xref: /linux/drivers/acpi/apei/erst.c (revision 20dfee95936413708701eb151f419597fdd9d948)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * APEI Error Record Serialization Table support
4  *
5  * ERST is a way provided by APEI to save and retrieve hardware error
6  * information to and from a persistent store.
7  *
8  * For more information about ERST, please refer to ACPI Specification
9  * version 4.0, section 17.4.
10  *
11  * Copyright 2010 Intel Corp.
12  *   Author: Huang Ying <ying.huang@intel.com>
13  */
14 
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/delay.h>
19 #include <linux/io.h>
20 #include <linux/acpi.h>
21 #include <linux/uaccess.h>
22 #include <linux/cper.h>
23 #include <linux/nmi.h>
24 #include <linux/hardirq.h>
25 #include <linux/pstore.h>
26 #include <linux/vmalloc.h>
27 #include <linux/mm.h> /* kvfree() */
28 #include <acpi/apei.h>
29 
30 #include "apei-internal.h"
31 
32 #undef pr_fmt
33 #define pr_fmt(fmt) "ERST: " fmt
34 
35 /* ERST command status */
36 #define ERST_STATUS_SUCCESS			0x0
37 #define ERST_STATUS_NOT_ENOUGH_SPACE		0x1
38 #define ERST_STATUS_HARDWARE_NOT_AVAILABLE	0x2
39 #define ERST_STATUS_FAILED			0x3
40 #define ERST_STATUS_RECORD_STORE_EMPTY		0x4
41 #define ERST_STATUS_RECORD_NOT_FOUND		0x5
42 
43 #define ERST_TAB_ENTRY(tab)						\
44 	((struct acpi_whea_header *)((char *)(tab) +			\
45 				     sizeof(struct acpi_table_erst)))
46 
47 #define SPIN_UNIT		100			/* 100ns */
48 /* Firmware should respond within 1 milliseconds */
49 #define FIRMWARE_TIMEOUT	(1 * NSEC_PER_MSEC)
50 #define FIRMWARE_MAX_STALL	50			/* 50us */
51 
52 int erst_disable;
53 EXPORT_SYMBOL_GPL(erst_disable);
54 
55 static struct acpi_table_erst *erst_tab;
56 
57 /* ERST Error Log Address Range attributes */
58 #define ERST_RANGE_RESERVED	0x0001
59 #define ERST_RANGE_NVRAM	0x0002
60 #define ERST_RANGE_SLOW		0x0004
61 
62 /* ERST Exec max timings */
63 #define ERST_EXEC_TIMING_MAX_MASK      0xFFFFFFFF00000000
64 #define ERST_EXEC_TIMING_MAX_SHIFT     32
65 
66 /*
67  * ERST Error Log Address Range, used as buffer for reading/writing
68  * error records.
69  */
70 static struct erst_erange {
71 	u64 base;
72 	u64 size;
73 	void __iomem *vaddr;
74 	u32 attr;
75 	u64 timings;
76 } erst_erange;
77 
78 /*
79  * Prevent ERST interpreter to run simultaneously, because the
80  * corresponding firmware implementation may not work properly when
81  * invoked simultaneously.
82  *
83  * It is used to provide exclusive accessing for ERST Error Log
84  * Address Range too.
85  */
86 static DEFINE_RAW_SPINLOCK(erst_lock);
87 
88 static inline int erst_errno(int command_status)
89 {
90 	switch (command_status) {
91 	case ERST_STATUS_SUCCESS:
92 		return 0;
93 	case ERST_STATUS_HARDWARE_NOT_AVAILABLE:
94 		return -ENODEV;
95 	case ERST_STATUS_NOT_ENOUGH_SPACE:
96 		return -ENOSPC;
97 	case ERST_STATUS_RECORD_STORE_EMPTY:
98 	case ERST_STATUS_RECORD_NOT_FOUND:
99 		return -ENOENT;
100 	default:
101 		return -EINVAL;
102 	}
103 }
104 
105 static inline u64 erst_get_timeout(void)
106 {
107 	u64 timeout = FIRMWARE_TIMEOUT;
108 
109 	if (erst_erange.attr & ERST_RANGE_SLOW) {
110 		timeout = ((erst_erange.timings & ERST_EXEC_TIMING_MAX_MASK) >>
111 			ERST_EXEC_TIMING_MAX_SHIFT) * NSEC_PER_MSEC;
112 		if (timeout < FIRMWARE_TIMEOUT)
113 			timeout = FIRMWARE_TIMEOUT;
114 	}
115 	return timeout;
116 }
117 
118 static int erst_timedout(u64 *t, u64 spin_unit)
119 {
120 	if ((s64)*t < spin_unit) {
121 		pr_warn(FW_WARN "Firmware does not respond in time.\n");
122 		return 1;
123 	}
124 	*t -= spin_unit;
125 	ndelay(spin_unit);
126 	touch_nmi_watchdog();
127 	return 0;
128 }
129 
130 static int erst_exec_load_var1(struct apei_exec_context *ctx,
131 			       struct acpi_whea_header *entry)
132 {
133 	return __apei_exec_read_register(entry, &ctx->var1);
134 }
135 
136 static int erst_exec_load_var2(struct apei_exec_context *ctx,
137 			       struct acpi_whea_header *entry)
138 {
139 	return __apei_exec_read_register(entry, &ctx->var2);
140 }
141 
142 static int erst_exec_store_var1(struct apei_exec_context *ctx,
143 				struct acpi_whea_header *entry)
144 {
145 	return __apei_exec_write_register(entry, ctx->var1);
146 }
147 
148 static int erst_exec_add(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_subtract(struct apei_exec_context *ctx,
156 			      struct acpi_whea_header *entry)
