1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Implement the AER root port service driver. The driver registers an IRQ
4 * handler. When a root port triggers an AER interrupt, the IRQ handler
5 * collects root port status and schedules work.
6 *
7 * Copyright (C) 2006 Intel Corp.
8 * Tom Long Nguyen (tom.l.nguyen@intel.com)
9 * Zhang Yanmin (yanmin.zhang@intel.com)
10 *
11 * (C) Copyright 2009 Hewlett-Packard Development Company, L.P.
12 * Andrew Patterson <andrew.patterson@hp.com>
13 */
14
15 #define pr_fmt(fmt) "AER: " fmt
16 #define dev_fmt pr_fmt
17
18 #include <linux/bitops.h>
19 #include <linux/cper.h>
20 #include <linux/pci.h>
21 #include <linux/pci-acpi.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/errno.h>
25 #include <linux/pm.h>
26 #include <linux/init.h>
27 #include <linux/interrupt.h>
28 #include <linux/delay.h>
29 #include <linux/kfifo.h>
30 #include <linux/slab.h>
31 #include <acpi/apei.h>
32 #include <acpi/ghes.h>
33 #include <ras/ras_event.h>
34
35 #include "../pci.h"
36 #include "portdrv.h"
37
38 #define AER_ERROR_SOURCES_MAX 128
39
40 #define AER_MAX_TYPEOF_COR_ERRS 16 /* as per PCI_ERR_COR_STATUS */
41 #define AER_MAX_TYPEOF_UNCOR_ERRS 27 /* as per PCI_ERR_UNCOR_STATUS*/
42
43 struct aer_err_source {
44 unsigned int status;
45 unsigned int id;
46 };
47
48 struct aer_rpc {
49 struct pci_dev *rpd; /* Root Port device */
50 DECLARE_KFIFO(aer_fifo, struct aer_err_source, AER_ERROR_SOURCES_MAX);
51 };
52
53 /* AER stats for the device */
54 struct aer_stats {
55
56 /*
57 * Fields for all AER capable devices. They indicate the errors
58 * "as seen by this device". Note that this may mean that if an
59 * end point is causing problems, the AER counters may increment
60 * at its link partner (e.g. root port) because the errors will be
61 * "seen" by the link partner and not the problematic end point
62 * itself (which may report all counters as 0 as it never saw any
63 * problems).
64 */
65 /* Counters for different type of correctable errors */
66 u64 dev_cor_errs[AER_MAX_TYPEOF_COR_ERRS];
67 /* Counters for different type of fatal uncorrectable errors */
68 u64 dev_fatal_errs[AER_MAX_TYPEOF_UNCOR_ERRS];
69 /* Counters for different type of nonfatal uncorrectable errors */
70 u64 dev_nonfatal_errs[AER_MAX_TYPEOF_UNCOR_ERRS];
71 /* Total number of ERR_COR sent by this device */
72 u64 dev_total_cor_errs;
73 /* Total number of ERR_FATAL sent by this device */
74 u64 dev_total_fatal_errs;
75 /* Total number of ERR_NONFATAL sent by this device */
76 u64 dev_total_nonfatal_errs;
77
78 /*
79 * Fields for Root ports & root complex event collectors only, these
80 * indicate the total number of ERR_COR, ERR_FATAL, and ERR_NONFATAL
81 * messages received by the root port / event collector, INCLUDING the
82 * ones that are generated internally (by the rootport itself)
83 */
84 u64 rootport_total_cor_errs;
85 u64 rootport_total_fatal_errs;
86 u64 rootport_total_nonfatal_errs;
87 };
88
89 #define AER_LOG_TLP_MASKS (PCI_ERR_UNC_POISON_TLP| \
90 PCI_ERR_UNC_ECRC| \
91 PCI_ERR_UNC_UNSUP| \
92 PCI_ERR_UNC_COMP_ABORT| \
93 PCI_ERR_UNC_UNX_COMP| \
94 PCI_ERR_UNC_MALF_TLP)
95
96 #define SYSTEM_ERROR_INTR_ON_MESG_MASK (PCI_EXP_RTCTL_SECEE| \
97 PCI_EXP_RTCTL_SENFEE| \
98 PCI_EXP_RTCTL_SEFEE)
99 #define ROOT_PORT_INTR_ON_MESG_MASK (PCI_ERR_ROOT_CMD_COR_EN| \
100 PCI_ERR_ROOT_CMD_NONFATAL_EN| \
101 PCI_ERR_ROOT_CMD_FATAL_EN)
102 #define ERR_COR_ID(d) (d & 0xffff)
103 #define ERR_UNCOR_ID(d) (d >> 16)
104
105 #define AER_ERR_STATUS_MASK (PCI_ERR_ROOT_UNCOR_RCV | \
106 PCI_ERR_ROOT_COR_RCV | \
107 PCI_ERR_ROOT_MULTI_COR_RCV | \
108 PCI_ERR_ROOT_MULTI_UNCOR_RCV)
109
110 static int pcie_aer_disable;
111 static pci_ers_result_t aer_root_reset(struct pci_dev *dev);
112
pci_no_aer(void)113 void pci_no_aer(void)
114 {
115 pcie_aer_disable = 1;
116 }
117
pci_aer_available(void)118 bool pci_aer_available(void)
119 {
120 return !pcie_aer_disable && pci_msi_enabled();
121 }
122
123 #ifdef CONFIG_PCIE_ECRC
124
125 #define ECRC_POLICY_DEFAULT 0 /* ECRC set by BIOS */
126 #define ECRC_POLICY_OFF 1 /* ECRC off for performance */
127 #define ECRC_POLICY_ON 2 /* ECRC on for data integrity */
128
129 static int ecrc_policy = ECRC_POLICY_DEFAULT;
130
131 static const char * const ecrc_policy_str[] = {
132 [ECRC_POLICY_DEFAULT] = "bios",
133 [ECRC_POLICY_OFF] = "off",
134 [ECRC_POLICY_ON] = "on"
135 };
136
137 /**
138 * enable_ecrc_checking - enable PCIe ECRC checking for a device
139 * @dev: the PCI device
140 *
141 * Returns 0 on success, or negative on failure.
142 */
enable_ecrc_checking(struct pci_dev * dev)143 static int enable_ecrc_checking(struct pci_dev *dev)
144 {
145 int aer = dev->aer_cap;
146 u32 reg32;
147
148 if (!aer)
149 return -ENODEV;
150
151 pci_read_config_dword(dev, aer + PCI_ERR_CAP, ®32);
152 if (reg32 & PCI_ERR_CAP_ECRC_GENC)
153 reg32 |= PCI_ERR_CAP_ECRC_GENE;
154 if (reg32 & PCI_ERR_CAP_ECRC_CHKC)
155 reg32 |= PCI_ERR_CAP_ECRC_CHKE;
156 pci_write_config_dword(dev, aer + PCI_ERR_CAP, reg32);
157
158 return 0;
159 }
160
161 /**
162 * disable_ecrc_checking - disables PCIe ECRC checking for a device
163 * @dev: the PCI device
164 *
165 * Returns 0 on success, or negative on failure.
