1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * APEI Error INJection support
4 *
5 * EINJ provides a hardware error injection mechanism, this is useful
6 * for debugging and testing of other APEI and RAS features.
7 *
8 * For more information about EINJ, please refer to ACPI Specification
9 * version 4.0, section 17.5.
10 *
11 * Copyright 2009-2010 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/io.h>
19 #include <linux/debugfs.h>
20 #include <linux/seq_file.h>
21 #include <linux/nmi.h>
22 #include <linux/delay.h>
23 #include <linux/mm.h>
24 #include <linux/device/faux.h>
25 #include <linux/unaligned.h>
26
27 #include "apei-internal.h"
28
29 #undef pr_fmt
30 #define pr_fmt(fmt) "EINJ: " fmt
31
32 #define SLEEP_UNIT_MIN 1000 /* 1ms */
33 #define SLEEP_UNIT_MAX 5000 /* 5ms */
34 /* Firmware should respond within 1 seconds */
35 #define FIRMWARE_TIMEOUT (1 * USEC_PER_SEC)
36 #define COMPONENT_LEN 16
37 #define ACPI65_EINJV2_SUPP BIT(30)
38 #define ACPI5_VENDOR_BIT BIT(31)
39 #define MEM_ERROR_MASK (ACPI_EINJ_MEMORY_CORRECTABLE | \
40 ACPI_EINJ_MEMORY_UNCORRECTABLE | \
41 ACPI_EINJ_MEMORY_FATAL)
42 #define CXL_ERROR_MASK (ACPI_EINJ_CXL_CACHE_CORRECTABLE | \
43 ACPI_EINJ_CXL_CACHE_UNCORRECTABLE | \
44 ACPI_EINJ_CXL_CACHE_FATAL | \
45 ACPI_EINJ_CXL_MEM_CORRECTABLE | \
46 ACPI_EINJ_CXL_MEM_UNCORRECTABLE | \
47 ACPI_EINJ_CXL_MEM_FATAL)
48
49 /*
50 * ACPI version 5 provides a SET_ERROR_TYPE_WITH_ADDRESS action.
51 */
52 static int acpi5;
53
54 struct syndrome_array {
55 union {
56 u8 acpi_id[COMPONENT_LEN];
57 u8 device_id[COMPONENT_LEN];
58 u8 pcie_sbdf[COMPONENT_LEN];
59 u8 vendor_id[COMPONENT_LEN];
60 } comp_id;
61 union {
62 u8 proc_synd[COMPONENT_LEN];
63 u8 mem_synd[COMPONENT_LEN];
64 u8 pcie_synd[COMPONENT_LEN];
65 u8 vendor_synd[COMPONENT_LEN];
66 } comp_synd;
67 };
68
69 struct einjv2_extension_struct {
70 u32 length;
71 u16 revision;
72 u16 component_arr_count;
73 struct syndrome_array component_arr[] __counted_by(component_arr_count);
74 };
75
76 struct set_error_type_with_address {
77 u32 type;
78 u32 vendor_extension;
79 u32 flags;
80 u32 apicid;
81 u64 memory_address;
82 u64 memory_address_range;
83 u32 pcie_sbdf;
84 struct einjv2_extension_struct einjv2_struct;
85 };
86 enum {
87 SETWA_FLAGS_APICID = 1,
88 SETWA_FLAGS_MEM = 2,
89 SETWA_FLAGS_PCIE_SBDF = 4,
90 SETWA_FLAGS_EINJV2 = 8,
91 };
92
93 /*
94 * Vendor extensions for platform specific operations
95 */
96 struct vendor_error_type_extension {
97 u32 length;
98 u32 pcie_sbdf;
99 u16 vendor_id;
100 u16 device_id;
101 u8 rev_id;
102 u8 reserved[3];
103 };
104
105 static u32 notrigger;
106
107 static u32 vendor_flags;
108 static struct debugfs_blob_wrapper vendor_blob;
109 static struct debugfs_blob_wrapper vendor_errors;
110 static char vendor_dev[64];
111
112 static u32 max_nr_components;
113 static u32 available_error_type;
114 static u32 available_error_type_v2;
115 static struct syndrome_array *syndrome_data;
116
117 /*
118 * Some BIOSes allow parameters to the SET_ERROR_TYPE entries in the
119 * EINJ table through an unpublished extension. Use with caution as
120 * most will ignore the parameter and make their own choice of address
121 * for error injection. This extension is used only if
122 * param_extension module parameter is specified.
123 */
124 struct einj_parameter {
125 u64 type;
126 u64 reserved1;
127 u64 reserved2;
128 u64 param1;
129 u64 param2;
130 };
131
132 #define EINJ_OP_BUSY 0x1
133 #define EINJ_STATUS_SUCCESS 0x0
134 #define EINJ_STATUS_FAIL 0x1
135 #define EINJ_STATUS_INVAL 0x2
136
137 #define EINJ_TAB_ENTRY(tab) \
138 ((struct acpi_whea_header *)((char *)(tab) + \
139 sizeof(struct acpi_table_einj)))
140
141 static bool param_extension;
142 module_param(param_extension, bool, 0);
143
144 static struct acpi_table_einj *einj_tab;
145
146 static struct apei_resources einj_resources;
147
148 static struct apei_exec_ins_type einj_ins_type[] = {
149 [ACPI_EINJ_READ_REGISTER] = {
150 .flags = APEI_EXEC_INS_ACCESS_REGISTER,
151 .run = apei_exec_read_register,
152 },
153 [ACPI_EINJ_READ_REGISTER_VALUE] = {
154 .flags = APEI_EXEC_INS_ACCESS_REGISTER,
155 .run = apei_exec_read_register_value,
156 },
157 [ACPI_EINJ_WRITE_REGISTER] = {
158 .flags = APEI_EXEC_INS_ACCESS_REGISTER,
159 .run = apei_exec_write_register,
160 },
161 [ACPI_EINJ_WRITE_REGISTER_VALUE] = {
162 .flags = APEI_EXEC_INS_ACCESS_REGISTER,
163 .run = apei_exec_write_register_value,
164 },
165 [ACPI_EINJ_NOOP] = {
166 .flags = 0,
167 .run = apei_exec_noop,
168 },
169 };
170
171 /*
172 * Prevent EINJ interpreter to run simultaneously, because the
173 * corresponding firmware implementation may not work properly when
174 * invoked simultaneously.
