]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/pci/pcie/aer/aerdrv_core.c
PCI: Fix whitespace, capitalization, and spelling errors
[mirror_ubuntu-bionic-kernel.git] / drivers / pci / pcie / aer / aerdrv_core.c
CommitLineData
6c2b374d
ZY
1/*
2 * drivers/pci/pcie/aer/aerdrv_core.c
3 *
4 * This file is subject to the terms and conditions of the GNU General Public
5 * License. See the file "COPYING" in the main directory of this archive
6 * for more details.
7 *
8 * This file implements the core part of PCI-Express AER. When an pci-express
9 * error is delivered, an error message will be collected and printed to
10 * console, then, an error recovery procedure will be executed by following
11 * the pci error recovery rules.
12 *
13 * Copyright (C) 2006 Intel Corp.
14 * Tom Long Nguyen (tom.l.nguyen@intel.com)
15 * Zhang Yanmin (yanmin.zhang@intel.com)
16 *
17 */
18
19#include <linux/module.h>
20#include <linux/pci.h>
21#include <linux/kernel.h>
22#include <linux/errno.h>
23#include <linux/pm.h>
24#include <linux/suspend.h>
6c2b374d 25#include <linux/delay.h>
5a0e3ad6 26#include <linux/slab.h>
0918472c 27#include <linux/kfifo.h>
6c2b374d
ZY
28#include "aerdrv.h"
29
90ab5ee9
RR
30static bool forceload;
31static bool nosourceid;
6c2b374d 32module_param(forceload, bool, 0);
28eb27cf 33module_param(nosourceid, bool, 0);
6c2b374d 34
43bd4ee8
JL
35#define PCI_EXP_AER_FLAGS (PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \
36 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE)
37
6c2b374d
ZY
38int pci_enable_pcie_error_reporting(struct pci_dev *dev)
39{
affb72c3 40 if (pcie_aer_get_firmware_first(dev))
05843961
MD
41 return -EIO;
42
43bd4ee8 43 if (!pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR))
0927678f
JB
44 return -EIO;
45
43bd4ee8 46 return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
6c2b374d 47}
c9a91883 48EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
6c2b374d
ZY
49
50int pci_disable_pcie_error_reporting(struct pci_dev *dev)
51{
affb72c3 52 if (pcie_aer_get_firmware_first(dev))
05843961
MD
53 return -EIO;
54
43bd4ee8
JL
55 return pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
56 PCI_EXP_AER_FLAGS);
6c2b374d 57}
c9a91883 58EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
6c2b374d
ZY
59
60int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
61{
62 int pos;
6cdfd995 63 u32 status;
6c2b374d 64
0927678f 65 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
6c2b374d
ZY
66 if (!pos)
67 return -EIO;
68
69 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
6cdfd995
AP
70 if (status)
71 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
6c2b374d
ZY
72
73 return 0;
74}
c9a91883 75EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
6c2b374d 76
4a0c096e
HS
77/**
78 * add_error_device - list device to be handled
79 * @e_info: pointer to error info
80 * @dev: pointer to pci_dev to be added
81 */
3d5505c5
ZY
82static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
83{
84 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
85 e_info->dev[e_info->error_dev_num] = dev;
86 e_info->error_dev_num++;
4a0c096e 87 return 0;
c9a91883 88 }
4a0c096e 89 return -ENOSPC;
3d5505c5
ZY
90}
91
c887275e
HS
92/**
93 * is_error_source - check whether the device is source of reported error
94 * @dev: pointer to pci_dev to be checked
95 * @e_info: pointer to reported error info
96 */
97static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info)
28eb27cf
ZY
98{
99 int pos;
c887275e 100 u32 status, mask;
28eb27cf 101 u16 reg16;
28eb27cf
ZY
102
103 /*
104 * When bus id is equal to 0, it might be a bad id
105 * reported by root port.
106 */
fff0ee36 107 if (!nosourceid && (PCI_BUS_NUM(e_info->id) != 0)) {
bd17d474
HS
108 /* Device ID match? */
109 if (e_info->id == ((dev->bus->number << 8) | dev->devfn))
c887275e 110 return true;
3d5505c5 111
c887275e 112 /* Continue id comparing if there is no multiple error */
273024de 113 if (!e_info->multi_error_valid)
c887275e 114 return false;
28eb27cf
ZY
115 }
116
117 /*
3d5505c5
ZY
118 * When either
119 * 1) nosourceid==y;
120 * 2) bus id is equal to 0. Some ports might lose the bus
121 * id of error source id;
122 * 3) There are multiple errors and prior id comparing fails;
c887275e 123 * We check AER status registers to find possible reporter.