157 {
158 	ctx->var1 -= ctx->var2;
159 	return 0;
160 }
161 
162 static int erst_exec_add_value(struct apei_exec_context *ctx,
163 			       struct acpi_whea_header *entry)
164 {
165 	int rc;
166 	u64 val;
167 
168 	rc = __apei_exec_read_register(entry, &val);
169 	if (rc)
170 		return rc;
171 	val += ctx->value;
172 	rc = __apei_exec_write_register(entry, val);
173 	return rc;
174 }
175 
176 static int erst_exec_subtract_value(struct apei_exec_context *ctx,
177 				    struct acpi_whea_header *entry)
178 {
179 	int rc;
180 	u64 val;
181 
182 	rc = __apei_exec_read_register(entry, &val);
183 	if (rc)
184 		return rc;
185 	val -= ctx->value;
186 	rc = __apei_exec_write_register(entry, val);
187 	return rc;
188 }
189 
190 static int erst_exec_stall(struct apei_exec_context *ctx,
191 			   struct acpi_whea_header *entry)
192 {
193 	u64 stall_time;
194 
195 	if (ctx->value > FIRMWARE_MAX_STALL) {
196 		if (!in_nmi())
197 			pr_warn(FW_WARN
198 			"Too long stall time for stall instruction: 0x%llx.\n",
199 				   ctx->value);
200 		stall_time = FIRMWARE_MAX_STALL;
201 	} else
202 		stall_time = ctx->value;
203 	udelay(stall_time);
204 	return 0;
205 }
206 
207 static int erst_exec_stall_while_true(struct apei_exec_context *ctx,
208 				      struct acpi_whea_header *entry)
209 {
210 	int rc;
211 	u64 val;
212 	u64 timeout;
213 	u64 stall_time;
214 
215 	timeout = erst_get_timeout();
216 
217 	if (ctx->var1 > FIRMWARE_MAX_STALL) {
218 		if (!in_nmi())
219 			pr_warn(FW_WARN
220 		"Too long stall time for stall while true instruction: 0x%llx.\n",
221 				   ctx->var1);
222 		stall_time = FIRMWARE_MAX_STALL;
223 	} else
224 		stall_time = ctx->var1;
225 
226 	for (;;) {
227 		rc = __apei_exec_read_register(entry, &val);
228 		if (rc)
229 			return rc;
230 		if (val != ctx->value)
231 			break;
232 		if (erst_timedout(&timeout, stall_time * NSEC_PER_USEC))
233 			return -EIO;
234 	}
235 	return 0;
236 }
237 
238 static int erst_exec_skip_next_instruction_if_true(
239 	struct apei_exec_context *ctx,
240 	struct acpi_whea_header *entry)
241 {
242 	int rc;
243 	u64 val;
244 
245 	rc = __apei_exec_read_register(entry, &val);
246 	if (rc)
247 		return rc;
248 	if (val == ctx->value) {
249 		ctx->ip += 2;
250 		return APEI_EXEC_SET_IP;
251 	}
252 
253 	return 0;
254 }
255 
256 static int erst_exec_goto(struct apei_exec_context *ctx,
257 			  struct acpi_whea_header *entry)
258 {
259 	ctx->ip = ctx->value;
260 	return APEI_EXEC_SET_IP;
261 }
262 
263 static int erst_exec_set_src_address_base(struct apei_exec_context *ctx,
264 					  struct acpi_whea_header *entry)
265 {
266 	return __apei_exec_read_register(entry, &ctx->src_base);
267 }
268 
269 static int erst_exec_set_dst_address_base(struct apei_exec_context *ctx,
270 					  struct acpi_whea_header *entry)
271 {
272 	return __apei_exec_read_register(entry, &ctx->dst_base);
273 }
274 
275 static int erst_exec_move_data(struct apei_exec_context *ctx,
276 			       struct acpi_whea_header *entry)
277 {
278 	int rc;
279 	u64 offset;
280 	void *src, *dst;
281 
282 	/* ioremap does not work in interrupt context */
283 	if (in_interrupt()) {
284 		pr_warn("MOVE_DATA can not be used in interrupt context.\n");
285 		return -EBUSY;
286 	}
287 
288 	rc = __apei_exec_read_register(entry, &offset);
289 	if (rc)
290 		return rc;
291 
292 	src = ioremap(ctx->src_base + offset, ctx->var2);
293 	if (!src)
294 		return -ENOMEM;
295 	dst = ioremap(ctx->dst_base + offset, ctx->var2);
296 	if (!dst) {
297 		iounmap(src);
298 		return -ENOMEM;
299 	}
300 
301 	memmove(dst, src, ctx->var2);
302 
303 	iounmap(src);
304 	iounmap(dst);
305 
306 	return 0;
307 }
308 
309 static struct apei_exec_ins_type erst_ins_type[] = {
310 	[ACPI_ERST_READ_REGISTER] = {
311 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
312 		.run = apei_exec_read_register,
313 	},
314 	[ACPI_ERST_READ_REGISTER_VALUE] = {
315 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
316 		.run = apei_exec_read_register_value,
317 	},
318 	[ACPI_ERST_WRITE_REGISTER] = {
319 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
320 		.run = apei_exec_write_register,
321 	},
322 	[ACPI_ERST_WRITE_REGISTER_VALUE] = {
323 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
324 		.run = apei_exec_write_register_value,
325 	},
326 	[ACPI_ERST_NOOP] = {
327 		.flags = 0,
328 		.run = apei_exec_noop,
