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
is_memory_injection(u32 type,u32 flags)405 static bool is_memory_injection(u32 type, u32 flags)
406 {
407 if (flags & SETWA_FLAGS_EINJV2)
408 return !!(type & ACPI_EINJV2_MEMORY);
409 if (type & ACPI5_VENDOR_BIT)
410 return !!(vendor_flags & SETWA_FLAGS_MEM);
411 return !!(type & MEM_ERROR_MASK) || !!(flags & SETWA_FLAGS_MEM);
412 }
413
414 /* Execute instructions in trigger error action table */
__einj_error_trigger(u64 trigger_paddr,u32 type,u32 flags,u64 param1,u64 param2)415 static int __einj_error_trigger(u64 trigger_paddr, u32 type, u32 flags,
416 u64 param1, u64 param2)
417 {
418 struct acpi_einj_trigger trigger_tab;
419 struct acpi_einj_trigger *full_trigger_tab;
420 struct apei_exec_context trigger_ctx;
421 struct apei_resources trigger_resources;
422 struct acpi_whea_header *trigger_entry;
423 struct resource *r;
424 u32 table_size;
425 int rc = -EIO;
426 struct acpi_generic_address *trigger_param_region = NULL;
427 struct acpi_einj_trigger __iomem *p = NULL;
428
429 r = request_mem_region(trigger_paddr, sizeof(trigger_tab),
430 "APEI EINJ Trigger Table");
431 if (!r) {
432 pr_err("Can not request [mem %#010llx-%#010llx] for Trigger table\n",
433 (unsigned long long)trigger_paddr,
434 (unsigned long long)trigger_paddr +
435 sizeof(trigger_tab) - 1);
436 goto out;
437 }
438 p = ioremap_cache(trigger_paddr, sizeof(*p));
439 if (!p) {
440 pr_err("Failed to map trigger table!\n");
441 goto out_rel_header;
442 }
443 memcpy_fromio(&trigger_tab, p, sizeof(trigger_tab));
444 rc = einj_check_trigger_header(&trigger_tab);
445 if (rc) {
446 pr_warn(FW_BUG "Invalid trigger error action table.\n");
447 goto out_rel_header;
448 }
449
450 /* No action structures in the TRIGGER_ERROR table, nothing to do */
451 if (!trigger_tab.entry_count)
452 goto out_rel_header;
453
454 rc = -EIO;
455 table_size = trigger_tab.table_size;
456 full_trigger_tab = kmalloc(table_size, GFP_KERNEL);
457 if (!full_trigger_tab)
458 goto out_rel_header;
459 r = request_mem_region(trigger_paddr + sizeof(trigger_tab),
460 table_size - sizeof(trigger_tab),
461 "APEI EINJ Trigger Table");
462 if (!r) {
463 pr_err("Can not request [mem %#010llx-%#010llx] for Trigger Table Entry\n",
464 (unsigned long long)trigger_paddr + sizeof(trigger_tab),
465 (unsigned long long)trigger_paddr + table_size - 1);
466 goto out_free_trigger_tab;
467 }
468 iounmap(p);
469 p = ioremap_cache(trigger_paddr, table_size);
470 if (!p) {
471 pr_err("Failed to map trigger table!\n");
472 goto out_rel_entry;
473 }
474 memcpy_fromio(full_trigger_tab, p, table_size);
475 trigger_entry = (struct acpi_whea_header *)
476 ((char *)full_trigger_tab + sizeof(struct acpi_einj_trigger));
477 apei_resources_init(&trigger_resources);
478 apei_exec_ctx_init(&trigger_ctx, einj_ins_type,
479 ARRAY_SIZE(einj_ins_type),
480 trigger_entry, trigger_tab.entry_count);
481 rc = apei_exec_collect_resources(&trigger_ctx, &trigger_resources);
482 if (rc)
483 goto out_fini;
484 rc = apei_resources_sub(&trigger_resources, &einj_resources);
485 if (rc)
486 goto out_fini;
487 /*
488 * Some firmware will access target address specified in
489 * param1 to trigger the error when injecting memory error.
490 * This will cause resource conflict with regular memory. So
491 * remove it from trigger table resources.