28eb27cf
ZY
124 */
125 if (atomic_read(&dev->enable_cnt) == 0)
c887275e 126 return false;
c887275e 127
28eb27cf 128 /* Check if AER is enabled */
43bd4ee8
JL
129 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &reg16);
130 if (!(reg16 & PCI_EXP_AER_FLAGS))
c887275e 131 return false;
43bd4ee8 132
28eb27cf
ZY
133 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
134 if (!pos)
c887275e 135 return false;
28eb27cf 136
c887275e 137 /* Check if error is recorded */
28eb27cf 138 if (e_info->severity == AER_CORRECTABLE) {
0d90c3ac
HS
139 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
140 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, &mask);
28eb27cf 141 } else {
0d90c3ac
HS
142 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
143 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
6c2b374d 144 }
c887275e
HS
145 if (status & ~mask)
146 return true;
6c2b374d 147
c887275e
HS
148 return false;
149}
3d5505c5 150
c887275e
HS
151static int find_device_iter(struct pci_dev *dev, void *data)
152{
153 struct aer_err_info *e_info = (struct aer_err_info *)data;
154
155 if (is_error_source(dev, e_info)) {
4a0c096e
HS
156 /* List this device */
157 if (add_error_device(e_info, dev)) {
158 /* We cannot handle more... Stop iteration */
159 /* TODO: Should print error message here? */
160 return 1;
161 }
c887275e
HS
162
163 /* If there is only a single error, stop iteration */
164 if (!e_info->multi_error_valid)
165 return 1;
166 }
167 return 0;
6c2b374d
ZY
168}
169
170/**
171 * find_source_device - search through device hierarchy for source device
d885c6b7 172 * @parent: pointer to Root Port pci_dev data structure
98ca3964 173 * @e_info: including detailed error information such like id
6c2b374d 174 *
98ca3964
HS
175 * Return true if found.
176 *
177 * Invoked by DPC when error is detected at the Root Port.
7c4ec94f
HS
178 * Caller of this function must set id, severity, and multi_error_valid of
179 * struct aer_err_info pointed by @e_info properly. This function must fill
180 * e_info->error_dev_num and e_info->dev[], based on the given information.
d885c6b7 181 */
98ca3964 182static bool find_source_device(struct pci_dev *parent,
28eb27cf 183 struct aer_err_info *e_info)
6c2b374d
ZY
184{
185 struct pci_dev *dev = parent;
28eb27cf 186 int result;
6c2b374d 187
7c4ec94f
HS
188 /* Must reset in this function */
189 e_info->error_dev_num = 0;
190
6c2b374d 191 /* Is Root Port an agent that sends error message? */
28eb27cf
ZY
192 result = find_device_iter(dev, e_info);
193 if (result)
98ca3964 194 return true;
6c2b374d 195
28eb27cf 196 pci_walk_bus(parent->subordinate, find_device_iter, e_info);
98ca3964
HS
197
198 if (!e_info->error_dev_num) {
199 dev_printk(KERN_DEBUG, &parent->dev,
200 "can't find device of ID%04x\n",
201 e_info->id);
202 return false;
203 }
204 return true;
6c2b374d
ZY
205}
206
70298c6e 207static int report_error_detected(struct pci_dev *dev, void *data)
6c2b374d
ZY
208{
209 pci_ers_result_t vote;
49453028 210 const struct pci_error_handlers *err_handler;
6c2b374d
ZY
211 struct aer_broadcast_data *result_data;
212 result_data = (struct aer_broadcast_data *) data;
213
90b5c1d7 214 device_lock(&dev->dev);
6c2b374d
ZY
215 dev->error_state = result_data->state;
216
217 if (!dev->driver ||
218 !dev->driver->err_handler ||
219 !dev->driver->err_handler->error_detected) {
220 if (result_data->state == pci_channel_io_frozen &&
221 !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
222 /*
223 * In case of fatal recovery, if one of down-
224 * stream device has no driver. We might be
225 * unable to recover because a later insmod
226 * of a driver for this device is unaware of
227 * its hw state.
228 */
531f254e
BH
229 dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
230 dev->driver ?
