | ||
374 | return NULL; | |
375 | } | |
376 | ||
377 | static void *eeh_pe_detach_dev(void *data, void *userdata) | |
378 | { | |
379 | struct eeh_pe *pe = (struct eeh_pe *)data; | |
380 | struct eeh_dev *edev, *tmp; | |
381 | ||
382 | eeh_pe_for_each_dev(pe, edev, tmp) { | |
383 | if (!(edev->mode & EEH_DEV_DISCONNECTED)) | |
384 | continue; | |
385 | ||
386 | edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED); | |
387 | eeh_rmv_from_parent_pe(edev); | |
388 | } | |
389 | ||
390 | return NULL; | |
391 | } | |
392 | ||
77bd7415 | 393 | /** |
29f8bf1b | 394 | * eeh_reset_device - Perform actual reset of a pci slot |
9b3c76f0 | 395 | * @pe: EEH PE |
29f8bf1b | 396 | * @bus: PCI bus corresponding to the isolcated slot |
77bd7415 | 397 | * |
29f8bf1b GS |
398 | * This routine must be called to do reset on the indicated PE. |
399 | * During the reset, udev might be invoked because those affected | |
400 | * PCI devices will be removed and then added. | |
77bd7415 | 401 | */ |
9b3c76f0 | 402 | static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus) |
77bd7415 | 403 | { |
f5c57710 | 404 | struct pci_bus *frozen_bus = eeh_pe_bus_get(pe); |
5a71978e | 405 | struct timeval tstamp; |
f5c57710 | 406 | int cnt, rc, removed = 0; |
42405456 LV |
407 | |
408 | /* pcibios will clear the counter; save the value */ | |
9b3c76f0 | 409 | cnt = pe->freeze_count; |
5a71978e | 410 | tstamp = pe->tstamp; |
42405456 | 411 | |
20ee6a97 GS |
412 | /* |
413 | * We don't remove the corresponding PE instances because | |
414 | * we need the information afterwords. The attached EEH | |
415 | * devices are expected to be attached soon when calling | |
416 | * into pcibios_add_pci_devices(). | |
417 | */ | |
f5c57710 GS |
418 | eeh_pe_state_mark(pe, EEH_PE_KEEP); |
419 | if (bus) | |
807a827d | 420 | pcibios_remove_pci_devices(bus); |
f5c57710 GS |
421 | else if (frozen_bus) |
422 | eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed); | |
77bd7415 LV |
423 | |
424 | /* Reset the pci controller. (Asserts RST#; resets config space). | |
b6495c0c | 425 | * Reconfigure bridges and devices. Don't try to bring the system |
29f8bf1b GS |
426 | * up if the reset failed for some reason. |
427 | */ | |
9b3c76f0 | 428 | rc = eeh_reset_pe(pe); |
b6495c0c LV |
429 | if (rc) |
430 | return rc; | |
77bd7415 | 431 | |
9b3c76f0 GS |
432 | /* Restore PE */ |
433 | eeh_ops->configure_bridge(pe); | |
434 | eeh_pe_restore_bars(pe); | |
77bd7415 LV |
435 | |
436 | /* Give the system 5 seconds to finish running the user-space | |
a84f273c GS |
437 | * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, |
438 | * this is a hack, but if we don't do this, and try to bring | |
439 | * the device up before the scripts have taken it down, | |
77bd7415 LV |
440 | * potentially weird things happen. |
441 | */ | |
442 | if (bus) { | |
f5c57710 | 443 | pr_info("EEH: Sleep 5s ahead of complete hotplug\n"); |
29f8bf1b | 444 | ssleep(5); |
f5c57710 GS |
445 | |
446 | /* | |
447 | * The EEH device is still connected with its parent | |
448 | * PE. We should disconnect it so the binding can be | |
449 | * rebuilt when adding PCI devices. | |
450 | */ | |
451 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); | |
77bd7415 | 452 | pcibios_add_pci_devices(bus); |
f5c57710 GS |
453 | } else if (frozen_bus && removed) { |
454 | pr_info("EEH: Sleep 5s ahead of partial hotplug\n"); | |
455 | ssleep(5); | |
456 | ||
457 | eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); | |
458 | pcibios_add_pci_devices(frozen_bus); | |
77bd7415 | 459 | } |
f5c57710 | 460 | eeh_pe_state_clear(pe, EEH_PE_KEEP); |
5a71978e GS |
461 | |
462 | pe->tstamp = tstamp; | |
