xref: /openbmc/linux/drivers/acpi/apei/erst.c (revision d6472302)
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 <linux/vmalloc.h>
39 #include <acpi/apei.h>
40 
41 #include "apei-internal.h"
42 
43 #undef pr_fmt
44 #define pr_fmt(fmt) "ERST: " fmt
45 
46 /* ERST command status */
47 #define ERST_STATUS_SUCCESS			0x0
48 #define ERST_STATUS_NOT_ENOUGH_SPACE		0x1
49 #define ERST_STATUS_HARDWARE_NOT_AVAILABLE	0x2
50 #define ERST_STATUS_FAILED			0x3
51 #define ERST_STATUS_RECORD_STORE_EMPTY		0x4
52 #define ERST_STATUS_RECORD_NOT_FOUND		0x5
53 
54 #define ERST_TAB_ENTRY(tab)						\
55 	((struct acpi_whea_header *)((char *)(tab) +			\
56 				     sizeof(struct acpi_table_erst)))
57 
58 #define SPIN_UNIT		100			/* 100ns */
59 /* Firmware should respond within 1 milliseconds */
60 #define FIRMWARE_TIMEOUT	(1 * NSEC_PER_MSEC)
61 #define FIRMWARE_MAX_STALL	50			/* 50us */
62 
63 int erst_disable;
64 EXPORT_SYMBOL_GPL(erst_disable);
65 
66 static struct acpi_table_erst *erst_tab;
67 
68 /* ERST Error Log Address Range atrributes */
69 #define ERST_RANGE_RESERVED	0x0001
70 #define ERST_RANGE_NVRAM	0x0002
71 #define ERST_RANGE_SLOW		0x0004
72 
73 /*
74  * ERST Error Log Address Range, used as buffer for reading/writing
75  * error records.
76  */
77 static struct erst_erange {
78 	u64 base;
79 	u64 size;
80 	void __iomem *vaddr;
81 	u32 attr;
82 } erst_erange;
83 
84 /*
85  * Prevent ERST interpreter to run simultaneously, because the
86  * corresponding firmware implementation may not work properly when
87  * invoked simultaneously.
88  *
89  * It is used to provide exclusive accessing for ERST Error Log
90  * Address Range too.
91  */
92 static DEFINE_RAW_SPINLOCK(erst_lock);
93 
94 static inline int erst_errno(int command_status)
95 {
96 	switch (command_status) {
97 	case ERST_STATUS_SUCCESS:
98 		return 0;
99 	case ERST_STATUS_HARDWARE_NOT_AVAILABLE:
100 		return -ENODEV;
101 	case ERST_STATUS_NOT_ENOUGH_SPACE:
102 		return -ENOSPC;
103 	case ERST_STATUS_RECORD_STORE_EMPTY:
104 	case ERST_STATUS_RECORD_NOT_FOUND:
105 		return -ENOENT;
106 	default:
107 		return -EINVAL;
108 	}
109 }
110 
111 static int erst_timedout(u64 *t, u64 spin_unit)
112 {
113 	if ((s64)*t < spin_unit) {
114 		pr_warn(FW_WARN "Firmware does not respond in time.\n");
115 		return 1;
116 	}
117 	*t -= spin_unit;
118 	ndelay(spin_unit);
119 	touch_nmi_watchdog();
120 	return 0;
121 }
122 
123 static int erst_exec_load_var1(struct apei_exec_context *ctx,
124 			       struct acpi_whea_header *entry)
125 {
126 	return __apei_exec_read_register(entry, &ctx->var1);
127 }
128 
129 static int erst_exec_load_var2(struct apei_exec_context *ctx,
130 			       struct acpi_whea_header *entry)
131 {
132 	return __apei_exec_read_register(entry, &ctx->var2);
133 }
134 
135 static int erst_exec_store_var1(struct apei_exec_context *ctx,
136 				struct acpi_whea_header *entry)
137 {
138 	return __apei_exec_write_register(entry, ctx->var1);
139 }
140 
141 static int erst_exec_add(struct apei_exec_context *ctx,
142 			 struct acpi_whea_header *entry)
143 {
144 	ctx->var1 += ctx->var2;
145 	return 0;
146 }
147 
148 static int erst_exec_subtract(struct apei_exec_context *ctx,
149 			      struct acpi_whea_header *entry)
150 {
151 	ctx->var1 -= ctx->var2;
152 	return 0;
153 }
154 
155 static int erst_exec_add_value(struct apei_exec_context *ctx,
156 			       struct acpi_whea_header *entry)
157 {
158 	int rc;
159 	u64 val;
160 
161 	rc = __apei_exec_read_register(entry, &val);
162 	if (rc)