492 */
493 if ((param_extension || acpi5) && is_memory_injection(type, flags)) {
494 struct apei_resources addr_resources;
495
496 apei_resources_init(&addr_resources);
497 trigger_param_region = einj_get_trigger_parameter_region(
498 full_trigger_tab, param1, param2);
499 if (trigger_param_region) {
500 rc = apei_resources_add(&addr_resources,
501 trigger_param_region->address,
502 trigger_param_region->bit_width/8, true);
503 if (rc)
504 goto out_fini;
505 rc = apei_resources_sub(&trigger_resources,
506 &addr_resources);
507 }
508 apei_resources_fini(&addr_resources);
509 if (rc)
510 goto out_fini;
511 }
512 rc = apei_resources_request(&trigger_resources, "APEI EINJ Trigger");
513 if (rc)
514 goto out_fini;
515 rc = apei_exec_pre_map_gars(&trigger_ctx);
516 if (rc)
517 goto out_release;
518
519 rc = apei_exec_run(&trigger_ctx, ACPI_EINJ_TRIGGER_ERROR);
520
521 apei_exec_post_unmap_gars(&trigger_ctx);
522 out_release:
523 apei_resources_release(&trigger_resources);
524 out_fini:
525 apei_resources_fini(&trigger_resources);
526 out_rel_entry:
527 release_mem_region(trigger_paddr + sizeof(trigger_tab),
528 table_size - sizeof(trigger_tab));
529 out_free_trigger_tab:
530 kfree(full_trigger_tab);
531 out_rel_header:
532 release_mem_region(trigger_paddr, sizeof(trigger_tab));
533 out:
534 if (p)
535 iounmap(p);
536
537 return rc;
538 }
539
is_end_of_list(u8 * val)540 static bool is_end_of_list(u8 *val)
541 {
542 for (int i = 0; i < COMPONENT_LEN; ++i) {
543 if (val[i] != 0xFF)
544 return false;
545 }
546 return true;
547 }
__einj_error_inject(u32 type,u32 flags,u64 param1,u64 param2,u64 param3,u64 param4)548 static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
549 u64 param3, u64 param4)
550 {
551 struct apei_exec_context ctx;
552 u32 param_size = is_v2 ? v66param_size : v5param_size;
553 u64 val, trigger_paddr, timeout = FIRMWARE_TIMEOUT;
554 int i, rc;
555
556 einj_exec_ctx_init(&ctx);
557
558 rc = apei_exec_run_optional(&ctx, ACPI_EINJ_BEGIN_OPERATION);
559 if (rc)
560 return rc;
561 apei_exec_ctx_set_input(&ctx, type);
562 if (acpi5) {
563 struct set_error_type_with_address *v5param;
564
565 v5param = kmalloc(param_size, GFP_KERNEL);
566 if (!v5param)
567 return -ENOMEM;
568
569 memcpy_fromio(v5param, einj_param, param_size);
570 v5param->type = type;
571 if (type & ACPI5_VENDOR_BIT) {
572 switch (vendor_flags) {
573 case SETWA_FLAGS_APICID:
574 v5param->apicid = param1;
575 break;
576 case SETWA_FLAGS_MEM:
577 v5param->memory_address = param1;
578 v5param->memory_address_range = param2;
579 break;
580 case SETWA_FLAGS_PCIE_SBDF:
581 v5param->pcie_sbdf = param1;
582 break;
583 }
584 v5param->flags = vendor_flags;
585 } else if (flags) {
586 v5param->flags = flags;
587 v5param->memory_address = param1;
588 v5param->memory_address_range = param2;
589
590 if (is_v2) {
591 for (i = 0; i < max_nr_components; i++) {
592 if (is_end_of_list(syndrome_data[i].comp_id.acpi_id))