9b3c76f0 | 463 | pe->freeze_count = cnt; |
b6495c0c LV |
464 | |
465 | return 0; | |
77bd7415 LV |
466 | } |
467 | ||
468 | /* The longest amount of time to wait for a pci device | |
469 | * to come back on line, in seconds. | |
470 | */ | |
fa1be476 | 471 | #define MAX_WAIT_FOR_RECOVERY 150 |
77bd7415 | 472 | |
8a6b1bc7 | 473 | static void eeh_handle_normal_event(struct eeh_pe *pe) |
77bd7415 | 474 | { |
77bd7415 | 475 | struct pci_bus *frozen_bus; |
b6495c0c | 476 | int rc = 0; |
18eb3b39 | 477 | enum pci_ers_result result = PCI_ERS_RESULT_NONE; |
77bd7415 | 478 | |
9b3c76f0 | 479 | frozen_bus = eeh_pe_bus_get(pe); |
77bd7415 | 480 | if (!frozen_bus) { |
9b3c76f0 GS |
481 | pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n", |
482 | __func__, pe->phb->global_number, pe->addr); | |
483 | return; | |
77bd7415 LV |
484 | } |
485 | ||
5a71978e | 486 | eeh_pe_update_time_stamp(pe); |
9b3c76f0 GS |
487 | pe->freeze_count++; |
488 | if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES) | |
8df83028 | 489 | goto excess_failures; |
9b3c76f0 GS |
490 | pr_warning("EEH: This PCI device has failed %d times in the last hour\n", |
491 | pe->freeze_count); | |
77bd7415 LV |
492 | |
493 | /* Walk the various device drivers attached to this slot through | |
494 | * a reset sequence, giving each an opportunity to do what it needs | |
495 | * to accomplish the reset. Each child gets a report of the | |
496 | * status ... if any child can't handle the reset, then the entire | |
497 | * slot is dlpar removed and added. | |
498 | */ | |
56ca4fde | 499 | pr_info("EEH: Notify device drivers to shutdown\n"); |
9b3c76f0 | 500 | eeh_pe_dev_traverse(pe, eeh_report_error, &result); |
77bd7415 | 501 | |
5f1a7c81 | 502 | /* Get the current PCI slot state. This can take a long time, |
29f8bf1b GS |
503 | * sometimes over 3 seconds for certain systems. |
504 | */ | |
9b3c76f0 | 505 | rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000); |
eb594a47 | 506 | if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) { |
56ca4fde | 507 | pr_warning("EEH: Permanent failure\n"); |
5f1a7c81 LV |
508 | goto hard_fail; |
509 | } | |
510 | ||
ede8ca26 LV |
511 | /* Since rtas may enable MMIO when posting the error log, |
512 | * don't post the error log until after all dev drivers | |
17213c3b LV |
513 | * have been informed. |
514 | */ | |
56ca4fde | 515 | pr_info("EEH: Collect temporary log\n"); |
9b3c76f0 | 516 | eeh_slot_error_detail(pe, EEH_LOG_TEMP); |
ede8ca26 | 517 | |
77bd7415 LV |
518 | /* If all device drivers were EEH-unaware, then shut |
519 | * down all of the device drivers, and hope they | |
520 | * go down willingly, without panicing the system. | |
521 | */ | |
18eb3b39 | 522 | if (result == PCI_ERS_RESULT_NONE) { |
56ca4fde | 523 | pr_info("EEH: Reset with hotplug activity\n"); |
9b3c76f0 | 524 | rc = eeh_reset_device(pe, frozen_bus); |
e0f90b64 | 525 | if (rc) { |
56ca4fde GS |
526 | pr_warning("%s: Unable to reset, err=%d\n", |
527 | __func__, rc); | |
b6495c0c | 528 | goto hard_fail; |
e0f90b64 | 529 | } |
77bd7415 LV |
530 | } |
531 | ||
6a1ca373 LV |
532 | /* If all devices reported they can proceed, then re-enable MMIO */ |
533 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { | |
56ca4fde | 534 | pr_info("EEH: Enable I/O for affected devices\n"); |
9b3c76f0 | 535 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO); |
6a1ca373 | 536 | |
fa1be476 LV |
537 | if (rc < 0) |
538 | goto hard_fail; | |
6a1ca373 LV |
539 | if (rc) { |
540 | result = PCI_ERS_RESULT_NEED_RESET; | |
541 | } else { | |
56ca4fde | 542 | pr_info("EEH: Notify device drivers to resume I/O\n"); |