163 		return rc;
164 	val += ctx->value;
165 	rc = __apei_exec_write_register(entry, val);
166 	return rc;
167 }
168 
169 static int erst_exec_subtract_value(struct apei_exec_context *ctx,
170 				    struct acpi_whea_header *entry)
171 {
172 	int rc;
173 	u64 val;
174 
175 	rc = __apei_exec_read_register(entry, &val);
176 	if (rc)
177 		return rc;
178 	val -= ctx->value;
179 	rc = __apei_exec_write_register(entry, val);
180 	return rc;
181 }
182 
183 static int erst_exec_stall(struct apei_exec_context *ctx,
184 			   struct acpi_whea_header *entry)
185 {
186 	u64 stall_time;
187 
188 	if (ctx->value > FIRMWARE_MAX_STALL) {
189 		if (!in_nmi())
190 			pr_warn(FW_WARN
191 			"Too long stall time for stall instruction: 0x%llx.\n",
192 				   ctx->value);
193 		stall_time = FIRMWARE_MAX_STALL;
194 	} else
195 		stall_time = ctx->value;
196 	udelay(stall_time);
197 	return 0;
198 }
199 
200 static int erst_exec_stall_while_true(struct apei_exec_context *ctx,
201 				      struct acpi_whea_header *entry)
202 {
203 	int rc;
204 	u64 val;
205 	u64 timeout = FIRMWARE_TIMEOUT;
206 	u64 stall_time;
207 
208 	if (ctx->var1 > FIRMWARE_MAX_STALL) {
209 		if (!in_nmi())
210 			pr_warn(FW_WARN
211 		"Too long stall time for stall while true instruction: 0x%llx.\n",
212 				   ctx->var1);
213 		stall_time = FIRMWARE_MAX_STALL;
214 	} else
215 		stall_time = ctx->var1;
216 
217 	for (;;) {
218 		rc = __apei_exec_read_register(entry, &val);
219 		if (rc)
220 			return rc;
221 		if (val != ctx->value)
222 			break;
223 		if (erst_timedout(&timeout, stall_time * NSEC_PER_USEC))
224 			return -EIO;
225 	}
226 	return 0;
227 }
228 
229 static int erst_exec_skip_next_instruction_if_true(
230 	struct apei_exec_context *ctx,
231 	struct acpi_whea_header *entry)
232 {
233 	int rc;
234 	u64 val;
235 
236 	rc = __apei_exec_read_register(entry, &val);
237 	if (rc)
238 		return rc;
239 	if (val == ctx->value) {
240 		ctx->ip += 2;
241 		return APEI_EXEC_SET_IP;
242 	}
243 
244 	return 0;
245 }
246 
247 static int erst_exec_goto(struct apei_exec_context *ctx,
248 			  struct acpi_whea_header *entry)
249 {
250 	ctx->ip = ctx->value;
251 	return APEI_EXEC_SET_IP;
252 }
253 
254 static int erst_exec_set_src_address_base(struct apei_exec_context *ctx,
255 					  struct acpi_whea_header *entry)
256 {
257 	return __apei_exec_read_register(entry, &ctx->src_base);
258 }
259 
260 static int erst_exec_set_dst_address_base(struct apei_exec_context *ctx,
261 					  struct acpi_whea_header *entry)
262 {
263 	return __apei_exec_read_register(entry, &ctx->dst_base);
264 }
265 
266 static int erst_exec_move_data(struct apei_exec_context *ctx,
267 			       struct acpi_whea_header *entry)
268 {
269 	int rc;
270 	u64 offset;
271 	void *src, *dst;
272 
273 	/* ioremap does not work in interrupt context */
274 	if (in_interrupt()) {
275 		pr_warn("MOVE_DATA can not be used in interrupt context.\n");
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 		iounmap(src);
289 		return -ENOMEM;
290 	}
291 
292 	memmove(dst, src, ctx->var2);
293 
294 	iounmap(src);
295 	iounmap(dst);
296 
297 	return 0;
298 }
299 
300 static struct apei_exec_ins_type erst_ins_type[] = {
301 	[ACPI_ERST_READ_REGISTER] = {
302 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
303 		.run = apei_exec_read_register,
304 	},
305 	[ACPI_ERST_READ_REGISTER_VALUE] = {
306 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
307 		.run = apei_exec_read_register_value,
308 	},
309 	[ACPI_ERST_WRITE_REGISTER] = {
