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