6a1ca373 | 543 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 544 | eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result); |
6a1ca373 | 545 | } |
77bd7415 LV |
546 | } |
547 | ||
6a1ca373 | 548 | /* If all devices reported they can proceed, then re-enable DMA */ |
18eb3b39 | 549 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { |
56ca4fde | 550 | pr_info("EEH: Enabled DMA for affected devices\n"); |
9b3c76f0 | 551 | rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA); |
6a1ca373 | 552 | |
fa1be476 LV |
553 | if (rc < 0) |
554 | goto hard_fail; | |
6a1ca373 LV |
555 | if (rc) |
556 | result = PCI_ERS_RESULT_NEED_RESET; | |
d0e70341 LV |
557 | else |
558 | result = PCI_ERS_RESULT_RECOVERED; | |
6a1ca373 LV |
559 | } |
560 | ||
561 | /* If any device has a hard failure, then shut off everything. */ | |
e0f90b64 | 562 | if (result == PCI_ERS_RESULT_DISCONNECT) { |
56ca4fde | 563 | pr_warning("EEH: Device driver gave up\n"); |
6a1ca373 | 564 | goto hard_fail; |
e0f90b64 | 565 | } |
6a1ca373 LV |
566 | |
567 | /* If any device called out for a reset, then reset the slot */ | |
568 | if (result == PCI_ERS_RESULT_NEED_RESET) { | |
56ca4fde | 569 | pr_info("EEH: Reset without hotplug activity\n"); |
9b3c76f0 | 570 | rc = eeh_reset_device(pe, NULL); |
e0f90b64 | 571 | if (rc) { |
56ca4fde GS |
572 | pr_warning("%s: Cannot reset, err=%d\n", |
573 | __func__, rc); | |
b6495c0c | 574 | goto hard_fail; |
e0f90b64 | 575 | } |
56ca4fde GS |
576 | |
577 | pr_info("EEH: Notify device drivers " | |
578 | "the completion of reset\n"); | |
6a1ca373 | 579 | result = PCI_ERS_RESULT_NONE; |
9b3c76f0 | 580 | eeh_pe_dev_traverse(pe, eeh_report_reset, &result); |
77bd7415 LV |
581 | } |
582 | ||
6a1ca373 | 583 | /* All devices should claim they have recovered by now. */ |
90fdd613 LV |
584 | if ((result != PCI_ERS_RESULT_RECOVERED) && |
585 | (result != PCI_ERS_RESULT_NONE)) { | |
56ca4fde | 586 | pr_warning("EEH: Not recovered\n"); |
6a1ca373 | 587 | goto hard_fail; |
e0f90b64 | 588 | } |
6a1ca373 | 589 | |
77bd7415 | 590 | /* Tell all device drivers that they can resume operations */ |
56ca4fde | 591 | pr_info("EEH: Notify device driver to resume\n"); |
9b3c76f0 | 592 | eeh_pe_dev_traverse(pe, eeh_report_resume, NULL); |
b6495c0c | 593 | |
9b3c76f0 | 594 | return; |
a84f273c | 595 | |
8df83028 | 596 | excess_failures: |
b6495c0c LV |
597 | /* |
598 | * About 90% of all real-life EEH failures in the field | |
599 | * are due to poorly seated PCI cards. Only 10% or so are | |
600 | * due to actual, failed cards. | |
601 | */ | |
9b3c76f0 GS |
602 | pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n" |
603 | "last hour and has been permanently disabled.\n" | |
604 | "Please try reseating or replacing it.\n", | |
605 | pe->phb->global_number, pe->addr, | |
606 | pe->freeze_count); | |
8df83028 LV |
607 | goto perm_error; |
608 | ||
609 | hard_fail: | |
9b3c76f0 GS |
610 | pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n" |
611 | "Please try reseating or replacing it\n", | |
612 | pe->phb->global_number, pe->addr); | |
b6495c0c | 613 | |
8df83028 | 614 | perm_error: |
9b3c76f0 | 615 | eeh_slot_error_detail(pe, EEH_LOG_PERM); |
b6495c0c LV |
616 | |
617 | /* Notify all devices that they're about to go down. */ | |
9b3c76f0 | 618 | eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); |
b6495c0c LV |
619 | |
620 | /* Shut down the device drivers for good. */ | |
9b3c76f0 GS |
621 | if (frozen_bus) |
622 | pcibios_remove_pci_devices(frozen_bus); | |
77bd7415 | 623 | } |
8a6b1bc7 GS |
624 | |
625 | static void eeh_handle_special_event(void) | |
626 | { | |
627 | struct eeh_pe *pe, *phb_pe; | |
