1 // SPDX-License-Identifier: GPL-2.0-only
3 * APEI Error INJection support
5 * EINJ provides a hardware error injection mechanism, this is useful
6 * for debugging and testing of other APEI and RAS features.
8 * For more information about EINJ, please refer to ACPI Specification
9 * version 4.0, section 17.5.
11 * Copyright 2009-2010 Intel Corp.
12 * Author: Huang Ying <ying.huang@intel.com>
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/init.h>
19 #include <linux/debugfs.h>
20 #include <linux/seq_file.h>
21 #include <linux/nmi.h>
22 #include <linux/delay.h>
24 #include <asm/unaligned.h>
26 #include "apei-internal.h"
29 #define pr_fmt(fmt) "EINJ: " fmt
31 #define SLEEP_UNIT_MIN 1000 /* 1ms */
32 #define SLEEP_UNIT_MAX 5000 /* 5ms */
33 /* Firmware should respond within 1 seconds */
34 #define FIRMWARE_TIMEOUT (1 * USEC_PER_SEC)
35 #define ACPI5_VENDOR_BIT BIT(31)
36 #define MEM_ERROR_MASK (ACPI_EINJ_MEMORY_CORRECTABLE | \
37 ACPI_EINJ_MEMORY_UNCORRECTABLE | \
38 ACPI_EINJ_MEMORY_FATAL)
41 * ACPI version 5 provides a SET_ERROR_TYPE_WITH_ADDRESS action.
45 struct set_error_type_with_address {
51 u64 memory_address_range;
55 SETWA_FLAGS_APICID = 1,
57 SETWA_FLAGS_PCIE_SBDF = 4,
61 * Vendor extensions for platform specific operations
63 struct vendor_error_type_extension {
74 static u32 vendor_flags;
75 static struct debugfs_blob_wrapper vendor_blob;
76 static char vendor_dev[64];
79 * Some BIOSes allow parameters to the SET_ERROR_TYPE entries in the
80 * EINJ table through an unpublished extension. Use with caution as
81 * most will ignore the parameter and make their own choice of address
82 * for error injection. This extension is used only if
83 * param_extension module parameter is specified.
85 struct einj_parameter {
93 #define EINJ_OP_BUSY 0x1
94 #define EINJ_STATUS_SUCCESS 0x0
95 #define EINJ_STATUS_FAIL 0x1
96 #define EINJ_STATUS_INVAL 0x2
98 #define EINJ_TAB_ENTRY(tab) \
99 ((struct acpi_whea_header *)((char *)(tab) + \
100 sizeof(struct acpi_table_einj)))
102 static bool param_extension;
103 module_param(param_extension, bool, 0);
105 static struct acpi_table_einj *einj_tab;
107 static struct apei_resources einj_resources;
109 static struct apei_exec_ins_type einj_ins_type[] = {
110 [ACPI_EINJ_READ_REGISTER] = {
111 .flags = APEI_EXEC_INS_ACCESS_REGISTER,
112 .run = apei_exec_read_register,
114 [ACPI_EINJ_READ_REGISTER_VALUE] = {
115 .flags = APEI_EXEC_INS_ACCESS_REGISTER,
116 .run = apei_exec_read_register_value,
118 [ACPI_EINJ_WRITE_REGISTER] = {
119 .flags = APEI_EXEC_INS_ACCESS_REGISTER,
120 .run = apei_exec_write_register,
122 [ACPI_EINJ_WRITE_REGISTER_VALUE] = {
123 .flags = APEI_EXEC_INS_ACCESS_REGISTER,
124 .run = apei_exec_write_register_value,
128 .run = apei_exec_noop,
133 * Prevent EINJ interpreter to run simultaneously, because the
134 * corresponding firmware implementation may not work properly when
135 * invoked simultaneously.