166 */
disable_ecrc_checking(struct pci_dev * dev)167 static int disable_ecrc_checking(struct pci_dev *dev)
168 {
169 int aer = dev->aer_cap;
170 u32 reg32;
171
172 if (!aer)
173 return -ENODEV;
174
175 pci_read_config_dword(dev, aer + PCI_ERR_CAP, ®32);
176 reg32 &= ~(PCI_ERR_CAP_ECRC_GENE | PCI_ERR_CAP_ECRC_CHKE);
177 pci_write_config_dword(dev, aer + PCI_ERR_CAP, reg32);
178
179 return 0;
180 }
181
182 /**
183 * pcie_set_ecrc_checking - set/unset PCIe ECRC checking for a device based on global policy
184 * @dev: the PCI device
185 */
pcie_set_ecrc_checking(struct pci_dev * dev)186 void pcie_set_ecrc_checking(struct pci_dev *dev)
187 {
188 if (!pcie_aer_is_native(dev))
189 return;
190
191 switch (ecrc_policy) {
192 case ECRC_POLICY_DEFAULT:
193 return;
194 case ECRC_POLICY_OFF:
195 disable_ecrc_checking(dev);
196 break;
197 case ECRC_POLICY_ON:
198 enable_ecrc_checking(dev);
199 break;
200 default:
201 return;
202 }
203 }
204
205 /**
206 * pcie_ecrc_get_policy - parse kernel command-line ecrc option
207 * @str: ECRC policy from kernel command line to use
208 */
pcie_ecrc_get_policy(char * str)209 void pcie_ecrc_get_policy(char *str)
210 {
211 int i;
212
213 i = match_string(ecrc_policy_str, ARRAY_SIZE(ecrc_policy_str), str);
214 if (i < 0)
215 return;
216
217 ecrc_policy = i;
218 }
219 #endif /* CONFIG_PCIE_ECRC */
220
221 #define PCI_EXP_AER_FLAGS (PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \
222 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE)
223
pcie_aer_is_native(struct pci_dev * dev)224 int pcie_aer_is_native(struct pci_dev *dev)
225 {
226 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
227
228 if (!dev->aer_cap)
229 return 0;
230
231 return pcie_ports_native || host->native_aer;
232 }
233 EXPORT_SYMBOL_NS_GPL(pcie_aer_is_native, CXL);
234
pci_enable_pcie_error_reporting(struct pci_dev * dev)235 static int pci_enable_pcie_error_reporting(struct pci_dev *dev)
236 {
237 int rc;
238
239 if (!pcie_aer_is_native(dev))
240 return -EIO;
241
242 rc = pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
243 return pcibios_err_to_errno(rc);
244 }
245
pci_aer_clear_nonfatal_status(struct pci_dev * dev)246 int pci_aer_clear_nonfatal_status(struct pci_dev *dev)
247 {
248 int aer = dev->aer_cap;
249 u32 status, sev;
250
251 if (!pcie_aer_is_native(dev))
252 return -EIO;
253
254 /* Clear status bits for ERR_NONFATAL errors only */
255 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status);
256 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, &sev);
257 status &= ~sev;
258 if (status)
259 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status);
260
261 return 0;
262 }
263 EXPORT_SYMBOL_GPL(pci_aer_clear_nonfatal_status);
264
pci_aer_clear_fatal_status(struct pci_dev * dev)265 void pci_aer_clear_fatal_status(struct pci_dev *dev)
266 {
267 int aer = dev->aer_cap;
268 u32 status, sev;
269
270 if (!pcie_aer_is_native(dev))
271 return;
272
273 /* Clear status bits for ERR_FATAL errors only */
274 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status);
275 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, &sev);
276 status &= sev;
277 if (status)
278 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status);
279 }
280
281 /**
282 * pci_aer_raw_clear_status - Clear AER error registers.
283 * @dev: the PCI device
284 *
285 * Clearing AER error status registers unconditionally, regardless of
286 * whether they're owned by firmware or the OS.
287 *
288 * Returns 0 on success, or negative on failure.
289 */
pci_aer_raw_clear_status(struct pci_dev * dev)290 int pci_aer_raw_clear_status(struct pci_dev *dev)
291 {
292 int aer = dev->aer_cap;
293 u32 status;
294 int port_type;
295
296 if (!aer)
297 return -EIO;
298
299 port_type = pci_pcie_type(dev);
300 if (port_type == PCI_EXP_TYPE_ROOT_PORT ||
301 port_type == PCI_EXP_TYPE_RC_EC) {
302 pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &status);
303 pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, status);
304 }
305
306 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS, &status);
307 pci_write_config_dword(dev, aer + PCI_ERR_COR_STATUS, status);
308
309 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status);
310 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status);
311
312 return 0;
313 }
314
pci_aer_clear_status(struct pci_dev * dev)315 int pci_aer_clear_status(struct pci_dev *dev)
316 {
317 if (!pcie_aer_is_native(dev))
318 return -EIO;
319
320 return pci_aer_raw_clear_status(dev);
321 }
322
pci_save_aer_state(struct pci_dev * dev)323 void pci_save_aer_state(struct pci_dev *dev)
324 {
325 int aer = dev->aer_cap;
326 struct pci_cap_saved_state *save_state;
327 u32 *cap;
328
329 if (!aer)
330 return;
331
332 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_ERR);
333 if (!save_state)
334 return;
335
336 cap = &save_state->cap.data[0];
337 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, cap++);
338 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, cap++);
339 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK, cap++);
340 pci_read_config_dword(dev, aer + PCI_ERR_CAP, cap++);
341 if (pcie_cap_has_rtctl(dev))
342 pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, cap++);
343 }
344
pci_restore_aer_state(struct pci_dev * dev)345 void pci_restore_aer_state(struct pci_dev *dev)
346 {
347 int aer = dev->aer_cap;
348 struct pci_cap_saved_state *save_state;
349 u32 *cap;
350
351 if (!aer)
352 return;
353
354 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_ERR);
355 if (!save_state)
356 return;
357
358 cap = &save_state->cap.data[0];
359 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, *cap++);
360 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, *cap++);
361 pci_write_config_dword(dev, aer + PCI_ERR_COR_MASK, *cap++);
362 pci_write_config_dword(dev, aer + PCI_ERR_CAP, *cap++);
363 if (pcie_cap_has_rtctl(dev))
364 pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, *cap++);
365 }
366
pci_aer_init(struct pci_dev * dev)367 void pci_aer_init(struct pci_dev *dev)
368 {
369 int n;
370
371 dev->aer_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
372 if (!dev->aer_cap)
373 return;
374
375 dev->aer_stats = kzalloc(sizeof(struct aer_stats), GFP_KERNEL);
376
377 /*
378 * We save/restore PCI_ERR_UNCOR_MASK, PCI_ERR_UNCOR_SEVER,
379 * PCI_ERR_COR_MASK, and PCI_ERR_CAP. Root and Root Complex Event
380 * Collectors also implement PCI_ERR_ROOT_COMMAND (PCIe r5.0, sec
381 * 7.8.4).
