Commit | Line | Data |
---|---|---|
77bd7415 LV |
1 | /* |
2 | * PCI Error Recovery Driver for RPA-compliant PPC64 platform. | |
3 | * Copyright (C) 2004, 2005 Linas Vepstas <linas@linas.org> | |
4 | * | |
5 | * All rights reserved. | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or (at | |
10 | * your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, but | |
13 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or | |
15 | * NON INFRINGEMENT. See the GNU General Public License for more | |
16 | * details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | * | |
22 | * Send feedback to <linas@us.ibm.com> | |
23 | * | |
24 | */ | |
25 | #include <linux/delay.h> | |
77bd7415 | 26 | #include <linux/interrupt.h> |
ac325acd | 27 | #include <linux/irq.h> |
77bd7415 LV |
28 | #include <linux/pci.h> |
29 | #include <asm/eeh.h> | |
30 | #include <asm/eeh_event.h> | |
31 | #include <asm/ppc-pci.h> | |
32 | #include <asm/pci-bridge.h> | |
33 | #include <asm/prom.h> | |
34 | #include <asm/rtas.h> | |
35 | ||
36 | ||
37 | static inline const char * pcid_name (struct pci_dev *pdev) | |
38 | { | |
273d2803 | 39 | if (pdev && pdev->dev.driver) |
77bd7415 LV |
40 | return pdev->dev.driver->name; |
41 | return ""; | |
42 | } | |
43 | ||
77bd7415 LV |
44 | #ifdef DEBUG |
45 | static void print_device_node_tree (struct pci_dn *pdn, int dent) | |
46 | { | |
47 | int i; | |
48 | if (!pdn) return; | |
49 | for (i=0;i<dent; i++) | |
50 | printk(" "); | |
51 | printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n", | |
52 | pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr, | |
53 | pdn->eeh_pe_config_addr, pdn->node->full_name); | |
54 | dent += 3; | |
55 | struct device_node *pc = pdn->node->child; | |
56 | while (pc) { | |
57 | print_device_node_tree(PCI_DN(pc), dent); | |
58 | pc = pc->sibling; | |
59 | } | |
60 | } | |
61 | #endif | |
62 | ||
63 | /** | |
64 | * irq_in_use - return true if this irq is being used | |
65 | */ | |
66 | static int irq_in_use(unsigned int irq) | |
67 | { | |
68 | int rc = 0; | |
69 | unsigned long flags; | |
70 | struct irq_desc *desc = irq_desc + irq; | |
71 | ||
72 | spin_lock_irqsave(&desc->lock, flags); | |
73 | if (desc->action) | |
74 | rc = 1; | |
75 | spin_unlock_irqrestore(&desc->lock, flags); | |
76 | return rc; | |
77 | } | |
78 | ||
79 | /* ------------------------------------------------------- */ | |
cb5b5624 LV |
80 | /** |
81 | * eeh_report_error - report pci error to each device driver | |
82 | * | |
83 | * Report an EEH error to each device driver, collect up and | |
84 | * merge the device driver responses. Cumulative response | |
85 | * passed back in "userdata". | |
77bd7415 LV |
86 | */ |
87 | ||
88 | static void eeh_report_error(struct pci_dev *dev, void *userdata) | |
89 | { | |
18eb3b39 | 90 | enum pci_ers_result rc, *res = userdata; |
77bd7415 LV |
91 | struct pci_driver *driver = dev->driver; |
92 | ||
93 | dev->error_state = pci_channel_io_frozen; | |
94 | ||
95 | if (!driver) | |
96 | return; | |
97 | ||
98 | if (irq_in_use (dev->irq)) { | |
99 | struct device_node *dn = pci_device_to_OF_node(dev); | |
100 | PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED; | |
101 | disable_irq_nosync(dev->irq); | |
102 | } | |
6a1ca373 LV |
103 | if (!driver->err_handler || |
104 | !driver->err_handler->error_detected) | |
77bd7415 LV |
105 | return; |
106 | ||
107 | rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen); | |
18eb3b39 | 108 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
6a1ca373 LV |
109 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
110 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
111 | } | |
112 | ||
113 | /** | |
114 | * eeh_report_mmio_enabled - tell drivers that MMIO has been enabled | |
115 | * | |
116 | * Report an EEH error to each device driver, collect up and | |