175 */
176 static DEFINE_MUTEX(einj_mutex);
177
178 /*
179 * Exported APIs use this flag to exit early if einj_probe() failed.
180 */
181 bool einj_initialized __ro_after_init;
182
183 static void __iomem *einj_param;
184 static u32 v5param_size;
185 static u32 v66param_size;
186 static bool is_v2;
187
einj_exec_ctx_init(struct apei_exec_context * ctx)188 static void einj_exec_ctx_init(struct apei_exec_context *ctx)
189 {
190 apei_exec_ctx_init(ctx, einj_ins_type, ARRAY_SIZE(einj_ins_type),
191 EINJ_TAB_ENTRY(einj_tab), einj_tab->entries);
192 }
193
__einj_get_available_error_type(u32 * type,int einj_action)194 static int __einj_get_available_error_type(u32 *type, int einj_action)
195 {
196 struct apei_exec_context ctx;
197 int rc;
198
199 einj_exec_ctx_init(&ctx);
200 rc = apei_exec_run(&ctx, einj_action);
201 if (rc)
202 return rc;
203 *type = apei_exec_ctx_get_output(&ctx);
204
205 return 0;
206 }
207
208 /* Get error injection capabilities of the platform */
einj_get_available_error_type(u32 * type,int einj_action)209 int einj_get_available_error_type(u32 *type, int einj_action)
210 {
211 int rc;
212
213 mutex_lock(&einj_mutex);
214 rc = __einj_get_available_error_type(type, einj_action);
215 mutex_unlock(&einj_mutex);
216
217 return rc;
218 }
219
einj_get_available_error_types(u32 * type1,u32 * type2)220 static int einj_get_available_error_types(u32 *type1, u32 *type2)
221 {
222 int rc;
223
224 rc = einj_get_available_error_type(type1, ACPI_EINJ_GET_ERROR_TYPE);
225 if (rc)
226 return rc;
227 if (*type1 & ACPI65_EINJV2_SUPP) {
228 rc = einj_get_available_error_type(type2,
229 ACPI_EINJV2_GET_ERROR_TYPE);
230 if (rc)
231 return rc;
232 }
233
234 return 0;
235 }
236
einj_timedout(u64 * t)237 static int einj_timedout(u64 *t)
238 {
239 if ((s64)*t < SLEEP_UNIT_MIN) {
240 pr_warn(FW_WARN "Firmware does not respond in time\n");
241 return 1;
242 }
243 *t -= SLEEP_UNIT_MIN;
244 usleep_range(SLEEP_UNIT_MIN, SLEEP_UNIT_MAX);
245
246 return 0;
247 }
248
get_oem_vendor_struct(u64 paddr,int offset,struct vendor_error_type_extension * v)249 static void get_oem_vendor_struct(u64 paddr, int offset,
250 struct vendor_error_type_extension *v)
251 {
252 unsigned long vendor_size;
253 u64 target_pa = paddr + offset + sizeof(struct vendor_error_type_extension);
254
255 vendor_size = v->length - sizeof(struct vendor_error_type_extension);
256
257 if (vendor_size)
258 vendor_errors.data = acpi_os_map_memory(target_pa, vendor_size);
259
260 if (vendor_errors.data)
261 vendor_errors.size = vendor_size;
262 }
263
check_vendor_extension(u64 paddr,struct set_error_type_with_address * v5param)264 static void check_vendor_extension(u64 paddr,
265 struct set_error_type_with_address *v5param)
266 {
267 int offset = v5param->vendor_extension;
268 struct vendor_error_type_extension v;
269 struct vendor_error_type_extension __iomem *p;
270 u32 sbdf;
271
272 if (!offset)
273 return;
274 p = acpi_os_map_iomem(paddr + offset, sizeof(*p));
275 if (!p)
276 return;
277 memcpy_fromio(&v, p, sizeof(v));
278 get_oem_vendor_struct(paddr, offset, &v);
279 sbdf = v.pcie_sbdf;
280 sprintf(vendor_dev, "%x:%x:%x.%x vendor_id=%x device_id=%x rev_id=%x\n",
281 sbdf >> 24, (sbdf >> 16) & 0xff,
282 (sbdf >> 11) & 0x1f, (sbdf >> 8) & 0x7,
283 v.vendor_id, v.device_id, v.rev_id);
284 acpi_os_unmap_iomem(p, sizeof(v));
285 }
286
einjv2_init(struct einjv2_extension_struct * e)287 static u32 einjv2_init(struct einjv2_extension_struct *e)
288 {
289 if (e->revision != 1) {
290 pr_info("Unknown v2 extension revision %u\n", e->revision);
291 return 0;
292 }
293 if (e->length < sizeof(*e) || e->length > PAGE_SIZE) {
294 pr_info(FW_BUG "Bad1 v2 extension length %u\n", e->length);
295 return 0;
296 }
297 if ((e->length - sizeof(*e)) % sizeof(e->component_arr[0])) {
298 pr_info(FW_BUG "Bad2 v2 extension length %u\n", e->length);
299 return 0;
300 }
301
302 return (e->length - sizeof(*e)) / sizeof(e->component_arr[0]);
303 }
304
einj_get_parameter_address(void)305 static void __iomem *einj_get_parameter_address(void)
306 {
307 int i;
308 u64 pa_v4 = 0, pa_v5 = 0;
309 struct acpi_whea_header *entry;
310
311 entry = EINJ_TAB_ENTRY(einj_tab);
312 for (i = 0; i < einj_tab->entries; i++) {
313 if (entry->action == ACPI_EINJ_SET_ERROR_TYPE &&
314 entry->instruction == ACPI_EINJ_WRITE_REGISTER &&
315 entry->register_region.space_id ==
316 ACPI_ADR_SPACE_SYSTEM_MEMORY)
317 pa_v4 = get_unaligned(&entry->register_region.address);
318 if (entry->action == ACPI_EINJ_SET_ERROR_TYPE_WITH_ADDRESS &&
319 entry->instruction == ACPI_EINJ_WRITE_REGISTER &&
320 entry->register_region.space_id ==
321 ACPI_ADR_SPACE_SYSTEM_MEMORY)
322 pa_v5 = get_unaligned(&entry->register_region.address);
323 entry++;
324 }
325 if (pa_v5) {
326 struct set_error_type_with_address v5param;
327 struct set_error_type_with_address __iomem *p;
328
329 v5param_size = sizeof(v5param);
330 p = acpi_os_map_iomem(pa_v5, sizeof(*p));
331 if (p) {
332 memcpy_fromio(&v5param, p, v5param_size);
333 acpi5 = 1;
334 check_vendor_extension(pa_v5, &v5param);