382 */
383 n = pcie_cap_has_rtctl(dev) ? 5 : 4;
384 pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_ERR, sizeof(u32) * n);
385
386 pci_aer_clear_status(dev);
387
388 if (pci_aer_available())
389 pci_enable_pcie_error_reporting(dev);
390
391 pcie_set_ecrc_checking(dev);
392 }
393
pci_aer_exit(struct pci_dev * dev)394 void pci_aer_exit(struct pci_dev *dev)
395 {
396 kfree(dev->aer_stats);
397 dev->aer_stats = NULL;
398 }
399
400 #define AER_AGENT_RECEIVER 0
401 #define AER_AGENT_REQUESTER 1
402 #define AER_AGENT_COMPLETER 2
403 #define AER_AGENT_TRANSMITTER 3
404
405 #define AER_AGENT_REQUESTER_MASK(t) ((t == AER_CORRECTABLE) ? \
406 0 : (PCI_ERR_UNC_COMP_TIME|PCI_ERR_UNC_UNSUP))
407 #define AER_AGENT_COMPLETER_MASK(t) ((t == AER_CORRECTABLE) ? \
408 0 : PCI_ERR_UNC_COMP_ABORT)
409 #define AER_AGENT_TRANSMITTER_MASK(t) ((t == AER_CORRECTABLE) ? \
410 (PCI_ERR_COR_REP_ROLL|PCI_ERR_COR_REP_TIMER) : 0)
411
412 #define AER_GET_AGENT(t, e) \
413 ((e & AER_AGENT_COMPLETER_MASK(t)) ? AER_AGENT_COMPLETER : \
414 (e & AER_AGENT_REQUESTER_MASK(t)) ? AER_AGENT_REQUESTER : \
415 (e & AER_AGENT_TRANSMITTER_MASK(t)) ? AER_AGENT_TRANSMITTER : \
416 AER_AGENT_RECEIVER)
417
418 #define AER_PHYSICAL_LAYER_ERROR 0
419 #define AER_DATA_LINK_LAYER_ERROR 1
420 #define AER_TRANSACTION_LAYER_ERROR 2
421
422 #define AER_PHYSICAL_LAYER_ERROR_MASK(t) ((t == AER_CORRECTABLE) ? \
423 PCI_ERR_COR_RCVR : 0)
424 #define AER_DATA_LINK_LAYER_ERROR_MASK(t) ((t == AER_CORRECTABLE) ? \
425 (PCI_ERR_COR_BAD_TLP| \
426 PCI_ERR_COR_BAD_DLLP| \
427 PCI_ERR_COR_REP_ROLL| \
428 PCI_ERR_COR_REP_TIMER) : PCI_ERR_UNC_DLP)
429
430 #define AER_GET_LAYER_ERROR(t, e) \
431 ((e & AER_PHYSICAL_LAYER_ERROR_MASK(t)) ? AER_PHYSICAL_LAYER_ERROR : \
432 (e & AER_DATA_LINK_LAYER_ERROR_MASK(t)) ? AER_DATA_LINK_LAYER_ERROR : \
433 AER_TRANSACTION_LAYER_ERROR)
434
435 /*
436 * AER error strings
437 */
438 static const char *aer_error_severity_string[] = {
439 "Uncorrected (Non-Fatal)",
440 "Uncorrected (Fatal)",
441 "Corrected"
442 };
443
444 static const char *aer_error_layer[] = {
445 "Physical Layer",
446 "Data Link Layer",
447 "Transaction Layer"
448 };
449
450 static const char *aer_correctable_error_string[] = {
451 "RxErr", /* Bit Position 0 */
452 NULL,
453 NULL,
454 NULL,
455 NULL,
456 NULL,
457 "BadTLP", /* Bit Position 6 */
458 "BadDLLP", /* Bit Position 7 */
459 "Rollover", /* Bit Position 8 */
460 NULL,
461 NULL,
462 NULL,
463 "Timeout", /* Bit Position 12 */
464 "NonFatalErr", /* Bit Position 13 */
465 "CorrIntErr", /* Bit Position 14 */
466 "HeaderOF", /* Bit Position 15 */
467 NULL, /* Bit Position 16 */
468 NULL, /* Bit Position 17 */
469 NULL, /* Bit Position 18 */
470 NULL, /* Bit Position 19 */
471 NULL, /* Bit Position 20 */
472 NULL, /* Bit Position 21 */
473 NULL, /* Bit Position 22 */
474 NULL, /* Bit Position 23 */
475 NULL, /* Bit Position 24 */
476 NULL, /* Bit Position 25 */
477 NULL, /* Bit Position 26 */
478 NULL, /* Bit Position 27 */
479 NULL, /* Bit Position 28 */
480 NULL, /* Bit Position 29 */
481 NULL, /* Bit Position 30 */
482 NULL, /* Bit Position 31 */
483 };
484
485 static const char *aer_uncorrectable_error_string[] = {
486 "Undefined", /* Bit Position 0 */
487 NULL,
488 NULL,
489 NULL,
490 "DLP", /* Bit Position 4 */
491 "SDES", /* Bit Position 5 */
492 NULL,
493 NULL,
494 NULL,
495 NULL,
496 NULL,
497 NULL,
498 "TLP", /* Bit Position 12 */
499 "FCP", /* Bit Position 13 */
500 "CmpltTO", /* Bit Position 14 */
501 "CmpltAbrt", /* Bit Position 15 */
502 "UnxCmplt", /* Bit Position 16 */
503 "RxOF", /* Bit Position 17 */
504 "MalfTLP", /* Bit Position 18 */
505 "ECRC", /* Bit Position 19 */
506 "UnsupReq", /* Bit Position 20 */
507 "ACSViol", /* Bit Position 21 */
508 "UncorrIntErr", /* Bit Position 22 */
509 "BlockedTLP", /* Bit Position 23 */
510 "AtomicOpBlocked", /* Bit Position 24 */
511 "TLPBlockedErr", /* Bit Position 25 */
512 "PoisonTLPBlocked", /* Bit Position 26 */
513 NULL, /* Bit Position 27 */
514 NULL, /* Bit Position 28 */
515 NULL, /* Bit Position 29 */
516 NULL, /* Bit Position 30 */
517 NULL, /* Bit Position 31 */
518 };
519
520 static const char *aer_agent_string[] = {
521 "Receiver ID",
522 "Requester ID",
523 "Completer ID",
524 "Transmitter ID"
525 };
526
527 #define aer_stats_dev_attr(name, stats_array, strings_array, \
528 total_string, total_field) \
529 static ssize_t \
530 name##_show(struct device *dev, struct device_attribute *attr, \
531 char *buf) \
532 { \
533 unsigned int i; \
534 struct pci_dev *pdev = to_pci_dev(dev); \
535 u64 *stats = pdev->aer_stats->stats_array; \
536 size_t len = 0; \
537 \
538 for (i = 0; i < ARRAY_SIZE(pdev->aer_stats->stats_array); i++) {\
539 if (strings_array[i]) \
540 len += sysfs_emit_at(buf, len, "%s %llu\n", \
541 strings_array[i], \
542 stats[i]); \
543 else if (stats[i]) \
544 len += sysfs_emit_at(buf, len, \
545 #stats_array "_bit[%d] %llu\n",\
546 i, stats[i]); \
547 } \
548 len += sysfs_emit_at(buf, len, "TOTAL_%s %llu\n", total_string, \