329 	},
330 	[ACPI_ERST_LOAD_VAR1] = {
331 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
332 		.run = erst_exec_load_var1,
333 	},
334 	[ACPI_ERST_LOAD_VAR2] = {
335 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
336 		.run = erst_exec_load_var2,
337 	},
338 	[ACPI_ERST_STORE_VAR1] = {
339 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
340 		.run = erst_exec_store_var1,
341 	},
342 	[ACPI_ERST_ADD] = {
343 		.flags = 0,
344 		.run = erst_exec_add,
345 	},
346 	[ACPI_ERST_SUBTRACT] = {
347 		.flags = 0,
348 		.run = erst_exec_subtract,
349 	},
350 	[ACPI_ERST_ADD_VALUE] = {
351 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
352 		.run = erst_exec_add_value,
353 	},
354 	[ACPI_ERST_SUBTRACT_VALUE] = {
355 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
356 		.run = erst_exec_subtract_value,
357 	},
358 	[ACPI_ERST_STALL] = {
359 		.flags = 0,
360 		.run = erst_exec_stall,
361 	},
362 	[ACPI_ERST_STALL_WHILE_TRUE] = {
363 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
364 		.run = erst_exec_stall_while_true,
365 	},
366 	[ACPI_ERST_SKIP_NEXT_IF_TRUE] = {
367 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
368 		.run = erst_exec_skip_next_instruction_if_true,
369 	},
370 	[ACPI_ERST_GOTO] = {
371 		.flags = 0,
372 		.run = erst_exec_goto,
373 	},
374 	[ACPI_ERST_SET_SRC_ADDRESS_BASE] = {
375 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
376 		.run = erst_exec_set_src_address_base,
377 	},
378 	[ACPI_ERST_SET_DST_ADDRESS_BASE] = {
379 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
380 		.run = erst_exec_set_dst_address_base,
381 	},
382 	[ACPI_ERST_MOVE_DATA] = {
383 		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
384 		.run = erst_exec_move_data,
385 	},
386 };
387 
388 static inline void erst_exec_ctx_init(struct apei_exec_context *ctx)
389 {
390 	apei_exec_ctx_init(ctx, erst_ins_type, ARRAY_SIZE(erst_ins_type),
391 			   ERST_TAB_ENTRY(erst_tab), erst_tab->entries);
392 }
393 
394 static int erst_get_erange(struct erst_erange *range)
395 {
396 	struct apei_exec_context ctx;
397 	int rc;
398 
399 	erst_exec_ctx_init(&ctx);
400 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_RANGE);
401 	if (rc)
402 		return rc;
403 	range->base = apei_exec_ctx_get_output(&ctx);
404 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_LENGTH);
405 	if (rc)
406 		return rc;
407 	range->size = apei_exec_ctx_get_output(&ctx);
408 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_ATTRIBUTES);
409 	if (rc)
410 		return rc;
411 	range->attr = apei_exec_ctx_get_output(&ctx);
412 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_TIMINGS);
413 	if (rc == 0)
414 		range->timings = apei_exec_ctx_get_output(&ctx);
415 	else if (rc == -ENOENT)
416 		range->timings = 0;
417 	else
418 		return rc;
419 
420 	return 0;
421 }
422 
423 static ssize_t __erst_get_record_count(void)
424 {
425 	struct apei_exec_context ctx;
426 	int rc;
427 
428 	erst_exec_ctx_init(&ctx);
429 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_RECORD_COUNT);
430 	if (rc)
431 		return rc;
432 	return apei_exec_ctx_get_output(&ctx);
433 }
434 
435 ssize_t erst_get_record_count(void)
436 {
437 	ssize_t count;
438 	unsigned long flags;
439 
440 	if (erst_disable)
441 		return -ENODEV;
442 
443 	raw_spin_lock_irqsave(&erst_lock, flags);
444 	count = __erst_get_record_count();
445 	raw_spin_unlock_irqrestore(&erst_lock, flags);
446 
447 	return count;
448 }
449 EXPORT_SYMBOL_GPL(erst_get_record_count);
450 
451 #define ERST_RECORD_ID_CACHE_SIZE_MIN	16
452 #define ERST_RECORD_ID_CACHE_SIZE_MAX	1024
453 
454 struct erst_record_id_cache {
455 	struct mutex lock;
456 	u64 *entries;
457 	int len;
458 	int size;
459 	int refcount;
460 };
461 
462 static struct erst_record_id_cache erst_record_id_cache = {
463 	.lock = __MUTEX_INITIALIZER(erst_record_id_cache.lock),
464 	.refcount = 0,
465 };
466 
467 static int __erst_get_next_record_id(u64 *record_id)
468 {
469 	struct apei_exec_context ctx;
470 	int rc;
471 
472 	erst_exec_ctx_init(&ctx);
473 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_RECORD_ID);
474 	if (rc)
475 		return rc;
476 	*record_id = apei_exec_ctx_get_output(&ctx);
477 
478 	return 0;
479 }
480 
481 int erst_get_record_id_begin(int *pos)
482 {
483 	int rc;
484 
485 	if (erst_disable)
486 		return -ENODEV;
487 