310 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
311 		.run = apei_exec_write_register,
312 	},
313 	[ACPI_ERST_WRITE_REGISTER_VALUE] = {
314 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
315 		.run = apei_exec_write_register_value,
316 	},
317 	[ACPI_ERST_NOOP] = {
318 		.flags = 0,
319 		.run = apei_exec_noop,
320 	},
321 	[ACPI_ERST_LOAD_VAR1] = {
322 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
323 		.run = erst_exec_load_var1,
324 	},
325 	[ACPI_ERST_LOAD_VAR2] = {
326 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
327 		.run = erst_exec_load_var2,
328 	},
329 	[ACPI_ERST_STORE_VAR1] = {
330 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
331 		.run = erst_exec_store_var1,
332 	},
333 	[ACPI_ERST_ADD] = {
334 		.flags = 0,
335 		.run = erst_exec_add,
336 	},
337 	[ACPI_ERST_SUBTRACT] = {
338 		.flags = 0,
339 		.run = erst_exec_subtract,
340 	},
341 	[ACPI_ERST_ADD_VALUE] = {
342 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
343 		.run = erst_exec_add_value,
344 	},
345 	[ACPI_ERST_SUBTRACT_VALUE] = {
346 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
347 		.run = erst_exec_subtract_value,
348 	},
349 	[ACPI_ERST_STALL] = {
350 		.flags = 0,
351 		.run = erst_exec_stall,
352 	},
353 	[ACPI_ERST_STALL_WHILE_TRUE] = {
354 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
355 		.run = erst_exec_stall_while_true,
356 	},
357 	[ACPI_ERST_SKIP_NEXT_IF_TRUE] = {
358 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
359 		.run = erst_exec_skip_next_instruction_if_true,
360 	},
361 	[ACPI_ERST_GOTO] = {
362 		.flags = 0,
363 		.run = erst_exec_goto,
364 	},
365 	[ACPI_ERST_SET_SRC_ADDRESS_BASE] = {
366 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
367 		.run = erst_exec_set_src_address_base,
368 	},
369 	[ACPI_ERST_SET_DST_ADDRESS_BASE] = {
370 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
371 		.run = erst_exec_set_dst_address_base,
372 	},
373 	[ACPI_ERST_MOVE_DATA] = {
374 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
375 		.run = erst_exec_move_data,
376 	},
377 };
378 
379 static inline void erst_exec_ctx_init(struct apei_exec_context *ctx)
380 {
381 	apei_exec_ctx_init(ctx, erst_ins_type, ARRAY_SIZE(erst_ins_type),
382 			   ERST_TAB_ENTRY(erst_tab), erst_tab->entries);
383 }
384 
385 static int erst_get_erange(struct erst_erange *range)
386 {
387 	struct apei_exec_context ctx;
388 	int rc;
389 
390 	erst_exec_ctx_init(&ctx);
391 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_RANGE);
392 	if (rc)
393 		return rc;
394 	range->base = apei_exec_ctx_get_output(&ctx);
395 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_LENGTH);
396 	if (rc)
397 		return rc;
398 	range->size = apei_exec_ctx_get_output(&ctx);
399 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_ATTRIBUTES);
400 	if (rc)
401 		return rc;
402 	range->attr = apei_exec_ctx_get_output(&ctx);
403 
404 	return 0;
405 }
406 
407 static ssize_t __erst_get_record_count(void)
408 {
409 	struct apei_exec_context ctx;
410 	int rc;
411 
412 	erst_exec_ctx_init(&ctx);
413 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_RECORD_COUNT);
414 	if (rc)
415 		return rc;
416 	return apei_exec_ctx_get_output(&ctx);
417 }
418 
419 ssize_t erst_get_record_count(void)
420 {
421 	ssize_t count;
422 	unsigned long flags;
423 
424 	if (erst_disable)
425 		return -ENODEV;
426 
427 	raw_spin_lock_irqsave(&erst_lock, flags);
428 	count = __erst_get_record_count();
429 	raw_spin_unlock_irqrestore(&erst_lock, flags);
430 
