xref: /openbmc/linux/drivers/acpi/apei/erst.c (revision 56280682)
1 /*
2  * APEI Error Record Serialization Table support
3  *
4  * ERST is a way provided by APEI to save and retrieve hardware error
5  * information to and from a persistent store.
6  *
7  * For more information about ERST, please refer to ACPI Specification
8  * version 4.0, section 17.4.
9  *
10  * Copyright 2010 Intel Corp.
11  *   Author: Huang Ying <ying.huang@intel.com>
12  *
13  * This program is free software; you can redistribute it and/or
14  * modify it under the terms of the GNU General Public License version
15  * 2 as published by the Free Software Foundation.
16  *
17  * This program is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20  * GNU General Public License for more details.
21  *
22  * You should have received a copy of the GNU General Public License
23  * along with this program; if not, write to the Free Software
24  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
25  */
26 
27 #include <linux/kernel.h>
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/delay.h>
31 #include <linux/io.h>
32 #include <linux/acpi.h>
33 #include <linux/uaccess.h>
34 #include <linux/cper.h>
35 #include <linux/nmi.h>
36 #include <linux/hardirq.h>
37 #include <linux/pstore.h>
38 #include <acpi/apei.h>
39 
40 #include "apei-internal.h"
41 
42 #define ERST_PFX "ERST: "
43 
44 /* ERST command status */
45 #define ERST_STATUS_SUCCESS			0x0
46 #define ERST_STATUS_NOT_ENOUGH_SPACE		0x1
47 #define ERST_STATUS_HARDWARE_NOT_AVAILABLE	0x2
48 #define ERST_STATUS_FAILED			0x3
49 #define ERST_STATUS_RECORD_STORE_EMPTY		0x4
50 #define ERST_STATUS_RECORD_NOT_FOUND		0x5
51 
52 #define ERST_TAB_ENTRY(tab)						\
53 	((struct acpi_whea_header *)((char *)(tab) +			\
54 				     sizeof(struct acpi_table_erst)))
55 
56 #define SPIN_UNIT		100			/* 100ns */
57 /* Firmware should respond within 1 milliseconds */
58 #define FIRMWARE_TIMEOUT	(1 * NSEC_PER_MSEC)
59 #define FIRMWARE_MAX_STALL	50			/* 50us */
60 
61 int erst_disable;
62 EXPORT_SYMBOL_GPL(erst_disable);
63 
64 static struct acpi_table_erst *erst_tab;
65 
66 /* ERST Error Log Address Range atrributes */
67 #define ERST_RANGE_RESERVED	0x0001
68 #define ERST_RANGE_NVRAM	0x0002
69 #define ERST_RANGE_SLOW		0x0004
70 
71 /*
72  * ERST Error Log Address Range, used as buffer for reading/writing
73  * error records.
74  */
75 static struct erst_erange {
76 	u64 base;
77 	u64 size;
78 	void __iomem *vaddr;
79 	u32 attr;
80 } erst_erange;
81 
82 /*
83  * Prevent ERST interpreter to run simultaneously, because the
84  * corresponding firmware implementation may not work properly when
85  * invoked simultaneously.
86  *
87  * It is used to provide exclusive accessing for ERST Error Log
88  * Address Range too.
89  */
90 static DEFINE_RAW_SPINLOCK(erst_lock);
91 
92 static inline int erst_errno(int command_status)
93 {
94 	switch (command_status) {
95 	case ERST_STATUS_SUCCESS:
96 		return 0;
97 	case ERST_STATUS_HARDWARE_NOT_AVAILABLE:
98 		return -ENODEV;
99 	case ERST_STATUS_NOT_ENOUGH_SPACE:
100 		return -ENOSPC;
101 	case ERST_STATUS_RECORD_STORE_EMPTY:
102 	case ERST_STATUS_RECORD_NOT_FOUND:
103 		return -ENOENT;
104 	default:
105 		return -EINVAL;
106 	}
107 }
108 
109 static int erst_timedout(u64 *t, u64 spin_unit)
110 {
111 	if ((s64)*t < spin_unit) {
112 		pr_warning(FW_WARN ERST_PFX
113 			   "Firmware does not respond in time\n");
114 		return 1;
115 	}
116 	*t -= spin_unit;
117 	ndelay(spin_unit);
118 	touch_nmi_watchdog();
119 	return 0;
120 }
121 
122 static int erst_exec_load_var1(struct apei_exec_context *ctx,
123 			       struct acpi_whea_header *entry)
124 {
125 	return __apei_exec_read_register(entry, &ctx->var1);
126 }
127 
128 static int erst_exec_load_var2(struct apei_exec_context *ctx,
129 			       struct acpi_whea_header *entry)
130 {
131 	return __apei_exec_read_register(entry, &ctx->var2);
132 }
133 
134 static int erst_exec_store_var1(struct apei_exec_context *ctx,
135 				struct acpi_whea_header *entry)
136 {
137 	return __apei_exec_write_register(entry, ctx->var1);
138 }
139 
140 static int erst_exec_add(struct apei_exec_context *ctx,
141 			 struct acpi_whea_header *entry)
142 {
143 	ctx->var1 += ctx->var2;
144 	return 0;
145 }
146 
147 static int erst_exec_subtract(struct apei_exec_context *ctx,
148 			      struct acpi_whea_header *entry)
