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
  52int erst_disable;
  53EXPORT_SYMBOL_GPL(erst_disable);
  54
  55static struct acpi_table_erst *erst_tab;
  56
  57/* ERST Error Log Address Range atrributes */
  58#define ERST_RANGE_RESERVED	0x0001
  59#define ERST_RANGE_NVRAM	0x0002
  60#define ERST_RANGE_SLOW		0x0004
  61
  62/*
  63 * ERST Error Log Address Range, used as buffer for reading/writing
  64 * error records.
  65 */
  66static struct erst_erange {
  67	u64 base;
  68	u64 size;
  69	void __iomem *vaddr;
  70	u32 attr;
  71} erst_erange;
  72
  73/*
  74 * Prevent ERST interpreter to run simultaneously, because the
  75 * corresponding firmware implementation may not work properly when
  76 * invoked simultaneously.
  77 *
  78 * It is used to provide exclusive accessing for ERST Error Log
  79 * Address Range too.
  80 */
  81static DEFINE_RAW_SPINLOCK(erst_lock);
  82
  83static inline int erst_errno(int command_status)
  84{
  85	switch (command_status) {
  86	case ERST_STATUS_SUCCESS:
  87		return 0;
  88	case ERST_STATUS_HARDWARE_NOT_AVAILABLE:
  89		return -ENODEV;
  90	case ERST_STATUS_NOT_ENOUGH_SPACE:
  91		return -ENOSPC;
  92	case ERST_STATUS_RECORD_STORE_EMPTY:
  93	case ERST_STATUS_RECORD_NOT_FOUND:
  94		return -ENOENT;
  95	default:
  96		return -EINVAL;
  97	}
  98}
  99
 100static int erst_timedout(u64 *t, u64 spin_unit)
 101{
 102	if ((s64)*t < spin_unit) {
 103		pr_warn(FW_WARN "Firmware does not respond in time.\n");
 104		return 1;
 105	}
 106	*t -= spin_unit;
 107	ndelay(spin_unit);
 108	touch_nmi_watchdog();
 109	return 0;
 110}
 111
 112static int erst_exec_load_var1(struct apei_exec_context *ctx,
 113			       struct acpi_whea_header *entry)
 114{
 115	return __apei_exec_read_register(entry, &ctx->var1);
 116}
 117
 118static int erst_exec_load_var2(struct apei_exec_context *ctx,
 119			       struct acpi_whea_header *entry)
 120{
 121	return __apei_exec_read_register(entry, &ctx->var2);
 122}
 123
 124static int erst_exec_store_var1(struct apei_exec_context *ctx,
 125				struct acpi_whea_header *entry)
 126{
 127	return __apei_exec_write_register(entry, ctx->var1);
 128}
 129
 130static int erst_exec_add(struct apei_exec_context *ctx,
 131			 struct acpi_whea_header *entry)
 132{
 133	ctx->var1 += ctx->var2;
 134	return 0;
 135}
 136
 137static int erst_exec_subtract(struct apei_exec_context *ctx,
 138			      struct acpi_whea_header *entry)
 139{
 140	ctx->var1 -= ctx->var2;
 141	return 0;
 142}
 143
 144static int erst_exec_add_value(struct apei_exec_context *ctx,
 145			       struct acpi_whea_header *entry)
 146{
 147	int rc;
 148	u64 val;
 149
 150	rc = __apei_exec_read_register(entry, &val);
 151	if (rc)
 152		return rc;
 153	val += ctx->value;
 154	rc = __apei_exec_write_register(entry, val);
 155	return rc;
 156}
 157
 158static int erst_exec_subtract_value(struct apei_exec_context *ctx,
 159				    struct acpi_whea_header *entry)
 160{
 161	int rc;
 162	u64 val;
 163
 164	rc = __apei_exec_read_register(entry, &val);
 165	if (rc)
 166		return rc;
 167	val -= ctx->value;
 168	rc = __apei_exec_write_register(entry, val);
 169	return rc;
 170}
 171
 172static int erst_exec_stall(struct apei_exec_context *ctx,
 173			   struct acpi_whea_header *entry)
 174{
 175	u64 stall_time;
 176
 177	if (ctx->value > FIRMWARE_MAX_STALL) {
 178		if (!in_nmi())
 179			pr_warn(FW_WARN
 180			"Too long stall time for stall instruction: 0x%llx.\n",
 181				   ctx->value);
 182		stall_time = FIRMWARE_MAX_STALL;
 183	} else
 184		stall_time = ctx->value;
 185	udelay(stall_time);
 186	return 0;
 187}
 188
 189static int erst_exec_stall_while_true(struct apei_exec_context *ctx,
 190				      struct acpi_whea_header *entry)
 191{
 192	int rc;
 193	u64 val;
 194	u64 timeout = FIRMWARE_TIMEOUT;
 195	u64 stall_time;