117 | * merge the device driver responses. Cumulative response | |
118 | * passed back in "userdata". | |
119 | */ | |
120 | ||
121 | static void eeh_report_mmio_enabled(struct pci_dev *dev, void *userdata) | |
122 | { | |
123 | enum pci_ers_result rc, *res = userdata; | |
124 | struct pci_driver *driver = dev->driver; | |
125 | ||
126 | // dev->error_state = pci_channel_mmio_enabled; | |
127 | ||
128 | if (!driver || | |
129 | !driver->err_handler || | |
130 | !driver->err_handler->mmio_enabled) | |
131 | return; | |
132 | ||
133 | rc = driver->err_handler->mmio_enabled (dev); | |
134 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; | |
18eb3b39 PM |
135 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
136 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
77bd7415 LV |
137 | } |
138 | ||
cb5b5624 LV |
139 | /** |
140 | * eeh_report_reset - tell device that slot has been reset | |
77bd7415 LV |
141 | */ |
142 | ||
143 | static void eeh_report_reset(struct pci_dev *dev, void *userdata) | |
144 | { | |
6a1ca373 | 145 | enum pci_ers_result rc, *res = userdata; |
77bd7415 LV |
146 | struct pci_driver *driver = dev->driver; |
147 | struct device_node *dn = pci_device_to_OF_node(dev); | |
148 | ||
149 | if (!driver) | |
150 | return; | |
151 | ||
152 | if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) { | |
153 | PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED; | |
154 | enable_irq(dev->irq); | |
155 | } | |
6a1ca373 LV |
156 | if (!driver->err_handler || |
157 | !driver->err_handler->slot_reset) | |
77bd7415 LV |
158 | return; |
159 | ||
6a1ca373 | 160 | rc = driver->err_handler->slot_reset(dev); |
5794dbcb LV |
161 | if ((*res == PCI_ERS_RESULT_NONE) || |
162 | (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; | |
6a1ca373 LV |
163 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
164 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; | |
77bd7415 LV |
165 | } |
166 | ||
cb5b5624 LV |
167 | /** |
168 | * eeh_report_resume - tell device to resume normal operations | |
169 | */ | |
170 | ||
77bd7415 LV |
171 | static void eeh_report_resume(struct pci_dev *dev, void *userdata) |
172 | { | |
173 | struct pci_driver *driver = dev->driver; | |
d0e70341 | 174 | struct device_node *dn = pci_device_to_OF_node(dev); |
77bd7415 LV |
175 | |
176 | dev->error_state = pci_channel_io_normal; | |
177 | ||
178 | if (!driver) | |
179 | return; | |
d0e70341 LV |
180 | |
181 | if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) { | |
182 | PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED; | |
183 | enable_irq(dev->irq); | |
184 | } | |
185 | if (!driver->err_handler || | |
186 | !driver->err_handler->resume) | |
77bd7415 LV |
187 | return; |
188 | ||
189 | driver->err_handler->resume(dev); | |
190 | } | |
191 | ||
cb5b5624 LV |
192 | /** |
193 | * eeh_report_failure - tell device driver that device is dead. | |
194 | * | |
195 | * This informs the device driver that the device is permanently | |
196 | * dead, and that no further recovery attempts will be made on it. | |
197 | */ | |
198 | ||
77bd7415 LV |
199 | static void eeh_report_failure(struct pci_dev *dev, void *userdata) |
200 | { | |
201 | struct pci_driver *driver = dev->driver; | |
202 | ||
203 | dev->error_state = pci_channel_io_perm_failure; | |
204 | ||
205 | if (!driver) | |
206 | return; | |
207 | ||
208 | if (irq_in_use (dev->irq)) { | |
209 | struct device_node *dn = pci_device_to_OF_node(dev); | |
210 | PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED; | |
211 | disable_irq_nosync(dev->irq); | |
212 | } | |
213 | if (!driver->err_handler) | |
214 | return; | |
215 | if (!driver->err_handler->error_detected) | |
216 | return; | |
217 | driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); | |
218 | } | |
219 | ||
220 | /* ------------------------------------------------------- */ | |
221 | /** | |