549 pdev->aer_stats->total_field); \
550 return len; \
551 } \
552 static DEVICE_ATTR_RO(name)
553
554 aer_stats_dev_attr(aer_dev_correctable, dev_cor_errs,
555 aer_correctable_error_string, "ERR_COR",
556 dev_total_cor_errs);
557 aer_stats_dev_attr(aer_dev_fatal, dev_fatal_errs,
558 aer_uncorrectable_error_string, "ERR_FATAL",
559 dev_total_fatal_errs);
560 aer_stats_dev_attr(aer_dev_nonfatal, dev_nonfatal_errs,
561 aer_uncorrectable_error_string, "ERR_NONFATAL",
562 dev_total_nonfatal_errs);
563
564 #define aer_stats_rootport_attr(name, field) \
565 static ssize_t \
566 name##_show(struct device *dev, struct device_attribute *attr, \
567 char *buf) \
568 { \
569 struct pci_dev *pdev = to_pci_dev(dev); \
570 return sysfs_emit(buf, "%llu\n", pdev->aer_stats->field); \
571 } \
572 static DEVICE_ATTR_RO(name)
573
574 aer_stats_rootport_attr(aer_rootport_total_err_cor,
575 rootport_total_cor_errs);
576 aer_stats_rootport_attr(aer_rootport_total_err_fatal,
577 rootport_total_fatal_errs);
578 aer_stats_rootport_attr(aer_rootport_total_err_nonfatal,
579 rootport_total_nonfatal_errs);
580
581 static struct attribute *aer_stats_attrs[] __ro_after_init = {
582 &dev_attr_aer_dev_correctable.attr,
583 &dev_attr_aer_dev_fatal.attr,
584 &dev_attr_aer_dev_nonfatal.attr,
585 &dev_attr_aer_rootport_total_err_cor.attr,
586 &dev_attr_aer_rootport_total_err_fatal.attr,
587 &dev_attr_aer_rootport_total_err_nonfatal.attr,
588 NULL
589 };
590
aer_stats_attrs_are_visible(struct kobject * kobj,struct attribute * a,int n)591 static umode_t aer_stats_attrs_are_visible(struct kobject *kobj,
592 struct attribute *a, int n)
593 {
594 struct device *dev = kobj_to_dev(kobj);
595 struct pci_dev *pdev = to_pci_dev(dev);
596
597 if (!pdev->aer_stats)
598 return 0;
599
600 if ((a == &dev_attr_aer_rootport_total_err_cor.attr ||
601 a == &dev_attr_aer_rootport_total_err_fatal.attr ||
602 a == &dev_attr_aer_rootport_total_err_nonfatal.attr) &&
603 ((pci_pcie_type(pdev) != PCI_EXP_TYPE_ROOT_PORT) &&
604 (pci_pcie_type(pdev) != PCI_EXP_TYPE_RC_EC)))
605 return 0;
606
607 return a->mode;
608 }
609
610 const struct attribute_group aer_stats_attr_group = {
611 .attrs = aer_stats_attrs,
612 .is_visible = aer_stats_attrs_are_visible,
613 };
614
pci_dev_aer_stats_incr(struct pci_dev * pdev,struct aer_err_info * info)615 static void pci_dev_aer_stats_incr(struct pci_dev *pdev,
616 struct aer_err_info *info)
617 {
618 unsigned long status = info->status & ~info->mask;
619 int i, max = -1;
620 u64 *counter = NULL;
621 struct aer_stats *aer_stats = pdev->aer_stats;
622
623 if (!aer_stats)
624 return;
625
626 switch (info->severity) {
627 case AER_CORRECTABLE:
628 aer_stats->dev_total_cor_errs++;
629 counter = &aer_stats->dev_cor_errs[0];
630 max = AER_MAX_TYPEOF_COR_ERRS;
631 break;
632 case AER_NONFATAL:
633 aer_stats->dev_total_nonfatal_errs++;
634 counter = &aer_stats->dev_nonfatal_errs[0];
635 max = AER_MAX_TYPEOF_UNCOR_ERRS;
636 break;
637 case AER_FATAL:
638 aer_stats->dev_total_fatal_errs++;
639 counter = &aer_stats->dev_fatal_errs[0];
640 max = AER_MAX_TYPEOF_UNCOR_ERRS;
641 break;
642 }
643
644 for_each_set_bit(i, &status, max)
645 counter[i]++;
646 }
647
pci_rootport_aer_stats_incr(struct pci_dev * pdev,struct aer_err_source * e_src)648 static void pci_rootport_aer_stats_incr(struct pci_dev *pdev,
649 struct aer_err_source *e_src)
650 {
651 struct aer_stats *aer_stats = pdev->aer_stats;
652
653 if (!aer_stats)
654 return;
655
656 if (e_src->status & PCI_ERR_ROOT_COR_RCV)
657 aer_stats->rootport_total_cor_errs++;
658
659 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
660 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
661 aer_stats->rootport_total_fatal_errs++;
662 else
663 aer_stats->rootport_total_nonfatal_errs++;
664 }
665 }
666
__print_tlp_header(struct pci_dev * dev,struct aer_header_log_regs * t)667 static void __print_tlp_header(struct pci_dev *dev,
668 struct aer_header_log_regs *t)
669 {
670 pci_err(dev, " TLP Header: %08x %08x %08x %08x\n",
671 t->dw0, t->dw1, t->dw2, t->dw3);
672 }
673
__aer_print_error(struct pci_dev * dev,struct aer_err_info * info)674 static void __aer_print_error(struct pci_dev *dev,
675 struct aer_err_info *info)
676 {
677 const char **strings;
678 unsigned long status = info->status & ~info->mask;
679 const char *level, *errmsg;
680 int i;
681
682 if (info->severity == AER_CORRECTABLE) {
683 strings = aer_correctable_error_string;
684 level = KERN_WARNING;
685 } else {
686 strings = aer_uncorrectable_error_string;
687 level = KERN_ERR;
688 }
689
690 for_each_set_bit(i, &status, 32) {
691 errmsg = strings[i];
692 if (!errmsg)
693 errmsg = "Unknown Error Bit";
694
695 pci_printk(level, dev, " [%2d] %-22s%s\n", i, errmsg,
696 info->first_error == i ? " (First)" : "");
697 }
698 pci_dev_aer_stats_incr(dev, info);
699 }
700
aer_print_error(struct pci_dev * dev,struct aer_err_info * info)701 void aer_print_error(struct pci_dev *dev, struct aer_err_info *info)