593 break;
594 v5param->einjv2_struct.component_arr[i].comp_id =
595 syndrome_data[i].comp_id;
596 v5param->einjv2_struct.component_arr[i].comp_synd =
597 syndrome_data[i].comp_synd;
598 }
599 v5param->einjv2_struct.component_arr_count = i;
600 } else {
601 v5param->apicid = param3;
602 v5param->pcie_sbdf = param4;
603 }
604 } else {
605 switch (type) {
606 case ACPI_EINJ_PROCESSOR_CORRECTABLE:
607 case ACPI_EINJ_PROCESSOR_UNCORRECTABLE:
608 case ACPI_EINJ_PROCESSOR_FATAL:
609 v5param->apicid = param1;
610 v5param->flags = SETWA_FLAGS_APICID;
611 break;
612 case ACPI_EINJ_MEMORY_CORRECTABLE:
613 case ACPI_EINJ_MEMORY_UNCORRECTABLE:
614 case ACPI_EINJ_MEMORY_FATAL:
615 v5param->memory_address = param1;
616 v5param->memory_address_range = param2;
617 v5param->flags = SETWA_FLAGS_MEM;
618 break;
619 case ACPI_EINJ_PCIX_CORRECTABLE:
620 case ACPI_EINJ_PCIX_UNCORRECTABLE:
621 case ACPI_EINJ_PCIX_FATAL:
622 v5param->pcie_sbdf = param1;
623 v5param->flags = SETWA_FLAGS_PCIE_SBDF;
624 break;
625 }
626 }
627 memcpy_toio(einj_param, v5param, param_size);
628 kfree(v5param);
629 } else {
630 rc = apei_exec_run(&ctx, ACPI_EINJ_SET_ERROR_TYPE);
631 if (rc)
632 return rc;
633 if (einj_param) {
634 struct einj_parameter v4param;
635
636 memcpy_fromio(&v4param, einj_param, sizeof(v4param));
637 v4param.param1 = param1;
638 v4param.param2 = param2;
639 memcpy_toio(einj_param, &v4param, sizeof(v4param));
640 }
641 }
642 rc = apei_exec_run(&ctx, ACPI_EINJ_EXECUTE_OPERATION);
643 if (rc)
644 return rc;
645 for (;;) {
646 rc = apei_exec_run(&ctx, ACPI_EINJ_CHECK_BUSY_STATUS);
647 if (rc)
648 return rc;
649 val = apei_exec_ctx_get_output(&ctx);
650 if (!(val & EINJ_OP_BUSY))
651 break;
652 if (einj_timedout(&timeout))
653 return -EIO;
654 }
655 rc = apei_exec_run(&ctx, ACPI_EINJ_GET_COMMAND_STATUS);
656 if (rc)
657 return rc;
658 val = apei_exec_ctx_get_output(&ctx);
659 if (val == EINJ_STATUS_FAIL)
660 return -EBUSY;
661 else if (val == EINJ_STATUS_INVAL)
662 return -EINVAL;
663
664 /*
665 * The error is injected into the platform successfully, then it needs
666 * to trigger the error.
667 */
668 rc = apei_exec_run(&ctx, ACPI_EINJ_GET_TRIGGER_TABLE);
669 if (rc)
670 return rc;
671 trigger_paddr = apei_exec_ctx_get_output(&ctx);
672 if (notrigger == 0) {
673 rc = __einj_error_trigger(trigger_paddr, type, flags, param1, param2);
674 if (rc)
675 return rc;
676 }
677 rc = apei_exec_run_optional(&ctx, ACPI_EINJ_END_OPERATION);
678
679 return rc;
680 }
681
682 /* Allow almost all types of address except MMIO. */
is_allowed_range(u64 base_addr,u64 size)683 static bool is_allowed_range(u64 base_addr, u64 size)
684 {
685 int i;
686 /*
687 * MMIO region is usually claimed with IORESOURCE_MEM + IORES_DESC_NONE.
688 * However, IORES_DESC_NONE is treated like a wildcard when we check if
689 * region intersects with known resource. So do an allow list check for
690 * IORES_DESCs that definitely or most likely not MMIO.