137 static DEFINE_MUTEX(einj_mutex);
139 static void *einj_param;
141 static void einj_exec_ctx_init(struct apei_exec_context *ctx)
143 apei_exec_ctx_init(ctx, einj_ins_type, ARRAY_SIZE(einj_ins_type),
144 EINJ_TAB_ENTRY(einj_tab), einj_tab->entries);
147 static int __einj_get_available_error_type(u32 *type)
149 struct apei_exec_context ctx;
152 einj_exec_ctx_init(&ctx);
153 rc = apei_exec_run(&ctx, ACPI_EINJ_GET_ERROR_TYPE);
156 *type = apei_exec_ctx_get_output(&ctx);
161 /* Get error injection capabilities of the platform */
162 static int einj_get_available_error_type(u32 *type)
166 mutex_lock(&einj_mutex);
167 rc = __einj_get_available_error_type(type);
168 mutex_unlock(&einj_mutex);
173 static int einj_timedout(u64 *t)
175 if ((s64)*t < SLEEP_UNIT_MIN) {
176 pr_warn(FW_WARN "Firmware does not respond in time\n");
179 *t -= SLEEP_UNIT_MIN;
180 usleep_range(SLEEP_UNIT_MIN, SLEEP_UNIT_MAX);
185 static void check_vendor_extension(u64 paddr,
186 struct set_error_type_with_address *v5param)
188 int offset = v5param->vendor_extension;
189 struct vendor_error_type_extension *v;
194 v = acpi_os_map_iomem(paddr + offset, sizeof(*v));
198 sprintf(vendor_dev, "%x:%x:%x.%x vendor_id=%x device_id=%x rev_id=%x\n",
199 sbdf >> 24, (sbdf >> 16) & 0xff,
200 (sbdf >> 11) & 0x1f, (sbdf >> 8) & 0x7,
201 v->vendor_id, v->device_id, v->rev_id);
202 acpi_os_unmap_iomem(v, sizeof(*v));
205 static void *einj_get_parameter_address(void)
208 u64 pa_v4 = 0, pa_v5 = 0;
209 struct acpi_whea_header *entry;
211 entry = EINJ_TAB_ENTRY(einj_tab);
212 for (i = 0; i < einj_tab->entries; i++) {
213 if (entry->action == ACPI_EINJ_SET_ERROR_TYPE &&
214 entry->instruction == ACPI_EINJ_WRITE_REGISTER &&
215 entry->register_region.space_id ==
216 ACPI_ADR_SPACE_SYSTEM_MEMORY)
217 pa_v4 = get_unaligned(&entry->register_region.address);
218 if (entry->action == ACPI_EINJ_SET_ERROR_TYPE_WITH_ADDRESS &&
219 entry->instruction == ACPI_EINJ_WRITE_REGISTER &&
220 entry->register_region.space_id ==
221 ACPI_ADR_SPACE_SYSTEM_MEMORY)
222 pa_v5 = get_unaligned(&entry->register_region.address);
226 struct set_error_type_with_address *v5param;
228 v5param = acpi_os_map_iomem(pa_v5, sizeof(*v5param));
231 check_vendor_extension(pa_v5, v5param);
235 if (param_extension && pa_v4) {
236 struct einj_parameter *v4param;
238 v4param = acpi_os_map_iomem(pa_v4, sizeof(*v4param));
241 if (v4param->reserved1 || v4param->reserved2) {
242 acpi_os_unmap_iomem(v4param, sizeof(*v4param));
251 /* do sanity check to trigger table */
252 static int einj_check_trigger_header(struct acpi_einj_trigger *trigger_tab)
254 if (trigger_tab->header_size != sizeof(struct acpi_einj_trigger))
256 if (trigger_tab->table_size > PAGE_SIZE ||
257 trigger_tab->table_size < trigger_tab->header_size)
259 if (trigger_tab->entry_count !=
260 (trigger_tab->table_size - trigger_tab->header_size) /