222 | * handle_eeh_events -- reset a PCI device after hard lockup. | |
223 | * | |
224 | * pSeries systems will isolate a PCI slot if the PCI-Host | |
225 | * bridge detects address or data parity errors, DMA's | |
80f7228b | 226 | * occurring to wild addresses (which usually happen due to |
77bd7415 LV |
227 | * bugs in device drivers or in PCI adapter firmware). |
228 | * Slot isolations also occur if #SERR, #PERR or other misc | |
229 | * PCI-related errors are detected. | |
230 | * | |
231 | * Recovery process consists of unplugging the device driver | |
232 | * (which generated hotplug events to userspace), then issuing | |
233 | * a PCI #RST to the device, then reconfiguring the PCI config | |
234 | * space for all bridges & devices under this slot, and then | |
235 | * finally restarting the device drivers (which cause a second | |
236 | * set of hotplug events to go out to userspace). | |
237 | */ | |
238 | ||
239 | /** | |
240 | * eeh_reset_device() -- perform actual reset of a pci slot | |
cb5b5624 | 241 | * @bus: pointer to the pci bus structure corresponding |
77bd7415 LV |
242 | * to the isolated slot. A non-null value will |
243 | * cause all devices under the bus to be removed | |
244 | * and then re-added. | |
cb5b5624 | 245 | * @pe_dn: pointer to a "Partionable Endpoint" device node. |
77bd7415 LV |
246 | * This is the top-level structure on which pci |
247 | * bus resets can be performed. | |
248 | */ | |
249 | ||
b6495c0c | 250 | static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus) |
77bd7415 | 251 | { |
42405456 LV |
252 | int cnt, rc; |
253 | ||
254 | /* pcibios will clear the counter; save the value */ | |
255 | cnt = pe_dn->eeh_freeze_count; | |
256 | ||
77bd7415 LV |
257 | if (bus) |
258 | pcibios_remove_pci_devices(bus); | |
259 | ||
260 | /* Reset the pci controller. (Asserts RST#; resets config space). | |
b6495c0c LV |
261 | * Reconfigure bridges and devices. Don't try to bring the system |
262 | * up if the reset failed for some reason. */ | |
263 | rc = rtas_set_slot_reset(pe_dn); | |
264 | if (rc) | |
265 | return rc; | |
77bd7415 | 266 | |
3914ac7b LV |
267 | /* New-style config addrs might be shared across multiple devices, |
268 | * Walk over all functions on this device */ | |
269 | if (pe_dn->eeh_pe_config_addr) { | |
270 | struct device_node *pe = pe_dn->node; | |
271 | pe = pe->parent->child; | |
272 | while (pe) { | |
273 | struct pci_dn *ppe = PCI_DN(pe); | |
274 | if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) { | |
275 | rtas_configure_bridge(ppe); | |
276 | eeh_restore_bars(ppe); | |
277 | } | |
278 | pe = pe->sibling; | |
279 | } | |
280 | } else { | |
281 | rtas_configure_bridge(pe_dn); | |
282 | eeh_restore_bars(pe_dn); | |
283 | } | |
77bd7415 LV |
284 | |
285 | /* Give the system 5 seconds to finish running the user-space | |
286 | * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, | |
287 | * this is a hack, but if we don't do this, and try to bring | |
288 | * the device up before the scripts have taken it down, | |
289 | * potentially weird things happen. | |
290 | */ | |
291 | if (bus) { | |
292 | ssleep (5); | |
293 | pcibios_add_pci_devices(bus); | |
294 | } | |
42405456 | 295 | pe_dn->eeh_freeze_count = cnt; |
b6495c0c LV |
296 | |
297 | return 0; | |
77bd7415 LV |
298 | } |
299 | ||
300 | /* The longest amount of time to wait for a pci device | |
301 | * to come back on line, in seconds. | |
302 | */ | |
303 | #define MAX_WAIT_FOR_RECOVERY 15 | |
304 | ||
ac325acd | 305 | struct pci_dn * handle_eeh_events (struct eeh_event *event) |
77bd7415 LV |
306 | { |
307 | struct device_node *frozen_dn; | |
308 | struct pci_dn *frozen_pdn; | |
309 | struct pci_bus *frozen_bus; | |
b6495c0c | 310 | int rc = 0; |
18eb3b39 | 311 | enum pci_ers_result result = PCI_ERS_RESULT_NONE; |
0aa8d15b | 312 | const char *location, *pci_str, *drv_str; |