691 */
692 static const int non_mmio_desc[] = {
693 IORES_DESC_CRASH_KERNEL,
694 IORES_DESC_ACPI_TABLES,
695 IORES_DESC_ACPI_NV_STORAGE,
696 IORES_DESC_PERSISTENT_MEMORY,
697 IORES_DESC_PERSISTENT_MEMORY_LEGACY,
698 /* Treat IORES_DESC_DEVICE_PRIVATE_MEMORY as MMIO. */
699 IORES_DESC_RESERVED,
700 IORES_DESC_SOFT_RESERVED,
701 };
702
703 if (region_intersects(base_addr, size, IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE)
704 == REGION_INTERSECTS)
705 return true;
706
707 for (i = 0; i < ARRAY_SIZE(non_mmio_desc); ++i) {
708 if (region_intersects(base_addr, size, IORESOURCE_MEM, non_mmio_desc[i])
709 == REGION_INTERSECTS)
710 return true;
711 }
712
713 if (arch_is_platform_page(base_addr))
714 return true;
715
716 return false;
717 }
718
719 /* Inject the specified hardware error */
einj_error_inject(u32 type,u32 flags,u64 param1,u64 param2,u64 param3,u64 param4)720 int einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2, u64 param3,
721 u64 param4)
722 {
723 int rc;
724 u64 base_addr, size;
725
726 /* If user manually set "flags", make sure it is legal */
727 if (flags && (flags & ~(SETWA_FLAGS_APICID | SETWA_FLAGS_MEM |
728 SETWA_FLAGS_PCIE_SBDF | SETWA_FLAGS_EINJV2)))
729 return -EINVAL;
730
731 /*
732 * Injections targeting a CXL 1.0/1.1 port have to be injected
733 * via the einj_cxl_rch_error_inject() path as that does the proper
734 * validation of the given RCRB base (MMIO) address.
735 */
736 if (einj_is_cxl_error_type(type) && (flags & SETWA_FLAGS_MEM))
737 return -EINVAL;
738
739 /* check if type is a valid EINJv2 error type */
740 if (is_v2) {
741 if (!(type & available_error_type_v2))
742 return -EINVAL;
743 }
744
745 /* ensure param1/param2 existed */
746 if (!(param_extension || acpi5))
747 goto inject;
748
749 /*
750 * We need extra sanity checks for memory errors.
751 * Other types leap directly to injection.
752 */
753 if (!is_memory_injection(type, flags))
754 goto inject;
755
756 /*
757 * Disallow crazy address masks that give BIOS leeway to pick
758 * injection address almost anywhere. Insist on page or
759 * better granularity and that target address is normal RAM or
760 * as long as is not MMIO.
761 */
762 base_addr = param1 & param2;
763 size = ~param2 + 1;
764
765 if ((param2 & PAGE_MASK) != PAGE_MASK)
766 return -EINVAL;
767
768 if (!is_allowed_range(base_addr, size))
769 return -EINVAL;
770
771 if (is_zero_pfn(base_addr >> PAGE_SHIFT))
772 return -EADDRINUSE;
773
774 inject:
775 mutex_lock(&einj_mutex);
776 rc = __einj_error_inject(type, flags, param1, param2, param3, param4);
777 mutex_unlock(&einj_mutex);
778
779 return rc;
780 }
781
einj_cxl_rch_error_inject(u32 type,u32 flags,u64 param1,u64 param2,u64 param3,u64 param4)782 int einj_cxl_rch_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
783 u64 param3, u64 param4)
784 {
785 int rc;
786
787 if (!(einj_is_cxl_error_type(type) && (flags & SETWA_FLAGS_MEM)))
788 return -EINVAL;
789
790 mutex_lock(&einj_mutex);
791 rc = __einj_error_inject(type, flags, param1, param2, param3, param4);
792 mutex_unlock(&einj_mutex);
793
794 return rc;
795 }
796
797 static u32 error_type;
798 static u32 error_flags;
799 static u64 error_param1;
800 static u64 error_param2;
801 static u64 error_param3;
802 static u64 error_param4;
803 static struct dentry *einj_debug_dir;
804 static char einj_buf[32];