149 {
150 	ctx->var1 -= ctx->var2;
151 	return 0;
152 }
153 
154 static int erst_exec_add_value(struct apei_exec_context *ctx,
155 			       struct acpi_whea_header *entry)
156 {
157 	int rc;
158 	u64 val;
159 
160 	rc = __apei_exec_read_register(entry, &val);
161 	if (rc)
162 		return rc;
163 	val += ctx->value;
164 	rc = __apei_exec_write_register(entry, val);
165 	return rc;
166 }
167 
168 static int erst_exec_subtract_value(struct apei_exec_context *ctx,
169 				    struct acpi_whea_header *entry)
170 {
171 	int rc;
172 	u64 val;
173 
174 	rc = __apei_exec_read_register(entry, &val);
175 	if (rc)
176 		return rc;
177 	val -= ctx->value;
178 	rc = __apei_exec_write_register(entry, val);
179 	return rc;
180 }
181 
182 static int erst_exec_stall(struct apei_exec_context *ctx,
183 			   struct acpi_whea_header *entry)
184 {
185 	u64 stall_time;
186 
187 	if (ctx->value > FIRMWARE_MAX_STALL) {
188 		if (!in_nmi())
189 			pr_warning(FW_WARN ERST_PFX
190 			"Too long stall time for stall instruction: %llx.\n",
191 				   ctx->value);
192 		stall_time = FIRMWARE_MAX_STALL;
193 	} else
194 		stall_time = ctx->value;
195 	udelay(stall_time);
196 	return 0;
197 }
198 
199 static int erst_exec_stall_while_true(struct apei_exec_context *ctx,
200 				      struct acpi_whea_header *entry)
201 {
202 	int rc;
203 	u64 val;
204 	u64 timeout = FIRMWARE_TIMEOUT;
205 	u64 stall_time;
206 
207 	if (ctx->var1 > FIRMWARE_MAX_STALL) {
208 		if (!in_nmi())
209 			pr_warning(FW_WARN ERST_PFX
210 		"Too long stall time for stall while true instruction: %llx.\n",
211 				   ctx->var1);
212 		stall_time = FIRMWARE_MAX_STALL;
213 	} else
214 		stall_time = ctx->var1;
215 
216 	for (;;) {
217 		rc = __apei_exec_read_register(entry, &val);
218 		if (rc)
219 			return rc;
220 		if (val != ctx->value)
221 			break;
222 		if (erst_timedout(&timeout, stall_time * NSEC_PER_USEC))
223 			return -EIO;
224 	}
225 	return 0;
226 }
227 
228 static int erst_exec_skip_next_instruction_if_true(
229 	struct apei_exec_context *ctx,
230 	struct acpi_whea_header *entry)
231 {
232 	int rc;
233 	u64 val;
234 
235 	rc = __apei_exec_read_register(entry, &val);
236 	if (rc)
237 		return rc;
238 	if (val == ctx->value) {
239 		ctx->ip += 2;
240 		return APEI_EXEC_SET_IP;
241 	}
242 
243 	return 0;
244 }
245 
246 static int erst_exec_goto(struct apei_exec_context *ctx,
247 			  struct acpi_whea_header *entry)
248 {
249 	ctx->ip = ctx->value;
250 	return APEI_EXEC_SET_IP;
251 }
252 
253 static int erst_exec_set_src_address_base(struct apei_exec_context *ctx,
254 					  struct acpi_whea_header *entry)
255 {
256 	return __apei_exec_read_register(entry, &ctx->src_base);
257 }
258 
259 static int erst_exec_set_dst_address_base(struct apei_exec_context *ctx,
260 					  struct acpi_whea_header *entry)
261 {
262 	return __apei_exec_read_register(entry, &ctx->dst_base);
263 }
264 
265 static int erst_exec_move_data(struct apei_exec_context *ctx,
266 			       struct acpi_whea_header *entry)
267 {
268 	int rc;
269 	u64 offset;
270 	void *src, *dst;
271 
272 	/* ioremap does not work in interrupt context */
273 	if (in_interrupt()) {
274 		pr_warning(ERST_PFX
275 			   "MOVE_DATA can not be used in interrupt context");
276 		return -EBUSY;
277 	}
278 
279 	rc = __apei_exec_read_register(entry, &offset);
280 	if (rc)
281 		return rc;
282 
283 	src = ioremap(ctx->src_base + offset, ctx->var2);
284 	if (!src)
285 		return -ENOMEM;
286 	dst = ioremap(ctx->dst_base + offset, ctx->var2);
287 	if (!dst)
288 		return -ENOMEM;
289 
290 	memmove(dst, src, ctx->var2);
291 
292 	iounmap(src);
293 	iounmap(dst);
294 
295 	return 0;
296 }
297 
298 static struct apei_exec_ins_type erst_ins_type[] = {
299 	[ACPI_ERST_READ_REGISTER] = {
300 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
301 		.run = apei_exec_read_register,
302 	},
303 	[ACPI_ERST_READ_REGISTER_VALUE] = {
304 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
305 		.run = apei_exec_read_register_value,
306 	},
307 	[ACPI_ERST_WRITE_REGISTER] = {
308 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
309 		.run = apei_exec_write_register,
310 	},
311 	[ACPI_ERST_WRITE_REGISTER_VALUE] = {
312 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
313 		.run = apei_exec_write_register_value,
314 	},
315 	[ACPI_ERST_NOOP] = {
316 		.flags = 0,
317 		.run = apei_exec_noop,
318 	},