335 if (available_error_type & ACPI65_EINJV2_SUPP) {
336 struct einjv2_extension_struct *e;
337
338 e = &v5param.einjv2_struct;
339 max_nr_components = einjv2_init(e);
340
341 /* remap including einjv2_extension_struct */
342 acpi_os_unmap_iomem(p, v5param_size);
343 v66param_size = v5param_size - sizeof(*e) + e->length;
344 p = acpi_os_map_iomem(pa_v5, v66param_size);
345 }
346
347 return p;
348 }
349 }
350 if (param_extension && pa_v4) {
351 struct einj_parameter v4param;
352 struct einj_parameter __iomem *p;
353
354 p = acpi_os_map_iomem(pa_v4, sizeof(*p));
355 if (!p)
356 return NULL;
357 memcpy_fromio(&v4param, p, sizeof(v4param));
358 if (v4param.reserved1 || v4param.reserved2) {
359 acpi_os_unmap_iomem(p, sizeof(v4param));
360 return NULL;
361 }
362 return p;
363 }
364
365 return NULL;
366 }
367
368 /* do sanity check to trigger table */
einj_check_trigger_header(struct acpi_einj_trigger * trigger_tab)369 static int einj_check_trigger_header(struct acpi_einj_trigger *trigger_tab)
370 {
371 if (trigger_tab->header_size != sizeof(struct acpi_einj_trigger))
372 return -EINVAL;
373 if (trigger_tab->table_size > PAGE_SIZE ||
374 trigger_tab->table_size < trigger_tab->header_size)
375 return -EINVAL;
376 if (trigger_tab->entry_count !=
377 (trigger_tab->table_size - trigger_tab->header_size) /
378 sizeof(struct acpi_einj_entry))
379 return -EINVAL;
380
381 return 0;
382 }
383
einj_get_trigger_parameter_region(struct acpi_einj_trigger * trigger_tab,u64 param1,u64 param2)384 static struct acpi_generic_address *einj_get_trigger_parameter_region(
385 struct acpi_einj_trigger *trigger_tab, u64 param1, u64 param2)
386 {
387 int i;
388 struct acpi_whea_header *entry;
389
390 entry = (struct acpi_whea_header *)
391 ((char *)trigger_tab + sizeof(struct acpi_einj_trigger));
392 for (i = 0; i < trigger_tab->entry_count; i++) {
393 if (entry->action == ACPI_EINJ_TRIGGER_ERROR &&
394 entry->instruction <= ACPI_EINJ_WRITE_REGISTER_VALUE &&
395 entry->register_region.space_id ==
396 ACPI_ADR_SPACE_SYSTEM_MEMORY &&
397 (entry->register_region.address & param2) == (param1 & param2))
398 return &entry->register_region;
399 entry++;
400 }
401
402 return NULL;
403 }
404 /* Execute instructions in trigger error action table */
__einj_error_trigger(u64 trigger_paddr,u32 type,u64 param1,u64 param2)405 static int __einj_error_trigger(u64 trigger_paddr, u32 type,
406 u64 param1, u64 param2)
407 {
408 struct acpi_einj_trigger trigger_tab;
409 struct acpi_einj_trigger *full_trigger_tab;
410 struct apei_exec_context trigger_ctx;
411 struct apei_resources trigger_resources;
412 struct acpi_whea_header *trigger_entry;
413 struct resource *r;
414 u32 table_size;
415 int rc = -EIO;
416 struct acpi_generic_address *trigger_param_region = NULL;
417 struct acpi_einj_trigger __iomem *p = NULL;
418
419 r = request_mem_region(trigger_paddr, sizeof(trigger_tab),
420 "APEI EINJ Trigger Table");
421 if (!r) {
422 pr_err("Can not request [mem %#010llx-%#010llx] for Trigger table\n",
423 (unsigned long long)trigger_paddr,
424 (unsigned long long)trigger_paddr +
425 sizeof(trigger_tab) - 1);
426 goto out;
427 }
428 p = ioremap_cache(trigger_paddr, sizeof(*p));
429 if (!p) {
430 pr_err("Failed to map trigger table!\n");
431 goto out_rel_header;
432 }
433 memcpy_fromio(&trigger_tab, p, sizeof(trigger_tab));
434 rc = einj_check_trigger_header(&trigger_tab);
435 if (rc) {
436 pr_warn(FW_BUG "Invalid trigger error action table.\n");
437 goto out_rel_header;
438 }
439
440 /* No action structures in the TRIGGER_ERROR table, nothing to do */
441 if (!trigger_tab.entry_count)
442 goto out_rel_header;
443
444 rc = -EIO;
445 table_size = trigger_tab.table_size;
446 full_trigger_tab = kmalloc(table_size, GFP_KERNEL);
447 if (!full_trigger_tab)
448 goto out_rel_header;
449 r = request_mem_region(trigger_paddr + sizeof(trigger_tab),
450 table_size - sizeof(trigger_tab),
451 "APEI EINJ Trigger Table");
452 if (!r) {
453 pr_err("Can not request [mem %#010llx-%#010llx] for Trigger Table Entry\n",
454 (unsigned long long)trigger_paddr + sizeof(trigger_tab),
455 (unsigned long long)trigger_paddr + table_size - 1);
456 goto out_free_trigger_tab;
457 }
458 iounmap(p);
459 p = ioremap_cache(trigger_paddr, table_size);
460 if (!p) {
461 pr_err("Failed to map trigger table!\n");
462 goto out_rel_entry;
463 }
464 memcpy_fromio(full_trigger_tab, p, table_size);
465 trigger_entry = (struct acpi_whea_header *)
466 ((char *)full_trigger_tab + sizeof(struct acpi_einj_trigger));
467 apei_resources_init(&trigger_resources);
468 apei_exec_ctx_init(&trigger_ctx, einj_ins_type,
469 ARRAY_SIZE(einj_ins_type),
470 trigger_entry, trigger_tab.entry_count);
471 rc = apei_exec_collect_resources(&trigger_ctx, &trigger_resources);
472 if (rc)
473 goto out_fini;
474 rc = apei_resources_sub(&trigger_resources, &einj_resources);
475 if (rc)
476 goto out_fini;
477 /*
478 * Some firmware will access target address specified in
479 * param1 to trigger the error when injecting memory error.