 196
 197	if (ctx->var1 > FIRMWARE_MAX_STALL) {
 198		if (!in_nmi())
 199			pr_warn(FW_WARN
 200		"Too long stall time for stall while true instruction: 0x%llx.\n",
 201				   ctx->var1);
 202		stall_time = FIRMWARE_MAX_STALL;
 203	} else
 204		stall_time = ctx->var1;
 205
 206	for (;;) {
 207		rc = __apei_exec_read_register(entry, &val);
 208		if (rc)
 209			return rc;
 210		if (val != ctx->value)
 211			break;
 212		if (erst_timedout(&timeout, stall_time * NSEC_PER_USEC))
 213			return -EIO;
 214	}
 215	return 0;
 216}
 217
 218static int erst_exec_skip_next_instruction_if_true(
 219	struct apei_exec_context *ctx,
 220	struct acpi_whea_header *entry)
 221{
 222	int rc;
 223	u64 val;
 224
 225	rc = __apei_exec_read_register(entry, &val);
 226	if (rc)
 227		return rc;
 228	if (val == ctx->value) {
 229		ctx->ip += 2;
 230		return APEI_EXEC_SET_IP;
 231	}
 232
 233	return 0;
 234}
 235
 236static int erst_exec_goto(struct apei_exec_context *ctx,
 237			  struct acpi_whea_header *entry)
 238{
 239	ctx->ip = ctx->value;
 240	return APEI_EXEC_SET_IP;
 241}
 242
 243static int erst_exec_set_src_address_base(struct apei_exec_context *ctx,
 244					  struct acpi_whea_header *entry)
 245{
 246	return __apei_exec_read_register(entry, &ctx->src_base);
 247}
 248
 249static int erst_exec_set_dst_address_base(struct apei_exec_context *ctx,
 250					  struct acpi_whea_header *entry)
 251{
 252	return __apei_exec_read_register(entry, &ctx->dst_base);
 253}
 254
 255static int erst_exec_move_data(struct apei_exec_context *ctx,
 256			       struct acpi_whea_header *entry)
 257{
 258	int rc;
 259	u64 offset;
 260	void *src, *dst;
 261
 262	/* ioremap does not work in interrupt context */
 263	if (in_interrupt()) {
 264		pr_warn("MOVE_DATA can not be used in interrupt context.\n");
 265		return -EBUSY;
 266	}
 267
 268	rc = __apei_exec_read_register(entry, &offset);
 269	if (rc)
 270		return rc;
 271
 272	src = ioremap(ctx->src_base + offset, ctx->var2);
 273	if (!src)
 274		return -ENOMEM;
 275	dst = ioremap(ctx->dst_base + offset, ctx->var2);
 276	if (!dst) {
 277		iounmap(src);
 278		return -ENOMEM;
 279	}
 280
 281	memmove(dst, src, ctx->var2);
 282
 283	iounmap(src);
 284	iounmap(dst);
 285
 286	return 0;
 287}
 288
 289static struct apei_exec_ins_type erst_ins_type[] = {
 290	[ACPI_ERST_READ_REGISTER] = {
 291		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 292		.run = apei_exec_read_register,
 293	},
 294	[ACPI_ERST_READ_REGISTER_VALUE] = {
 295		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 296		.run = apei_exec_read_register_value,
 297	},
 298	[ACPI_ERST_WRITE_REGISTER] = {
 299		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 300		.run = apei_exec_write_register,
 301	},
 302	[ACPI_ERST_WRITE_REGISTER_VALUE] = {
 303		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 304		.run = apei_exec_write_register_value,
 305	},
 306	[ACPI_ERST_NOOP] = {
 307		.flags = 0,
 308		.run = apei_exec_noop,
 309	},
 310	[ACPI_ERST_LOAD_VAR1] = {
 311		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 312		.run = erst_exec_load_var1,
 313	},
 314	[ACPI_ERST_LOAD_VAR2] = {
 315		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 316		.run = erst_exec_load_var2,
 317	},
 318	[ACPI_ERST_STORE_VAR1] = {
 319		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 320		.run = erst_exec_store_var1,
 321	},
 322	[ACPI_ERST_ADD] = {
 323		.flags = 0,
 324		.run = erst_exec_add,
 325	},
 326	[ACPI_ERST_SUBTRACT] = {
 327		.flags = 0,
 328		.run = erst_exec_subtract,
 329	},
 330	[ACPI_ERST_ADD_VALUE] = {
 331		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 332		.run = erst_exec_add_value,
 333	},
 334	[ACPI_ERST_SUBTRACT_VALUE] = {
 335		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 336		.run = erst_exec_subtract_value,