231 "no AER-aware driver" : "no driver");
6c2b374d 232 }
918b4053
VMP
233
234 /*
235 * If there's any device in the subtree that does not
236 * have an error_detected callback, returning
237 * PCI_ERS_RESULT_NO_AER_DRIVER prevents calling of
238 * the subsequent mmio_enabled/slot_reset/resume
239 * callbacks of "any" device in the subtree. All the
240 * devices in the subtree are left in the error state
241 * without recovery.
242 */
243
244 if (!(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
245 vote = PCI_ERS_RESULT_NO_AER_DRIVER;
246 else
247 vote = PCI_ERS_RESULT_NONE;
248 } else {
249 err_handler = dev->driver->err_handler;
250 vote = err_handler->error_detected(dev, result_data->state);
6c2b374d
ZY
251 }
252
6c2b374d 253 result_data->result = merge_result(result_data->result, vote);
90b5c1d7 254 device_unlock(&dev->dev);
70298c6e 255 return 0;
6c2b374d
ZY
256}
257
70298c6e 258static int report_mmio_enabled(struct pci_dev *dev, void *data)
6c2b374d
ZY
259{
260 pci_ers_result_t vote;
49453028 261 const struct pci_error_handlers *err_handler;
6c2b374d
ZY
262 struct aer_broadcast_data *result_data;
263 result_data = (struct aer_broadcast_data *) data;
264
90b5c1d7 265 device_lock(&dev->dev);
6c2b374d
ZY
266 if (!dev->driver ||
267 !dev->driver->err_handler ||
268 !dev->driver->err_handler->mmio_enabled)
90b5c1d7 269 goto out;
6c2b374d
ZY
270
271 err_handler = dev->driver->err_handler;
272 vote = err_handler->mmio_enabled(dev);
273 result_data->result = merge_result(result_data->result, vote);
90b5c1d7
HY
274out:
275 device_unlock(&dev->dev);
70298c6e 276 return 0;
6c2b374d
ZY
277}
278
70298c6e 279static int report_slot_reset(struct pci_dev *dev, void *data)
6c2b374d
ZY
280{
281 pci_ers_result_t vote;
49453028 282 const struct pci_error_handlers *err_handler;
6c2b374d
ZY
283 struct aer_broadcast_data *result_data;
284 result_data = (struct aer_broadcast_data *) data;
285
90b5c1d7 286 device_lock(&dev->dev);
6c2b374d
ZY
287 if (!dev->driver ||
288 !dev->driver->err_handler ||
289 !dev->driver->err_handler->slot_reset)
90b5c1d7 290 goto out;
6c2b374d
ZY
291
292 err_handler = dev->driver->err_handler;
293 vote = err_handler->slot_reset(dev);
294 result_data->result = merge_result(result_data->result, vote);
90b5c1d7
HY
295out:
296 device_unlock(&dev->dev);
70298c6e 297 return 0;
6c2b374d
ZY
298}
299
70298c6e 300static int report_resume(struct pci_dev *dev, void *data)
6c2b374d 301{
49453028 302 const struct pci_error_handlers *err_handler;
6c2b374d 303
90b5c1d7 304 device_lock(&dev->dev);
6c2b374d
ZY
305 dev->error_state = pci_channel_io_normal;
306
307 if (!dev->driver ||
308 !dev->driver->err_handler ||
b0b801dd 309 !dev->driver->err_handler->resume)
90b5c1d7 310 goto out;
6c2b374d
ZY
311
312 err_handler = dev->driver->err_handler;
313 err_handler->resume(dev);
90b5c1d7
HY
314out:
315 device_unlock(&dev->dev);
70298c6e 316 return 0;
6c2b374d
ZY
317}
318
319/**
320 * broadcast_error_message - handle message broadcast to downstream drivers
d885c6b7 321 * @dev: pointer to from where in a hierarchy message is broadcasted down
6c2b374d 322 * @state: error state
d885c6b7
RD
323 * @error_mesg: message to print
324 * @cb: callback to be broadcasted
6c2b374d
ZY
325 *
326 * Invoked during error recovery process. Once being invoked, the content
327 * of error severity will be broadcasted to all downstream drivers in a
328 * hierarchy in question.
d885c6b7 329 */
6c2b374d
ZY
330static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
331 enum pci_channel_state state,
332 char *error_mesg,
70298c6e 333 int (*cb)(struct pci_dev *, void *))
6c2b374d
ZY
334{
335 struct aer_broadcast_data result_data;
336
531f254e 337 dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
6c2b374d
ZY
338 result_data.state = state;
339 if (cb == report_error_detected)
340 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
341 else
342 result_data.result = PCI_ERS_RESULT_RECOVERED;
343
344 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
345 /*
346 * If the error is reported by a bridge, we think this error
347 * is related to the downstream link of the bridge, so we
348 * do error recovery on all subordinates of the bridge instead
349 * of the bridge and clear the error status of the bridge.