488 	rc = mutex_lock_interruptible(&erst_record_id_cache.lock);
489 	if (rc)
490 		return rc;
491 	erst_record_id_cache.refcount++;
492 	mutex_unlock(&erst_record_id_cache.lock);
493 
494 	*pos = 0;
495 
496 	return 0;
497 }
498 EXPORT_SYMBOL_GPL(erst_get_record_id_begin);
499 
500 /* erst_record_id_cache.lock must be held by caller */
501 static int __erst_record_id_cache_add_one(void)
502 {
503 	u64 id, prev_id, first_id;
504 	int i, rc;
505 	u64 *entries;
506 	unsigned long flags;
507 
508 	id = prev_id = first_id = APEI_ERST_INVALID_RECORD_ID;
509 retry:
510 	raw_spin_lock_irqsave(&erst_lock, flags);
511 	rc = __erst_get_next_record_id(&id);
512 	raw_spin_unlock_irqrestore(&erst_lock, flags);
513 	if (rc == -ENOENT)
514 		return 0;
515 	if (rc)
516 		return rc;
517 	if (id == APEI_ERST_INVALID_RECORD_ID)
518 		return 0;
519 	/* can not skip current ID, or loop back to first ID */
520 	if (id == prev_id || id == first_id)
521 		return 0;
522 	if (first_id == APEI_ERST_INVALID_RECORD_ID)
523 		first_id = id;
524 	prev_id = id;
525 
526 	entries = erst_record_id_cache.entries;
527 	for (i = 0; i < erst_record_id_cache.len; i++) {
528 		if (entries[i] == id)
529 			break;
530 	}
531 	/* record id already in cache, try next */
532 	if (i < erst_record_id_cache.len)
533 		goto retry;
534 	if (erst_record_id_cache.len >= erst_record_id_cache.size) {
535 		int new_size;
536 		u64 *new_entries;
537 
538 		new_size = erst_record_id_cache.size * 2;
539 		new_size = clamp_val(new_size, ERST_RECORD_ID_CACHE_SIZE_MIN,
540 				     ERST_RECORD_ID_CACHE_SIZE_MAX);
541 		if (new_size <= erst_record_id_cache.size) {
542 			if (printk_ratelimit())
543 				pr_warn(FW_WARN "too many record IDs!\n");
544 			return 0;
545 		}
546 		new_entries = kvmalloc_array(new_size, sizeof(entries[0]),
547 					     GFP_KERNEL);
548 		if (!new_entries)
549 			return -ENOMEM;
550 		memcpy(new_entries, entries,
551 		       erst_record_id_cache.len * sizeof(entries[0]));
552 		kvfree(entries);
553 		erst_record_id_cache.entries = entries = new_entries;
554 		erst_record_id_cache.size = new_size;
555 	}
556 	entries[i] = id;
557 	erst_record_id_cache.len++;
558 
559 	return 1;
560 }
561 
562 /*
563  * Get the record ID of an existing error record on the persistent
564  * storage. If there is no error record on the persistent storage, the
565  * returned record_id is APEI_ERST_INVALID_RECORD_ID.
566  */
567 int erst_get_record_id_next(int *pos, u64 *record_id)
568 {
569 	int rc = 0;
570 	u64 *entries;
571 
572 	if (erst_disable)
573 		return -ENODEV;
574 
575 	/* must be enclosed by erst_get_record_id_begin/end */
576 	BUG_ON(!erst_record_id_cache.refcount);
577 	BUG_ON(*pos < 0 || *pos > erst_record_id_cache.len);
578 
579 	mutex_lock(&erst_record_id_cache.lock);
580 	entries = erst_record_id_cache.entries;
581 	for (; *pos < erst_record_id_cache.len; (*pos)++)
582 		if (entries[*pos] != APEI_ERST_INVALID_RECORD_ID)
583 			break;
584 	/* found next record id in cache */
585 	if (*pos < erst_record_id_cache.len) {
586 		*record_id = entries[*pos];
587 		(*pos)++;
588 		goto out_unlock;
589 	}
590 
591 	/* Try to add one more record ID to cache */
592 	rc = __erst_record_id_cache_add_one();
593 	if (rc < 0)
594 		goto out_unlock;
595 	/* successfully add one new ID */
596 	if (rc == 1) {
597 		*record_id = erst_record_id_cache.entries[*pos];
598 		(*pos)++;
599 		rc = 0;
600 	} else {
601 		*pos = -1;
602 		*record_id = APEI_ERST_INVALID_RECORD_ID;
603 	}
604 out_unlock:
605 	mutex_unlock(&erst_record_id_cache.lock);
606 
607 	return rc;
608 }
609 EXPORT_SYMBOL_GPL(erst_get_record_id_next);
610 
611 /* erst_record_id_cache.lock must be held by caller */
612 static void __erst_record_id_cache_compact(void)
613 {
614 	int i, wpos = 0;
615 	u64 *entries;
616 
617 	if (erst_record_id_cache.refcount)
618 		return;
619 
620 	entries = erst_record_id_cache.entries;
621 	for (i = 0; i < erst_record_id_cache.len; i++) {
622 		if (entries[i] == APEI_ERST_INVALID_RECORD_ID)
623 			continue;
624 		if (wpos != i)
625 			entries[wpos] = entries[i];
626 		wpos++;
627 	}
628 	erst_record_id_cache.len = wpos;
629 }
630 
631 void erst_get_record_id_end(void)
632 {
633 	/*
634 	 * erst_disable != 0 should be detected by invoker via the
635 	 * return value of erst_get_record_id_begin/next, so this
636 	 * function should not be called for erst_disable != 0.