319 	[ACPI_ERST_LOAD_VAR1] = {
320 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
321 		.run = erst_exec_load_var1,
322 	},
323 	[ACPI_ERST_LOAD_VAR2] = {
324 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
325 		.run = erst_exec_load_var2,
326 	},
327 	[ACPI_ERST_STORE_VAR1] = {
328 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
329 		.run = erst_exec_store_var1,
330 	},
331 	[ACPI_ERST_ADD] = {
332 		.flags = 0,
333 		.run = erst_exec_add,
334 	},
335 	[ACPI_ERST_SUBTRACT] = {
336 		.flags = 0,
337 		.run = erst_exec_subtract,
338 	},
339 	[ACPI_ERST_ADD_VALUE] = {
340 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
341 		.run = erst_exec_add_value,
342 	},
343 	[ACPI_ERST_SUBTRACT_VALUE] = {
344 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
345 		.run = erst_exec_subtract_value,
346 	},
347 	[ACPI_ERST_STALL] = {
348 		.flags = 0,
349 		.run = erst_exec_stall,
350 	},
351 	[ACPI_ERST_STALL_WHILE_TRUE] = {
352 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
353 		.run = erst_exec_stall_while_true,
354 	},
355 	[ACPI_ERST_SKIP_NEXT_IF_TRUE] = {
356 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
357 		.run = erst_exec_skip_next_instruction_if_true,
358 	},
359 	[ACPI_ERST_GOTO] = {
360 		.flags = 0,
361 		.run = erst_exec_goto,
362 	},
363 	[ACPI_ERST_SET_SRC_ADDRESS_BASE] = {
364 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
365 		.run = erst_exec_set_src_address_base,
366 	},
367 	[ACPI_ERST_SET_DST_ADDRESS_BASE] = {
368 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
369 		.run = erst_exec_set_dst_address_base,
370 	},
371 	[ACPI_ERST_MOVE_DATA] = {
372 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
373 		.run = erst_exec_move_data,
374 	},
375 };
376 
377 static inline void erst_exec_ctx_init(struct apei_exec_context *ctx)
378 {
379 	apei_exec_ctx_init(ctx, erst_ins_type, ARRAY_SIZE(erst_ins_type),
380 			   ERST_TAB_ENTRY(erst_tab), erst_tab->entries);
381 }
382 
383 static int erst_get_erange(struct erst_erange *range)
384 {
385 	struct apei_exec_context ctx;
386 	int rc;
387 
388 	erst_exec_ctx_init(&ctx);
389 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_RANGE);
390 	if (rc)
391 		return rc;
392 	range->base = apei_exec_ctx_get_output(&ctx);
393 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_LENGTH);
394 	if (rc)
395 		return rc;
396 	range->size = apei_exec_ctx_get_output(&ctx);
397 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_ATTRIBUTES);
398 	if (rc)
399 		return rc;
400 	range->attr = apei_exec_ctx_get_output(&ctx);
401 
402 	return 0;
403 }
404 
405 static ssize_t __erst_get_record_count(void)
406 {
407 	struct apei_exec_context ctx;
408 	int rc;
409 
410 	erst_exec_ctx_init(&ctx);
411 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_RECORD_COUNT);
412 	if (rc)
413 		return rc;
414 	return apei_exec_ctx_get_output(&ctx);
415 }
416 
417 ssize_t erst_get_record_count(void)
418 {
419 	ssize_t count;
420 	unsigned long flags;
421 
422 	if (erst_disable)
423 		return -ENODEV;
424 
425 	raw_spin_lock_irqsave(&erst_lock, flags);
426 	count = __erst_get_record_count();
427 	raw_spin_unlock_irqrestore(&erst_lock, flags);
428 
429 	return count;
430 }
431 EXPORT_SYMBOL_GPL(erst_get_record_count);
432 
433 #define ERST_RECORD_ID_CACHE_SIZE_MIN	16
434 #define ERST_RECORD_ID_CACHE_SIZE_MAX	1024
435 
436 struct erst_record_id_cache {
437 	struct mutex lock;
438 	u64 *entries;
439 	int len;
440 	int size;
441 	int refcount;
442 };
443 
444 static struct erst_record_id_cache erst_record_id_cache = {
445 	.lock = __MUTEX_INITIALIZER(erst_record_id_cache.lock),
446 	.refcount = 0,
447 };
448 
449 static int __erst_get_next_record_id(u64 *record_id)
450 {
451 	struct apei_exec_context ctx;
452 	int rc;
453 
454 	erst_exec_ctx_init(&ctx);
455 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_RECORD_ID);
456 	if (rc)
457 		return rc;
458 	*record_id = apei_exec_ctx_get_output(&ctx);
459 
460 	return 0;
461 }
462 
463 int erst_get_record_id_begin(int *pos)
464 {
465 	int rc;
466 
467 	if (erst_disable)
468 		return -ENODEV;
469 
470 	rc = mutex_lock_interruptible(&erst_record_id_cache.lock);
471 	if (rc)
472 		return rc;
473 	erst_record_id_cache.refcount++;
474 	mutex_unlock(&erst_record_id_cache.lock);
475 
476 	*pos = 0;
477 
478 	return 0;
479 }
480 EXPORT_SYMBOL_GPL(erst_get_record_id_begin);