702 {
703 int layer, agent;
704 int id = pci_dev_id(dev);
705 const char *level;
706
707 if (!info->status) {
708 pci_err(dev, "PCIe Bus Error: severity=%s, type=Inaccessible, (Unregistered Agent ID)\n",
709 aer_error_severity_string[info->severity]);
710 goto out;
711 }
712
713 layer = AER_GET_LAYER_ERROR(info->severity, info->status);
714 agent = AER_GET_AGENT(info->severity, info->status);
715
716 level = (info->severity == AER_CORRECTABLE) ? KERN_WARNING : KERN_ERR;
717
718 pci_printk(level, dev, "PCIe Bus Error: severity=%s, type=%s, (%s)\n",
719 aer_error_severity_string[info->severity],
720 aer_error_layer[layer], aer_agent_string[agent]);
721
722 pci_printk(level, dev, " device [%04x:%04x] error status/mask=%08x/%08x\n",
723 dev->vendor, dev->device, info->status, info->mask);
724
725 __aer_print_error(dev, info);
726
727 if (info->tlp_header_valid)
728 __print_tlp_header(dev, &info->tlp);
729
730 out:
731 if (info->id && info->error_dev_num > 1 && info->id == id)
732 pci_err(dev, " Error of this Agent is reported first\n");
733
734 trace_aer_event(dev_name(&dev->dev), (info->status & ~info->mask),
735 info->severity, info->tlp_header_valid, &info->tlp);
736 }
737
aer_print_port_info(struct pci_dev * dev,struct aer_err_info * info)738 static void aer_print_port_info(struct pci_dev *dev, struct aer_err_info *info)
739 {
740 u8 bus = info->id >> 8;
741 u8 devfn = info->id & 0xff;
742
743 pci_info(dev, "%s%s error message received from %04x:%02x:%02x.%d\n",
744 info->multi_error_valid ? "Multiple " : "",
745 aer_error_severity_string[info->severity],
746 pci_domain_nr(dev->bus), bus, PCI_SLOT(devfn),
747 PCI_FUNC(devfn));
748 }
749
750 #ifdef CONFIG_ACPI_APEI_PCIEAER
cper_severity_to_aer(int cper_severity)751 int cper_severity_to_aer(int cper_severity)
752 {
753 switch (cper_severity) {
754 case CPER_SEV_RECOVERABLE:
755 return AER_NONFATAL;
756 case CPER_SEV_FATAL:
757 return AER_FATAL;
758 default:
759 return AER_CORRECTABLE;
760 }
761 }
762 EXPORT_SYMBOL_GPL(cper_severity_to_aer);
763
cper_print_aer(struct pci_dev * dev,int aer_severity,struct aer_capability_regs * aer)764 void cper_print_aer(struct pci_dev *dev, int aer_severity,
765 struct aer_capability_regs *aer)
766 {
767 int layer, agent, tlp_header_valid = 0;
768 u32 status, mask;
769 struct aer_err_info info;
770
771 if (aer_severity == AER_CORRECTABLE) {
772 status = aer->cor_status;
773 mask = aer->cor_mask;
774 } else {
775 status = aer->uncor_status;
776 mask = aer->uncor_mask;
777 tlp_header_valid = status & AER_LOG_TLP_MASKS;
778 }
779
780 layer = AER_GET_LAYER_ERROR(aer_severity, status);
781 agent = AER_GET_AGENT(aer_severity, status);
782
783 memset(&info, 0, sizeof(info));
784 info.severity = aer_severity;
785 info.status = status;
786 info.mask = mask;
787 info.first_error = PCI_ERR_CAP_FEP(aer->cap_control);
788
789 pci_err(dev, "aer_status: 0x%08x, aer_mask: 0x%08x\n", status, mask);
790 __aer_print_error(dev, &info);
791 pci_err(dev, "aer_layer=%s, aer_agent=%s\n",
792 aer_error_layer[layer], aer_agent_string[agent]);
793
794 if (aer_severity != AER_CORRECTABLE)
795 pci_err(dev, "aer_uncor_severity: 0x%08x\n",
796 aer->uncor_severity);
797
798 if (tlp_header_valid)
799 __print_tlp_header(dev, &aer->header_log);
800
801 trace_aer_event(dev_name(&dev->dev), (status & ~mask),
802 aer_severity, tlp_header_valid, &aer->header_log);
803 }
804 #endif
805
806 /**
807 * add_error_device - list device to be handled
808 * @e_info: pointer to error info
809 * @dev: pointer to pci_dev to be added
810 */
add_error_device(struct aer_err_info * e_info,struct pci_dev * dev)811 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
812 {
813 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
814 e_info->dev[e_info->error_dev_num] = pci_dev_get(dev);
815 e_info->error_dev_num++;
816 return 0;
817 }
818 return -ENOSPC;
819 }
820
821 /**
822 * is_error_source - check whether the device is source of reported error
823 * @dev: pointer to pci_dev to be checked
824 * @e_info: pointer to reported error info
825 */
is_error_source(struct pci_dev * dev,struct aer_err_info * e_info)826 static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info)
827 {
828 int aer = dev->aer_cap;
829 u32 status, mask;
830 u16 reg16;
831
832 /*
833 * When bus id is equal to 0, it might be a bad id
834 * reported by root port.
835 */
836 if ((PCI_BUS_NUM(e_info->id) != 0) &&
837 !(dev->bus->bus_flags & PCI_BUS_FLAGS_NO_AERSID)) {
838 /* Device ID match? */
839 if (e_info->id == pci_dev_id(dev))
840 return true;
841
842 /* Continue id comparing if there is no multiple error */
843 if (!e_info->multi_error_valid)
844 return false;
845 }
846
847 /*
848 * When either
849 * 1) bus id is equal to 0. Some ports might lose the bus
850 * id of error source id;
851 * 2) bus flag PCI_BUS_FLAGS_NO_AERSID is set
852 * 3) There are multiple errors and prior ID comparing fails;
853 * We check AER status registers to find possible reporter.