350 */
351 if (cb == report_error_detected)
352 dev->error_state = state;
353 pci_walk_bus(dev->subordinate, cb, &result_data);
354 if (cb == report_resume) {
355 pci_cleanup_aer_uncorrect_error_status(dev);
356 dev->error_state = pci_channel_io_normal;
357 }
c9a91883 358 } else {
6c2b374d
ZY
359 /*
360 * If the error is reported by an end point, we think this
361 * error is related to the upstream link of the end point.
362 */
363 pci_walk_bus(dev->bus, cb, &result_data);
364 }
365
366 return result_data.result;
367}
368
89713422 369/**
081d0fe0
BD
370 * default_reset_link - default reset function
371 * @dev: pointer to pci_dev data structure
89713422 372 *
081d0fe0
BD
373 * Invoked when performing link reset on a Downstream Port or a
374 * Root Port with no aer driver.
89713422 375 */
081d0fe0 376static pci_ers_result_t default_reset_link(struct pci_dev *dev)
89713422 377{
1b95ce8f 378 pci_reset_bridge_secondary_bus(dev);
081d0fe0 379 dev_printk(KERN_DEBUG, &dev->dev, "downstream link has been reset\n");
89713422
HS
380 return PCI_ERS_RESULT_RECOVERED;
381}
382
6c2b374d
ZY
383static int find_aer_service_iter(struct device *device, void *data)
384{
517cae38 385 struct pcie_port_service_driver *service_driver, **drv;
6c2b374d 386
517cae38 387 drv = (struct pcie_port_service_driver **) data;
6c2b374d 388
4f7ccf6a
HS
389 if (device->bus == &pcie_port_bus_type && device->driver) {
390 service_driver = to_service_driver(device->driver);
391 if (service_driver->service == PCIE_PORT_SERVICE_AER) {
517cae38 392 *drv = service_driver;
4f7ccf6a 393 return 1;
6c2b374d
ZY
394 }
395 }
396
397 return 0;
398}
399
517cae38 400static struct pcie_port_service_driver *find_aer_service(struct pci_dev *dev)
6c2b374d 401{
517cae38
HS
402 struct pcie_port_service_driver *drv = NULL;
403
404 device_for_each_child(&dev->dev, &drv, find_aer_service_iter);
405
406 return drv;
6c2b374d
ZY
407}
408
0918472c 409static pci_ers_result_t reset_link(struct pci_dev *dev)
6c2b374d
ZY
410{
411 struct pci_dev *udev;
412 pci_ers_result_t status;
517cae38 413 struct pcie_port_service_driver *driver;
6c2b374d 414
89713422
HS
415 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
416 /* Reset this port for all subordinates */
6c2b374d 417 udev = dev;
89713422
HS
418 } else {
419 /* Reset the upstream component (likely downstream port) */
c9a91883 420 udev = dev->bus->self;
89713422 421 }
6c2b374d 422
517cae38
HS
423 /* Use the aer driver of the component firstly */
424 driver = find_aer_service(udev);
6c2b374d 425
89713422
HS
426 if (driver && driver->reset_link) {
427 status = driver->reset_link(udev);
081d0fe0
BD
428 } else if (pci_pcie_type(udev) == PCI_EXP_TYPE_DOWNSTREAM ||
429 pci_pcie_type(udev) == PCI_EXP_TYPE_ROOT_PORT) {
430 status = default_reset_link(udev);
89713422
HS
431 } else {
432 dev_printk(KERN_DEBUG, &dev->dev,
433 "no link-reset support at upstream device %s\n",
434 pci_name(udev));
435 return PCI_ERS_RESULT_DISCONNECT;
6c2b374d
ZY
436 }
437
6c2b374d 438 if (status != PCI_ERS_RESULT_RECOVERED) {
4f7ccf6a
HS
439 dev_printk(KERN_DEBUG, &dev->dev,
440 "link reset at upstream device %s failed\n",
441 pci_name(udev));
6c2b374d
ZY
442 return PCI_ERS_RESULT_DISCONNECT;
443 }
444
445 return status;
446}
447
448/**
449 * do_recovery - handle nonfatal/fatal error recovery process
6c2b374d
ZY
450 * @dev: pointer to a pci_dev data structure of agent detecting an error
451 * @severity: error severity type
452 *
453 * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
454 * error detected message to all downstream drivers within a hierarchy in
455 * question and return the returned code.