805 static bool einj_v2_enabled;
806 static struct { u32 mask; const char *str; } const einj_error_type_string[] = {
807 { BIT(0), "Processor Correctable" },
808 { BIT(1), "Processor Uncorrectable non-fatal" },
809 { BIT(2), "Processor Uncorrectable fatal" },
810 { BIT(3), "Memory Correctable" },
811 { BIT(4), "Memory Uncorrectable non-fatal" },
812 { BIT(5), "Memory Uncorrectable fatal" },
813 { BIT(6), "PCI Express Correctable" },
814 { BIT(7), "PCI Express Uncorrectable non-fatal" },
815 { BIT(8), "PCI Express Uncorrectable fatal" },
816 { BIT(9), "Platform Correctable" },
817 { BIT(10), "Platform Uncorrectable non-fatal" },
818 { BIT(11), "Platform Uncorrectable fatal"},
819 { BIT(31), "Vendor Defined Error Types" },
820 };
821
822 static struct { u32 mask; const char *str; } const einjv2_error_type_string[] = {
823 { BIT(0), "EINJV2 Processor Error" },
824 { BIT(1), "EINJV2 Memory Error" },
825 { BIT(2), "EINJV2 PCI Express Error" },
826 };
827
available_error_type_show(struct seq_file * m,void * v)828 static int available_error_type_show(struct seq_file *m, void *v)
829 {
830
831 for (int pos = 0; pos < ARRAY_SIZE(einj_error_type_string); pos++)
832 if (available_error_type & einj_error_type_string[pos].mask)
833 seq_printf(m, "0x%08x\t%s\n", einj_error_type_string[pos].mask,
834 einj_error_type_string[pos].str);
835 if ((available_error_type & ACPI65_EINJV2_SUPP) && einj_v2_enabled) {
836 for (int pos = 0; pos < ARRAY_SIZE(einjv2_error_type_string); pos++) {
837 if (available_error_type_v2 & einjv2_error_type_string[pos].mask)
838 seq_printf(m, "V2_0x%08x\t%s\n", einjv2_error_type_string[pos].mask,
839 einjv2_error_type_string[pos].str);
840 }
841 }
842 return 0;
843 }
844
845 DEFINE_SHOW_ATTRIBUTE(available_error_type);
846
error_type_get(struct file * file,char __user * buf,size_t count,loff_t * ppos)847 static ssize_t error_type_get(struct file *file, char __user *buf,
848 size_t count, loff_t *ppos)
849 {
850 return simple_read_from_buffer(buf, count, ppos, einj_buf, strlen(einj_buf));
851 }
852
einj_is_cxl_error_type(u64 type)853 bool einj_is_cxl_error_type(u64 type)
854 {
855 return (type & CXL_ERROR_MASK) && (!(type & ACPI5_VENDOR_BIT));
856 }
857
einj_validate_error_type(u64 type)858 int einj_validate_error_type(u64 type)
859 {
860 u32 tval, vendor;
861
862 /* Only low 32 bits for error type are valid */
863 if (type & GENMASK_ULL(63, 32))
864 return -EINVAL;
865
866 /*
867 * Vendor defined types have 0x80000000 bit set, and
868 * are not enumerated by ACPI_EINJ_GET_ERROR_TYPE
869 */
870 vendor = type & ACPI5_VENDOR_BIT;
871 tval = type & GENMASK(30, 0);
872
873 /* Only one error type can be specified */
874 if (tval & (tval - 1))
875 return -EINVAL;
876 if (!vendor)
877 if (!(type & (available_error_type | available_error_type_v2)))
878 return -EINVAL;
879
880 return 0;
881 }
882
error_type_set(struct file * file,const char __user * buf,size_t count,loff_t * ppos)883 static ssize_t error_type_set(struct file *file, const char __user *buf,
884 size_t count, loff_t *ppos)
885 {
886 int rc;
887 u64 val;
888
889 /* Leave the last character for the NUL terminator */
890 if (count > sizeof(einj_buf) - 1)
891 return -EINVAL;
892
893 memset(einj_buf, 0, sizeof(einj_buf));
894 if (copy_from_user(einj_buf, buf, count))
895 return -EFAULT;
896
897 if (strncmp(einj_buf, "V2_", 3) == 0) {