431 	return count;
432 }
433 EXPORT_SYMBOL_GPL(erst_get_record_count);
434 
435 #define ERST_RECORD_ID_CACHE_SIZE_MIN	16
436 #define ERST_RECORD_ID_CACHE_SIZE_MAX	1024
437 
438 struct erst_record_id_cache {
439 	struct mutex lock;
440 	u64 *entries;
441 	int len;
442 	int size;
443 	int refcount;
444 };
445 
446 static struct erst_record_id_cache erst_record_id_cache = {
447 	.lock = __MUTEX_INITIALIZER(erst_record_id_cache.lock),
448 	.refcount = 0,
449 };
450 
451 static int __erst_get_next_record_id(u64 *record_id)
452 {
453 	struct apei_exec_context ctx;
454 	int rc;
455 
456 	erst_exec_ctx_init(&ctx);
457 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_RECORD_ID);
458 	if (rc)
459 		return rc;
460 	*record_id = apei_exec_ctx_get_output(&ctx);
461 
462 	return 0;
463 }
464 
465 int erst_get_record_id_begin(int *pos)
466 {
467 	int rc;
468 
469 	if (erst_disable)
470 		return -ENODEV;
471 
472 	rc = mutex_lock_interruptible(&erst_record_id_cache.lock);
473 	if (rc)
474 		return rc;
475 	erst_record_id_cache.refcount++;
476 	mutex_unlock(&erst_record_id_cache.lock);
477 
478 	*pos = 0;
479 
480 	return 0;
481 }
482 EXPORT_SYMBOL_GPL(erst_get_record_id_begin);
483 
484 /* erst_record_id_cache.lock must be held by caller */
485 static int __erst_record_id_cache_add_one(void)
486 {
487 	u64 id, prev_id, first_id;
488 	int i, rc;
489 	u64 *entries;
490 	unsigned long flags;
491 
492 	id = prev_id = first_id = APEI_ERST_INVALID_RECORD_ID;
493 retry:
494 	raw_spin_lock_irqsave(&erst_lock, flags);
495 	rc = __erst_get_next_record_id(&id);
496 	raw_spin_unlock_irqrestore(&erst_lock, flags);
497 	if (rc == -ENOENT)
498 		return 0;
499 	if (rc)
500 		return rc;
501 	if (id == APEI_ERST_INVALID_RECORD_ID)
502 		return 0;
503 	/* can not skip current ID, or loop back to first ID */
504 	if (id == prev_id || id == first_id)
505 		return 0;
506 	if (first_id == APEI_ERST_INVALID_RECORD_ID)
507 		first_id = id;
508 	prev_id = id;
509 
510 	entries = erst_record_id_cache.entries;
511 	for (i = 0; i < erst_record_id_cache.len; i++) {
512 		if (entries[i] == id)
513 			break;
514 	}
515 	/* record id already in cache, try next */
516 	if (i < erst_record_id_cache.len)
517 		goto retry;
518 	if (erst_record_id_cache.len >= erst_record_id_cache.size) {
519 		int new_size, alloc_size;
520 		u64 *new_entries;
521 
522 		new_size = erst_record_id_cache.size * 2;
523 		new_size = clamp_val(new_size, ERST_RECORD_ID_CACHE_SIZE_MIN,
524 				     ERST_RECORD_ID_CACHE_SIZE_MAX);
525 		if (new_size <= erst_record_id_cache.size) {
526 			if (printk_ratelimit())
527 				pr_warn(FW_WARN "too many record IDs!\n");
528 			return 0;
529 		}
530 		alloc_size = new_size * sizeof(entries[0]);
531 		if (alloc_size < PAGE_SIZE)
532 			new_entries = kmalloc(alloc_size, GFP_KERNEL);
533 		else
534 			new_entries = vmalloc(alloc_size);
535 		if (!new_entries)
536 			return -ENOMEM;
537 		memcpy(new_entries, entries,
538 		       erst_record_id_cache.len * sizeof(entries[0]));
539 		if (erst_record_id_cache.size < PAGE_SIZE)
540 			kfree(entries);
541 		else
542 			vfree(entries);
543 		erst_record_id_cache.entries = entries = new_entries;
544 		erst_record_id_cache.size = new_size;
545 	}
546 	entries[i] = id;
547 	erst_record_id_cache.len++;
548 
549 	return 1;
550 }
551 
552 /*
553  * Get the record ID of an existing error record on the persistent
554  * storage. If there is no error record on the persistent storage, the
555  * returned record_id is APEI_ERST_INVALID_RECORD_ID.