d885c6b7 456 */
0918472c 457static void do_recovery(struct pci_dev *dev, int severity)
6c2b374d
ZY
458{
459 pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
460 enum pci_channel_state state;
461
462 if (severity == AER_FATAL)
463 state = pci_channel_io_frozen;
464 else
465 state = pci_channel_io_normal;
466
467 status = broadcast_error_message(dev,
468 state,
469 "error_detected",
470 report_error_detected);
471
472 if (severity == AER_FATAL) {
0918472c 473 result = reset_link(dev);
17e21854
HS
474 if (result != PCI_ERS_RESULT_RECOVERED)
475 goto failed;
6c2b374d
ZY
476 }
477
478 if (status == PCI_ERS_RESULT_CAN_RECOVER)
479 status = broadcast_error_message(dev,
480 state,
481 "mmio_enabled",
482 report_mmio_enabled);
483
484 if (status == PCI_ERS_RESULT_NEED_RESET) {
485 /*
486 * TODO: Should call platform-specific
487 * functions to reset slot before calling
488 * drivers' slot_reset callbacks?
489 */
490 status = broadcast_error_message(dev,
491 state,
492 "slot_reset",
493 report_slot_reset);
494 }
495
17e21854
HS
496 if (status != PCI_ERS_RESULT_RECOVERED)
497 goto failed;
498
499 broadcast_error_message(dev,
6c2b374d
ZY
500 state,
501 "resume",
502 report_resume);
503
be5ac3d3 504 dev_info(&dev->dev, "AER: Device recovery successful\n");
17e21854
HS
505 return;
506
507failed:
508 /* TODO: Should kernel panic here? */
be5ac3d3 509 dev_info(&dev->dev, "AER: Device recovery failed\n");
6c2b374d
ZY
510}
511
512/**
513 * handle_error_source - handle logging error into an event log
514 * @aerdev: pointer to pcie_device data structure of the root port
515 * @dev: pointer to pci_dev data structure of error source device
516 * @info: comprehensive error information
517 *
518 * Invoked when an error being detected by Root Port.
d885c6b7 519 */
c9a91883 520static void handle_error_source(struct pcie_device *aerdev,
6c2b374d 521 struct pci_dev *dev,
28eb27cf 522 struct aer_err_info *info)
6c2b374d 523{
6c2b374d
ZY
524 int pos;
525
28eb27cf 526 if (info->severity == AER_CORRECTABLE) {
6c2b374d 527 /*
f7625980 528 * Correctable error does not need software intervention.
6c2b374d
ZY
529 * No need to go through error recovery process.
530 */
0927678f 531 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
6c2b374d
ZY
532 if (pos)
533 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
28eb27cf 534 info->status);
17e21854 535 } else
0918472c 536 do_recovery(dev, info->severity);
6c2b374d
ZY
537}
538
0918472c
HY
539#ifdef CONFIG_ACPI_APEI_PCIEAER
540static void aer_recover_work_func(struct work_struct *work);
541
542#define AER_RECOVER_RING_ORDER 4
543#define AER_RECOVER_RING_SIZE (1 << AER_RECOVER_RING_ORDER)
544
545struct aer_recover_entry
546{
547 u8 bus;
548 u8 devfn;
549 u16 domain;
550 int severity;
37448adf 551 struct aer_capability_regs *regs;
0918472c
HY
552};
553
554static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
555 AER_RECOVER_RING_SIZE);
556/*
557 * Mutual exclusion for writers of aer_recover_ring, reader side don't
558 * need lock, because there is only one reader and lock is not needed
559 * between reader and writer.