637 	 */
638 	BUG_ON(erst_disable);
639 
640 	mutex_lock(&erst_record_id_cache.lock);
641 	erst_record_id_cache.refcount--;
642 	BUG_ON(erst_record_id_cache.refcount < 0);
643 	__erst_record_id_cache_compact();
644 	mutex_unlock(&erst_record_id_cache.lock);
645 }
646 EXPORT_SYMBOL_GPL(erst_get_record_id_end);
647 
648 static int __erst_write_to_storage(u64 offset)
649 {
650 	struct apei_exec_context ctx;
651 	u64 timeout;
652 	u64 val;
653 	int rc;
654 
655 	timeout = erst_get_timeout();
656 
657 	erst_exec_ctx_init(&ctx);
658 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_WRITE);
659 	if (rc)
660 		return rc;
661 	apei_exec_ctx_set_input(&ctx, offset);
662 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_OFFSET);
663 	if (rc)
664 		return rc;
665 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
666 	if (rc)
667 		return rc;
668 	for (;;) {
669 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
670 		if (rc)
671 			return rc;
672 		val = apei_exec_ctx_get_output(&ctx);
673 		if (!val)
674 			break;
675 		if (erst_timedout(&timeout, SPIN_UNIT))
676 			return -EIO;
677 	}
678 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
679 	if (rc)
680 		return rc;
681 	val = apei_exec_ctx_get_output(&ctx);
682 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
683 	if (rc)
684 		return rc;
685 
686 	return erst_errno(val);
687 }
688 
689 static int __erst_read_from_storage(u64 record_id, u64 offset)
690 {
691 	struct apei_exec_context ctx;
692 	u64 timeout;
693 	u64 val;
694 	int rc;
695 
696 	timeout = erst_get_timeout();
697 
698 	erst_exec_ctx_init(&ctx);
699 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_READ);
700 	if (rc)
701 		return rc;
702 	apei_exec_ctx_set_input(&ctx, offset);
703 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_OFFSET);
704 	if (rc)
705 		return rc;
706 	apei_exec_ctx_set_input(&ctx, record_id);
707 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_ID);
708 	if (rc)
709 		return rc;
710 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
711 	if (rc)
712 		return rc;
713 	for (;;) {
714 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
715 		if (rc)
716 			return rc;
717 		val = apei_exec_ctx_get_output(&ctx);
718 		if (!val)
719 			break;
720 		if (erst_timedout(&timeout, SPIN_UNIT))
721 			return -EIO;
722 	}
723 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
724 	if (rc)
725 		return rc;
726 	val = apei_exec_ctx_get_output(&ctx);
727 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
728 	if (rc)
729 		return rc;
730 
731 	return erst_errno(val);
732 }
733 
734 static int __erst_clear_from_storage(u64 record_id)
735 {
736 	struct apei_exec_context ctx;
737 	u64 timeout;
738 	u64 val;
739 	int rc;
740 
741 	timeout = erst_get_timeout();
742 
743 	erst_exec_ctx_init(&ctx);
744 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_CLEAR);
745 	if (rc)
746 		return rc;
747 	apei_exec_ctx_set_input(&ctx, record_id);
748 	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_ID);
749 	if (rc)
750 		return rc;
751 	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
752 	if (rc)
753 		return rc;
754 	for (;;) {
755 		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
756 		if (rc)
757 			return rc;
758 		val = apei_exec_ctx_get_output(&ctx);
759 		if (!val)
760 			break;
761 		if (erst_timedout(&timeout, SPIN_UNIT))
762 			return -EIO;
763 	}
764 	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
765 	if (rc)
766 		return rc;
767 	val = apei_exec_ctx_get_output(&ctx);
768 	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
769 	if (rc)
770 		return rc;
771 
772 	return erst_errno(val);
773 }
774 
775 /* NVRAM ERST Error Log Address Range is not supported yet */
776 static void pr_unimpl_nvram(void)
777 {
778 	if (printk_ratelimit())
779 		pr_warn("NVRAM ERST Log Address Range not implemented yet.\n");
780 }
781 
782 static int __erst_write_to_nvram(const struct cper_record_header *record)
783 {
784 	/* do not print message, because printk is not safe for NMI */
785 	return -ENOSYS;
786 }
787 
788 static int __erst_read_to_erange_from_nvram(u64 record_id, u64 *offset)
789 {
790 	pr_unimpl_nvram();
791 	return -ENOSYS;
792 }
793 
794 static int __erst_clear_from_nvram(u64 record_id)
795 {
796 	pr_unimpl_nvram();
797 	return -ENOSYS;
798 }
799 
800 int erst_write(const struct cper_record_header *record)
801 {
802 	int rc;
803 	unsigned long flags;
804 	struct cper_record_header *rcd_erange;
805 
806 	if (erst_disable)
807 		return -ENODEV;
808 
809 	if (memcmp(record->signature, CPER_SIG_RECORD, CPER_SIG_SIZE))
810 		return -EINVAL;
811 
812 	if (erst_erange.attr & ERST_RANGE_NVRAM) {
813 		if (!raw_spin_trylock_irqsave(&erst_lock, flags))
814 			return -EBUSY;