480 * This will cause resource conflict with regular memory. So
481 * remove it from trigger table resources.
482 */
483 if ((param_extension || acpi5) && (type & MEM_ERROR_MASK) && param2) {
484 struct apei_resources addr_resources;
485
486 apei_resources_init(&addr_resources);
487 trigger_param_region = einj_get_trigger_parameter_region(
488 full_trigger_tab, param1, param2);
489 if (trigger_param_region) {
490 rc = apei_resources_add(&addr_resources,
491 trigger_param_region->address,
492 trigger_param_region->bit_width/8, true);
493 if (rc)
494 goto out_fini;
495 rc = apei_resources_sub(&trigger_resources,
496 &addr_resources);
497 }
498 apei_resources_fini(&addr_resources);
499 if (rc)
500 goto out_fini;
501 }
502 rc = apei_resources_request(&trigger_resources, "APEI EINJ Trigger");
503 if (rc)
504 goto out_fini;
505 rc = apei_exec_pre_map_gars(&trigger_ctx);
506 if (rc)
507 goto out_release;
508
509 rc = apei_exec_run(&trigger_ctx, ACPI_EINJ_TRIGGER_ERROR);
510
511 apei_exec_post_unmap_gars(&trigger_ctx);
512 out_release:
513 apei_resources_release(&trigger_resources);
514 out_fini:
515 apei_resources_fini(&trigger_resources);
516 out_rel_entry:
517 release_mem_region(trigger_paddr + sizeof(trigger_tab),
518 table_size - sizeof(trigger_tab));
519 out_free_trigger_tab:
520 kfree(full_trigger_tab);
521 out_rel_header:
522 release_mem_region(trigger_paddr, sizeof(trigger_tab));
523 out:
524 if (p)
525 iounmap(p);
526
527 return rc;
528 }
529
is_end_of_list(u8 * val)530 static bool is_end_of_list(u8 *val)
531 {
532 for (int i = 0; i < COMPONENT_LEN; ++i) {
533 if (val[i] != 0xFF)
534 return false;
535 }
536 return true;
537 }
__einj_error_inject(u32 type,u32 flags,u64 param1,u64 param2,u64 param3,u64 param4)538 static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
539 u64 param3, u64 param4)
540 {
541 struct apei_exec_context ctx;
542 u32 param_size = is_v2 ? v66param_size : v5param_size;
543 u64 val, trigger_paddr, timeout = FIRMWARE_TIMEOUT;
544 int i, rc;
545
546 einj_exec_ctx_init(&ctx);
547
548 rc = apei_exec_run_optional(&ctx, ACPI_EINJ_BEGIN_OPERATION);
549 if (rc)
550 return rc;
551 apei_exec_ctx_set_input(&ctx, type);
552 if (acpi5) {
553 struct set_error_type_with_address *v5param;
554
555 v5param = kmalloc(param_size, GFP_KERNEL);
556 if (!v5param)
557 return -ENOMEM;
558
559 memcpy_fromio(v5param, einj_param, param_size);
560 v5param->type = type;
561 if (type & ACPI5_VENDOR_BIT) {
562 switch (vendor_flags) {
563 case SETWA_FLAGS_APICID:
564 v5param->apicid = param1;
565 break;
566 case SETWA_FLAGS_MEM:
567 v5param->memory_address = param1;
568 v5param->memory_address_range = param2;
569 break;
570 case SETWA_FLAGS_PCIE_SBDF:
571 v5param->pcie_sbdf = param1;
572 break;
573 }
574 v5param->flags = vendor_flags;
575 } else if (flags) {
576 v5param->flags = flags;
577 v5param->memory_address = param1;
578 v5param->memory_address_range = param2;
579
580 if (is_v2) {
581 for (i = 0; i < max_nr_components; i++) {
582 if (is_end_of_list(syndrome_data[i].comp_id.acpi_id))
583 break;
584 v5param->einjv2_struct.component_arr[i].comp_id =
585 syndrome_data[i].comp_id;
586 v5param->einjv2_struct.component_arr[i].comp_synd =
587 syndrome_data[i].comp_synd;
588 }
589 v5param->einjv2_struct.component_arr_count = i;
590 } else {
591 v5param->apicid = param3;
592 v5param->pcie_sbdf = param4;
593 }
594 } else {
595 switch (type) {
596 case ACPI_EINJ_PROCESSOR_CORRECTABLE:
597 case ACPI_EINJ_PROCESSOR_UNCORRECTABLE:
598 case ACPI_EINJ_PROCESSOR_FATAL:
599 v5param->apicid = param1;
600 v5param->flags = SETWA_FLAGS_APICID;
601 break;
602 case ACPI_EINJ_MEMORY_CORRECTABLE:
603 case ACPI_EINJ_MEMORY_UNCORRECTABLE:
604 case ACPI_EINJ_MEMORY_FATAL:
605 v5param->memory_address = param1;
606 v5param->memory_address_range = param2;
607 v5param->flags = SETWA_FLAGS_MEM;
608 break;
609 case ACPI_EINJ_PCIX_CORRECTABLE:
610 case ACPI_EINJ_PCIX_UNCORRECTABLE:
611 case ACPI_EINJ_PCIX_FATAL:
612 v5param->pcie_sbdf = param1;