556  */
557 int erst_get_record_id_next(int *pos, u64 *record_id)
558 {
559 	int rc = 0;
560 	u64 *entries;
561 
562 	if (erst_disable)
563 		return -ENODEV;
564 
565 	/* must be enclosed by erst_get_record_id_begin/end */
566 	BUG_ON(!erst_record_id_cache.refcount);
567 	BUG_ON(*pos < 0 || *pos > erst_record_id_cache.len);
568 
569 	mutex_lock(&erst_record_id_cache.lock);
570 	entries = erst_record_id_cache.entries;
571 	for (; *pos < erst_record_id_cache.len; (*pos)++)
572 		if (entries[*pos] != APEI_ERST_INVALID_RECORD_ID)
573 			break;
574 	/* found next record id in cache */
575 	if (*pos < erst_record_id_cache.len) {
576 		*record_id = entries[*pos];
577 		(*pos)++;
578 		goto out_unlock;
579 	}
580 
581 	/* Try to add one more record ID to cache */
582 	rc = __erst_record_id_cache_add_one();
583 	if (rc < 0)
584 		goto out_unlock;
585 	/* successfully add one new ID */
586 	if (rc == 1) {
587 		*record_id = erst_record_id_cache.entries[*pos];
588 		(*pos)++;
589 		rc = 0;
590 	} else {
591 		*pos = -1;
592 		*record_id = APEI_ERST_INVALID_RECORD_ID;
593 	}
594 out_unlock:
595 	mutex_unlock(&erst_record_id_cache.lock);
596 
597 	return rc;
598 }
599 EXPORT_SYMBOL_GPL(erst_get_record_id_next);
600 
601 /* erst_record_id_cache.lock must be held by caller */
602 static void __erst_record_id_cache_compact(void)
603 {
604 	int i, wpos = 0;
605 	u64 *entries;
606 
607 	if (erst_record_id_cache.refcount)
608 		return;
609 
610 	entries = erst_record_id_cache.entries;
611 	for (i = 0; i < erst_record_id_cache.len; i++) {
612 		if (entries[i] == APEI_ERST_INVALID_RECORD_ID)
613 			continue;
614 		if (wpos != i)
615 			entries[wpos] = entries[i];
616 		wpos++;
617 	}
618 	erst_record_id_cache.len = wpos;
619 }
620 
621 void erst_get_record_id_end(void)
622 {
623 	/*
624 	 * erst_disable != 0 should be detected by invoker via the
625 	 * return value of erst_get_record_id_begin/next, so this
626 	 * function should not be called for erst_disable != 0.
627 	 */
628 	BUG_ON(erst_disable);
629 
630 	mutex_lock(&erst_record_id_cache.lock);
631 	erst_record_id_cache.refcount--;
632 	BUG_ON(erst_record_id_cache.refcount < 0);
633 	__erst_record_id_cache_compact();
634 	mutex_unlock(&erst_record_id_cache.lock);
635 }
636 EXPORT_SYMBOL_GPL(erst_get_record_id_end);
637 
638 static int __erst_write_to_storage(u64 offset)
639 {
640 	struct apei_exec_context ctx;
641 	u64 timeout = FIRMWARE_TIMEOUT;
642 	u64 val;
643 	int rc;
644 
645 	erst_exec_ctx_init(&ctx);
646 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_WRITE);
647 	if (rc)
648 		return rc;
649 	apei_exec_ctx_set_input(&ctx, offset);
650 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_OFFSET);
651 	if (rc)
652 		return rc;
653 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
654 	if (rc)
655 		return rc;
656 	for (;;) {
657 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
658 		if (rc)
659 			return rc;
660 		val = apei_exec_ctx_get_output(&ctx);
661 		if (!val)
662 			break;
663 		if (erst_timedout(&timeout, SPIN_UNIT))
664 			return -EIO;
665 	}
666 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
667 	if (rc)
668 		return rc;
669 	val = apei_exec_ctx_get_output(&ctx);
670 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
671 	if (rc)
672 		return rc;
673 
674 	return erst_errno(val);
675 }
676 
677 static int __erst_read_from_storage(u64 record_id, u64 offset)
678 {
679 	struct apei_exec_context ctx;
680 	u64 timeout = FIRMWARE_TIMEOUT;
681 	u64 val;
682 	int rc;
683 
684 	erst_exec_ctx_init(&ctx);
685 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_READ);
686 	if (rc)
687 		return rc;
688 	apei_exec_ctx_set_input(&ctx, offset);
689 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_OFFSET);