560 */
561static DEFINE_SPINLOCK(aer_recover_ring_lock);
562static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
563
564void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
37448adf 565 int severity, struct aer_capability_regs *aer_regs)
0918472c
HY
566{
567 unsigned long flags;
568 struct aer_recover_entry entry = {
569 .bus = bus,
570 .devfn = devfn,
571 .domain = domain,
572 .severity = severity,
37448adf 573 .regs = aer_regs,
0918472c
HY
574 };
575
576 spin_lock_irqsave(&aer_recover_ring_lock, flags);
577 if (kfifo_put(&aer_recover_ring, &entry))
578 schedule_work(&aer_recover_work);
579 else
580 pr_err("AER recover: Buffer overflow when recovering AER for %04x:%02x:%02x:%x\n",
581 domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
582 spin_unlock_irqrestore(&aer_recover_ring_lock, flags);
583}
584EXPORT_SYMBOL_GPL(aer_recover_queue);
585
586static void aer_recover_work_func(struct work_struct *work)
587{
588 struct aer_recover_entry entry;
589 struct pci_dev *pdev;
590
591 while (kfifo_get(&aer_recover_ring, &entry)) {
592 pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
593 entry.devfn);
594 if (!pdev) {
595 pr_err("AER recover: Can not find pci_dev for %04x:%02x:%02x:%x\n",
596 entry.domain, entry.bus,
597 PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
598 continue;
599 }
37448adf 600 cper_print_aer(pdev, entry.severity, entry.regs);
0918472c 601 do_recovery(pdev, entry.severity);
a82b6af3 602 pci_dev_put(pdev);
0918472c
HY
603 }
604}
605#endif
606
b1c089b7
HS
607/**
608 * get_device_error_info - read error status from dev and store it to info
609 * @dev: pointer to the device expected to have a error record
610 * @info: pointer to structure to store the error record
611 *
612 * Return 1 on success, 0 on error.
7c4ec94f
HS
613 *
614 * Note that @info is reused among all error devices. Clear fields properly.
b1c089b7 615 */
6c2b374d
ZY
616static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
617{
e7a0d92b 618 int pos, temp;
6c2b374d 619
7c4ec94f 620 /* Must reset in this function */
1b4ffcf8 621 info->status = 0;
273024de 622 info->tlp_header_valid = 0;
1b4ffcf8 623
0927678f 624 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
6c2b374d
ZY
625
626 /* The device might not support AER */
627 if (!pos)
b1c089b7 628 return 1;
6c2b374d
ZY
629
630 if (info->severity == AER_CORRECTABLE) {
631 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
632 &info->status);
0d90c3ac
HS
633 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
634 &info->mask);
635 if (!(info->status & ~info->mask))
b1c089b7 636 return 0;
6c2b374d
ZY
637 } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
638 info->severity == AER_NONFATAL) {
639
640 /* Link is still healthy for IO reads */
641 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
642 &info->status);
0d90c3ac
HS
643 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
644 &info->mask);
645 if (!(info->status & ~info->mask))
b1c089b7 646 return 0;
6c2b374d 647
e7a0d92b
HS
648 /* Get First Error Pointer */
649 pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
273024de 650 info->first_error = PCI_ERR_CAP_FEP(temp);
e7a0d92b 651
6c2b374d 652 if (info->status & AER_LOG_TLP_MASKS) {
273024de 653 info->tlp_header_valid = 1;
6c2b374d
ZY
654 pci_read_config_dword(dev,
655 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
656 pci_read_config_dword(dev,
657 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
658 pci_read_config_dword(dev,
659 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
660 pci_read_config_dword(dev,
661 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
662 }
663 }
664
b1c089b7 665 return 1;
6c2b374d
ZY
666}
667
3d5505c5
ZY
668static inline void aer_process_err_devices(struct pcie_device *p_device,
669 struct aer_err_info *e_info)
670{
671 int i;
672
b1c089b7 673 /* Report all before handle them, not to lost records by reset etc. */
3d5505c5 674 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
b1c089b7 675 if (get_device_error_info(e_info->dev[i], e_info))
3d5505c5 676 aer_print_error(e_info->dev[i], e_info);
b1c089b7
HS
677 }
678 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
679 if (get_device_error_info(e_info->dev[i], e_info))
680 handle_error_source(p_device, e_info->dev[i], e_info);
3d5505c5
ZY
681 }
682}
683
6c2b374d
ZY
684/**
685 * aer_isr_one_error - consume an error detected by root port
686 * @p_device: pointer to error root port service device
687 * @e_src: pointer to an error source
d885c6b7 688 */
6c2b374d
ZY
689static void aer_isr_one_error(struct pcie_device *p_device,
690 struct aer_err_source *e_src)
691{
28eb27cf 692 struct aer_err_info *e_info;
28eb27cf
ZY
693
694 /* struct aer_err_info might be big, so we allocate it with slab */
695 e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL);
7c4ec94f 696 if (!e_info) {
28eb27cf
ZY
697 dev_printk(KERN_DEBUG, &p_device->port->dev,
698 "Can't allocate mem when processing AER errors\n");
699 return;
700 }
6c2b374d
ZY
701
702 /*
703 * There is a possibility that both correctable error and
704 * uncorrectable error being logged. Report correctable error first.