613 v5param->flags = SETWA_FLAGS_PCIE_SBDF;
614 break;
615 }
616 }
617 memcpy_toio(einj_param, v5param, param_size);
618 kfree(v5param);
619 } else {
620 rc = apei_exec_run(&ctx, ACPI_EINJ_SET_ERROR_TYPE);
621 if (rc)
622 return rc;
623 if (einj_param) {
624 struct einj_parameter v4param;
625
626 memcpy_fromio(&v4param, einj_param, sizeof(v4param));
627 v4param.param1 = param1;
628 v4param.param2 = param2;
629 memcpy_toio(einj_param, &v4param, sizeof(v4param));
630 }
631 }
632 rc = apei_exec_run(&ctx, ACPI_EINJ_EXECUTE_OPERATION);
633 if (rc)
634 return rc;
635 for (;;) {
636 rc = apei_exec_run(&ctx, ACPI_EINJ_CHECK_BUSY_STATUS);
637 if (rc)
638 return rc;
639 val = apei_exec_ctx_get_output(&ctx);
640 if (!(val & EINJ_OP_BUSY))
641 break;
642 if (einj_timedout(&timeout))
643 return -EIO;
644 }
645 rc = apei_exec_run(&ctx, ACPI_EINJ_GET_COMMAND_STATUS);
646 if (rc)
647 return rc;
648 val = apei_exec_ctx_get_output(&ctx);
649 if (val == EINJ_STATUS_FAIL)
650 return -EBUSY;
651 else if (val == EINJ_STATUS_INVAL)
652 return -EINVAL;
653
654 /*
655 * The error is injected into the platform successfully, then it needs
656 * to trigger the error.
657 */
658 rc = apei_exec_run(&ctx, ACPI_EINJ_GET_TRIGGER_TABLE);
659 if (rc)
660 return rc;
661 trigger_paddr = apei_exec_ctx_get_output(&ctx);
662 if (notrigger == 0) {
663 rc = __einj_error_trigger(trigger_paddr, type, param1, param2);
664 if (rc)
665 return rc;
666 }
667 rc = apei_exec_run_optional(&ctx, ACPI_EINJ_END_OPERATION);
668
669 return rc;
670 }
671
672 /* Allow almost all types of address except MMIO. */
is_allowed_range(u64 base_addr,u64 size)673 static bool is_allowed_range(u64 base_addr, u64 size)
674 {
675 int i;
676 /*
677 * MMIO region is usually claimed with IORESOURCE_MEM + IORES_DESC_NONE.
678 * However, IORES_DESC_NONE is treated like a wildcard when we check if
679 * region intersects with known resource. So do an allow list check for
680 * IORES_DESCs that definitely or most likely not MMIO.
681 */
682 int non_mmio_desc[] = {
683 IORES_DESC_CRASH_KERNEL,
684 IORES_DESC_ACPI_TABLES,
685 IORES_DESC_ACPI_NV_STORAGE,
686 IORES_DESC_PERSISTENT_MEMORY,
687 IORES_DESC_PERSISTENT_MEMORY_LEGACY,
688 /* Treat IORES_DESC_DEVICE_PRIVATE_MEMORY as MMIO. */
689 IORES_DESC_RESERVED,
690 IORES_DESC_SOFT_RESERVED,
691 };
692
693 if (region_intersects(base_addr, size, IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE)
694 == REGION_INTERSECTS)
695 return true;
696
697 for (i = 0; i < ARRAY_SIZE(non_mmio_desc); ++i) {
698 if (region_intersects(base_addr, size, IORESOURCE_MEM, non_mmio_desc[i])
699 == REGION_INTERSECTS)
700 return true;
701 }
702
703 if (arch_is_platform_page(base_addr))
704 return true;
705
706 return false;
707 }
708
709 /* Inject the specified hardware error */
einj_error_inject(u32 type,u32 flags,u64 param1,u64 param2,u64 param3,u64 param4)710 int einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2, u64 param3,
711 u64 param4)
712 {
713 int rc;
714 u64 base_addr, size;
715
716 /* If user manually set "flags", make sure it is legal */
717 if (flags && (flags & ~(SETWA_FLAGS_APICID | SETWA_FLAGS_MEM |
718 SETWA_FLAGS_PCIE_SBDF | SETWA_FLAGS_EINJV2)))
719 return -EINVAL;
720
721 /* check if type is a valid EINJv2 error type */
722 if (is_v2) {
723 if (!(type & available_error_type_v2))
724 return -EINVAL;
725 }
726 /*
727 * We need extra sanity checks for memory errors.
728 * Other types leap directly to injection.
729 */
730
731 /* ensure param1/param2 existed */
732 if (!(param_extension || acpi5))
733 goto inject;
734
735 /* ensure injection is memory related */
736 if (type & ACPI5_VENDOR_BIT) {
737 if (vendor_flags != SETWA_FLAGS_MEM)
738 goto inject;
739 } else if (!(type & MEM_ERROR_MASK) && !(flags & SETWA_FLAGS_MEM)) {
740 goto inject;
741 }
742
743 /*
744 * Injections targeting a CXL 1.0/1.1 port have to be injected
745 * via the einj_cxl_rch_error_inject() path as that does the proper
746 * validation of the given RCRB base (MMIO) address.