690 	if (rc)
691 		return rc;
692 	apei_exec_ctx_set_input(&ctx, record_id);
693 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_ID);
694 	if (rc)
695 		return rc;
696 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
697 	if (rc)
698 		return rc;
699 	for (;;) {
700 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
701 		if (rc)
702 			return rc;
703 		val = apei_exec_ctx_get_output(&ctx);
704 		if (!val)
705 			break;
706 		if (erst_timedout(&timeout, SPIN_UNIT))
707 			return -EIO;
708 	};
709 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
710 	if (rc)
711 		return rc;
712 	val = apei_exec_ctx_get_output(&ctx);
713 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
714 	if (rc)
715 		return rc;
716 
717 	return erst_errno(val);
718 }
719 
720 static int __erst_clear_from_storage(u64 record_id)
721 {
722 	struct apei_exec_context ctx;
723 	u64 timeout = FIRMWARE_TIMEOUT;
724 	u64 val;
725 	int rc;
726 
727 	erst_exec_ctx_init(&ctx);
728 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_CLEAR);
729 	if (rc)
730 		return rc;
731 	apei_exec_ctx_set_input(&ctx, record_id);
732 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_ID);
733 	if (rc)
734 		return rc;
735 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
736 	if (rc)
737 		return rc;
738 	for (;;) {
739 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
740 		if (rc)
741 			return rc;
742 		val = apei_exec_ctx_get_output(&ctx);
743 		if (!val)
744 			break;
745 		if (erst_timedout(&timeout, SPIN_UNIT))
746 			return -EIO;
747 	}
748 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
749 	if (rc)
750 		return rc;
751 	val = apei_exec_ctx_get_output(&ctx);
752 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
753 	if (rc)
754 		return rc;
755 
756 	return erst_errno(val);
757 }
758 
759 /* NVRAM ERST Error Log Address Range is not supported yet */
760 static void pr_unimpl_nvram(void)
761 {
762 	if (printk_ratelimit())
763 		pr_warn("NVRAM ERST Log Address Range 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_erst)))
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, int *count,
935 			   struct timespec *time, char **buf,
936 			   bool *compressed, struct pstore_info *psi);
937 static int erst_writer(enum pstore_type_id type, enum kmsg_dump_reason reason,
938 		       u64 *id, unsigned int part, int count, bool compressed,
939 		       size_t size, struct pstore_info *psi);
940 static int erst_clearer(enum pstore_type_id type, u64 id, int count,
941 			struct timespec time, struct pstore_info *psi);
942 
943 static struct pstore_info erst_info = {
944 	.owner		= THIS_MODULE,
945 	.name		= "erst",
946 	.flags		= PSTORE_FLAGS_FRAGILE,
947 	.open		= erst_open_pstore,
948 	.close		= erst_close_pstore,
949 	.read		= erst_reader,
950 	.write		= erst_writer,
951 	.erase		= erst_clearer
952 };
953 
954 #define CPER_CREATOR_PSTORE						\
955 	UUID_LE(0x75a574e3, 0x5052, 0x4b29, 0x8a, 0x8e, 0xbe, 0x2c,	\
956 		0x64, 0x90, 0xb8, 0x9d)
957 #define CPER_SECTION_TYPE_DMESG						\
958 	UUID_LE(0xc197e04e, 0xd545, 0x4a70, 0x9c, 0x17, 0xa5, 0x54,	\
959 		0x94, 0x19, 0xeb, 0x12)
960 #define CPER_SECTION_TYPE_DMESG_Z					\
961 	UUID_LE(0x4f118707, 0x04dd, 0x4055, 0xb5, 0xdd, 0x95, 0x6d,	\
962 		0x34, 0xdd, 0xfa, 0xc6)
963 #define CPER_SECTION_TYPE_MCE						\
964 	UUID_LE(0xfe08ffbe, 0x95e4, 0x4be7, 0xbc, 0x73, 0x40, 0x96,	\
965 		0x04, 0x4a, 0x38, 0xfc)
966 
967 struct cper_pstore_record {
968 	struct cper_record_header hdr;
969 	struct cper_section_descriptor sec_hdr;
970 	char data[];
971 } __packed;
972 
973 static int reader_pos;
974 
975 static int erst_open_pstore(struct pstore_info *psi)
976 {
977 	int rc;
978 
979 	if (erst_disable)
980 		return -ENODEV;
981 