628 | struct pci_bus *bus; | |
629 | struct pci_controller *hose, *tmp; | |
630 | unsigned long flags; | |
631 | int rc = 0; | |
632 | ||
633 | /* | |
634 | * The return value from next_error() has been classified as follows. | |
635 | * It might be good to enumerate them. However, next_error() is only | |
636 | * supported by PowerNV platform for now. So it would be fine to use | |
637 | * integer directly: | |
638 | * | |
639 | * 4 - Dead IOC 3 - Dead PHB | |
640 | * 2 - Fenced PHB 1 - Frozen PE | |
641 | * 0 - No error found | |
642 | * | |
643 | */ | |
644 | rc = eeh_ops->next_error(&pe); | |
645 | if (rc <= 0) | |
646 | return; | |
647 | ||
648 | switch (rc) { | |
649 | case 4: | |
650 | /* Mark all PHBs in dead state */ | |
651 | eeh_serialize_lock(&flags); | |
652 | list_for_each_entry_safe(hose, tmp, | |
653 | &hose_list, list_node) { | |
654 | phb_pe = eeh_phb_pe_get(hose); | |
655 | if (!phb_pe) continue; | |
656 | ||
657 | eeh_pe_state_mark(phb_pe, | |
658 | EEH_PE_ISOLATED | EEH_PE_PHB_DEAD); | |
659 | } | |
660 | eeh_serialize_unlock(flags); | |
661 | ||
662 | /* Purge all events */ | |
663 | eeh_remove_event(NULL); | |
664 | break; | |
665 | case 3: | |
666 | case 2: | |
667 | case 1: | |
668 | /* Mark the PE in fenced state */ | |
669 | eeh_serialize_lock(&flags); | |
670 | if (rc == 3) | |
671 | eeh_pe_state_mark(pe, | |
672 | EEH_PE_ISOLATED | EEH_PE_PHB_DEAD); | |
673 | else | |
674 | eeh_pe_state_mark(pe, | |
675 | EEH_PE_ISOLATED | EEH_PE_RECOVERING); | |
676 | eeh_serialize_unlock(flags); | |
677 | ||
678 | /* Purge all events of the PHB */ | |
679 | eeh_remove_event(pe); | |
680 | break; | |
681 | default: | |
682 | pr_err("%s: Invalid value %d from next_error()\n", | |
683 | __func__, rc); | |
684 | return; | |
685 | } | |
686 | ||
687 | /* | |
688 | * For fenced PHB and frozen PE, it's handled as normal | |
689 | * event. We have to remove the affected PHBs for dead | |
690 | * PHB and IOC | |
691 | */ | |
692 | if (rc == 2 || rc == 1) | |
693 | eeh_handle_normal_event(pe); | |
694 | else { | |
695 | list_for_each_entry_safe(hose, tmp, | |
696 | &hose_list, list_node) { | |
697 | phb_pe = eeh_phb_pe_get(hose); | |
698 | if (!phb_pe || !(phb_pe->state & EEH_PE_PHB_DEAD)) | |
699 | continue; | |
700 | ||
701 | bus = eeh_pe_bus_get(phb_pe); | |
702 | /* Notify all devices that they're about to go down. */ | |
703 | eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); | |
704 | pcibios_remove_pci_devices(bus); | |
705 | } | |
706 | } | |
707 | } | |
708 | ||
709 | /** | |
710 | * eeh_handle_event - Reset a PCI device after hard lockup. | |
711 | * @pe: EEH PE | |
712 | * | |
713 | * While PHB detects address or data parity errors on particular PCI | |
714 | * slot, the associated PE will be frozen. Besides, DMA's occurring | |
715 | * to wild addresses (which usually happen due to bugs in device | |
716 | * drivers or in PCI adapter firmware) can cause EEH error. #SERR, | |
717 | * #PERR or other misc PCI-related errors also can trigger EEH errors. | |
718 | * | |
719 | * Recovery process consists of unplugging the device driver (which | |
720 | * generated hotplug events to userspace), then issuing a PCI #RST to | |
721 | * the device, then reconfiguring the PCI config space for all bridges | |
722 | * & devices under this slot, and then finally restarting the device | |
723 | * drivers (which cause a second set of hotplug events to go out to | |
724 | * userspace). | |
725 | */ | |
726 | void eeh_handle_event(struct eeh_pe *pe) | |
727 | { | |
728 | if (pe) | |
729 | eeh_handle_normal_event(pe); | |
730 | else | |
731 | eeh_handle_special_event(); | |
732 | } |