815 		rc = __erst_write_to_nvram(record);
816 		raw_spin_unlock_irqrestore(&erst_lock, flags);
817 		return rc;
818 	}
819 
820 	if (record->record_length > erst_erange.size)
821 		return -EINVAL;
822 
823 	if (!raw_spin_trylock_irqsave(&erst_lock, flags))
824 		return -EBUSY;
825 	memcpy(erst_erange.vaddr, record, record->record_length);
826 	rcd_erange = erst_erange.vaddr;
827 	/* signature for serialization system */
828 	memcpy(&rcd_erange->persistence_information, "ER", 2);
829 
830 	rc = __erst_write_to_storage(0);
831 	raw_spin_unlock_irqrestore(&erst_lock, flags);
832 
833 	return rc;
834 }
835 EXPORT_SYMBOL_GPL(erst_write);
836 
837 static int __erst_read_to_erange(u64 record_id, u64 *offset)
838 {
839 	int rc;
840 
841 	if (erst_erange.attr & ERST_RANGE_NVRAM)
842 		return __erst_read_to_erange_from_nvram(
843 			record_id, offset);
844 
845 	rc = __erst_read_from_storage(record_id, 0);
846 	if (rc)
847 		return rc;
848 	*offset = 0;
849 
850 	return 0;
851 }
852 
853 static ssize_t __erst_read(u64 record_id, struct cper_record_header *record,
854 			   size_t buflen)
855 {
856 	int rc;
857 	u64 offset, len = 0;
858 	struct cper_record_header *rcd_tmp;
859 
860 	rc = __erst_read_to_erange(record_id, &offset);
861 	if (rc)
862 		return rc;
863 	rcd_tmp = erst_erange.vaddr + offset;
864 	len = rcd_tmp->record_length;
865 	if (len <= buflen)
866 		memcpy(record, rcd_tmp, len);
867 
868 	return len;
869 }
870 
871 /*
872  * If return value > buflen, the buffer size is not big enough,
873  * else if return value < 0, something goes wrong,
874  * else everything is OK, and return value is record length
875  */
876 ssize_t erst_read(u64 record_id, struct cper_record_header *record,
877 		  size_t buflen)
878 {
879 	ssize_t len;
880 	unsigned long flags;
881 
882 	if (erst_disable)
883 		return -ENODEV;
884 
885 	raw_spin_lock_irqsave(&erst_lock, flags);
886 	len = __erst_read(record_id, record, buflen);
887 	raw_spin_unlock_irqrestore(&erst_lock, flags);
888 	return len;
889 }
890 EXPORT_SYMBOL_GPL(erst_read);
891 
892 static void erst_clear_cache(u64 record_id)
893 {
894 	int i;
895 	u64 *entries;
896 
897 	mutex_lock(&erst_record_id_cache.lock);
898 
899 	entries = erst_record_id_cache.entries;
900 	for (i = 0; i < erst_record_id_cache.len; i++) {
901 		if (entries[i] == record_id)
902 			entries[i] = APEI_ERST_INVALID_RECORD_ID;
903 	}
904 	__erst_record_id_cache_compact();
905 
906 	mutex_unlock(&erst_record_id_cache.lock);
907 }
908 
909 ssize_t erst_read_record(u64 record_id, struct cper_record_header *record,
910 		size_t buflen, size_t recordlen, const guid_t *creatorid)
911 {
912 	ssize_t len;
913 
914 	/*
915 	 * if creatorid is NULL, read any record for erst-dbg module
916 	 */
917 	if (creatorid == NULL) {
918 		len = erst_read(record_id, record, buflen);
919 		if (len == -ENOENT)
920 			erst_clear_cache(record_id);
921 
922 		return len;
923 	}
924 
925 	len = erst_read(record_id, record, buflen);
926 	/*
927 	 * if erst_read return value is -ENOENT skip to next record_id,
928 	 * and clear the record_id cache.
929 	 */
930 	if (len == -ENOENT) {
931 		erst_clear_cache(record_id);
932 		goto out;
933 	}
934 
935 	if (len < 0)
936 		goto out;
937 
938 	/*
939 	 * if erst_read return value is less than record head length,
940 	 * consider it as -EIO, and clear the record_id cache.
941 	 */
942 	if (len < recordlen) {
943 		len = -EIO;
944 		erst_clear_cache(record_id);
945 		goto out;
946 	}
947 
948 	/*
949 	 * if creatorid is not wanted, consider it as not found,
950 	 * for skipping to next record_id.
951 	 */
952 	if (!guid_equal(&record->creator_id, creatorid))
953 		len = -ENOENT;
954 
955 out:
956 	return len;
957 }
958 EXPORT_SYMBOL_GPL(erst_read_record);
959 
960 int erst_clear(u64 record_id)
961 {
962 	int rc, i;
963 	unsigned long flags;
964 	u64 *entries;
965 
966 	if (erst_disable)
967 		return -ENODEV;
968 
969 	rc = mutex_lock_interruptible(&erst_record_id_cache.lock);
970 	if (rc)
971 		return rc;
972 	raw_spin_lock_irqsave(&erst_lock, flags);
973 	if (erst_erange.attr & ERST_RANGE_NVRAM)
974 		rc = __erst_clear_from_nvram(record_id);
975 	else
976 		rc = __erst_clear_from_storage(record_id);
977 	raw_spin_unlock_irqrestore(&erst_lock, flags);
978 	if (rc)
979 		goto out;
980 	entries = erst_record_id_cache.entries;
981 	for (i = 0; i < erst_record_id_cache.len; i++) {
982 		if (entries[i] == record_id)
983 			entries[i] = APEI_ERST_INVALID_RECORD_ID;
984 	}
985 	__erst_record_id_cache_compact();
986 out:
987 	mutex_unlock(&erst_record_id_cache.lock);
988 	return rc;
989 }
990 EXPORT_SYMBOL_GPL(erst_clear);
991 
992 static int __init setup_erst_disable(char *str)
993 {
994 	erst_disable = 1;
995 	return 1;
996 }
997 