481 
482 /* erst_record_id_cache.lock must be held by caller */
483 static int __erst_record_id_cache_add_one(void)
484 {
485 	u64 id, prev_id, first_id;
486 	int i, rc;
487 	u64 *entries;
488 	unsigned long flags;
489 
490 	id = prev_id = first_id = APEI_ERST_INVALID_RECORD_ID;
491 retry:
492 	raw_spin_lock_irqsave(&erst_lock, flags);
493 	rc = __erst_get_next_record_id(&id);
494 	raw_spin_unlock_irqrestore(&erst_lock, flags);
495 	if (rc == -ENOENT)
496 		return 0;
497 	if (rc)
498 		return rc;
499 	if (id == APEI_ERST_INVALID_RECORD_ID)
500 		return 0;
501 	/* can not skip current ID, or loop back to first ID */
502 	if (id == prev_id || id == first_id)
503 		return 0;
504 	if (first_id == APEI_ERST_INVALID_RECORD_ID)
505 		first_id = id;
506 	prev_id = id;
507 
508 	entries = erst_record_id_cache.entries;
509 	for (i = 0; i < erst_record_id_cache.len; i++) {
510 		if (entries[i] == id)
511 			break;
512 	}
513 	/* record id already in cache, try next */
514 	if (i < erst_record_id_cache.len)
515 		goto retry;
516 	if (erst_record_id_cache.len >= erst_record_id_cache.size) {
517 		int new_size, alloc_size;
518 		u64 *new_entries;
519 
520 		new_size = erst_record_id_cache.size * 2;
521 		new_size = clamp_val(new_size, ERST_RECORD_ID_CACHE_SIZE_MIN,
522 				     ERST_RECORD_ID_CACHE_SIZE_MAX);
523 		if (new_size <= erst_record_id_cache.size) {
524 			if (printk_ratelimit())
525 				pr_warning(FW_WARN ERST_PFX
526 					   "too many record ID!\n");
527 			return 0;
528 		}
529 		alloc_size = new_size * sizeof(entries[0]);
530 		if (alloc_size < PAGE_SIZE)
531 			new_entries = kmalloc(alloc_size, GFP_KERNEL);
532 		else
533 			new_entries = vmalloc(alloc_size);
534 		if (!new_entries)
535 			return -ENOMEM;
536 		memcpy(new_entries, entries,
537 		       erst_record_id_cache.len * sizeof(entries[0]));
538 		if (erst_record_id_cache.size < PAGE_SIZE)
539 			kfree(entries);
540 		else
541 			vfree(entries);
542 		erst_record_id_cache.entries = entries = new_entries;
543 		erst_record_id_cache.size = new_size;
544 	}
545 	entries[i] = id;
546 	erst_record_id_cache.len++;
547 
548 	return 1;
549 }
550 
551 /*
552  * Get the record ID of an existing error record on the persistent
553  * storage. If there is no error record on the persistent storage, the
554  * returned record_id is APEI_ERST_INVALID_RECORD_ID.
555  */
556 int erst_get_record_id_next(int *pos, u64 *record_id)
557 {
558 	int rc = 0;
559 	u64 *entries;
560 
561 	if (erst_disable)
562 		return -ENODEV;
563 
564 	/* must be enclosed by erst_get_record_id_begin/end */
565 	BUG_ON(!erst_record_id_cache.refcount);
566 	BUG_ON(*pos < 0 || *pos > erst_record_id_cache.len);
567 
568 	mutex_lock(&erst_record_id_cache.lock);
569 	entries = erst_record_id_cache.entries;
570 	for (; *pos < erst_record_id_cache.len; (*pos)++)
571 		if (entries[*pos] != APEI_ERST_INVALID_RECORD_ID)
572 			break;
573 	/* found next record id in cache */
574 	if (*pos < erst_record_id_cache.len) {
575 		*record_id = entries[*pos];
576 		(*pos)++;
577 		goto out_unlock;
578 	}
579 
580 	/* Try to add one more record ID to cache */
581 	rc = __erst_record_id_cache_add_one();
582 	if (rc < 0)
583 		goto out_unlock;
584 	/* successfully add one new ID */
585 	if (rc == 1) {
586 		*record_id = erst_record_id_cache.entries[*pos];
587 		(*pos)++;
588 		rc = 0;
589 	} else {
590 		*pos = -1;
591 		*record_id = APEI_ERST_INVALID_RECORD_ID;
592 	}
593 out_unlock:
594 	mutex_unlock(&erst_record_id_cache.lock);
595 
596 	return rc;
597 }
598 EXPORT_SYMBOL_GPL(erst_get_record_id_next);
599 
600 /* erst_record_id_cache.lock must be held by caller */
601 static void __erst_record_id_cache_compact(void)
602 {
603 	int i, wpos = 0;
604 	u64 *entries;
605 
606 	if (erst_record_id_cache.refcount)
607 		return;
608 
609 	entries = erst_record_id_cache.entries;
610 	for (i = 0; i < erst_record_id_cache.len; i++) {
611 		if (entries[i] == APEI_ERST_INVALID_RECORD_ID)
612 			continue;
613 		if (wpos != i)
614 			memcpy(&entries[wpos], &entries[i], sizeof(entries[i]));
615 		wpos++;
616 	}
617 	erst_record_id_cache.len = wpos;
618 }
619 
620 void erst_get_record_id_end(void)
621 {
622 	/*
623 	 * erst_disable != 0 should be detected by invoker via the
624 	 * return value of erst_get_record_id_begin/next, so this
625 	 * function should not be called for erst_disable != 0.