77bd7415 LV |
313 | |
314 | frozen_dn = find_device_pe(event->dn); | |
315 | frozen_bus = pcibios_find_pci_bus(frozen_dn); | |
316 | ||
317 | if (!frozen_dn) { | |
0aa8d15b | 318 | |
954a46e2 | 319 | location = get_property(event->dn, "ibm,loc-code", NULL); |
0aa8d15b LV |
320 | location = location ? location : "unknown"; |
321 | printk(KERN_ERR "EEH: Error: Cannot find partition endpoint " | |
322 | "for location=%s pci addr=%s\n", | |
323 | location, pci_name(event->dev)); | |
ac325acd | 324 | return NULL; |
77bd7415 | 325 | } |
954a46e2 | 326 | location = get_property(frozen_dn, "ibm,loc-code", NULL); |
0aa8d15b | 327 | location = location ? location : "unknown"; |
77bd7415 LV |
328 | |
329 | /* There are two different styles for coming up with the PE. | |
330 | * In the old style, it was the highest EEH-capable device | |
331 | * which was always an EADS pci bridge. In the new style, | |
332 | * there might not be any EADS bridges, and even when there are, | |
333 | * the firmware marks them as "EEH incapable". So another | |
334 | * two-step is needed to find the pci bus.. */ | |
335 | if (!frozen_bus) | |
336 | frozen_bus = pcibios_find_pci_bus (frozen_dn->parent); | |
337 | ||
338 | if (!frozen_bus) { | |
0aa8d15b LV |
339 | printk(KERN_ERR "EEH: Cannot find PCI bus " |
340 | "for location=%s dn=%s\n", | |
341 | location, frozen_dn->full_name); | |
ac325acd | 342 | return NULL; |
77bd7415 LV |
343 | } |
344 | ||
345 | #if 0 | |
346 | /* We may get "permanent failure" messages on empty slots. | |
347 | * These are false alarms. Empty slots have no child dn. */ | |
348 | if ((event->state == pci_channel_io_perm_failure) && (frozen_device == NULL)) | |
349 | return; | |
350 | #endif | |
351 | ||
352 | frozen_pdn = PCI_DN(frozen_dn); | |
353 | frozen_pdn->eeh_freeze_count++; | |
b4f382a3 | 354 | |
a219be2c LV |
355 | if (frozen_pdn->pcidev) { |
356 | pci_str = pci_name (frozen_pdn->pcidev); | |
357 | drv_str = pcid_name (frozen_pdn->pcidev); | |
358 | } else { | |
b4f382a3 LV |
359 | pci_str = pci_name (event->dev); |
360 | drv_str = pcid_name (event->dev); | |
361 | } | |
77bd7415 LV |
362 | |
363 | if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES) | |
8df83028 | 364 | goto excess_failures; |
77bd7415 LV |
365 | |
366 | /* If the reset state is a '5' and the time to reset is 0 (infinity) | |
367 | * or is more then 15 seconds, then mark this as a permanent failure. | |
368 | */ | |
369 | if ((event->state == pci_channel_io_perm_failure) && | |
370 | ((event->time_unavail <= 0) || | |
e0f90b64 LV |
371 | (event->time_unavail > MAX_WAIT_FOR_RECOVERY*1000))) { |
372 | printk(KERN_WARNING "EEH: Permanent failure\n"); | |
b6495c0c | 373 | goto hard_fail; |
e0f90b64 | 374 | } |
77bd7415 LV |
375 | |
376 | eeh_slot_error_detail(frozen_pdn, 1 /* Temporary Error */); | |
377 | printk(KERN_WARNING | |
0aa8d15b LV |
378 | "EEH: This PCI device has failed %d times since last reboot: " |
379 | "location=%s driver=%s pci addr=%s\n", | |
380 | frozen_pdn->eeh_freeze_count, location, drv_str, pci_str); | |
77bd7415 LV |
381 | |
382 | /* Walk the various device drivers attached to this slot through | |
383 | * a reset sequence, giving each an opportunity to do what it needs | |
384 | * to accomplish the reset. Each child gets a report of the | |
385 | * status ... if any child can't handle the reset, then the entire | |
386 | * slot is dlpar removed and added. | |
387 | */ | |
77bd7415 LV |
388 | pci_walk_bus(frozen_bus, eeh_report_error, &result); |
389 | ||
390 | /* If all device drivers were EEH-unaware, then shut | |
391 | * down all of the device drivers, and hope they | |
392 | * go down willingly, without panicing the system. | |
393 | */ | |
18eb3b39 | 394 | if (result == PCI_ERS_RESULT_NONE) { |
b6495c0c | 395 | rc = eeh_reset_device(frozen_pdn, frozen_bus); |
e0f90b64 LV |