898 if (!sscanf(einj_buf, "V2_%llx", &val))
899 return -EINVAL;
900 is_v2 = true;
901 } else {
902 if (!sscanf(einj_buf, "%llx", &val))
903 return -EINVAL;
904 is_v2 = false;
905 }
906
907 rc = einj_validate_error_type(val);
908 if (rc)
909 return rc;
910
911 error_type = val;
912
913 return count;
914 }
915
916 static const struct file_operations error_type_fops = {
917 .read = error_type_get,
918 .write = error_type_set,
919 };
920
error_inject_set(void * data,u64 val)921 static int error_inject_set(void *data, u64 val)
922 {
923 if (!error_type)
924 return -EINVAL;
925
926 if (is_v2)
927 error_flags |= SETWA_FLAGS_EINJV2;
928 else
929 error_flags &= ~SETWA_FLAGS_EINJV2;
930
931 return einj_error_inject(error_type, error_flags, error_param1, error_param2,
932 error_param3, error_param4);
933 }
934
935 DEFINE_DEBUGFS_ATTRIBUTE(error_inject_fops, NULL, error_inject_set, "%llu\n");
936
einj_check_table(struct acpi_table_einj * einj_tab)937 static int einj_check_table(struct acpi_table_einj *einj_tab)
938 {
939 if ((einj_tab->header_length !=
940 (sizeof(struct acpi_table_einj) - sizeof(einj_tab->header)))
941 && (einj_tab->header_length != sizeof(struct acpi_table_einj)))
942 return -EINVAL;
943 if (einj_tab->header.length < sizeof(struct acpi_table_einj))
944 return -EINVAL;
945 if (einj_tab->entries !=
946 (einj_tab->header.length - sizeof(struct acpi_table_einj)) /
947 sizeof(struct acpi_einj_entry))
948 return -EINVAL;
949
950 return 0;
951 }
952
u128_read(struct file * f,char __user * buf,size_t count,loff_t * off)953 static ssize_t u128_read(struct file *f, char __user *buf, size_t count, loff_t *off)
954 {
955 char output[2 * COMPONENT_LEN + 1];
956 u8 *data = f->f_inode->i_private;
957 int i;
958
959 if (*off >= sizeof(output))
960 return 0;
961
962 for (i = 0; i < COMPONENT_LEN; i++)
963 sprintf(output + 2 * i, "%.02x", data[COMPONENT_LEN - i - 1]);
964 output[2 * COMPONENT_LEN] = '\n';
965
966 return simple_read_from_buffer(buf, count, off, output, sizeof(output));
967 }
968
u128_write(struct file * f,const char __user * buf,size_t count,loff_t * off)969 static ssize_t u128_write(struct file *f, const char __user *buf, size_t count, loff_t *off)
970 {
971 char input[2 + 2 * COMPONENT_LEN + 2];
972 u8 *save = f->f_inode->i_private;
973 u8 tmp[COMPONENT_LEN];
974 char byte[3] = {};
975 char *s, *e;
976 ssize_t c;
977 long val;
978 int i;
979
980 /* Require that user supply whole input line in one write(2) syscall */
981 if (*off)
982 return -EINVAL;
983
984 c = simple_write_to_buffer(input, sizeof(input), off, buf, count);
985 if (c < 0)
986 return c;
987
988 if (c < 1 || input[c - 1] != '\n')
989 return -EINVAL;
990
991 /* Empty line means invalidate this entry */
992 if (c == 1) {
993 memset(save, 0xff, COMPONENT_LEN);
994 return c;
995 }
996
997 if (input[0] == '0' && (input[1] == 'x' || input[1] == 'X'))
998 s = input + 2;
999 else
1000 s = input;
1001 e = input + c - 1;
1002
1003 for (i = 0; i < COMPONENT_LEN; i++) {
1004 byte[1] = *--e;
1005 byte[0] = e > s ? *--e : '0';
1006 if (kstrtol(byte, 16, &val))
1007 return -EINVAL;
1008 tmp[i] = val;
1009 if (e <= s)
1010 break;
1011 }
1012 while (++i < COMPONENT_LEN)
1013 tmp[i] = 0;
1014
1015 memcpy(save, tmp, COMPONENT_LEN);
1016
1017 return c;
1018 }
1019
1020 static const struct file_operations u128_fops = {
1021 .read = u128_read,
1022 .write = u128_write,