626 	 */
627 	BUG_ON(erst_disable);
628 
629 	mutex_lock(&erst_record_id_cache.lock);
630 	erst_record_id_cache.refcount--;
631 	BUG_ON(erst_record_id_cache.refcount < 0);
632 	__erst_record_id_cache_compact();
633 	mutex_unlock(&erst_record_id_cache.lock);
634 }
635 EXPORT_SYMBOL_GPL(erst_get_record_id_end);
636 
637 static int __erst_write_to_storage(u64 offset)
638 {
639 	struct apei_exec_context ctx;
640 	u64 timeout = FIRMWARE_TIMEOUT;
641 	u64 val;
642 	int rc;
643 
644 	erst_exec_ctx_init(&ctx);
645 	rc = apei_exec_run(&ctx, ACPI_ERST_BEGIN_WRITE);
646 	if (rc)
647 		return rc;
648 	apei_exec_ctx_set_input(&ctx, offset);
649 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_OFFSET);
650 	if (rc)
651 		return rc;
652 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
653 	if (rc)
654 		return rc;
655 	for (;;) {
656 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
657 		if (rc)
658 			return rc;
659 		val = apei_exec_ctx_get_output(&ctx);
660 		if (!val)
661 			break;
662 		if (erst_timedout(&timeout, SPIN_UNIT))
663 			return -EIO;
664 	}
665 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
666 	if (rc)
667 		return rc;
668 	val = apei_exec_ctx_get_output(&ctx);
669 	rc = apei_exec_run(&ctx, ACPI_ERST_END);
670 	if (rc)
671 		return rc;
672 
673 	return erst_errno(val);
674 }
675 
676 static int __erst_read_from_storage(u64 record_id, u64 offset)
677 {
678 	struct apei_exec_context ctx;
679 	u64 timeout = FIRMWARE_TIMEOUT;
680 	u64 val;
681 	int rc;
682 
683 	erst_exec_ctx_init(&ctx);
684 	rc = apei_exec_run(&ctx, ACPI_ERST_BEGIN_READ);
685 	if (rc)
686 		return rc;
687 	apei_exec_ctx_set_input(&ctx, offset);
688 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_OFFSET);
689 	if (rc)
690 		return rc;
691 	apei_exec_ctx_set_input(&ctx, record_id);
692 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_ID);
693 	if (rc)
694 		return rc;
695 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
696 	if (rc)
697 		return rc;
698 	for (;;) {
699 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
700 		if (rc)
701 			return rc;
702 		val = apei_exec_ctx_get_output(&ctx);
703 		if (!val)
704 			break;
705 		if (erst_timedout(&timeout, SPIN_UNIT))
706 			return -EIO;
707 	};
708 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
709 	if (rc)
710 		return rc;
711 	val = apei_exec_ctx_get_output(&ctx);
712 	rc = apei_exec_run(&ctx, ACPI_ERST_END);
713 	if (rc)
714 		return rc;
715 
716 	return erst_errno(val);
717 }
718 
719 static int __erst_clear_from_storage(u64 record_id)
720 {
721 	struct apei_exec_context ctx;
722 	u64 timeout = FIRMWARE_TIMEOUT;
723 	u64 val;
724 	int rc;
725 
726 	erst_exec_ctx_init(&ctx);
727 	rc = apei_exec_run(&ctx, ACPI_ERST_BEGIN_CLEAR);
728 	if (rc)
729 		return rc;
730 	apei_exec_ctx_set_input(&ctx, record_id);
731 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_ID);
732 	if (rc)
733 		return rc;
734 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
735 	if (rc)
736 		return rc;
737 	for (;;) {
738 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
739 		if (rc)
740 			return rc;
741 		val = apei_exec_ctx_get_output(&ctx);
742 		if (!val)
743 			break;
744 		if (erst_timedout(&timeout, SPIN_UNIT))
745 			return -EIO;
746 	}
747 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
748 	if (rc)
749 		return rc;
750 	val = apei_exec_ctx_get_output(&ctx);
751 	rc = apei_exec_run(&ctx, ACPI_ERST_END);
752 	if (rc)
753 		return rc;
754 
755 	return erst_errno(val);
756 }
757 
758 /* NVRAM ERST Error Log Address Range is not supported yet */
759 static void pr_unimpl_nvram(void)
760 {
761 	if (printk_ratelimit())
762 		pr_warning(ERST_PFX
763 		"NVRAM ERST Log Address Range is not implemented yet\n");
764 }
765 
766 static int __erst_write_to_nvram(const struct cper_record_header *record)
767 {
768 	/* do not print message, because printk is not safe for NMI */
769 	return -ENOSYS;
770 }
771 
772 static int __erst_read_to_erange_from_nvram(u64 record_id, u64 *offset)
773 {
774 	pr_unimpl_nvram();
775 	return -ENOSYS;
776 }
777 