982 	rc = erst_get_record_id_begin(&reader_pos);
983 
984 	return rc;
985 }
986 
987 static int erst_close_pstore(struct pstore_info *psi)
988 {
989 	erst_get_record_id_end();
990 
991 	return 0;
992 }
993 
994 static ssize_t erst_reader(u64 *id, enum pstore_type_id *type, int *count,
995 			   struct timespec *time, char **buf,
996 			   bool *compressed, struct pstore_info *psi)
997 {
998 	int rc;
999 	ssize_t len = 0;
1000 	u64 record_id;
1001 	struct cper_pstore_record *rcd;
1002 	size_t rcd_len = sizeof(*rcd) + erst_info.bufsize;
1003 
1004 	if (erst_disable)
1005 		return -ENODEV;
1006 
1007 	rcd = kmalloc(rcd_len, GFP_KERNEL);
1008 	if (!rcd) {
1009 		rc = -ENOMEM;
1010 		goto out;
1011 	}
1012 skip:
1013 	rc = erst_get_record_id_next(&reader_pos, &record_id);
1014 	if (rc)
1015 		goto out;
1016 
1017 	/* no more record */
1018 	if (record_id == APEI_ERST_INVALID_RECORD_ID) {
1019 		rc = -EINVAL;
1020 		goto out;
1021 	}
1022 
1023 	len = erst_read(record_id, &rcd->hdr, rcd_len);
1024 	/* The record may be cleared by others, try read next record */
1025 	if (len == -ENOENT)
1026 		goto skip;
1027 	else if (len < sizeof(*rcd)) {
1028 		rc = -EIO;
1029 		goto out;
1030 	}
1031 	if (uuid_le_cmp(rcd->hdr.creator_id, CPER_CREATOR_PSTORE) != 0)
1032 		goto skip;
1033 
1034 	*buf = kmalloc(len, GFP_KERNEL);
1035 	if (*buf == NULL) {
1036 		rc = -ENOMEM;
1037 		goto out;
1038 	}
1039 	memcpy(*buf, rcd->data, len - sizeof(*rcd));
1040 	*id = record_id;
1041 	*compressed = false;
1042 	if (uuid_le_cmp(rcd->sec_hdr.section_type,
1043 			CPER_SECTION_TYPE_DMESG_Z) == 0) {
1044 		*type = PSTORE_TYPE_DMESG;
1045 		*compressed = true;
1046 	} else if (uuid_le_cmp(rcd->sec_hdr.section_type,
1047 			CPER_SECTION_TYPE_DMESG) == 0)
1048 		*type = PSTORE_TYPE_DMESG;
1049 	else if (uuid_le_cmp(rcd->sec_hdr.section_type,
1050 			     CPER_SECTION_TYPE_MCE) == 0)
1051 		*type = PSTORE_TYPE_MCE;
1052 	else
1053 		*type = PSTORE_TYPE_UNKNOWN;
1054 
1055 	if (rcd->hdr.validation_bits & CPER_VALID_TIMESTAMP)
1056 		time->tv_sec = rcd->hdr.timestamp;
1057 	else
1058 		time->tv_sec = 0;
1059 	time->tv_nsec = 0;
1060 
1061 out:
1062 	kfree(rcd);
1063 	return (rc < 0) ? rc : (len - sizeof(*rcd));
1064 }
1065 
1066 static int erst_writer(enum pstore_type_id type, enum kmsg_dump_reason reason,
1067 		       u64 *id, unsigned int part, int count, bool compressed,
1068 		       size_t size, struct pstore_info *psi)
1069 {
1070 	struct cper_pstore_record *rcd = (struct cper_pstore_record *)
1071 					(erst_info.buf - sizeof(*rcd));
1072 	int ret;
1073 
1074 	memset(rcd, 0, sizeof(*rcd));
1075 	memcpy(rcd->hdr.signature, CPER_SIG_RECORD, CPER_SIG_SIZE);
1076 	rcd->hdr.revision = CPER_RECORD_REV;
1077 	rcd->hdr.signature_end = CPER_SIG_END;
1078 	rcd->hdr.section_count = 1;
1079 	rcd->hdr.error_severity = CPER_SEV_FATAL;
1080 	/* timestamp valid. platform_id, partition_id are invalid */
1081 	rcd->hdr.validation_bits = CPER_VALID_TIMESTAMP;
1082 	rcd->hdr.timestamp = get_seconds();
1083 	rcd->hdr.record_length = sizeof(*rcd) + size;
1084 	rcd->hdr.creator_id = CPER_CREATOR_PSTORE;
1085 	rcd->hdr.notification_type = CPER_NOTIFY_MCE;
1086 	rcd->hdr.record_id = cper_next_record_id();
1087 	rcd->hdr.flags = CPER_HW_ERROR_FLAGS_PREVERR;
1088 
1089 	rcd->sec_hdr.section_offset = sizeof(*rcd);
1090 	rcd->sec_hdr.section_length = size;
1091 	rcd->sec_hdr.revision = CPER_SEC_REV;
1092 	/* fru_id and fru_text is invalid */
1093 	rcd->sec_hdr.validation_bits = 0;
1094 	rcd->sec_hdr.flags = CPER_SEC_PRIMARY;
1095 	switch (type) {
1096 	case PSTORE_TYPE_DMESG:
1097 		if (compressed)
1098 			rcd->sec_hdr.section_type = CPER_SECTION_TYPE_DMESG_Z;
1099 		else
1100 			rcd->sec_hdr.section_type = CPER_SECTION_TYPE_DMESG;