261 sizeof(struct acpi_einj_entry))
267 static struct acpi_generic_address *einj_get_trigger_parameter_region(
268 struct acpi_einj_trigger *trigger_tab, u64 param1, u64 param2)
271 struct acpi_whea_header *entry;
273 entry = (struct acpi_whea_header *)
274 ((char *)trigger_tab + sizeof(struct acpi_einj_trigger));
275 for (i = 0; i < trigger_tab->entry_count; i++) {
276 if (entry->action == ACPI_EINJ_TRIGGER_ERROR &&
277 entry->instruction <= ACPI_EINJ_WRITE_REGISTER_VALUE &&
278 entry->register_region.space_id ==
279 ACPI_ADR_SPACE_SYSTEM_MEMORY &&
280 (entry->register_region.address & param2) == (param1 & param2))
281 return &entry->register_region;
287 /* Execute instructions in trigger error action table */
288 static int __einj_error_trigger(u64 trigger_paddr, u32 type,
289 u64 param1, u64 param2)
291 struct acpi_einj_trigger *trigger_tab = NULL;
292 struct apei_exec_context trigger_ctx;
293 struct apei_resources trigger_resources;
294 struct acpi_whea_header *trigger_entry;
298 struct acpi_generic_address *trigger_param_region = NULL;
300 r = request_mem_region(trigger_paddr, sizeof(*trigger_tab),
301 "APEI EINJ Trigger Table");
303 pr_err("Can not request [mem %#010llx-%#010llx] for Trigger table\n",
304 (unsigned long long)trigger_paddr,
305 (unsigned long long)trigger_paddr +
306 sizeof(*trigger_tab) - 1);
309 trigger_tab = ioremap_cache(trigger_paddr, sizeof(*trigger_tab));
311 pr_err("Failed to map trigger table!\n");
314 rc = einj_check_trigger_header(trigger_tab);
316 pr_warn(FW_BUG "Invalid trigger error action table.\n");
320 /* No action structures in the TRIGGER_ERROR table, nothing to do */
321 if (!trigger_tab->entry_count)
325 table_size = trigger_tab->table_size;
326 r = request_mem_region(trigger_paddr + sizeof(*trigger_tab),
327 table_size - sizeof(*trigger_tab),
328 "APEI EINJ Trigger Table");
330 pr_err("Can not request [mem %#010llx-%#010llx] for Trigger Table Entry\n",
331 (unsigned long long)trigger_paddr + sizeof(*trigger_tab),
332 (unsigned long long)trigger_paddr + table_size - 1);
335 iounmap(trigger_tab);
336 trigger_tab = ioremap_cache(trigger_paddr, table_size);
338 pr_err("Failed to map trigger table!\n");
341 trigger_entry = (struct acpi_whea_header *)
342 ((char *)trigger_tab + sizeof(struct acpi_einj_trigger));
343 apei_resources_init(&trigger_resources);
344 apei_exec_ctx_init(&trigger_ctx, einj_ins_type,
345 ARRAY_SIZE(einj_ins_type),
346 trigger_entry, trigger_tab->entry_count);
347 rc = apei_exec_collect_resources(&trigger_ctx, &trigger_resources);
350 rc = apei_resources_sub(&trigger_resources, &einj_resources);
354 * Some firmware will access target address specified in
355 * param1 to trigger the error when injecting memory error.
356 * This will cause resource conflict with regular memory. So
357 * remove it from trigger table resources.