854 */
855 if (atomic_read(&dev->enable_cnt) == 0)
856 return false;
857
858 /* Check if AER is enabled */
859 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, ®16);
860 if (!(reg16 & PCI_EXP_AER_FLAGS))
861 return false;
862
863 if (!aer)
864 return false;
865
866 /* Check if error is recorded */
867 if (e_info->severity == AER_CORRECTABLE) {
868 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS, &status);
869 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK, &mask);
870 } else {
871 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status);
872 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &mask);
873 }
874 if (status & ~mask)
875 return true;
876
877 return false;
878 }
879
find_device_iter(struct pci_dev * dev,void * data)880 static int find_device_iter(struct pci_dev *dev, void *data)
881 {
882 struct aer_err_info *e_info = (struct aer_err_info *)data;
883
884 if (is_error_source(dev, e_info)) {
885 /* List this device */
886 if (add_error_device(e_info, dev)) {
887 /* We cannot handle more... Stop iteration */
888 /* TODO: Should print error message here? */
889 return 1;
890 }
891
892 /* If there is only a single error, stop iteration */
893 if (!e_info->multi_error_valid)
894 return 1;
895 }
896 return 0;
897 }
898
899 /**
900 * find_source_device - search through device hierarchy for source device
901 * @parent: pointer to Root Port pci_dev data structure
902 * @e_info: including detailed error information such like id
903 *
904 * Return true if found.
905 *
906 * Invoked by DPC when error is detected at the Root Port.
907 * Caller of this function must set id, severity, and multi_error_valid of
908 * struct aer_err_info pointed by @e_info properly. This function must fill
909 * e_info->error_dev_num and e_info->dev[], based on the given information.
910 */
find_source_device(struct pci_dev * parent,struct aer_err_info * e_info)911 static bool find_source_device(struct pci_dev *parent,
912 struct aer_err_info *e_info)
913 {
914 struct pci_dev *dev = parent;
915 int result;
916
917 /* Must reset in this function */
918 e_info->error_dev_num = 0;
919
920 /* Is Root Port an agent that sends error message? */
921 result = find_device_iter(dev, e_info);
922 if (result)
923 return true;
924
925 if (pci_pcie_type(parent) == PCI_EXP_TYPE_RC_EC)
926 pcie_walk_rcec(parent, find_device_iter, e_info);
927 else
928 pci_walk_bus(parent->subordinate, find_device_iter, e_info);
929
930 if (!e_info->error_dev_num) {
931 u8 bus = e_info->id >> 8;
932 u8 devfn = e_info->id & 0xff;
933
934 pci_info(parent, "found no error details for %04x:%02x:%02x.%d\n",
935 pci_domain_nr(parent->bus), bus, PCI_SLOT(devfn),
936 PCI_FUNC(devfn));
937 return false;
938 }
939 return true;
940 }
941
942 /**
943 * handle_error_source - handle logging error into an event log
944 * @dev: pointer to pci_dev data structure of error source device
945 * @info: comprehensive error information
946 *
947 * Invoked when an error being detected by Root Port.
948 */
handle_error_source(struct pci_dev * dev,struct aer_err_info * info)949 static void handle_error_source(struct pci_dev *dev, struct aer_err_info *info)
950 {
951 int aer = dev->aer_cap;
952
953 if (info->severity == AER_CORRECTABLE) {
954 /*
955 * Correctable error does not need software intervention.
956 * No need to go through error recovery process.
957 */
958 if (aer)
959 pci_write_config_dword(dev, aer + PCI_ERR_COR_STATUS,
960 info->status);
961 if (pcie_aer_is_native(dev)) {
962 struct pci_driver *pdrv = dev->driver;
963
964 if (pdrv && pdrv->err_handler &&
965 pdrv->err_handler->cor_error_detected)
966 pdrv->err_handler->cor_error_detected(dev);
967 pcie_clear_device_status(dev);
968 }
969 } else if (info->severity == AER_NONFATAL)
970 pcie_do_recovery(dev, pci_channel_io_normal, aer_root_reset);
971 else if (info->severity == AER_FATAL)
972 pcie_do_recovery(dev, pci_channel_io_frozen, aer_root_reset);
973 pci_dev_put(dev);
974 }
975
976 #ifdef CONFIG_ACPI_APEI_PCIEAER
977
978 #define AER_RECOVER_RING_SIZE 16
979
980 struct aer_recover_entry {
981 u8 bus;
982 u8 devfn;
983 u16 domain;
984 int severity;
985 struct aer_capability_regs *regs;
986 };
987
988 static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
989 AER_RECOVER_RING_SIZE);
990
aer_recover_work_func(struct work_struct * work)991 static void aer_recover_work_func(struct work_struct *work)
992 {
993 struct aer_recover_entry entry;
994 struct pci_dev *pdev;
995
996 while (kfifo_get(&aer_recover_ring, &entry)) {
997 pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
998 entry.devfn);
999 if (!pdev) {
1000 pr_err("no pci_dev for %04x:%02x:%02x.%x\n",
1001 entry.domain, entry.bus,
1002 PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
1003 continue;
1004 }
1005 cper_print_aer(pdev, entry.severity, entry.regs);
1006 /*
1007 * Memory for aer_capability_regs(entry.regs) is being allocated from the
1008 * ghes_estatus_pool to protect it from overwriting when multiple sections
1009 * are present in the error status. Thus free the same after processing
1010 * the data.
1011 */
1012 ghes_estatus_pool_region_free((unsigned long)entry.regs,
1013 sizeof(struct aer_capability_regs));
1014
1015 if (entry.severity == AER_NONFATAL)
1016 pcie_do_recovery(pdev, pci_channel_io_normal,
1017 aer_root_reset);
1018 else if (entry.severity == AER_FATAL)
1019 pcie_do_recovery(pdev, pci_channel_io_frozen,
1020 aer_root_reset);
1021 pci_dev_put(pdev);
1022 }
1023 }
1024
1025 /*
1026 * Mutual exclusion for writers of aer_recover_ring, reader side don't
1027 * need lock, because there is only one reader and lock is not needed
1028 * between reader and writer.
1029 */
1030 static DEFINE_SPINLOCK(aer_recover_ring_lock);
1031 static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
1032
aer_recover_queue(int domain,unsigned int bus,unsigned int devfn,int severity,struct aer_capability_regs * aer_regs)1033 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
1034 int severity, struct aer_capability_regs *aer_regs)
1035 {
1036 struct aer_recover_entry entry = {
1037 .bus = bus,
1038 .devfn = devfn,
1039 .domain = domain,
1040 .severity = severity,
1041 .regs = aer_regs,
1042 };
1043
1044 if (kfifo_in_spinlocked(&aer_recover_ring, &entry, 1,
1045 &aer_recover_ring_lock))
1046 schedule_work(&aer_recover_work);
1047 else
1048 pr_err("buffer overflow in recovery for %04x:%02x:%02x.%x\n",
1049 domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
1050 }
1051 EXPORT_SYMBOL_GPL(aer_recover_queue);
1052 #endif
1053
1054 /**
1055 * aer_get_device_error_info - read error status from dev and store it to info
1056 * @dev: pointer to the device expected to have a error record
1057 * @info: pointer to structure to store the error record
1058 *
1059 * Return 1 on success, 0 on error.
1060 *
1061 * Note that @info is reused among all error devices. Clear fields properly.