 337	},
 338	[ACPI_ERST_STALL] = {
 339		.flags = 0,
 340		.run = erst_exec_stall,
 341	},
 342	[ACPI_ERST_STALL_WHILE_TRUE] = {
 343		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 344		.run = erst_exec_stall_while_true,
 345	},
 346	[ACPI_ERST_SKIP_NEXT_IF_TRUE] = {
 347		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 348		.run = erst_exec_skip_next_instruction_if_true,
 349	},
 350	[ACPI_ERST_GOTO] = {
 351		.flags = 0,
 352		.run = erst_exec_goto,
 353	},
 354	[ACPI_ERST_SET_SRC_ADDRESS_BASE] = {
 355		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 356		.run = erst_exec_set_src_address_base,
 357	},
 358	[ACPI_ERST_SET_DST_ADDRESS_BASE] = {
 359		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 360		.run = erst_exec_set_dst_address_base,
 361	},
 362	[ACPI_ERST_MOVE_DATA] = {
 363		.flags = APEI_EXEC_INS_ACCESS_REGISTER,
 364		.run = erst_exec_move_data,
 365	},
 366};
 367
 368static inline void erst_exec_ctx_init(struct apei_exec_context *ctx)
 369{
 370	apei_exec_ctx_init(ctx, erst_ins_type, ARRAY_SIZE(erst_ins_type),
 371			   ERST_TAB_ENTRY(erst_tab), erst_tab->entries);
 372}
 373
 374static int erst_get_erange(struct erst_erange *range)
 375{
 376	struct apei_exec_context ctx;
 377	int rc;
 378
 379	erst_exec_ctx_init(&ctx);
 380	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_RANGE);
 381	if (rc)
 382		return rc;
 383	range->base = apei_exec_ctx_get_output(&ctx);
 384	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_LENGTH);
 385	if (rc)
 386		return rc;
 387	range->size = apei_exec_ctx_get_output(&ctx);
 388	rc = apei_exec_run(&ctx, ACPI_ERST_GET_ERROR_ATTRIBUTES);
 389	if (rc)
 390		return rc;
 391	range->attr = apei_exec_ctx_get_output(&ctx);
 392
 393	return 0;
 394}
 395
 396static ssize_t __erst_get_record_count(void)
 397{
 398	struct apei_exec_context ctx;
 399	int rc;
 400
 401	erst_exec_ctx_init(&ctx);
 402	rc = apei_exec_run(&ctx, ACPI_ERST_GET_RECORD_COUNT);
 403	if (rc)
 404		return rc;
 405	return apei_exec_ctx_get_output(&ctx);
 406}
 407
 408ssize_t erst_get_record_count(void)
 409{
 410	ssize_t count;
 411	unsigned long flags;
 412
 413	if (erst_disable)
 414		return -ENODEV;
 415
 416	raw_spin_lock_irqsave(&erst_lock, flags);
 417	count = __erst_get_record_count();
 418	raw_spin_unlock_irqrestore(&erst_lock, flags);
 419
 420	return count;
 421}
 422EXPORT_SYMBOL_GPL(erst_get_record_count);
 423
 424#define ERST_RECORD_ID_CACHE_SIZE_MIN	16
 425#define ERST_RECORD_ID_CACHE_SIZE_MAX	1024
 426
 427struct erst_record_id_cache {
 428	struct mutex lock;
 429	u64 *entries;
 430	int len;
 431	int size;
 432	int refcount;
 433};
 434
 435static struct erst_record_id_cache erst_record_id_cache = {
 436	.lock = __MUTEX_INITIALIZER(erst_record_id_cache.lock),
 437	.refcount = 0,
 438};
 439
 440static int __erst_get_next_record_id(u64 *record_id)
 441{
 442	struct apei_exec_context ctx;
 443	int rc;
 444
 445	erst_exec_ctx_init(&ctx);
 446	rc = apei_exec_run(&ctx, ACPI_ERST_GET_RECORD_ID);
 447	if (rc)
 448		return rc;
 449	*record_id = apei_exec_ctx_get_output(&ctx);
 450
 451	return 0;
 452}
 453
 454int erst_get_record_id_begin(int *pos)
 455{
 456	int rc;
 457
 458	if (erst_disable)
 459		return -ENODEV;
 460
 461	rc = mutex_lock_interruptible(&erst_record_id_cache.lock);
 462	if (rc)
 463		return rc;
 464	erst_record_id_cache.refcount++;
 465	mutex_unlock(&erst_record_id_cache.lock);
 466
 467	*pos = 0;
 468
 469	return 0;
 470}
 471EXPORT_SYMBOL_GPL(erst_get_record_id_begin);
 472
 473/* erst_record_id_cache.lock must be held by caller */
 474static int __erst_record_id_cache_add_one(void)
 475{
 476	u64 id, prev_id, first_id;
 477	int i, rc;
 478	u64 *entries;
 479	unsigned long flags;
 480
 481	id = prev_id = first_id = APEI_ERST_INVALID_RECORD_ID;
 482retry:
 483	raw_spin_lock_irqsave(&erst_lock, flags);
 484	rc = __erst_get_next_record_id(&id);
 485	raw_spin_unlock_irqrestore(&erst_lock, flags);
 486	if (rc == -ENOENT)
 487		return 0;
 488	if (rc)
 489		return rc;
 490	if (id == APEI_ERST_INVALID_RECORD_ID)
 491		return 0;
 492	/* can not skip current ID, or loop back to first ID */
 493	if (id == prev_id || id == first_id)
 494		return 0;
 495	if (first_id == APEI_ERST_INVALID_RECORD_ID)
 496		first_id = id;
 497	prev_id = id;
 498
 499	entries = erst_record_id_cache.entries;
 500	for (i = 0; i < erst_record_id_cache.len; i++) {
 501		if (entries[i] == id)
 502			break;
 503	}
 504	/* record id already in cache, try next */
 505	if (i < erst_record_id_cache.len)
 506		goto retry;
 507	if (erst_record_id_cache.len >= erst_record_id_cache.size) {
 508		int new_size;
 509		u64 *new_entries;
 510
 511		new_size = erst_record_id_cache.size * 2;
 512		new_size = clamp_val(new_size, ERST_RECORD_ID_CACHE_SIZE_MIN,
 513				     ERST_RECORD_ID_CACHE_SIZE_MAX);
 514		if (new_size <= erst_record_id_cache.size) {
 515			if (printk_ratelimit())
 516				pr_warn(FW_WARN "too many record IDs!\n");
 517			return 0;
 518		}
 519		new_entries = kvmalloc_array(new_size, sizeof(entries[0]),
 520					     GFP_KERNEL);
 521		if (!new_entries)
 522			return -ENOMEM;
 523		memcpy(new_entries, entries,
 524		       erst_record_id_cache.len * sizeof(entries[0]));
 525		kvfree(entries);
 526		erst_record_id_cache.entries = entries = new_entries;
 527		erst_record_id_cache.size = new_size;
 528	}
 529	entries[i] = id;
 530	erst_record_id_cache.len++;
 531
 532	return 1;
 533}
 534
 535/*
 536 * Get the record ID of an existing error record on the persistent
 537 * storage. If there is no error record on the persistent storage, the
 538 * returned record_id is APEI_ERST_INVALID_RECORD_ID.
 539 */
 540int erst_get_record_id_next(int *pos, u64 *record_id)
 541{
 542	int rc = 0;
 543	u64 *entries;
 544
 545	if (erst_disable)
 546		return -ENODEV;
 547
 548	/* must be enclosed by erst_get_record_id_begin/end */
 549	BUG_ON(!erst_record_id_cache.refcount);
 550	BUG_ON(*pos < 0 || *pos > erst_record_id_cache.len);
 551
 552	mutex_lock(&erst_record_id_cache.lock);
 553	entries = erst_record_id_cache.entries;
 554	for (; *pos < erst_record_id_cache.len; (*pos)++)
 555		if (entries[*pos] != APEI_ERST_INVALID_RECORD_ID)
 556			break;
 557	/* found next record id in cache */
 558	if (*pos < erst_record_id_cache.len) {
 559		*record_id = entries[*pos];
 560		(*pos)++;
 561		goto out_unlock;
 562	}
 563
 564	/* Try to add one more record ID to cache */
 565	rc = __erst_record_id_cache_add_one();
 566	if (rc < 0)
 567		goto out_unlock;
 568	/* successfully add one new ID */
 569	if (rc == 1) {
 570		*record_id = erst_record_id_cache.entries[*pos];
 571		(*pos)++;
 572		rc = 0;
 573	} else {
 574		*pos = -1;
 575		*record_id = APEI_ERST_INVALID_RECORD_ID;
 576	}
 577out_unlock:
 578	mutex_unlock(&erst_record_id_cache.lock);
 579
 580	return rc;
 581}
 582EXPORT_SYMBOL_GPL(erst_get_record_id_next);
 583
 584/* erst_record_id_cache.lock must be held by caller */
 585static void __erst_record_id_cache_compact(void)
 586{
 587	int i, wpos = 0;
 588	u64 *entries;
 589
 590	if (erst_record_id_cache.refcount)
 591		return;
 592
 593	entries = erst_record_id_cache.entries;
 594	for (i = 0; i < erst_record_id_cache.len; i++) {
 595		if (entries[i] == APEI_ERST_INVALID_RECORD_ID)
 596			continue;
 597		if (wpos != i)
 598			entries[wpos] = entries[i];
 599		wpos++;
 600	}
 601	erst_record_id_cache.len = wpos;
 602}
 603
 604void erst_get_record_id_end(void)
 605{
 606	/*
 607	 * erst_disable != 0 should be detected by invoker via the
 608	 * return value of erst_get_record_id_begin/next, so this
 609	 * function should not be called for erst_disable != 0.