778 static int __erst_clear_from_nvram(u64 record_id)
779 {
780 	pr_unimpl_nvram();
781 	return -ENOSYS;
782 }
783 
784 int erst_write(const struct cper_record_header *record)
785 {
786 	int rc;
787 	unsigned long flags;
788 	struct cper_record_header *rcd_erange;
789 
790 	if (erst_disable)
791 		return -ENODEV;
792 
793 	if (memcmp(record->signature, CPER_SIG_RECORD, CPER_SIG_SIZE))
794 		return -EINVAL;
795 
796 	if (erst_erange.attr & ERST_RANGE_NVRAM) {
797 		if (!raw_spin_trylock_irqsave(&erst_lock, flags))
798 			return -EBUSY;
799 		rc = __erst_write_to_nvram(record);
800 		raw_spin_unlock_irqrestore(&erst_lock, flags);
801 		return rc;
802 	}
803 
804 	if (record->record_length > erst_erange.size)
805 		return -EINVAL;
806 
807 	if (!raw_spin_trylock_irqsave(&erst_lock, flags))
808 		return -EBUSY;
809 	memcpy(erst_erange.vaddr, record, record->record_length);
810 	rcd_erange = erst_erange.vaddr;
811 	/* signature for serialization system */
812 	memcpy(&rcd_erange->persistence_information, "ER", 2);
813 
814 	rc = __erst_write_to_storage(0);
815 	raw_spin_unlock_irqrestore(&erst_lock, flags);
816 
817 	return rc;
818 }
819 EXPORT_SYMBOL_GPL(erst_write);
820 
821 static int __erst_read_to_erange(u64 record_id, u64 *offset)
822 {
823 	int rc;
824 
825 	if (erst_erange.attr & ERST_RANGE_NVRAM)
826 		return __erst_read_to_erange_from_nvram(
827 			record_id, offset);
828 
829 	rc = __erst_read_from_storage(record_id, 0);
830 	if (rc)
831 		return rc;
832 	*offset = 0;
833 
834 	return 0;
835 }
836 
837 static ssize_t __erst_read(u64 record_id, struct cper_record_header *record,
838 			   size_t buflen)
839 {
840 	int rc;
841 	u64 offset, len = 0;
842 	struct cper_record_header *rcd_tmp;
843 
844 	rc = __erst_read_to_erange(record_id, &offset);
845 	if (rc)
846 		return rc;
847 	rcd_tmp = erst_erange.vaddr + offset;
848 	len = rcd_tmp->record_length;
849 	if (len <= buflen)
850 		memcpy(record, rcd_tmp, len);
851 
852 	return len;
853 }
854 
855 /*
856  * If return value > buflen, the buffer size is not big enough,
857  * else if return value < 0, something goes wrong,
858  * else everything is OK, and return value is record length
859  */
860 ssize_t erst_read(u64 record_id, struct cper_record_header *record,
861 		  size_t buflen)
862 {
863 	ssize_t len;
864 	unsigned long flags;
865 
866 	if (erst_disable)
867 		return -ENODEV;
868 
869 	raw_spin_lock_irqsave(&erst_lock, flags);
870 	len = __erst_read(record_id, record, buflen);
871 	raw_spin_unlock_irqrestore(&erst_lock, flags);
872 	return len;
873 }
874 EXPORT_SYMBOL_GPL(erst_read);
875 
876 int erst_clear(u64 record_id)
877 {
878 	int rc, i;
879 	unsigned long flags;
880 	u64 *entries;
881 
882 	if (erst_disable)
883 		return -ENODEV;
884 
885 	rc = mutex_lock_interruptible(&erst_record_id_cache.lock);
886 	if (rc)
887 		return rc;
888 	raw_spin_lock_irqsave(&erst_lock, flags);
889 	if (erst_erange.attr & ERST_RANGE_NVRAM)
890 		rc = __erst_clear_from_nvram(record_id);
891 	else
892 		rc = __erst_clear_from_storage(record_id);
893 	raw_spin_unlock_irqrestore(&erst_lock, flags);
894 	if (rc)
895 		goto out;
896 	entries = erst_record_id_cache.entries;
897 	for (i = 0; i < erst_record_id_cache.len; i++) {
898 		if (entries[i] == record_id)
899 			entries[i] = APEI_ERST_INVALID_RECORD_ID;
900 	}
901 	__erst_record_id_cache_compact();
902 out:
903 	mutex_unlock(&erst_record_id_cache.lock);
904 	return rc;
905 }
906 EXPORT_SYMBOL_GPL(erst_clear);
907 
908 static int __init setup_erst_disable(char *str)
909 {
910 	erst_disable = 1;
911 	return 0;
912 }
913 
914 __setup("erst_disable", setup_erst_disable);
915 
916 static int erst_check_table(struct acpi_table_erst *erst_tab)
917 {
918 	if ((erst_tab->header_length !=
919 	     (sizeof(struct acpi_table_erst) - sizeof(erst_tab->header)))
920 	    && (erst_tab->header_length != sizeof(struct acpi_table_einj)))
921 		return -EINVAL;
922 	if (erst_tab->header.length < sizeof(struct acpi_table_erst))
923 		return -EINVAL;
924 	if (erst_tab->entries !=
925 	    (erst_tab->header.length - sizeof(struct acpi_table_erst)) /