359 if ((param_extension || acpi5) && (type & MEM_ERROR_MASK) && param2) {
360 struct apei_resources addr_resources;
361 apei_resources_init(&addr_resources);
362 trigger_param_region = einj_get_trigger_parameter_region(
363 trigger_tab, param1, param2);
364 if (trigger_param_region) {
365 rc = apei_resources_add(&addr_resources,
366 trigger_param_region->address,
367 trigger_param_region->bit_width/8, true);
370 rc = apei_resources_sub(&trigger_resources,
373 apei_resources_fini(&addr_resources);
377 rc = apei_resources_request(&trigger_resources, "APEI EINJ Trigger");
380 rc = apei_exec_pre_map_gars(&trigger_ctx);
384 rc = apei_exec_run(&trigger_ctx, ACPI_EINJ_TRIGGER_ERROR);
386 apei_exec_post_unmap_gars(&trigger_ctx);
388 apei_resources_release(&trigger_resources);
390 apei_resources_fini(&trigger_resources);
392 release_mem_region(trigger_paddr + sizeof(*trigger_tab),
393 table_size - sizeof(*trigger_tab));
395 release_mem_region(trigger_paddr, sizeof(*trigger_tab));
398 iounmap(trigger_tab);
403 static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
404 u64 param3, u64 param4)
406 struct apei_exec_context ctx;
407 u64 val, trigger_paddr, timeout = FIRMWARE_TIMEOUT;
410 einj_exec_ctx_init(&ctx);
412 rc = apei_exec_run_optional(&ctx, ACPI_EINJ_BEGIN_OPERATION);
415 apei_exec_ctx_set_input(&ctx, type);
417 struct set_error_type_with_address *v5param = einj_param;
419 v5param->type = type;
420 if (type & ACPI5_VENDOR_BIT) {
421 switch (vendor_flags) {
422 case SETWA_FLAGS_APICID:
423 v5param->apicid = param1;
425 case SETWA_FLAGS_MEM:
426 v5param->memory_address = param1;
427 v5param->memory_address_range = param2;
429 case SETWA_FLAGS_PCIE_SBDF:
430 v5param->pcie_sbdf = param1;
433 v5param->flags = vendor_flags;
435 v5param->flags = flags;
436 v5param->memory_address = param1;
437 v5param->memory_address_range = param2;
438 v5param->apicid = param3;
439 v5param->pcie_sbdf = param4;
442 case ACPI_EINJ_PROCESSOR_CORRECTABLE:
443 case ACPI_EINJ_PROCESSOR_UNCORRECTABLE:
444 case ACPI_EINJ_PROCESSOR_FATAL:
445 v5param->apicid = param1;
446 v5param->flags = SETWA_FLAGS_APICID;
448 case ACPI_EINJ_MEMORY_CORRECTABLE:
449 case ACPI_EINJ_MEMORY_UNCORRECTABLE:
450 case ACPI_EINJ_MEMORY_FATAL:
451 v5param->memory_address = param1;
452 v5param->memory_address_range = param2;
453 v5param->flags = SETWA_FLAGS_MEM;
455 case ACPI_EINJ_PCIX_CORRECTABLE:
456 case ACPI_EINJ_PCIX_UNCORRECTABLE:
457 case ACPI_EINJ_PCIX_FATAL:
458 v5param->pcie_sbdf = param1;
459 v5param->flags = SETWA_FLAGS_PCIE_SBDF;
464 rc = apei_exec_run(&ctx, ACPI_EINJ_SET_ERROR_TYPE);
468 struct einj_parameter *v4param = einj_param;
469 v4param->param1 = param1;
470 v4param->param2 = param2;
473 rc = apei_exec_run(&ctx, ACPI_EINJ_EXECUTE_OPERATION);
477 rc = apei_exec_run(&ctx, ACPI_EINJ_CHECK_BUSY_STATUS);
480 val = apei_exec_ctx_get_output(&ctx);
481 if (!(val & EINJ_OP_BUSY))
483 if (einj_timedout(&timeout))
486 rc = apei_exec_run(&ctx, ACPI_EINJ_GET_COMMAND_STATUS);
489 val = apei_exec_ctx_get_output(&ctx);
490 if (val != EINJ_STATUS_SUCCESS)
493 rc = apei_exec_run(&ctx, ACPI_EINJ_GET_TRIGGER_TABLE);
496 trigger_paddr = apei_exec_ctx_get_output(&ctx);
497 if (notrigger == 0) {
498 rc = __einj_error_trigger(trigger_paddr, type, param1, param2);
502 rc = apei_exec_run_optional(&ctx, ACPI_EINJ_END_OPERATION);
507 /* Inject the specified hardware error */
508 static int einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
509 u64 param3, u64 param4)
514 /* If user manually set "flags", make sure it is legal */
515 if (flags && (flags &
516 ~(SETWA_FLAGS_APICID|SETWA_FLAGS_MEM|SETWA_FLAGS_PCIE_SBDF)))
520 * We need extra sanity checks for memory errors.
521 * Other types leap directly to injection.