 610	 */
 611	BUG_ON(erst_disable);
 612
 613	mutex_lock(&erst_record_id_cache.lock);
 614	erst_record_id_cache.refcount--;
 615	BUG_ON(erst_record_id_cache.refcount < 0);
 616	__erst_record_id_cache_compact();
 617	mutex_unlock(&erst_record_id_cache.lock);
 618}
 619EXPORT_SYMBOL_GPL(erst_get_record_id_end);
 620
 621static int __erst_write_to_storage(u64 offset)
 622{
 623	struct apei_exec_context ctx;
 624	u64 timeout = FIRMWARE_TIMEOUT;
 625	u64 val;
 626	int rc;
 627
 628	erst_exec_ctx_init(&ctx);
 629	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_WRITE);
 630	if (rc)
 631		return rc;
 632	apei_exec_ctx_set_input(&ctx, offset);
 633	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_OFFSET);
 634	if (rc)
 635		return rc;
 636	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
 637	if (rc)
 638		return rc;
 639	for (;;) {
 640		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
 641		if (rc)
 642			return rc;
 643		val = apei_exec_ctx_get_output(&ctx);
 644		if (!val)
 645			break;
 646		if (erst_timedout(&timeout, SPIN_UNIT))
 647			return -EIO;
 648	}
 649	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
 650	if (rc)
 651		return rc;
 652	val = apei_exec_ctx_get_output(&ctx);
 653	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
 654	if (rc)
 655		return rc;
 656
 657	return erst_errno(val);
 658}
 659
 660static int __erst_read_from_storage(u64 record_id, u64 offset)
 661{
 662	struct apei_exec_context ctx;
 663	u64 timeout = FIRMWARE_TIMEOUT;
 664	u64 val;
 665	int rc;
 666
 667	erst_exec_ctx_init(&ctx);
 668	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_READ);
 669	if (rc)
 670		return rc;
 671	apei_exec_ctx_set_input(&ctx, offset);
 672	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_OFFSET);
 673	if (rc)
 674		return rc;
 675	apei_exec_ctx_set_input(&ctx, record_id);
 676	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_ID);
 677	if (rc)
 678		return rc;
 679	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
 680	if (rc)
 681		return rc;
 682	for (;;) {
 683		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
 684		if (rc)
 685			return rc;
 686		val = apei_exec_ctx_get_output(&ctx);
 687		if (!val)
 688			break;
 689		if (erst_timedout(&timeout, SPIN_UNIT))
 690			return -EIO;
 691	};
 692	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
 693	if (rc)
 694		return rc;
 695	val = apei_exec_ctx_get_output(&ctx);
 696	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
 697	if (rc)
 698		return rc;
 699
 700	return erst_errno(val);
 701}
 702
 703static int __erst_clear_from_storage(u64 record_id)
 704{
 705	struct apei_exec_context ctx;
 706	u64 timeout = FIRMWARE_TIMEOUT;
 707	u64 val;
 708	int rc;
 709
 710	erst_exec_ctx_init(&ctx);
 711	rc = apei_exec_run_optional(&ctx, ACPI_ERST_BEGIN_CLEAR);
 712	if (rc)
 713		return rc;
 714	apei_exec_ctx_set_input(&ctx, record_id);
 715	rc = apei_exec_run(&ctx, ACPI_ERST_SET_RECORD_ID);
 716	if (rc)
 717		return rc;
 718	rc = apei_exec_run(&ctx, ACPI_ERST_EXECUTE_OPERATION);
 719	if (rc)
 720		return rc;
 721	for (;;) {
 722		rc = apei_exec_run(&ctx, ACPI_ERST_CHECK_BUSY_STATUS);
 723		if (rc)
 724			return rc;
 725		val = apei_exec_ctx_get_output(&ctx);
 726		if (!val)
 727			break;
 728		if (erst_timedout(&timeout, SPIN_UNIT))
 729			return -EIO;
 730	}
 731	rc = apei_exec_run(&ctx, ACPI_ERST_GET_COMMAND_STATUS);
 732	if (rc)
 733		return rc;
 734	val = apei_exec_ctx_get_output(&ctx);
 735	rc = apei_exec_run_optional(&ctx, ACPI_ERST_END);
 736	if (rc)
 737		return rc;
 738
 739	return erst_errno(val);
 740}
 741
 742/* NVRAM ERST Error Log Address Range is not supported yet */
 743static void pr_unimpl_nvram(void)
 744{
 745	if (printk_ratelimit())
 746		pr_warn("NVRAM ERST Log Address Range not implemented yet.\n");
 747}
 748
 749static int __erst_write_to_nvram(const struct cper_record_header *record)
 750{
 751	/* do not print message, because printk is not safe for NMI */
 752	return -ENOSYS;
 753}
 754
 755static int __erst_read_to_erange_from_nvram(u64 record_id, u64 *offset)
 756{
 757	pr_unimpl_nvram();