1062 */
aer_get_device_error_info(struct pci_dev * dev,struct aer_err_info * info)1063 int aer_get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
1064 {
1065 int type = pci_pcie_type(dev);
1066 int aer = dev->aer_cap;
1067 int temp;
1068
1069 /* Must reset in this function */
1070 info->status = 0;
1071 info->tlp_header_valid = 0;
1072
1073 /* The device might not support AER */
1074 if (!aer)
1075 return 0;
1076
1077 if (info->severity == AER_CORRECTABLE) {
1078 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS,
1079 &info->status);
1080 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK,
1081 &info->mask);
1082 if (!(info->status & ~info->mask))
1083 return 0;
1084 } else if (type == PCI_EXP_TYPE_ROOT_PORT ||
1085 type == PCI_EXP_TYPE_RC_EC ||
1086 type == PCI_EXP_TYPE_DOWNSTREAM ||
1087 info->severity == AER_NONFATAL) {
1088
1089 /* Link is still healthy for IO reads */
1090 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS,
1091 &info->status);
1092 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK,
1093 &info->mask);
1094 if (!(info->status & ~info->mask))
1095 return 0;
1096
1097 /* Get First Error Pointer */
1098 pci_read_config_dword(dev, aer + PCI_ERR_CAP, &temp);
1099 info->first_error = PCI_ERR_CAP_FEP(temp);
1100
1101 if (info->status & AER_LOG_TLP_MASKS) {
1102 info->tlp_header_valid = 1;
1103 pci_read_config_dword(dev,
1104 aer + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
1105 pci_read_config_dword(dev,
1106 aer + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
1107 pci_read_config_dword(dev,
1108 aer + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
1109 pci_read_config_dword(dev,
1110 aer + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
1111 }
1112 }
1113
1114 return 1;
1115 }
1116
aer_process_err_devices(struct aer_err_info * e_info)1117 static inline void aer_process_err_devices(struct aer_err_info *e_info)
1118 {
1119 int i;
1120
1121 /* Report all before handle them, not to lost records by reset etc. */
1122 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
1123 if (aer_get_device_error_info(e_info->dev[i], e_info))
1124 aer_print_error(e_info->dev[i], e_info);
1125 }
1126 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
1127 if (aer_get_device_error_info(e_info->dev[i], e_info))
1128 handle_error_source(e_info->dev[i], e_info);
1129 }
1130 }
1131
1132 /**
1133 * aer_isr_one_error - consume an error detected by root port
1134 * @rpc: pointer to the root port which holds an error
1135 * @e_src: pointer to an error source
1136 */
aer_isr_one_error(struct aer_rpc * rpc,struct aer_err_source * e_src)1137 static void aer_isr_one_error(struct aer_rpc *rpc,
1138 struct aer_err_source *e_src)
1139 {
1140 struct pci_dev *pdev = rpc->rpd;
1141 struct aer_err_info e_info;
1142
1143 pci_rootport_aer_stats_incr(pdev, e_src);
1144
1145 /*
1146 * There is a possibility that both correctable error and
1147 * uncorrectable error being logged. Report correctable error first.
1148 */
1149 if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
1150 e_info.id = ERR_COR_ID(e_src->id);
1151 e_info.severity = AER_CORRECTABLE;
1152
1153 if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
1154 e_info.multi_error_valid = 1;
1155 else
1156 e_info.multi_error_valid = 0;
1157 aer_print_port_info(pdev, &e_info);
1158
1159 if (find_source_device(pdev, &e_info))
1160 aer_process_err_devices(&e_info);
1161 }
1162
1163 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
1164 e_info.id = ERR_UNCOR_ID(e_src->id);
1165
1166 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
1167 e_info.severity = AER_FATAL;
1168 else
1169 e_info.severity = AER_NONFATAL;
1170
1171 if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
1172 e_info.multi_error_valid = 1;
1173 else
1174 e_info.multi_error_valid = 0;
1175
1176 aer_print_port_info(pdev, &e_info);
1177
1178 if (find_source_device(pdev, &e_info))
1179 aer_process_err_devices(&e_info);
1180 }
1181 }
1182
1183 /**
1184 * aer_isr - consume errors detected by root port
1185 * @irq: IRQ assigned to Root Port
1186 * @context: pointer to Root Port data structure
1187 *
1188 * Invoked, as DPC, when root port records new detected error
1189 */
aer_isr(int irq,void * context)1190 static irqreturn_t aer_isr(int irq, void *context)
1191 {
1192 struct pcie_device *dev = (struct pcie_device *)context;
1193 struct aer_rpc *rpc = get_service_data(dev);
1194 struct aer_err_source e_src;
1195
1196 if (kfifo_is_empty(&rpc->aer_fifo))
1197 return IRQ_NONE;
1198
1199 while (kfifo_get(&rpc->aer_fifo, &e_src))
1200 aer_isr_one_error(rpc, &e_src);
1201 return IRQ_HANDLED;
1202 }
1203
1204 /**
1205 * aer_irq - Root Port's ISR
1206 * @irq: IRQ assigned to Root Port
1207 * @context: pointer to Root Port data structure
1208 *
1209 * Invoked when Root Port detects AER messages.
1210 */
aer_irq(int irq,void * context)1211 static irqreturn_t aer_irq(int irq, void *context)
1212 {
1213 struct pcie_device *pdev = (struct pcie_device *)context;
1214 struct aer_rpc *rpc = get_service_data(pdev);
1215 struct pci_dev *rp = rpc->rpd;
1216 int aer = rp->aer_cap;
1217 struct aer_err_source e_src = {};
1218
1219 pci_read_config_dword(rp, aer + PCI_ERR_ROOT_STATUS, &e_src.status);
1220 if (!(e_src.status & AER_ERR_STATUS_MASK))
1221 return IRQ_NONE;
1222
1223 pci_read_config_dword(rp, aer + PCI_ERR_ROOT_ERR_SRC, &e_src.id);
1224 pci_write_config_dword(rp, aer + PCI_ERR_ROOT_STATUS, e_src.status);
1225
1226 if (!kfifo_put(&rpc->aer_fifo, e_src))
1227 return IRQ_HANDLED;
1228
1229 return IRQ_WAKE_THREAD;
1230 }
1231
1232 /**
1233 * aer_enable_rootport - enable Root Port's interrupts when receiving messages
1234 * @rpc: pointer to a Root Port data structure
1235 *
1236 * Invoked when PCIe bus loads AER service driver.