747 */
748 if (einj_is_cxl_error_type(type) && (flags & SETWA_FLAGS_MEM))
749 return -EINVAL;
750
751 /*
752 * Disallow crazy address masks that give BIOS leeway to pick
753 * injection address almost anywhere. Insist on page or
754 * better granularity and that target address is normal RAM or
755 * as long as is not MMIO.
756 */
757 base_addr = param1 & param2;
758 size = ~param2 + 1;
759
760 if ((param2 & PAGE_MASK) != PAGE_MASK)
761 return -EINVAL;
762
763 if (!is_allowed_range(base_addr, size))
764 return -EINVAL;
765
766 if (is_zero_pfn(base_addr >> PAGE_SHIFT))
767 return -EADDRINUSE;
768
769 inject:
770 mutex_lock(&einj_mutex);
771 rc = __einj_error_inject(type, flags, param1, param2, param3, param4);
772 mutex_unlock(&einj_mutex);
773
774 return rc;
775 }
776
einj_cxl_rch_error_inject(u32 type,u32 flags,u64 param1,u64 param2,u64 param3,u64 param4)777 int einj_cxl_rch_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
778 u64 param3, u64 param4)
779 {
780 int rc;
781
782 if (!(einj_is_cxl_error_type(type) && (flags & SETWA_FLAGS_MEM)))
783 return -EINVAL;
784
785 mutex_lock(&einj_mutex);
786 rc = __einj_error_inject(type, flags, param1, param2, param3, param4);
787 mutex_unlock(&einj_mutex);
788
789 return rc;
790 }
791
792 static u32 error_type;
793 static u32 error_flags;
794 static u64 error_param1;
795 static u64 error_param2;
796 static u64 error_param3;
797 static u64 error_param4;
798 static struct dentry *einj_debug_dir;
799 static char einj_buf[32];
800 static bool einj_v2_enabled;
801 static struct { u32 mask; const char *str; } const einj_error_type_string[] = {
802 { BIT(0), "Processor Correctable" },
803 { BIT(1), "Processor Uncorrectable non-fatal" },
804 { BIT(2), "Processor Uncorrectable fatal" },
805 { BIT(3), "Memory Correctable" },
806 { BIT(4), "Memory Uncorrectable non-fatal" },
807 { BIT(5), "Memory Uncorrectable fatal" },
808 { BIT(6), "PCI Express Correctable" },
809 { BIT(7), "PCI Express Uncorrectable non-fatal" },
810 { BIT(8), "PCI Express Uncorrectable fatal" },
811 { BIT(9), "Platform Correctable" },
812 { BIT(10), "Platform Uncorrectable non-fatal" },
813 { BIT(11), "Platform Uncorrectable fatal"},
814 { BIT(31), "Vendor Defined Error Types" },
815 };
816
817 static struct { u32 mask; const char *str; } const einjv2_error_type_string[] = {
818 { BIT(0), "EINJV2 Processor Error" },
819 { BIT(1), "EINJV2 Memory Error" },
820 { BIT(2), "EINJV2 PCI Express Error" },
821 };
822
available_error_type_show(struct seq_file * m,void * v)823 static int available_error_type_show(struct seq_file *m, void *v)
824 {
825
826 for (int pos = 0; pos < ARRAY_SIZE(einj_error_type_string); pos++)
827 if (available_error_type & einj_error_type_string[pos].mask)
828 seq_printf(m, "0x%08x\t%s\n", einj_error_type_string[pos].mask,
829 einj_error_type_string[pos].str);
830 if ((available_error_type & ACPI65_EINJV2_SUPP) && einj_v2_enabled) {
831 for (int pos = 0; pos < ARRAY_SIZE(einjv2_error_type_string); pos++) {
832 if (available_error_type_v2 & einjv2_error_type_string[pos].mask)
833 seq_printf(m, "V2_0x%08x\t%s\n", einjv2_error_type_string[pos].mask,
834 einjv2_error_type_string[pos].str);
835 }
836 }
837 return 0;
838 }
839
840 DEFINE_SHOW_ATTRIBUTE(available_error_type);
841
error_type_get(struct file * file,char __user * buf,size_t count,loff_t * ppos)842 static ssize_t error_type_get(struct file *file, char __user *buf,
843 size_t count, loff_t *ppos)
844 {
845 return simple_read_from_buffer(buf, count, ppos, einj_buf, strlen(einj_buf));
846 }
847
einj_is_cxl_error_type(u64 type)848 bool einj_is_cxl_error_type(u64 type)
849 {
850 return (type & CXL_ERROR_MASK) && (!(type & ACPI5_VENDOR_BIT));
851 }
852
einj_validate_error_type(u64 type)853 int einj_validate_error_type(u64 type)
854 {
855 u32 tval, vendor;
856
857 /* Only low 32 bits for error type are valid */
858 if (type & GENMASK_ULL(63, 32))
859 return -EINVAL;
860
861 /*
862 * Vendor defined types have 0x80000000 bit set, and
863 * are not enumerated by ACPI_EINJ_GET_ERROR_TYPE
864 */
865 vendor = type & ACPI5_VENDOR_BIT;
866 tval = type & GENMASK(30, 0);
867
868 /* Only one error type can be specified */
869 if (tval & (tval - 1))
870 return -EINVAL;
871 if (!vendor)
872 if (!(type & (available_error_type | available_error_type_v2)))
873 return -EINVAL;
874
875 return 0;
876 }
877
error_type_set(struct file * file,const char __user * buf,size_t count,loff_t * ppos)878 static ssize_t error_type_set(struct file *file, const char __user *buf,
879 size_t count, loff_t *ppos)
880 {
881 int rc;
882 u64 val;
883