 758	return -ENOSYS;
 759}
 760
 761static int __erst_clear_from_nvram(u64 record_id)
 762{
 763	pr_unimpl_nvram();
 764	return -ENOSYS;
 765}
 766
 767int erst_write(const struct cper_record_header *record)
 768{
 769	int rc;
 770	unsigned long flags;
 771	struct cper_record_header *rcd_erange;
 772
 773	if (erst_disable)
 774		return -ENODEV;
 775
 776	if (memcmp(record->signature, CPER_SIG_RECORD, CPER_SIG_SIZE))
 777		return -EINVAL;
 778
 779	if (erst_erange.attr & ERST_RANGE_NVRAM) {
 780		if (!raw_spin_trylock_irqsave(&erst_lock, flags))
 781			return -EBUSY;
 782		rc = __erst_write_to_nvram(record);
 783		raw_spin_unlock_irqrestore(&erst_lock, flags);
 784		return rc;
 785	}
 786
 787	if (record->record_length > erst_erange.size)
 788		return -EINVAL;
 789
 790	if (!raw_spin_trylock_irqsave(&erst_lock, flags))
 791		return -EBUSY;
 792	memcpy(erst_erange.vaddr, record, record->record_length);
 793	rcd_erange = erst_erange.vaddr;
 794	/* signature for serialization system */
 795	memcpy(&rcd_erange->persistence_information, "ER", 2);
 796
 797	rc = __erst_write_to_storage(0);
 798	raw_spin_unlock_irqrestore(&erst_lock, flags);
 799
 800	return rc;
 801}
 802EXPORT_SYMBOL_GPL(erst_write);
 803
 804static int __erst_read_to_erange(u64 record_id, u64 *offset)
 805{
 806	int rc;
 807
 808	if (erst_erange.attr & ERST_RANGE_NVRAM)
 809		return __erst_read_to_erange_from_nvram(
 810			record_id, offset);
 811
 812	rc = __erst_read_from_storage(record_id, 0);
 813	if (rc)
 814		return rc;
 815	*offset = 0;
 816
 817	return 0;
 818}
 819
 820static ssize_t __erst_read(u64 record_id, struct cper_record_header *record,
 821			   size_t buflen)
 822{
 823	int rc;
 824	u64 offset, len = 0;
 825	struct cper_record_header *rcd_tmp;
 826
 827	rc = __erst_read_to_erange(record_id, &offset);
 828	if (rc)
 829		return rc;
 830	rcd_tmp = erst_erange.vaddr + offset;
 831	len = rcd_tmp->record_length;
 832	if (len <= buflen)
 833		memcpy(record, rcd_tmp, len);
 834
 835	return len;
 836}
 837
 838/*
 839 * If return value > buflen, the buffer size is not big enough,
 840 * else if return value < 0, something goes wrong,
 841 * else everything is OK, and return value is record length
 842 */
 843ssize_t erst_read(u64 record_id, struct cper_record_header *record,
 844		  size_t buflen)
 845{
 846	ssize_t len;
 847	unsigned long flags;
 848
 849	if (erst_disable)
 850		return -ENODEV;
 851
 852	raw_spin_lock_irqsave(&erst_lock, flags);
 853	len = __erst_read(record_id, record, buflen);
 854	raw_spin_unlock_irqrestore(&erst_lock, flags);
 855	return len;
 856}
 857EXPORT_SYMBOL_GPL(erst_read);
 858
 859int erst_clear(u64 record_id)
 860{
 861	int rc, i;
 862	unsigned long flags;
 863	u64 *entries;
 864
 865	if (erst_disable)
 866		return -ENODEV;
 867
 868	rc = mutex_lock_interruptible(&erst_record_id_cache.lock);
 869	if (rc)
 870		return rc;
 871	raw_spin_lock_irqsave(&erst_lock, flags);
 872	if (erst_erange.attr & ERST_RANGE_NVRAM)
 873		rc = __erst_clear_from_nvram(record_id);
 874	else
 875		rc = __erst_clear_from_storage(record_id);
 876	raw_spin_unlock_irqrestore(&erst_lock, flags);
 877	if (rc)
 878		goto out;
 879	entries = erst_record_id_cache.entries;
 880	for (i = 0; i < erst_record_id_cache.len; i++) {
 881		if (entries[i] == record_id)
 882			entries[i] = APEI_ERST_INVALID_RECORD_ID;
 883	}
 884	__erst_record_id_cache_compact();
 885out:
 886	mutex_unlock(&erst_record_id_cache.lock);
 887	return rc;
 888}
 889EXPORT_SYMBOL_GPL(erst_clear);
 890
 891static int __init setup_erst_disable(char *str)
 892{
 893	erst_disable = 1;
 894	return 0;
 895}
 896
 897__setup("erst_disable", setup_erst_disable);
 898
 899static int erst_check_table(struct acpi_table_erst *erst_tab)
 900{
 901	if ((erst_tab->header_length !=
 902	     (sizeof(struct acpi_table_erst) - sizeof(erst_tab->header)))
 903	    && (erst_tab->header_length != sizeof(struct acpi_table_erst)))
 904		return -EINVAL;
 905	if (erst_tab->header.length < sizeof(struct acpi_table_erst))
 906		return -EINVAL;
 907	if (erst_tab->entries !=