705 */
7c4ec94f
HS
706 if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
707 e_info->id = ERR_COR_ID(e_src->id);
708 e_info->severity = AER_CORRECTABLE;
709
710 if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
711 e_info->multi_error_valid = 1;
712 else
713 e_info->multi_error_valid = 0;
714
715 aer_print_port_info(p_device->port, e_info);
716
717 if (find_source_device(p_device->port, e_info))
718 aer_process_err_devices(p_device, e_info);
719 }
720
721 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
722 e_info->id = ERR_UNCOR_ID(e_src->id);
723
724 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
725 e_info->severity = AER_FATAL;
726 else
727 e_info->severity = AER_NONFATAL;
728
729 if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
273024de 730 e_info->multi_error_valid = 1;
7c4ec94f
HS
731 else
732 e_info->multi_error_valid = 0;
28eb27cf 733
79e4b89b
HS
734 aer_print_port_info(p_device->port, e_info);
735
98ca3964
HS
736 if (find_source_device(p_device->port, e_info))
737 aer_process_err_devices(p_device, e_info);
6c2b374d 738 }
28eb27cf
ZY
739
740 kfree(e_info);
6c2b374d
ZY
741}
742
88da13bf
HS
743/**
744 * get_e_source - retrieve an error source
745 * @rpc: pointer to the root port which holds an error
746 * @e_src: pointer to store retrieved error source
747 *
748 * Return 1 if an error source is retrieved, otherwise 0.
749 *
750 * Invoked by DPC handler to consume an error.
751 */
752static int get_e_source(struct aer_rpc *rpc, struct aer_err_source *e_src)
753{
754 unsigned long flags;
88da13bf
HS
755
756 /* Lock access to Root error producer/consumer index */
757 spin_lock_irqsave(&rpc->e_lock, flags);
f6735590
LT
758 if (rpc->prod_idx == rpc->cons_idx) {
759 spin_unlock_irqrestore(&rpc->e_lock, flags);
760 return 0;
88da13bf 761 }
f6735590
LT
762
763 *e_src = rpc->e_sources[rpc->cons_idx];
764 rpc->cons_idx++;
765 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
766 rpc->cons_idx = 0;
88da13bf
HS
767 spin_unlock_irqrestore(&rpc->e_lock, flags);
768
f6735590 769 return 1;
88da13bf
HS
770}
771
6c2b374d
ZY
772/**
773 * aer_isr - consume errors detected by root port
65f27f38 774 * @work: definition of this work item
6c2b374d
ZY
775 *
776 * Invoked, as DPC, when root port records new detected error
d885c6b7 777 */
65f27f38 778void aer_isr(struct work_struct *work)
6c2b374d 779{
65f27f38
DH
780 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
781 struct pcie_device *p_device = rpc->rpd;
50c1126e 782 struct aer_err_source uninitialized_var(e_src);
6c2b374d
ZY
783
784 mutex_lock(&rpc->rpc_mutex);
88da13bf
HS
785 while (get_e_source(rpc, &e_src))
786 aer_isr_one_error(p_device, &e_src);
6c2b374d
ZY
787 mutex_unlock(&rpc->rpc_mutex);
788
789 wake_up(&rpc->wait_release);
790}
791
6c2b374d
ZY
792/**
793 * aer_init - provide AER initialization
794 * @dev: pointer to AER pcie device
795 *
796 * Invoked when AER service driver is loaded.
d885c6b7 797 */
6c2b374d
ZY
798int aer_init(struct pcie_device *dev)
799{
05843961
MD
800 if (forceload) {
801 dev_printk(KERN_DEBUG, &dev->device,
802 "aerdrv forceload requested.\n");
affb72c3 803 pcie_aer_force_firmware_first(dev->port, 0);
05843961 804 }
28eb5f27 805 return 0;
6c2b374d 806}