926 	    sizeof(struct acpi_erst_entry))
927 		return -EINVAL;
928 
929 	return 0;
930 }
931 
932 static int erst_open_pstore(struct pstore_info *psi);
933 static int erst_close_pstore(struct pstore_info *psi);
934 static ssize_t erst_reader(u64 *id, enum pstore_type_id *type,
935 			   struct timespec *time, struct pstore_info *psi);
936 static u64 erst_writer(enum pstore_type_id type, int part, size_t size,
937 		       struct pstore_info *psi);
938 static int erst_clearer(enum pstore_type_id type, u64 id,
939 			struct pstore_info *psi);
940 
941 static struct pstore_info erst_info = {
942 	.owner		= THIS_MODULE,
943 	.name		= "erst",
944 	.open		= erst_open_pstore,
945 	.close		= erst_close_pstore,
946 	.read		= erst_reader,
947 	.write		= erst_writer,
948 	.erase		= erst_clearer
949 };
950 
951 #define CPER_CREATOR_PSTORE						\
952 	UUID_LE(0x75a574e3, 0x5052, 0x4b29, 0x8a, 0x8e, 0xbe, 0x2c,	\
953 		0x64, 0x90, 0xb8, 0x9d)
954 #define CPER_SECTION_TYPE_DMESG						\
955 	UUID_LE(0xc197e04e, 0xd545, 0x4a70, 0x9c, 0x17, 0xa5, 0x54,	\
956 		0x94, 0x19, 0xeb, 0x12)
957 #define CPER_SECTION_TYPE_MCE						\
958 	UUID_LE(0xfe08ffbe, 0x95e4, 0x4be7, 0xbc, 0x73, 0x40, 0x96,	\
959 		0x04, 0x4a, 0x38, 0xfc)
960 
961 struct cper_pstore_record {
962 	struct cper_record_header hdr;
963 	struct cper_section_descriptor sec_hdr;
964 	char data[];
965 } __packed;
966 
967 static int reader_pos;
968 
969 static int erst_open_pstore(struct pstore_info *psi)
970 {
971 	int rc;
972 
973 	if (erst_disable)
974 		return -ENODEV;
975 
976 	rc = erst_get_record_id_begin(&reader_pos);
977 
978 	return rc;
979 }
980 
981 static int erst_close_pstore(struct pstore_info *psi)
982 {
983 	erst_get_record_id_end();
984 
985 	return 0;
986 }
987 
988 static ssize_t erst_reader(u64 *id, enum pstore_type_id *type,
989 			   struct timespec *time, struct pstore_info *psi)
990 {
991 	int rc;
992 	ssize_t len = 0;
993 	u64 record_id;
994 	struct cper_pstore_record *rcd = (struct cper_pstore_record *)
995 					(erst_info.buf - sizeof(*rcd));
996 
997 	if (erst_disable)
998 		return -ENODEV;
999 
1000 skip:
1001 	rc = erst_get_record_id_next(&reader_pos, &record_id);
1002 	if (rc)
1003 		goto out;
1004 
1005 	/* no more record */
1006 	if (record_id == APEI_ERST_INVALID_RECORD_ID) {
1007 		rc = -1;
1008 		goto out;
1009 	}
1010 
1011 	len = erst_read(record_id, &rcd->hdr, sizeof(*rcd) +
1012 			erst_info.bufsize);
1013 	/* The record may be cleared by others, try read next record */
1014 	if (len == -ENOENT)
1015 		goto skip;
1016 	else if (len < 0) {
1017 		rc = -1;
1018 		goto out;
1019 	}
1020 	if (uuid_le_cmp(rcd->hdr.creator_id, CPER_CREATOR_PSTORE) != 0)
1021 		goto skip;
1022 
1023 	*id = record_id;
1024 	if (uuid_le_cmp(rcd->sec_hdr.section_type,
1025 			CPER_SECTION_TYPE_DMESG) == 0)
1026 		*type = PSTORE_TYPE_DMESG;
1027 	else if (uuid_le_cmp(rcd->sec_hdr.section_type,
1028 			     CPER_SECTION_TYPE_MCE) == 0)
1029 		*type = PSTORE_TYPE_MCE;
1030 	else
1031 		*type = PSTORE_TYPE_UNKNOWN;
1032 
1033 	if (rcd->hdr.validation_bits & CPER_VALID_TIMESTAMP)
1034 		time->tv_sec = rcd->hdr.timestamp;
1035 	else
1036 		time->tv_sec = 0;
1037 	time->tv_nsec = 0;
1038 
1039 out:
1040 	return (rc < 0) ? rc : (len - sizeof(*rcd));
1041 }
1042 
1043 static u64 erst_writer(enum pstore_type_id type, int part, size_t size,
1044 		       struct pstore_info *psi)
1045 {
1046 	struct cper_pstore_record *rcd = (struct cper_pstore_record *)
1047 					(erst_info.buf - sizeof(*rcd));
1048 
1049 	memset(rcd, 0, sizeof(*rcd));
1050 	memcpy(rcd->hdr.signature, CPER_SIG_RECORD, CPER_SIG_SIZE);
1051 	rcd->hdr.revision = CPER_RECORD_REV;
1052 	rcd->hdr.signature_end = CPER_SIG_END;
1053 	rcd->hdr.section_count = 1;
1054 	rcd->hdr.error_severity = CPER_SEV_FATAL;
1055 	/* timestamp valid. platform_id, partition_id are invalid */
1056 	rcd->hdr.validation_bits = CPER_VALID_TIMESTAMP;
1057 	rcd->hdr.timestamp = get_seconds();
1058 	rcd->hdr.record_length = sizeof(*rcd) + size;
1059 	rcd->hdr.creator_id = CPER_CREATOR_PSTORE;
1060 	rcd->hdr.notification_type = CPER_NOTIFY_MCE;