 908	    (erst_tab->header.length - sizeof(struct acpi_table_erst)) /
 909	    sizeof(struct acpi_erst_entry))
 910		return -EINVAL;
 911
 912	return 0;
 913}
 914
 915static int erst_open_pstore(struct pstore_info *psi);
 916static int erst_close_pstore(struct pstore_info *psi);
 917static ssize_t erst_reader(struct pstore_record *record);
 918static int erst_writer(struct pstore_record *record);
 919static int erst_clearer(struct pstore_record *record);
 920
 921static struct pstore_info erst_info = {
 922	.owner		= THIS_MODULE,
 923	.name		= "erst",
 924	.flags		= PSTORE_FLAGS_DMESG,
 925	.open		= erst_open_pstore,
 926	.close		= erst_close_pstore,
 927	.read		= erst_reader,
 928	.write		= erst_writer,
 929	.erase		= erst_clearer
 930};
 931
 932#define CPER_CREATOR_PSTORE						\
 933	GUID_INIT(0x75a574e3, 0x5052, 0x4b29, 0x8a, 0x8e, 0xbe, 0x2c,	\
 934		  0x64, 0x90, 0xb8, 0x9d)
 935#define CPER_SECTION_TYPE_DMESG						\
 936	GUID_INIT(0xc197e04e, 0xd545, 0x4a70, 0x9c, 0x17, 0xa5, 0x54,	\
 937		  0x94, 0x19, 0xeb, 0x12)
 938#define CPER_SECTION_TYPE_DMESG_Z					\
 939	GUID_INIT(0x4f118707, 0x04dd, 0x4055, 0xb5, 0xdd, 0x95, 0x6d,	\
 940		  0x34, 0xdd, 0xfa, 0xc6)
 941#define CPER_SECTION_TYPE_MCE						\
 942	GUID_INIT(0xfe08ffbe, 0x95e4, 0x4be7, 0xbc, 0x73, 0x40, 0x96,	\
 943		  0x04, 0x4a, 0x38, 0xfc)
 944
 945struct cper_pstore_record {
 946	struct cper_record_header hdr;
 947	struct cper_section_descriptor sec_hdr;
 948	char data[];
 949} __packed;
 950
 951static int reader_pos;
 952
 953static int erst_open_pstore(struct pstore_info *psi)
 954{
 955	int rc;
 956
 957	if (erst_disable)
 958		return -ENODEV;
 959
 960	rc = erst_get_record_id_begin(&reader_pos);
 961
 962	return rc;
 963}
 964
 965static int erst_close_pstore(struct pstore_info *psi)
 966{
 967	erst_get_record_id_end();
 968
 969	return 0;
 970}
 971
 972static ssize_t erst_reader(struct pstore_record *record)
 973{
 974	int rc;
 975	ssize_t len = 0;
 976	u64 record_id;
 977	struct cper_pstore_record *rcd;
 978	size_t rcd_len = sizeof(*rcd) + erst_info.bufsize;
 979
 980	if (erst_disable)
 981		return -ENODEV;
 982
 983	rcd = kmalloc(rcd_len, GFP_KERNEL);
 984	if (!rcd) {
 985		rc = -ENOMEM;
 986		goto out;
 987	}
 988skip:
 989	rc = erst_get_record_id_next(&reader_pos, &record_id);
 990	if (rc)
 991		goto out;
 992
 993	/* no more record */
 994	if (record_id == APEI_ERST_INVALID_RECORD_ID) {
 995		rc = -EINVAL;
 996		goto out;
 997	}
 998
 999	len = erst_read(record_id, &rcd->hdr, rcd_len);
1000	/* The record may be cleared by others, try read next record */
1001	if (len == -ENOENT)
1002		goto skip;
1003	else if (len < 0 || len < sizeof(*rcd)) {
1004		rc = -EIO;
1005		goto out;
1006	}
1007	if (!guid_equal(&rcd->hdr.creator_id, &CPER_CREATOR_PSTORE))
1008		goto skip;
1009
1010	record->buf = kmalloc(len, GFP_KERNEL);
1011	if (record->buf == NULL) {
1012		rc = -ENOMEM;
1013		goto out;
1014	}
1015	memcpy(record->buf, rcd->data, len - sizeof(*rcd));
1016	record->id = record_id;
1017	record->compressed = false;
1018	record->ecc_notice_size = 0;
1019	if (guid_equal(&rcd->sec_hdr.section_type, &CPER_SECTION_TYPE_DMESG_Z)) {
1020		record->type = PSTORE_TYPE_DMESG;
1021		record->compressed = true;
1022	} else if (guid_equal(&rcd->sec_hdr.section_type, &CPER_SECTION_TYPE_DMESG))
1023		record->type = PSTORE_TYPE_DMESG;
1024	else if (guid_equal(&rcd->sec_hdr.section_type, &CPER_SECTION_TYPE_MCE))
1025		record->type = PSTORE_TYPE_MCE;
1026	else
1027		record->type = PSTORE_TYPE_MAX;
1028
1029	if (rcd->hdr.validation_bits & CPER_VALID_TIMESTAMP)
1030		record->time.tv_sec = rcd->hdr.timestamp;
1031	else
1032		record->time.tv_sec = 0;
1033	record->time.tv_nsec = 0;
1034
1035out:
1036	kfree(rcd);
1037	return (rc < 0) ? rc : (len - sizeof(*rcd));
1038}
1039
1040static int erst_writer(struct pstore_record *record)
1041{
1042	struct cper_pstore_record *rcd = (struct cper_pstore_record *)
1043					(erst_info.buf - sizeof(*rcd));
1044	int ret;
1045
1046	memset(rcd, 0, sizeof(*rcd));
1047	memcpy(rcd->hdr.signature, CPER_SIG_RECORD, CPER_SIG_SIZE);
1048	rcd->hdr.revision = CPER_RECORD_REV;