884 /* Leave the last character for the NUL terminator */
885 if (count > sizeof(einj_buf) - 1)
886 return -EINVAL;
887
888 memset(einj_buf, 0, sizeof(einj_buf));
889 if (copy_from_user(einj_buf, buf, count))
890 return -EFAULT;
891
892 if (strncmp(einj_buf, "V2_", 3) == 0) {
893 if (!sscanf(einj_buf, "V2_%llx", &val))
894 return -EINVAL;
895 is_v2 = true;
896 } else {
897 if (!sscanf(einj_buf, "%llx", &val))
898 return -EINVAL;
899 is_v2 = false;
900 }
901
902 rc = einj_validate_error_type(val);
903 if (rc)
904 return rc;
905
906 error_type = val;
907
908 return count;
909 }
910
911 static const struct file_operations error_type_fops = {
912 .read = error_type_get,
913 .write = error_type_set,
914 };
915
error_inject_set(void * data,u64 val)916 static int error_inject_set(void *data, u64 val)
917 {
918 if (!error_type)
919 return -EINVAL;
920
921 if (is_v2)
922 error_flags |= SETWA_FLAGS_EINJV2;
923 else
924 error_flags &= ~SETWA_FLAGS_EINJV2;
925
926 return einj_error_inject(error_type, error_flags, error_param1, error_param2,
927 error_param3, error_param4);
928 }
929
930 DEFINE_DEBUGFS_ATTRIBUTE(error_inject_fops, NULL, error_inject_set, "%llu\n");
931
einj_check_table(struct acpi_table_einj * einj_tab)932 static int einj_check_table(struct acpi_table_einj *einj_tab)
933 {
934 if ((einj_tab->header_length !=
935 (sizeof(struct acpi_table_einj) - sizeof(einj_tab->header)))
936 && (einj_tab->header_length != sizeof(struct acpi_table_einj)))
937 return -EINVAL;
938 if (einj_tab->header.length < sizeof(struct acpi_table_einj))
939 return -EINVAL;
940 if (einj_tab->entries !=
941 (einj_tab->header.length - sizeof(struct acpi_table_einj)) /
942 sizeof(struct acpi_einj_entry))
943 return -EINVAL;
944
945 return 0;
946 }
947
u128_read(struct file * f,char __user * buf,size_t count,loff_t * off)948 static ssize_t u128_read(struct file *f, char __user *buf, size_t count, loff_t *off)
949 {
950 char output[2 * COMPONENT_LEN + 1];
951 u8 *data = f->f_inode->i_private;
952 int i;
953
954 if (*off >= sizeof(output))
955 return 0;
956
957 for (i = 0; i < COMPONENT_LEN; i++)
958 sprintf(output + 2 * i, "%.02x", data[COMPONENT_LEN - i - 1]);
959 output[2 * COMPONENT_LEN] = '\n';
960
961 return simple_read_from_buffer(buf, count, off, output, sizeof(output));
962 }
963
u128_write(struct file * f,const char __user * buf,size_t count,loff_t * off)964 static ssize_t u128_write(struct file *f, const char __user *buf, size_t count, loff_t *off)
965 {
966 char input[2 + 2 * COMPONENT_LEN + 2];
967 u8 *save = f->f_inode->i_private;
968 u8 tmp[COMPONENT_LEN];
969 char byte[3] = {};
970 char *s, *e;
971 ssize_t c;
972 long val;
973 int i;
974
975 /* Require that user supply whole input line in one write(2) syscall */
976 if (*off)
977 return -EINVAL;
978
979 c = simple_write_to_buffer(input, sizeof(input), off, buf, count);
980 if (c < 0)
981 return c;
982
983 if (c < 1 || input[c - 1] != '\n')
984 return -EINVAL;
985
986 /* Empty line means invalidate this entry */
987 if (c == 1) {
988 memset(save, 0xff, COMPONENT_LEN);
989 return c;
990 }
991
992 if (input[0] == '0' && (input[1] == 'x' || input[1] == 'X'))
993 s = input + 2;
994 else
995 s = input;
996 e = input + c - 1;
997
998 for (i = 0; i < COMPONENT_LEN; i++) {
999 byte[1] = *--e;
1000 byte[0] = e > s ? *--e : '0';
1001 if (kstrtol(byte, 16, &val))
1002 return -EINVAL;
1003 tmp[i] = val;
1004 if (e <= s)
1005 break;
1006 }
1007 while (++i < COMPONENT_LEN)
1008 tmp[i] = 0;
1009
1010 memcpy(save, tmp, COMPONENT_LEN);
1011
1012 return c;
1013 }
1014
1015 static const struct file_operations u128_fops = {
1016 .read = u128_read,
1017 .write = u128_write,
1018 };
1019
setup_einjv2_component_files(void)1020 static bool setup_einjv2_component_files(void)
1021 {
1022 char name[32];
1023
1024 syndrome_data = kcalloc(max_nr_components, sizeof(syndrome_data[0]), GFP_KERNEL);
1025 if (!syndrome_data)
1026 return false;
1027
1028 for (int i = 0; i < max_nr_components; i++) {
1029 sprintf(name, "component_id%d", i);
1030 debugfs_create_file(name, 0600, einj_debug_dir,
1031 &syndrome_data[i].comp_id, &u128_fops);
1032 sprintf(name, "component_syndrome%d", i);
1033 debugfs_create_file(name, 0600, einj_debug_dir,
1034 &syndrome_data[i].comp_synd, &u128_fops);
1035 }
1036
1037 return true;
1038 }
1039
einj_probe(struct faux_device * fdev)1040 static int __init einj_probe(struct faux_device *fdev)
1041 {
1042 int rc;
1043 acpi_status status;
1044 struct apei_exec_context ctx;
1045
1046 status = acpi_get_table(ACPI_SIG_EINJ, 0,
1047 (struct acpi_table_header **)&einj_tab);