1237 */
aer_enable_rootport(struct aer_rpc * rpc)1238 static void aer_enable_rootport(struct aer_rpc *rpc)
1239 {
1240 struct pci_dev *pdev = rpc->rpd;
1241 int aer = pdev->aer_cap;
1242 u16 reg16;
1243 u32 reg32;
1244
1245 /* Clear PCIe Capability's Device Status */
1246 pcie_capability_read_word(pdev, PCI_EXP_DEVSTA, ®16);
1247 pcie_capability_write_word(pdev, PCI_EXP_DEVSTA, reg16);
1248
1249 /* Disable system error generation in response to error messages */
1250 pcie_capability_clear_word(pdev, PCI_EXP_RTCTL,
1251 SYSTEM_ERROR_INTR_ON_MESG_MASK);
1252
1253 /* Clear error status */
1254 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, ®32);
1255 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, reg32);
1256 pci_read_config_dword(pdev, aer + PCI_ERR_COR_STATUS, ®32);
1257 pci_write_config_dword(pdev, aer + PCI_ERR_COR_STATUS, reg32);
1258 pci_read_config_dword(pdev, aer + PCI_ERR_UNCOR_STATUS, ®32);
1259 pci_write_config_dword(pdev, aer + PCI_ERR_UNCOR_STATUS, reg32);
1260
1261 /* Enable Root Port's interrupt in response to error messages */
1262 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, ®32);
1263 reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
1264 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, reg32);
1265 }
1266
1267 /**
1268 * aer_disable_rootport - disable Root Port's interrupts when receiving messages
1269 * @rpc: pointer to a Root Port data structure
1270 *
1271 * Invoked when PCIe bus unloads AER service driver.
1272 */
aer_disable_rootport(struct aer_rpc * rpc)1273 static void aer_disable_rootport(struct aer_rpc *rpc)
1274 {
1275 struct pci_dev *pdev = rpc->rpd;
1276 int aer = pdev->aer_cap;
1277 u32 reg32;
1278
1279 /* Disable Root's interrupt in response to error messages */
1280 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, ®32);
1281 reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
1282 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, reg32);
1283
1284 /* Clear Root's error status reg */
1285 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, ®32);
1286 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, reg32);
1287 }
1288
1289 /**
1290 * aer_remove - clean up resources
1291 * @dev: pointer to the pcie_dev data structure
1292 *
1293 * Invoked when PCI Express bus unloads or AER probe fails.
1294 */
aer_remove(struct pcie_device * dev)1295 static void aer_remove(struct pcie_device *dev)
1296 {
1297 struct aer_rpc *rpc = get_service_data(dev);
1298
1299 aer_disable_rootport(rpc);
1300 }
1301
1302 /**
1303 * aer_probe - initialize resources
1304 * @dev: pointer to the pcie_dev data structure
1305 *
1306 * Invoked when PCI Express bus loads AER service driver.
1307 */
aer_probe(struct pcie_device * dev)1308 static int aer_probe(struct pcie_device *dev)
1309 {
1310 int status;
1311 struct aer_rpc *rpc;
1312 struct device *device = &dev->device;
1313 struct pci_dev *port = dev->port;
1314
1315 BUILD_BUG_ON(ARRAY_SIZE(aer_correctable_error_string) <
1316 AER_MAX_TYPEOF_COR_ERRS);
1317 BUILD_BUG_ON(ARRAY_SIZE(aer_uncorrectable_error_string) <
1318 AER_MAX_TYPEOF_UNCOR_ERRS);
1319
1320 /* Limit to Root Ports or Root Complex Event Collectors */
1321 if ((pci_pcie_type(port) != PCI_EXP_TYPE_RC_EC) &&
1322 (pci_pcie_type(port) != PCI_EXP_TYPE_ROOT_PORT))
1323 return -ENODEV;
1324
1325 rpc = devm_kzalloc(device, sizeof(struct aer_rpc), GFP_KERNEL);
1326 if (!rpc)
1327 return -ENOMEM;
1328
1329 rpc->rpd = port;
1330 INIT_KFIFO(rpc->aer_fifo);
1331 set_service_data(dev, rpc);
1332
1333 status = devm_request_threaded_irq(device, dev->irq, aer_irq, aer_isr,
1334 IRQF_SHARED, "aerdrv", dev);
1335 if (status) {
1336 pci_err(port, "request AER IRQ %d failed\n", dev->irq);
1337 return status;
1338 }
1339
1340 aer_enable_rootport(rpc);
1341 pci_info(port, "enabled with IRQ %d\n", dev->irq);
1342 return 0;
1343 }
1344
1345 /**
1346 * aer_root_reset - reset Root Port hierarchy, RCEC, or RCiEP
1347 * @dev: pointer to Root Port, RCEC, or RCiEP
1348 *
1349 * Invoked by Port Bus driver when performing reset.
1350 */
aer_root_reset(struct pci_dev * dev)1351 static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
1352 {
1353 int type = pci_pcie_type(dev);
1354 struct pci_dev *root;
1355 int aer;
1356 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
1357 u32 reg32;
1358 int rc;
1359
1360 /*
1361 * Only Root Ports and RCECs have AER Root Command and Root Status
1362 * registers. If "dev" is an RCiEP, the relevant registers are in
1363 * the RCEC.
1364 */
1365 if (type == PCI_EXP_TYPE_RC_END)
1366 root = dev->rcec;
1367 else
1368 root = pcie_find_root_port(dev);
1369
1370 /*
1371 * If the platform retained control of AER, an RCiEP may not have
1372 * an RCEC visible to us, so dev->rcec ("root") may be NULL. In
1373 * that case, firmware is responsible for these registers.
1374 */
1375 aer = root ? root->aer_cap : 0;
1376
1377 if ((host->native_aer || pcie_ports_native) && aer) {
1378 /* Disable Root's interrupt in response to error messages */
1379 pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, ®32);
1380 reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
1381 pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
1382 }
1383
1384 if (type == PCI_EXP_TYPE_RC_EC || type == PCI_EXP_TYPE_RC_END) {
1385 rc = pcie_reset_flr(dev, PCI_RESET_DO_RESET);
1386 if (!rc)
1387 pci_info(dev, "has been reset\n");
1388 else
1389 pci_info(dev, "not reset (no FLR support: %d)\n", rc);
1390 } else {
1391 rc = pci_bus_error_reset(dev);
1392 pci_info(dev, "%s Port link has been reset (%d)\n",
1393 pci_is_root_bus(dev->bus) ? "Root" : "Downstream", rc);
1394 }
1395
1396 if ((host->native_aer || pcie_ports_native) && aer) {
1397 /* Clear Root Error Status */
1398 pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, ®32);
1399 pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
1400
1401 /* Enable Root Port's interrupt in response to error messages */
1402 pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, ®32);
1403 reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
1404 pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
1405 }
1406
1407 return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
1408 }
1409
1410 static struct pcie_port_service_driver aerdriver = {
1411 .name = "aer",
1412 .port_type = PCIE_ANY_PORT,
1413 .service = PCIE_PORT_SERVICE_AER,
1414
1415 .probe = aer_probe,
1416 .remove = aer_remove,
1417 };
1418
1419 /**
1420 * pcie_aer_init - register AER root service driver
1421 *
1422 * Invoked when AER root service driver is loaded.
1423 */
pcie_aer_init(void)1424 int __init pcie_aer_init(void)
1425 {
1426 if (!pci_aer_available())
1427 return -ENXIO;
1428 return pcie_port_service_register(&aerdriver);
1429 }
1430