f07d849cfc84837e47ee560a19746387b7c424f7
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / powerpc / platforms / pseries / eeh_driver.c
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>
26 #include <linux/interrupt.h>
27 #include <linux/irq.h>
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 {
39 if (pdev && pdev->dev.driver)
40 return pdev->dev.driver->name;
41 return "";
42 }
43
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 /* ------------------------------------------------------- */
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".
86 */
87
88 static void eeh_report_error(struct pci_dev *dev, void *userdata)
89 {
90 enum pci_ers_result rc, *res = userdata;
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 }
103 if (!driver->err_handler ||
104 !driver->err_handler->error_detected)
105 return;
106
107 rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
108 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
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;
135 if (*res == PCI_ERS_RESULT_DISCONNECT &&
136 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
137 }
138
139 /**
140 * eeh_report_reset - tell device that slot has been reset
141 */
142
143 static void eeh_report_reset(struct pci_dev *dev, void *userdata)
144 {
145 enum pci_ers_result rc, *res = userdata;
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 }
156 if (!driver->err_handler ||
157 !driver->err_handler->slot_reset)
158 return;
159
160 rc = driver->err_handler->slot_reset(dev);
161 if ((*res == PCI_ERS_RESULT_NONE) ||
162 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
163 if (*res == PCI_ERS_RESULT_DISCONNECT &&
164 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
165 }
166
167 /**
168 * eeh_report_resume - tell device to resume normal operations
169 */
170
171 static void eeh_report_resume(struct pci_dev *dev, void *userdata)
172 {
173 struct pci_driver *driver = dev->driver;
174 struct device_node *dn = pci_device_to_OF_node(dev);
175
176 dev->error_state = pci_channel_io_normal;
177
178 if (!driver)
179 return;
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)
187 return;
188
189 driver->err_handler->resume(dev);
190 }
191
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
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
226 * occurring to wild addresses (which usually happen due to
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
241 * @bus: pointer to the pci bus structure corresponding
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.
245 * @pe_dn: pointer to a "Partionable Endpoint" device node.
246 * This is the top-level structure on which pci
247 * bus resets can be performed.
248 */
249
250 static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
251 {
252 struct device_node *dn;
253 int cnt, rc;
254
255 /* pcibios will clear the counter; save the value */
256 cnt = pe_dn->eeh_freeze_count;
257
258 if (bus)
259 pcibios_remove_pci_devices(bus);
260
261 /* Reset the pci controller. (Asserts RST#; resets config space).
262 * Reconfigure bridges and devices. Don't try to bring the system
263 * up if the reset failed for some reason. */
264 rc = rtas_set_slot_reset(pe_dn);
265 if (rc)
266 return rc;
267
268 /* Walk over all functions on this device. */
269 dn = pe_dn->node;
270 if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
271 dn = dn->parent->child;
272
273 while (dn) {
274 struct pci_dn *ppe = PCI_DN(dn);
275 /* On Power4, always true because eeh_pe_config_addr=0 */
276 if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) {
277 rtas_configure_bridge(ppe);
278 eeh_restore_bars(ppe);
279 }
280 dn = dn->sibling;
281 }
282
283 /* Give the system 5 seconds to finish running the user-space
284 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
285 * this is a hack, but if we don't do this, and try to bring
286 * the device up before the scripts have taken it down,
287 * potentially weird things happen.
288 */
289 if (bus) {
290 ssleep (5);
291 pcibios_add_pci_devices(bus);
292 }
293 pe_dn->eeh_freeze_count = cnt;
294
295 return 0;
296 }
297
298 /* The longest amount of time to wait for a pci device
299 * to come back on line, in seconds.
300 */
301 #define MAX_WAIT_FOR_RECOVERY 150
302
303 struct pci_dn * handle_eeh_events (struct eeh_event *event)
304 {
305 struct device_node *frozen_dn;
306 struct pci_dn *frozen_pdn;
307 struct pci_bus *frozen_bus;
308 int rc = 0;
309 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
310 const char *location, *pci_str, *drv_str;
311
312 frozen_dn = find_device_pe(event->dn);
313 frozen_bus = pcibios_find_pci_bus(frozen_dn);
314
315 if (!frozen_dn) {
316
317 location = of_get_property(event->dn, "ibm,loc-code", NULL);
318 location = location ? location : "unknown";
319 printk(KERN_ERR "EEH: Error: Cannot find partition endpoint "
320 "for location=%s pci addr=%s\n",
321 location, pci_name(event->dev));
322 return NULL;
323 }
324 location = of_get_property(frozen_dn, "ibm,loc-code", NULL);
325 location = location ? location : "unknown";
326
327 /* There are two different styles for coming up with the PE.