1101 		break;
1102 	case PSTORE_TYPE_MCE:
1103 		rcd->sec_hdr.section_type = CPER_SECTION_TYPE_MCE;
1104 		break;
1105 	default:
1106 		return -EINVAL;
1107 	}
1108 	rcd->sec_hdr.section_severity = CPER_SEV_FATAL;
1109 
1110 	ret = erst_write(&rcd->hdr);
1111 	*id = rcd->hdr.record_id;
1112 
1113 	return ret;
1114 }
1115 
1116 static int erst_clearer(enum pstore_type_id type, u64 id, int count,
1117 			struct timespec time, struct pstore_info *psi)
1118 {
1119 	return erst_clear(id);
1120 }
1121 
1122 static int __init erst_init(void)
1123 {
1124 	int rc = 0;
1125 	acpi_status status;
1126 	struct apei_exec_context ctx;
1127 	struct apei_resources erst_resources;
1128 	struct resource *r;
1129 	char *buf;
1130 
1131 	if (acpi_disabled)
1132 		goto err;
1133 
1134 	if (erst_disable) {
1135 		pr_info(
1136 	"Error Record Serialization Table (ERST) support is disabled.\n");
1137 		goto err;
1138 	}
1139 
1140 	status = acpi_get_table(ACPI_SIG_ERST, 0,
1141 				(struct acpi_table_header **)&erst_tab);
1142 	if (status == AE_NOT_FOUND)
1143 		goto err;
1144 	else if (ACPI_FAILURE(status)) {
1145 		const char *msg = acpi_format_exception(status);
1146 		pr_err("Failed to get table, %s\n", msg);
1147 		rc = -EINVAL;
1148 		goto err;
1149 	}
1150 
1151 	rc = erst_check_table(erst_tab);
1152 	if (rc) {
1153 		pr_err(FW_BUG "ERST table is invalid.\n");
1154 		goto err;
1155 	}
1156 
1157 	apei_resources_init(&erst_resources);
1158 	erst_exec_ctx_init(&ctx);
1159 	rc = apei_exec_collect_resources(&ctx, &erst_resources);
1160 	if (rc)
1161 		goto err_fini;
1162 	rc = apei_resources_request(&erst_resources, "APEI ERST");
1163 	if (rc)
1164 		goto err_fini;
1165 	rc = apei_exec_pre_map_gars(&ctx);
1166 	if (rc)
1167 		goto err_release;
1168 	rc = erst_get_erange(&erst_erange);
1169 	if (rc) {
1170 		if (rc == -ENODEV)
1171 			pr_info(
1172 	"The corresponding hardware device or firmware implementation "
1173 	"is not available.\n");
1174 		else
1175 			pr_err("Failed to get Error Log Address Range.\n");
1176 		goto err_unmap_reg;
1177 	}
1178 
1179 	r = request_mem_region(erst_erange.base, erst_erange.size, "APEI ERST");
1180 	if (!r) {
1181 		pr_err("Can not request [mem %#010llx-%#010llx] for ERST.\n",
1182 		       (unsigned long long)erst_erange.base,
1183 		       (unsigned long long)erst_erange.base + erst_erange.size - 1);
1184 		rc = -EIO;
1185 		goto err_unmap_reg;
1186 	}
1187 	rc = -ENOMEM;
1188 	erst_erange.vaddr = ioremap_cache(erst_erange.base,
1189 					  erst_erange.size);
1190 	if (!erst_erange.vaddr)
1191 		goto err_release_erange;
1192 
1193 	pr_info(
1194 	"Error Record Serialization Table (ERST) support is initialized.\n");
1195 
1196 	buf = kmalloc(erst_erange.size, GFP_KERNEL);
1197 	spin_lock_init(&erst_info.buf_lock);
1198 	if (buf) {
1199 		erst_info.buf = buf + sizeof(struct cper_pstore_record);
1200 		erst_info.bufsize = erst_erange.size -
1201 				    sizeof(struct cper_pstore_record);
1202 		rc = pstore_register(&erst_info);
1203 		if (rc) {
1204 			if (rc != -EPERM)
1205 				pr_info(
1206 				"Could not register with persistent store.\n");
1207 			erst_info.buf = NULL;
1208 			erst_info.bufsize = 0;
1209 			kfree(buf);
1210 		}
1211 	} else
1212 		pr_err(
1213 		"Failed to allocate %lld bytes for persistent store error log.\n",
1214 		erst_erange.size);
1215 
1216 	return 0;
1217 
1218 err_release_erange:
1219 	release_mem_region(erst_erange.base, erst_erange.size);
1220 err_unmap_reg:
1221 	apei_exec_post_unmap_gars(&ctx);
1222 err_release:
1223 	apei_resources_release(&erst_resources);
1224 err_fini:
1225 	apei_resources_fini(&erst_resources);
1226 err:
1227 	erst_disable = 1;
1228 	return rc;
1229 }
1230 
1231 device_initcall(erst_init);
1232