1061 	rcd->hdr.record_id = cper_next_record_id();
1062 	rcd->hdr.flags = CPER_HW_ERROR_FLAGS_PREVERR;
1063 
1064 	rcd->sec_hdr.section_offset = sizeof(*rcd);
1065 	rcd->sec_hdr.section_length = size;
1066 	rcd->sec_hdr.revision = CPER_SEC_REV;
1067 	/* fru_id and fru_text is invalid */
1068 	rcd->sec_hdr.validation_bits = 0;
1069 	rcd->sec_hdr.flags = CPER_SEC_PRIMARY;
1070 	switch (type) {
1071 	case PSTORE_TYPE_DMESG:
1072 		rcd->sec_hdr.section_type = CPER_SECTION_TYPE_DMESG;
1073 		break;
1074 	case PSTORE_TYPE_MCE:
1075 		rcd->sec_hdr.section_type = CPER_SECTION_TYPE_MCE;
1076 		break;
1077 	default:
1078 		return -EINVAL;
1079 	}
1080 	rcd->sec_hdr.section_severity = CPER_SEV_FATAL;
1081 
1082 	erst_write(&rcd->hdr);
1083 
1084 	return rcd->hdr.record_id;
1085 }
1086 
1087 static int erst_clearer(enum pstore_type_id type, u64 id,
1088 			struct pstore_info *psi)
1089 {
1090 	return erst_clear(id);
1091 }
1092 
1093 static int __init erst_init(void)
1094 {
1095 	int rc = 0;
1096 	acpi_status status;
1097 	struct apei_exec_context ctx;
1098 	struct apei_resources erst_resources;
1099 	struct resource *r;
1100 	char *buf;
1101 
1102 	if (acpi_disabled)
1103 		goto err;
1104 
1105 	if (erst_disable) {
1106 		pr_info(ERST_PFX
1107 	"Error Record Serialization Table (ERST) support is disabled.\n");
1108 		goto err;
1109 	}
1110 
1111 	status = acpi_get_table(ACPI_SIG_ERST, 0,
1112 				(struct acpi_table_header **)&erst_tab);
1113 	if (status == AE_NOT_FOUND) {
1114 		pr_info(ERST_PFX "Table is not found!\n");
1115 		goto err;
1116 	} else if (ACPI_FAILURE(status)) {
1117 		const char *msg = acpi_format_exception(status);
1118 		pr_err(ERST_PFX "Failed to get table, %s\n", msg);
1119 		rc = -EINVAL;
1120 		goto err;
1121 	}
1122 
1123 	rc = erst_check_table(erst_tab);
1124 	if (rc) {
1125 		pr_err(FW_BUG ERST_PFX "ERST table is invalid\n");
1126 		goto err;
1127 	}
1128 
1129 	apei_resources_init(&erst_resources);
1130 	erst_exec_ctx_init(&ctx);
1131 	rc = apei_exec_collect_resources(&ctx, &erst_resources);
1132 	if (rc)
1133 		goto err_fini;
1134 	rc = apei_resources_request(&erst_resources, "APEI ERST");
1135 	if (rc)
1136 		goto err_fini;
1137 	rc = apei_exec_pre_map_gars(&ctx);
1138 	if (rc)
1139 		goto err_release;
1140 	rc = erst_get_erange(&erst_erange);
1141 	if (rc) {
1142 		if (rc == -ENODEV)
1143 			pr_info(ERST_PFX
1144 	"The corresponding hardware device or firmware implementation "
1145 	"is not available.\n");
1146 		else
1147 			pr_err(ERST_PFX
1148 			       "Failed to get Error Log Address Range.\n");
1149 		goto err_unmap_reg;
1150 	}
1151 
1152 	r = request_mem_region(erst_erange.base, erst_erange.size, "APEI ERST");
1153 	if (!r) {
1154 		pr_err(ERST_PFX
1155 		"Can not request iomem region <0x%16llx-0x%16llx> for ERST.\n",
1156 		(unsigned long long)erst_erange.base,
1157 		(unsigned long long)erst_erange.base + erst_erange.size);
1158 		rc = -EIO;
1159 		goto err_unmap_reg;
1160 	}
1161 	rc = -ENOMEM;
1162 	erst_erange.vaddr = ioremap_cache(erst_erange.base,
1163 					  erst_erange.size);
1164 	if (!erst_erange.vaddr)
1165 		goto err_release_erange;
1166 
1167 	buf = kmalloc(erst_erange.size, GFP_KERNEL);
1168 	mutex_init(&erst_info.buf_mutex);
1169 	if (buf) {
1170 		erst_info.buf = buf + sizeof(struct cper_pstore_record);
1171 		erst_info.bufsize = erst_erange.size -
1172 				    sizeof(struct cper_pstore_record);
1173 		if (pstore_register(&erst_info)) {
1174 			pr_info(ERST_PFX "Could not register with persistent store\n");
1175 			kfree(buf);
1176 		}
1177 	}
1178 
1179 	pr_info(ERST_PFX
1180 	"Error Record Serialization Table (ERST) support is initialized.\n");
1181 
1182 	return 0;
1183 
1184 err_release_erange:
1185 	release_mem_region(erst_erange.base, erst_erange.size);
1186 err_unmap_reg:
1187 	apei_exec_post_unmap_gars(&ctx);
1188 err_release:
1189 	apei_resources_release(&erst_resources);
1190 err_fini:
1191 	apei_resources_fini(&erst_resources);
1192 err:
1193 	erst_disable = 1;
1194 	return rc;
1195 }
1196 
1197 device_initcall(erst_init);
1198