524 /* ensure param1/param2 existed */
525 if (!(param_extension || acpi5))
528 /* ensure injection is memory related */
529 if (type & ACPI5_VENDOR_BIT) {
530 if (vendor_flags != SETWA_FLAGS_MEM)
532 } else if (!(type & MEM_ERROR_MASK) && !(flags & SETWA_FLAGS_MEM))
536 * Disallow crazy address masks that give BIOS leeway to pick
537 * injection address almost anywhere. Insist on page or
538 * better granularity and that target address is normal RAM or
541 base_addr = param1 & param2;
544 if (((param2 & PAGE_MASK) != PAGE_MASK) ||
545 ((region_intersects(base_addr, size, IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE)
546 != REGION_INTERSECTS) &&
547 (region_intersects(base_addr, size, IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY)
548 != REGION_INTERSECTS) &&
549 !arch_is_platform_page(base_addr)))
552 if (is_zero_pfn(base_addr >> PAGE_SHIFT))
556 mutex_lock(&einj_mutex);
557 rc = __einj_error_inject(type, flags, param1, param2, param3, param4);
558 mutex_unlock(&einj_mutex);
563 static u32 error_type;
564 static u32 error_flags;
565 static u64 error_param1;
566 static u64 error_param2;
567 static u64 error_param3;
568 static u64 error_param4;
569 static struct dentry *einj_debug_dir;
571 static int available_error_type_show(struct seq_file *m, void *v)
574 u32 available_error_type = 0;
576 rc = einj_get_available_error_type(&available_error_type);
579 if (available_error_type & 0x0001)
580 seq_printf(m, "0x00000001\tProcessor Correctable\n");
581 if (available_error_type & 0x0002)
582 seq_printf(m, "0x00000002\tProcessor Uncorrectable non-fatal\n");
583 if (available_error_type & 0x0004)
584 seq_printf(m, "0x00000004\tProcessor Uncorrectable fatal\n");
585 if (available_error_type & 0x0008)
586 seq_printf(m, "0x00000008\tMemory Correctable\n");
587 if (available_error_type & 0x0010)
588 seq_printf(m, "0x00000010\tMemory Uncorrectable non-fatal\n");
589 if (available_error_type & 0x0020)
590 seq_printf(m, "0x00000020\tMemory Uncorrectable fatal\n");
591 if (available_error_type & 0x0040)
592 seq_printf(m, "0x00000040\tPCI Express Correctable\n");
593 if (available_error_type & 0x0080)
594 seq_printf(m, "0x00000080\tPCI Express Uncorrectable non-fatal\n");
595 if (available_error_type & 0x0100)
596 seq_printf(m, "0x00000100\tPCI Express Uncorrectable fatal\n");
597 if (available_error_type & 0x0200)
598 seq_printf(m, "0x00000200\tPlatform Correctable\n");
599 if (available_error_type & 0x0400)
600 seq_printf(m, "0x00000400\tPlatform Uncorrectable non-fatal\n");
601 if (available_error_type & 0x0800)
602 seq_printf(m, "0x00000800\tPlatform Uncorrectable fatal\n");
607 DEFINE_SHOW_ATTRIBUTE(available_error_type);
609 static int error_type_get(void *data, u64 *val)
616 static int error_type_set(void *data, u64 val)
619 u32 available_error_type = 0;
623 * Vendor defined types have 0x80000000 bit set, and
624 * are not enumerated by ACPI_EINJ_GET_ERROR_TYPE
626 vendor = val & ACPI5_VENDOR_BIT;
627 tval = val & 0x7fffffff;
629 /* Only one error type can be specified */
630 if (tval & (tval - 1))
633 rc = einj_get_available_error_type(&available_error_type);
636 if (!(val & available_error_type))
644 DEFINE_DEBUGFS_ATTRIBUTE(error_type_fops, error_type_get, error_type_set,
647 static int error_inject_set(void *data, u64 val)
652 return einj_error_inject(error_type, error_flags, error_param1, error_param2,
653 error_param3, error_param4);