998 __setup("erst_disable", setup_erst_disable);
999 
1000 static int erst_check_table(struct acpi_table_erst *erst_tab)
1001 {
1002 	if ((erst_tab->header_length !=
1003 	     (sizeof(struct acpi_table_erst) - sizeof(erst_tab->header)))
1004 	    && (erst_tab->header_length != sizeof(struct acpi_table_erst)))
1005 		return -EINVAL;
1006 	if (erst_tab->header.length < sizeof(struct acpi_table_erst))
1007 		return -EINVAL;
1008 	if (erst_tab->entries !=
1009 	    (erst_tab->header.length - sizeof(struct acpi_table_erst)) /
1010 	    sizeof(struct acpi_erst_entry))
1011 		return -EINVAL;
1012 
1013 	return 0;
1014 }
1015 
1016 static int erst_open_pstore(struct pstore_info *psi);
1017 static int erst_close_pstore(struct pstore_info *psi);
1018 static ssize_t erst_reader(struct pstore_record *record);
1019 static int erst_writer(struct pstore_record *record);
1020 static int erst_clearer(struct pstore_record *record);
1021 
1022 static struct pstore_info erst_info = {
1023 	.owner		= THIS_MODULE,
1024 	.name		= "erst",
1025 	.flags		= PSTORE_FLAGS_DMESG,
1026 	.open		= erst_open_pstore,
1027 	.close		= erst_close_pstore,
1028 	.read		= erst_reader,
1029 	.write		= erst_writer,
1030 	.erase		= erst_clearer
1031 };
1032 
1033 #define CPER_CREATOR_PSTORE						\
1034 	GUID_INIT(0x75a574e3, 0x5052, 0x4b29, 0x8a, 0x8e, 0xbe, 0x2c,	\
1035 		  0x64, 0x90, 0xb8, 0x9d)
1036 #define CPER_SECTION_TYPE_DMESG						\
1037 	GUID_INIT(0xc197e04e, 0xd545, 0x4a70, 0x9c, 0x17, 0xa5, 0x54,	\
1038 		  0x94, 0x19, 0xeb, 0x12)
1039 #define CPER_SECTION_TYPE_DMESG_Z					\
1040 	GUID_INIT(0x4f118707, 0x04dd, 0x4055, 0xb5, 0xdd, 0x95, 0x6d,	\
1041 		  0x34, 0xdd, 0xfa, 0xc6)
1042 #define CPER_SECTION_TYPE_MCE						\
1043 	GUID_INIT(0xfe08ffbe, 0x95e4, 0x4be7, 0xbc, 0x73, 0x40, 0x96,	\
1044 		  0x04, 0x4a, 0x38, 0xfc)
1045 
1046 struct cper_pstore_record {
1047 	struct cper_record_header hdr;
1048 	struct cper_section_descriptor sec_hdr;
1049 	char data[];
1050 } __packed;
1051 
1052 static int reader_pos;
1053 
1054 static int erst_open_pstore(struct pstore_info *psi)
1055 {
1056 	if (erst_disable)
1057 		return -ENODEV;
1058 
1059 	return erst_get_record_id_begin(&reader_pos);
1060 }
1061 
1062 static int erst_close_pstore(struct pstore_info *psi)
1063 {
1064 	erst_get_record_id_end();
1065 
1066 	return 0;
1067 }
1068 
1069 static ssize_t erst_reader(struct pstore_record *record)
1070 {
1071 	int rc;
1072 	ssize_t len = 0;
1073 	u64 record_id;
1074 	struct cper_pstore_record *rcd;
1075 	size_t rcd_len = sizeof(*rcd) + erst_info.bufsize;
1076 
1077 	if (erst_disable)
1078 		return -ENODEV;
1079 
1080 	rcd = kmalloc(rcd_len, GFP_KERNEL);
1081 	if (!rcd) {
1082 		rc = -ENOMEM;
1083 		goto out;
1084 	}
1085 skip:
1086 	rc = erst_get_record_id_next(&reader_pos, &record_id);
1087 	if (rc)
1088 		goto out;
1089 
1090 	/* no more record */
1091 	if (record_id == APEI_ERST_INVALID_RECORD_ID) {
1092 		rc = -EINVAL;
1093 		goto out;
1094 	}
1095 
1096 	len = erst_read_record(record_id, &rcd->hdr, rcd_len, sizeof(*rcd),
1097 			&CPER_CREATOR_PSTORE);
1098 	/* The record may be cleared by others, try read next record */
1099 	if (len == -ENOENT)
1100 		goto skip;
1101 	else if (len < 0)
1102 		goto out;
1103 
1104 	record->buf = kmalloc(len, GFP_KERNEL);
1105 	if (record->buf == NULL) {
1106 		rc = -ENOMEM;
1107 		goto out;
1108 	}
1109 	memcpy(record->buf, rcd->data, len - sizeof(*rcd));
1110 	record->id = record_id;
1111 	record->compressed = false;
1112 	record->ecc_notice_size = 0;
1113 	if (guid_equal(&rcd->sec_hdr.section_type, &CPER_SECTION_TYPE_DMESG_Z)) {
1114 		record->type = PSTORE_TYPE_DMESG;
1115 		record->compressed = true;
1116 	} else if (guid_equal(&rcd->sec_hdr.section_type, &CPER_SECTION_TYPE_DMESG))
1117 		record->type = PSTORE_TYPE_DMESG;
1118 	else if (guid_equal(&rcd->sec_hdr.section_type, &CPER_SECTION_TYPE_MCE))
1119 		record->type = PSTORE_TYPE_MCE;
1120 	else
1121 		record->type = PSTORE_TYPE_MAX;
1122 
1123 	if (rcd->hdr.validation_bits & CPER_VALID_TIMESTAMP)
1124 		record->time.tv_sec = rcd->hdr.timestamp;
1125 	else
1126 		record->time.tv_sec = 0;
1127 	record->time.tv_nsec = 0;
1128 
1129 out:
1130 	kfree(rcd);
1131 	return (rc < 0) ? rc : (len - sizeof(*rcd));
1132 }
1133 
1134 static int erst_writer(struct pstore_record *record)
1135 {
1136 	struct cper_pstore_record *rcd = (struct cper_pstore_record *)
1137 					(erst_info.buf - sizeof(*rcd));
1138 	int ret;
1139 
1140 	memset(rcd, 0, sizeof(*rcd));
1141 	memcpy(rcd->hdr.signature, CPER_SIG_RECORD, CPER_SIG_SIZE);
1142 	rcd->hdr.revision = CPER_RECORD_REV;
1143 	rcd->hdr.signature_end = CPER_SIG_END;
1144 	rcd->hdr.section_count = 1;
1145 	rcd->hdr.error_severity = CPER_SEV_FATAL;