1048 if (status == AE_NOT_FOUND) {
1049 pr_debug("EINJ table not found.\n");
1050 return -ENODEV;
1051 } else if (ACPI_FAILURE(status)) {
1052 pr_err("Failed to get EINJ table: %s\n",
1053 acpi_format_exception(status));
1054 return -EINVAL;
1055 }
1056
1057 rc = einj_check_table(einj_tab);
1058 if (rc) {
1059 pr_warn(FW_BUG "Invalid EINJ table.\n");
1060 goto err_put_table;
1061 }
1062
1063 rc = einj_get_available_error_types(&available_error_type, &available_error_type_v2);
1064 if (rc)
1065 goto err_put_table;
1066
1067 rc = -ENOMEM;
1068 einj_debug_dir = debugfs_create_dir("einj", apei_get_debugfs_dir());
1069
1070 debugfs_create_file("available_error_type", S_IRUSR, einj_debug_dir,
1071 NULL, &available_error_type_fops);
1072 debugfs_create_file_unsafe("error_type", 0600, einj_debug_dir,
1073 NULL, &error_type_fops);
1074 debugfs_create_file_unsafe("error_inject", 0200, einj_debug_dir,
1075 NULL, &error_inject_fops);
1076
1077 apei_resources_init(&einj_resources);
1078 einj_exec_ctx_init(&ctx);
1079 rc = apei_exec_collect_resources(&ctx, &einj_resources);
1080 if (rc) {
1081 pr_err("Error collecting EINJ resources.\n");
1082 goto err_fini;
1083 }
1084
1085 rc = apei_resources_request(&einj_resources, "APEI EINJ");
1086 if (rc) {
1087 pr_err("Error requesting memory/port resources.\n");
1088 goto err_fini;
1089 }
1090
1091 rc = apei_exec_pre_map_gars(&ctx);
1092 if (rc) {
1093 pr_err("Error pre-mapping GARs.\n");
1094 goto err_release;
1095 }
1096
1097 einj_param = einj_get_parameter_address();
1098 if ((param_extension || acpi5) && einj_param) {
1099 debugfs_create_x32("flags", S_IRUSR | S_IWUSR, einj_debug_dir,
1100 &error_flags);
1101 debugfs_create_x64("param1", S_IRUSR | S_IWUSR, einj_debug_dir,
1102 &error_param1);
1103 debugfs_create_x64("param2", S_IRUSR | S_IWUSR, einj_debug_dir,
1104 &error_param2);
1105 debugfs_create_x64("param3", S_IRUSR | S_IWUSR, einj_debug_dir,
1106 &error_param3);
1107 debugfs_create_x64("param4", S_IRUSR | S_IWUSR, einj_debug_dir,
1108 &error_param4);
1109 debugfs_create_x32("notrigger", S_IRUSR | S_IWUSR,
1110 einj_debug_dir, ¬rigger);
1111 if (available_error_type & ACPI65_EINJV2_SUPP)
1112 einj_v2_enabled = setup_einjv2_component_files();
1113 }
1114
1115 if (vendor_dev[0]) {
1116 vendor_blob.data = vendor_dev;
1117 vendor_blob.size = strlen(vendor_dev);
1118 debugfs_create_blob("vendor", S_IRUSR, einj_debug_dir,
1119 &vendor_blob);
1120 debugfs_create_x32("vendor_flags", S_IRUSR | S_IWUSR,
1121 einj_debug_dir, &vendor_flags);
1122 }
1123
1124 if (vendor_errors.size)
1125 debugfs_create_blob("oem_error", 0600, einj_debug_dir,
1126 &vendor_errors);
1127
1128 pr_info("Error INJection is initialized.\n");
1129
1130 return 0;
1131
1132 err_release:
1133 apei_resources_release(&einj_resources);
1134 err_fini:
1135 apei_resources_fini(&einj_resources);
1136 debugfs_remove_recursive(einj_debug_dir);
1137 err_put_table:
1138 acpi_put_table((struct acpi_table_header *)einj_tab);
1139
1140 return rc;
1141 }
1142
einj_remove(struct faux_device * fdev)1143 static void einj_remove(struct faux_device *fdev)
1144 {
1145 struct apei_exec_context ctx;
1146
1147 if (einj_param) {
1148 acpi_size size;
1149
1150 if (v66param_size)
1151 size = v66param_size;
1152 else if (acpi5)
1153 size = v5param_size;
1154 else
1155 size = sizeof(struct einj_parameter);
1156
1157 acpi_os_unmap_iomem(einj_param, size);
1158 if (vendor_errors.size)
1159 acpi_os_unmap_memory(vendor_errors.data, vendor_errors.size);
1160 }
1161 einj_exec_ctx_init(&ctx);
1162 apei_exec_post_unmap_gars(&ctx);
1163 apei_resources_release(&einj_resources);
1164 apei_resources_fini(&einj_resources);
1165 debugfs_remove_recursive(einj_debug_dir);
1166 kfree(syndrome_data);
1167 acpi_put_table((struct acpi_table_header *)einj_tab);
1168 }
1169
1170 static struct faux_device *einj_dev;
1171 static struct faux_device_ops einj_device_ops = {
1172 .probe = einj_probe,
1173 .remove = einj_remove,
1174 };
1175
einj_init(void)1176 static int __init einj_init(void)
1177 {
1178 if (acpi_disabled) {
1179 pr_debug("ACPI disabled.\n");
1180 return -ENODEV;
1181 }
1182
1183 einj_dev = faux_device_create("acpi-einj", NULL, &einj_device_ops);
1184
1185 if (einj_dev)
1186 einj_initialized = true;
1187
1188 return 0;
1189 }
1190
einj_exit(void)1191 static void __exit einj_exit(void)
1192 {
1193 faux_device_destroy(einj_dev);
1194 }
1195
1196 module_init(einj_init);
1197 module_exit(einj_exit);
1198
1199 MODULE_AUTHOR("Huang Ying");
1200 MODULE_DESCRIPTION("APEI Error INJection support");
1201 MODULE_LICENSE("GPL");
1202