396 | if (rc) { |
397 | printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc); | |
b6495c0c | 398 | goto hard_fail; |
e0f90b64 | 399 | } |
77bd7415 LV |
400 | } |
401 | ||
6a1ca373 LV |
402 | /* If all devices reported they can proceed, then re-enable MMIO */ |
403 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { | |
404 | rc = rtas_pci_enable(frozen_pdn, EEH_THAW_MMIO); | |
405 | ||
406 | if (rc) { | |
407 | result = PCI_ERS_RESULT_NEED_RESET; | |
408 | } else { | |
409 | result = PCI_ERS_RESULT_NONE; | |
410 | pci_walk_bus(frozen_bus, eeh_report_mmio_enabled, &result); | |
411 | } | |
77bd7415 LV |
412 | } |
413 | ||
6a1ca373 | 414 | /* If all devices reported they can proceed, then re-enable DMA */ |
18eb3b39 | 415 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { |
6a1ca373 LV |
416 | rc = rtas_pci_enable(frozen_pdn, EEH_THAW_DMA); |
417 | ||
418 | if (rc) | |
419 | result = PCI_ERS_RESULT_NEED_RESET; | |
d0e70341 LV |
420 | else |
421 | result = PCI_ERS_RESULT_RECOVERED; | |
6a1ca373 LV |
422 | } |
423 | ||
424 | /* If any device has a hard failure, then shut off everything. */ | |
e0f90b64 LV |
425 | if (result == PCI_ERS_RESULT_DISCONNECT) { |
426 | printk(KERN_WARNING "EEH: Device driver gave up\n"); | |
6a1ca373 | 427 | goto hard_fail; |
e0f90b64 | 428 | } |
6a1ca373 LV |
429 | |
430 | /* If any device called out for a reset, then reset the slot */ | |
431 | if (result == PCI_ERS_RESULT_NEED_RESET) { | |
b6495c0c | 432 | rc = eeh_reset_device(frozen_pdn, NULL); |
e0f90b64 LV |
433 | if (rc) { |
434 | printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc); | |
b6495c0c | 435 | goto hard_fail; |
e0f90b64 | 436 | } |
6a1ca373 LV |
437 | result = PCI_ERS_RESULT_NONE; |
438 | pci_walk_bus(frozen_bus, eeh_report_reset, &result); | |
77bd7415 LV |
439 | } |
440 | ||
6a1ca373 | 441 | /* All devices should claim they have recovered by now. */ |
90fdd613 LV |
442 | if ((result != PCI_ERS_RESULT_RECOVERED) && |
443 | (result != PCI_ERS_RESULT_NONE)) { | |
e0f90b64 | 444 | printk(KERN_WARNING "EEH: Not recovered\n"); |
6a1ca373 | 445 | goto hard_fail; |
e0f90b64 | 446 | } |
6a1ca373 | 447 | |
77bd7415 | 448 | /* Tell all device drivers that they can resume operations */ |
d04e4e11 | 449 | pci_walk_bus(frozen_bus, eeh_report_resume, NULL); |
b6495c0c | 450 | |
ac325acd | 451 | return frozen_pdn; |
b6495c0c | 452 | |
8df83028 | 453 | excess_failures: |
b6495c0c LV |
454 | /* |
455 | * About 90% of all real-life EEH failures in the field | |
456 | * are due to poorly seated PCI cards. Only 10% or so are | |
457 | * due to actual, failed cards. | |
458 | */ | |
459 | printk(KERN_ERR | |
0aa8d15b | 460 | "EEH: PCI device at location=%s driver=%s pci addr=%s \n" |
a885902d LV |
461 | "has failed %d times in the last hour " |
462 | "and has been permanently disabled. \n" | |
0aa8d15b LV |
463 | "Please try reseating this device or replacing it.\n", |
464 | location, drv_str, pci_str, frozen_pdn->eeh_freeze_count); | |
8df83028 LV |
465 | goto perm_error; |
466 | ||
467 | hard_fail: | |
468 | printk(KERN_ERR | |
0aa8d15b LV |
469 | "EEH: Unable to recover from failure of PCI device " |
470 | "at location=%s driver=%s pci addr=%s \n" | |
8df83028 | 471 | "Please try reseating this device or replacing it.\n", |
0aa8d15b | 472 | location, drv_str, pci_str); |
b6495c0c | 473 | |
8df83028 | 474 | perm_error: |
b6495c0c LV |
475 | eeh_slot_error_detail(frozen_pdn, 2 /* Permanent Error */); |
476 | ||
477 | /* Notify all devices that they're about to go down. */ | |
d04e4e11 | 478 | pci_walk_bus(frozen_bus, eeh_report_failure, NULL); |
b6495c0c LV |
479 | |
480 | /* Shut down the device drivers for good. */ | |
481 | pcibios_remove_pci_devices(frozen_bus); | |
ac325acd LV |
482 | |
483 | return NULL; | |
77bd7415 LV |
484 | } |
485 | ||
486 | /* ---------- end of file ---------- */ |