328 * In the old style, it was the highest EEH-capable device
329 * which was always an EADS pci bridge. In the new style,
330 * there might not be any EADS bridges, and even when there are,
331 * the firmware marks them as "EEH incapable". So another
332 * two-step is needed to find the pci bus.. */
333 if (!frozen_bus)
334 frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
335
336 if (!frozen_bus) {
337 printk(KERN_ERR "EEH: Cannot find PCI bus "
338 "for location=%s dn=%s\n",
339 location, frozen_dn->full_name);
340 return NULL;
341 }
342
343 frozen_pdn = PCI_DN(frozen_dn);
344 frozen_pdn->eeh_freeze_count++;
345
346 if (frozen_pdn->pcidev) {
347 pci_str = pci_name (frozen_pdn->pcidev);
348 drv_str = pcid_name (frozen_pdn->pcidev);
349 } else {
350 pci_str = pci_name (event->dev);
351 drv_str = pcid_name (event->dev);
352 }
353
354 if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
355 goto excess_failures;
356
357 /* Get the current PCI slot state. */
358 rc = eeh_wait_for_slot_status (frozen_pdn, MAX_WAIT_FOR_RECOVERY*1000);
359 if (rc < 0) {
360 printk(KERN_WARNING "EEH: Permanent failure\n");
361 goto hard_fail;
362 }
363
364 printk(KERN_WARNING
365 "EEH: This PCI device has failed %d times in the last hour:\n",
366 frozen_pdn->eeh_freeze_count);
367 printk(KERN_WARNING
368 "EEH: location=%s driver=%s pci addr=%s\n",
369 location, drv_str, pci_str);
370
371 /* Walk the various device drivers attached to this slot through
372 * a reset sequence, giving each an opportunity to do what it needs
373 * to accomplish the reset. Each child gets a report of the
374 * status ... if any child can't handle the reset, then the entire
375 * slot is dlpar removed and added.
376 */
377 pci_walk_bus(frozen_bus, eeh_report_error, &result);
378
379 /* Since rtas may enable MMIO when posting the error log,
380 * don't post the error log until after all dev drivers
381 * have been informed. */
382 eeh_slot_error_detail(frozen_pdn, 1 /* Temporary Error */);
383
384 /* If all device drivers were EEH-unaware, then shut
385 * down all of the device drivers, and hope they
386 * go down willingly, without panicing the system.
387 */
388 if (result == PCI_ERS_RESULT_NONE) {
389 rc = eeh_reset_device(frozen_pdn, frozen_bus);
390 if (rc) {
391 printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc);
392 goto hard_fail;
393 }
394 }
395
396 /* If all devices reported they can proceed, then re-enable MMIO */
397 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
398 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_MMIO);
399
400 if (rc < 0)
401 goto hard_fail;
402 if (rc) {
403 result = PCI_ERS_RESULT_NEED_RESET;
404 } else {
405 result = PCI_ERS_RESULT_NONE;
406 pci_walk_bus(frozen_bus, eeh_report_mmio_enabled, &result);
407 }
408 }
409
410 /* If all devices reported they can proceed, then re-enable DMA */
411 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
412 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_DMA);
413
414 if (rc < 0)
415 goto hard_fail;
416 if (rc)
417 result = PCI_ERS_RESULT_NEED_RESET;
418 else
419 result = PCI_ERS_RESULT_RECOVERED;
420 }
421
422 /* If any device has a hard failure, then shut off everything. */
423 if (result == PCI_ERS_RESULT_DISCONNECT) {
424 printk(KERN_WARNING "EEH: Device driver gave up\n");
425 goto hard_fail;
426 }
427
428 /* If any device called out for a reset, then reset the slot */
429 if (result == PCI_ERS_RESULT_NEED_RESET) {
430 rc = eeh_reset_device(frozen_pdn, NULL);
431 if (rc) {
432 printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc);
433 goto hard_fail;
434 }
435 result = PCI_ERS_RESULT_NONE;
436 pci_walk_bus(frozen_bus, eeh_report_reset, &result);
437 }
438
439 /* All devices should claim they have recovered by now. */
440 if ((result != PCI_ERS_RESULT_RECOVERED) &&
441 (result != PCI_ERS_RESULT_NONE)) {
442 printk(KERN_WARNING "EEH: Not recovered\n");
443 goto hard_fail;
444 }
445
446 /* Tell all device drivers that they can resume operations */
447 pci_walk_bus(frozen_bus, eeh_report_resume, NULL);
448
449 return frozen_pdn;
450
451 excess_failures:
452 /*
453 * About 90% of all real-life EEH failures in the field
454 * are due to poorly seated PCI cards. Only 10% or so are
455 * due to actual, failed cards.
456 */
457 printk(KERN_ERR
458 "EEH: PCI device at location=%s driver=%s pci addr=%s \n"
459 "has failed %d times in the last hour "
460 "and has been permanently disabled. \n"
461 "Please try reseating this device or replacing it.\n",
462 location, drv_str, pci_str, frozen_pdn->eeh_freeze_count);
463 goto perm_error;
464
465 hard_fail:
466 printk(KERN_ERR
467 "EEH: Unable to recover from failure of PCI device "
468 "at location=%s driver=%s pci addr=%s \n"
469 "Please try reseating this device or replacing it.\n",
470 location, drv_str, pci_str);
471
472 perm_error:
473 eeh_slot_error_detail(frozen_pdn, 2 /* Permanent Error */);
474
475 /* Notify all devices that they're about to go down. */
476 pci_walk_bus(frozen_bus, eeh_report_failure, NULL);
477
478 /* Shut down the device drivers for good. */
479 pcibios_remove_pci_devices(frozen_bus);
480
481 return NULL;
482 }
483
484 /* ---------- end of file ---------- */