656 DEFINE_DEBUGFS_ATTRIBUTE(error_inject_fops, NULL, error_inject_set, "%llu\n");
658 static int einj_check_table(struct acpi_table_einj *einj_tab)
660 if ((einj_tab->header_length !=
661 (sizeof(struct acpi_table_einj) - sizeof(einj_tab->header)))
662 && (einj_tab->header_length != sizeof(struct acpi_table_einj)))
664 if (einj_tab->header.length < sizeof(struct acpi_table_einj))
666 if (einj_tab->entries !=
667 (einj_tab->header.length - sizeof(struct acpi_table_einj)) /
668 sizeof(struct acpi_einj_entry))
674 static int __init einj_init(void)
678 struct apei_exec_context ctx;
681 pr_info("ACPI disabled.\n");
685 status = acpi_get_table(ACPI_SIG_EINJ, 0,
686 (struct acpi_table_header **)&einj_tab);
687 if (status == AE_NOT_FOUND) {
688 pr_warn("EINJ table not found.\n");
691 else if (ACPI_FAILURE(status)) {
692 pr_err("Failed to get EINJ table: %s\n",
693 acpi_format_exception(status));
697 rc = einj_check_table(einj_tab);
699 pr_warn(FW_BUG "Invalid EINJ table.\n");
704 einj_debug_dir = debugfs_create_dir("einj", apei_get_debugfs_dir());
706 debugfs_create_file("available_error_type", S_IRUSR, einj_debug_dir,
707 NULL, &available_error_type_fops);
708 debugfs_create_file_unsafe("error_type", 0600, einj_debug_dir,
709 NULL, &error_type_fops);
710 debugfs_create_file_unsafe("error_inject", 0200, einj_debug_dir,
711 NULL, &error_inject_fops);
713 apei_resources_init(&einj_resources);
714 einj_exec_ctx_init(&ctx);
715 rc = apei_exec_collect_resources(&ctx, &einj_resources);
717 pr_err("Error collecting EINJ resources.\n");
721 rc = apei_resources_request(&einj_resources, "APEI EINJ");
723 pr_err("Error requesting memory/port resources.\n");
727 rc = apei_exec_pre_map_gars(&ctx);
729 pr_err("Error pre-mapping GARs.\n");
733 einj_param = einj_get_parameter_address();
734 if ((param_extension || acpi5) && einj_param) {
735 debugfs_create_x32("flags", S_IRUSR | S_IWUSR, einj_debug_dir,
737 debugfs_create_x64("param1", S_IRUSR | S_IWUSR, einj_debug_dir,
739 debugfs_create_x64("param2", S_IRUSR | S_IWUSR, einj_debug_dir,
741 debugfs_create_x64("param3", S_IRUSR | S_IWUSR, einj_debug_dir,
743 debugfs_create_x64("param4", S_IRUSR | S_IWUSR, einj_debug_dir,
745 debugfs_create_x32("notrigger", S_IRUSR | S_IWUSR,
746 einj_debug_dir, ¬rigger);
750 vendor_blob.data = vendor_dev;
751 vendor_blob.size = strlen(vendor_dev);
752 debugfs_create_blob("vendor", S_IRUSR, einj_debug_dir,
754 debugfs_create_x32("vendor_flags", S_IRUSR | S_IWUSR,
755 einj_debug_dir, &vendor_flags);
758 pr_info("Error INJection is initialized.\n");
763 apei_resources_release(&einj_resources);
765 apei_resources_fini(&einj_resources);
766 debugfs_remove_recursive(einj_debug_dir);
768 acpi_put_table((struct acpi_table_header *)einj_tab);
773 static void __exit einj_exit(void)
775 struct apei_exec_context ctx;
778 acpi_size size = (acpi5) ?
779 sizeof(struct set_error_type_with_address) :
780 sizeof(struct einj_parameter);
782 acpi_os_unmap_iomem(einj_param, size);
784 einj_exec_ctx_init(&ctx);
785 apei_exec_post_unmap_gars(&ctx);
786 apei_resources_release(&einj_resources);
787 apei_resources_fini(&einj_resources);
788 debugfs_remove_recursive(einj_debug_dir);
789 acpi_put_table((struct acpi_table_header *)einj_tab);
792 module_init(einj_init);
793 module_exit(einj_exit);
795 MODULE_AUTHOR("Huang Ying");
796 MODULE_DESCRIPTION("APEI Error INJection support");
797 MODULE_LICENSE("GPL");