1049	rcd->hdr.signature_end = CPER_SIG_END;
1050	rcd->hdr.section_count = 1;
1051	rcd->hdr.error_severity = CPER_SEV_FATAL;
1052	/* timestamp valid. platform_id, partition_id are invalid */
1053	rcd->hdr.validation_bits = CPER_VALID_TIMESTAMP;
1054	rcd->hdr.timestamp = ktime_get_real_seconds();
1055	rcd->hdr.record_length = sizeof(*rcd) + record->size;
1056	rcd->hdr.creator_id = CPER_CREATOR_PSTORE;
1057	rcd->hdr.notification_type = CPER_NOTIFY_MCE;
1058	rcd->hdr.record_id = cper_next_record_id();
1059	rcd->hdr.flags = CPER_HW_ERROR_FLAGS_PREVERR;
1060
1061	rcd->sec_hdr.section_offset = sizeof(*rcd);
1062	rcd->sec_hdr.section_length = record->size;
1063	rcd->sec_hdr.revision = CPER_SEC_REV;
1064	/* fru_id and fru_text is invalid */
1065	rcd->sec_hdr.validation_bits = 0;
1066	rcd->sec_hdr.flags = CPER_SEC_PRIMARY;
1067	switch (record->type) {
1068	case PSTORE_TYPE_DMESG:
1069		if (record->compressed)
1070			rcd->sec_hdr.section_type = CPER_SECTION_TYPE_DMESG_Z;
1071		else
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	ret = erst_write(&rcd->hdr);
1083	record->id = rcd->hdr.record_id;
1084
1085	return ret;
1086}
1087
1088static int erst_clearer(struct pstore_record *record)
1089{
1090	return erst_clear(record->id);
1091}
1092
1093static 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(
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		goto err;
1115	else if (ACPI_FAILURE(status)) {
1116		const char *msg = acpi_format_exception(status);
1117		pr_err("Failed to get table, %s\n", msg);
1118		rc = -EINVAL;
1119		goto err;
1120	}
1121
1122	rc = erst_check_table(erst_tab);
1123	if (rc) {
1124		pr_err(FW_BUG "ERST table is invalid.\n");
1125		goto err_put_erst_tab;
1126	}
1127
1128	apei_resources_init(&erst_resources);
1129	erst_exec_ctx_init(&ctx);
1130	rc = apei_exec_collect_resources(&ctx, &erst_resources);
1131	if (rc)
1132		goto err_fini;
1133	rc = apei_resources_request(&erst_resources, "APEI ERST");
1134	if (rc)
1135		goto err_fini;
1136	rc = apei_exec_pre_map_gars(&ctx);
1137	if (rc)
1138		goto err_release;
1139	rc = erst_get_erange(&erst_erange);
1140	if (rc) {
1141		if (rc == -ENODEV)
1142			pr_info(
1143	"The corresponding hardware device or firmware implementation "
1144	"is not available.\n");
1145		else
1146			pr_err("Failed to get Error Log Address Range.\n");
1147		goto err_unmap_reg;
1148	}
1149
1150	r = request_mem_region(erst_erange.base, erst_erange.size, "APEI ERST");
1151	if (!r) {
1152		pr_err("Can not request [mem %#010llx-%#010llx] for ERST.\n",
1153		       (unsigned long long)erst_erange.base,
1154		       (unsigned long long)erst_erange.base + erst_erange.size - 1);
1155		rc = -EIO;
1156		goto err_unmap_reg;
1157	}
1158	rc = -ENOMEM;
1159	erst_erange.vaddr = ioremap_cache(erst_erange.base,
1160					  erst_erange.size);
1161	if (!erst_erange.vaddr)
1162		goto err_release_erange;
1163
1164	pr_info(
1165	"Error Record Serialization Table (ERST) support is initialized.\n");
1166
1167	buf = kmalloc(erst_erange.size, GFP_KERNEL);
1168	if (buf) {
1169		erst_info.buf = buf + sizeof(struct cper_pstore_record);
1170		erst_info.bufsize = erst_erange.size -
1171				    sizeof(struct cper_pstore_record);
1172		rc = pstore_register(&erst_info);
1173		if (rc) {
1174			if (rc != -EPERM)
1175				pr_info(
1176				"Could not register with persistent store.\n");
1177			erst_info.buf = NULL;
1178			erst_info.bufsize = 0;
1179			kfree(buf);
1180		}
1181	} else
1182		pr_err(
1183		"Failed to allocate %lld bytes for persistent store error log.\n",
1184		erst_erange.size);
1185
1186	/* Cleanup ERST Resources */
1187	apei_resources_fini(&erst_resources);
1188
1189	return 0;
1190
1191err_release_erange:
1192	release_mem_region(erst_erange.base, erst_erange.size);
1193err_unmap_reg:
1194	apei_exec_post_unmap_gars(&ctx);
1195err_release:
1196	apei_resources_release(&erst_resources);
1197err_fini:
1198	apei_resources_fini(&erst_resources);
1199err_put_erst_tab:
1200	acpi_put_table((struct acpi_table_header *)erst_tab);
1201err:
1202	erst_disable = 1;
1203	return rc;
1204}
1205
1206device_initcall(erst_init);