1146 	/* timestamp valid. platform_id, partition_id are invalid */
1147 	rcd->hdr.validation_bits = CPER_VALID_TIMESTAMP;
1148 	rcd->hdr.timestamp = ktime_get_real_seconds();
1149 	rcd->hdr.record_length = sizeof(*rcd) + record->size;
1150 	rcd->hdr.creator_id = CPER_CREATOR_PSTORE;
1151 	rcd->hdr.notification_type = CPER_NOTIFY_MCE;
1152 	rcd->hdr.record_id = cper_next_record_id();
1153 	rcd->hdr.flags = CPER_HW_ERROR_FLAGS_PREVERR;
1154 
1155 	rcd->sec_hdr.section_offset = sizeof(*rcd);
1156 	rcd->sec_hdr.section_length = record->size;
1157 	rcd->sec_hdr.revision = CPER_SEC_REV;
1158 	/* fru_id and fru_text is invalid */
1159 	rcd->sec_hdr.validation_bits = 0;
1160 	rcd->sec_hdr.flags = CPER_SEC_PRIMARY;
1161 	switch (record->type) {
1162 	case PSTORE_TYPE_DMESG:
1163 		if (record->compressed)
1164 			rcd->sec_hdr.section_type = CPER_SECTION_TYPE_DMESG_Z;
1165 		else
1166 			rcd->sec_hdr.section_type = CPER_SECTION_TYPE_DMESG;
1167 		break;
1168 	case PSTORE_TYPE_MCE:
1169 		rcd->sec_hdr.section_type = CPER_SECTION_TYPE_MCE;
1170 		break;
1171 	default:
1172 		return -EINVAL;
1173 	}
1174 	rcd->sec_hdr.section_severity = CPER_SEV_FATAL;
1175 
1176 	ret = erst_write(&rcd->hdr);
1177 	record->id = rcd->hdr.record_id;
1178 
1179 	return ret;
1180 }
1181 
1182 static int erst_clearer(struct pstore_record *record)
1183 {
1184 	return erst_clear(record->id);
1185 }
1186 
1187 static int __init erst_init(void)
1188 {
1189 	int rc = 0;
1190 	acpi_status status;
1191 	struct apei_exec_context ctx;
1192 	struct apei_resources erst_resources;
1193 	struct resource *r;
1194 	char *buf;
1195 
1196 	if (acpi_disabled)
1197 		goto err;
1198 
1199 	if (erst_disable) {
1200 		pr_info(
1201 	"Error Record Serialization Table (ERST) support is disabled.\n");
1202 		goto err;
1203 	}
1204 
1205 	status = acpi_get_table(ACPI_SIG_ERST, 0,
1206 				(struct acpi_table_header **)&erst_tab);
1207 	if (status == AE_NOT_FOUND)
1208 		goto err;
1209 	else if (ACPI_FAILURE(status)) {
1210 		const char *msg = acpi_format_exception(status);
1211 		pr_err("Failed to get table, %s\n", msg);
1212 		rc = -EINVAL;
1213 		goto err;
1214 	}
1215 
1216 	rc = erst_check_table(erst_tab);
1217 	if (rc) {
1218 		pr_err(FW_BUG "ERST table is invalid.\n");
1219 		goto err_put_erst_tab;
1220 	}
1221 
1222 	apei_resources_init(&erst_resources);
1223 	erst_exec_ctx_init(&ctx);
1224 	rc = apei_exec_collect_resources(&ctx, &erst_resources);
1225 	if (rc)
1226 		goto err_fini;
1227 	rc = apei_resources_request(&erst_resources, "APEI ERST");
1228 	if (rc)
1229 		goto err_fini;
1230 	rc = apei_exec_pre_map_gars(&ctx);
1231 	if (rc)
1232 		goto err_release;
1233 	rc = erst_get_erange(&erst_erange);
1234 	if (rc) {
1235 		if (rc == -ENODEV)
1236 			pr_info(
1237 	"The corresponding hardware device or firmware implementation "
1238 	"is not available.\n");
1239 		else
1240 			pr_err("Failed to get Error Log Address Range.\n");
1241 		goto err_unmap_reg;
1242 	}
1243 
1244 	r = request_mem_region(erst_erange.base, erst_erange.size, "APEI ERST");
1245 	if (!r) {
1246 		pr_err("Can not request [mem %#010llx-%#010llx] for ERST.\n",
1247 		       (unsigned long long)erst_erange.base,
1248 		       (unsigned long long)erst_erange.base + erst_erange.size - 1);
1249 		rc = -EIO;
1250 		goto err_unmap_reg;
1251 	}
1252 	rc = -ENOMEM;
1253 	erst_erange.vaddr = ioremap_cache(erst_erange.base,
1254 					  erst_erange.size);
1255 	if (!erst_erange.vaddr)
1256 		goto err_release_erange;
1257 
1258 	pr_info(
1259 	"Error Record Serialization Table (ERST) support is initialized.\n");
1260 
1261 	buf = kmalloc(erst_erange.size, GFP_KERNEL);
1262 	if (buf) {
1263 		erst_info.buf = buf + sizeof(struct cper_pstore_record);
1264 		erst_info.bufsize = erst_erange.size -
1265 				    sizeof(struct cper_pstore_record);
1266 		rc = pstore_register(&erst_info);
1267 		if (rc) {
1268 			if (rc != -EPERM)
1269 				pr_info(
1270 				"Could not register with persistent store.\n");
1271 			erst_info.buf = NULL;
1272 			erst_info.bufsize = 0;
1273 			kfree(buf);
1274 		}
1275 	} else
1276 		pr_err(
1277 		"Failed to allocate %lld bytes for persistent store error log.\n",
1278 		erst_erange.size);
1279 
1280 	/* Cleanup ERST Resources */
1281 	apei_resources_fini(&erst_resources);
1282 
1283 	return 0;
1284 
1285 err_release_erange:
1286 	release_mem_region(erst_erange.base, erst_erange.size);
1287 err_unmap_reg:
1288 	apei_exec_post_unmap_gars(&ctx);
1289 err_release:
1290 	apei_resources_release(&erst_resources);
1291 err_fini:
1292 	apei_resources_fini(&erst_resources);
1293 err_put_erst_tab:
1294 	acpi_put_table((struct acpi_table_header *)erst_tab);
1295 err:
1296 	erst_disable = 1;
1297 	return rc;
1298 }
1299 
1300 device_initcall(erst_init);
1301