1023 };
1024
setup_einjv2_component_files(void)1025 static bool setup_einjv2_component_files(void)
1026 {
1027 char name[32];
1028
1029 syndrome_data = kzalloc_objs(syndrome_data[0], max_nr_components);
1030 if (!syndrome_data)
1031 return false;
1032
1033 for (int i = 0; i < max_nr_components; i++) {
1034 sprintf(name, "component_id%d", i);
1035 debugfs_create_file(name, 0600, einj_debug_dir,
1036 &syndrome_data[i].comp_id, &u128_fops);
1037 sprintf(name, "component_syndrome%d", i);
1038 debugfs_create_file(name, 0600, einj_debug_dir,
1039 &syndrome_data[i].comp_synd, &u128_fops);
1040 }
1041
1042 return true;
1043 }
1044
einj_probe(struct faux_device * fdev)1045 static int __init einj_probe(struct faux_device *fdev)
1046 {
1047 int rc;
1048 acpi_status status;
1049 struct apei_exec_context ctx;
1050
1051 status = acpi_get_table(ACPI_SIG_EINJ, 0,
1052 (struct acpi_table_header **)&einj_tab);
1053 if (status == AE_NOT_FOUND) {
1054 pr_debug("EINJ table not found.\n");
1055 return -ENODEV;
1056 } else if (ACPI_FAILURE(status)) {
1057 pr_err("Failed to get EINJ table: %s\n",
1058 acpi_format_exception(status));
1059 return -EINVAL;
1060 }
1061
1062 rc = einj_check_table(einj_tab);
1063 if (rc) {
1064 pr_warn(FW_BUG "Invalid EINJ table.\n");
1065 goto err_put_table;
1066 }
1067
1068 rc = einj_get_available_error_types(&available_error_type, &available_error_type_v2);
1069 if (rc)
1070 goto err_put_table;
1071
1072 rc = -ENOMEM;
1073 einj_debug_dir = debugfs_create_dir("einj", apei_get_debugfs_dir());
1074
1075 debugfs_create_file("available_error_type", S_IRUSR, einj_debug_dir,
1076 NULL, &available_error_type_fops);
1077 debugfs_create_file_unsafe("error_type", 0600, einj_debug_dir,
1078 NULL, &error_type_fops);
1079 debugfs_create_file_unsafe("error_inject", 0200, einj_debug_dir,
1080 NULL, &error_inject_fops);
1081
1082 apei_resources_init(&einj_resources);
1083 einj_exec_ctx_init(&ctx);
1084 rc = apei_exec_collect_resources(&ctx, &einj_resources);
1085 if (rc) {
1086 pr_err("Error collecting EINJ resources.\n");
1087 goto err_fini;
1088 }
1089
1090 rc = apei_resources_request(&einj_resources, "APEI EINJ");
1091 if (rc) {
1092 pr_err("Error requesting memory/port resources.\n");
1093 goto err_fini;
1094 }
1095
1096 rc = apei_exec_pre_map_gars(&ctx);
1097 if (rc) {
1098 pr_err("Error pre-mapping GARs.\n");
1099 goto err_release;
1100 }
1101
1102 einj_param = einj_get_parameter_address();
1103 if ((param_extension || acpi5) && einj_param) {
1104 debugfs_create_x32("flags", S_IRUSR | S_IWUSR, einj_debug_dir,
1105 &error_flags);
1106 debugfs_create_x64("param1", S_IRUSR | S_IWUSR, einj_debug_dir,
1107 &error_param1);
1108 debugfs_create_x64("param2", S_IRUSR | S_IWUSR, einj_debug_dir,
1109 &error_param2);
1110 debugfs_create_x64("param3", S_IRUSR | S_IWUSR, einj_debug_dir,
1111 &error_param3);
1112 debugfs_create_x64("param4", S_IRUSR | S_IWUSR, einj_debug_dir,
1113 &error_param4);
1114 debugfs_create_x32("notrigger", S_IRUSR | S_IWUSR,
1115 einj_debug_dir, ¬rigger);
1116 if (available_error_type & ACPI65_EINJV2_SUPP)
1117 einj_v2_enabled = setup_einjv2_component_files();
1118 }
1119
1120 if (vendor_dev[0]) {
1121 vendor_blob.data = vendor_dev;
1122 vendor_blob.size = strlen(vendor_dev);
1123 debugfs_create_blob("vendor", S_IRUSR, einj_debug_dir,
1124 &vendor_blob);
1125 debugfs_create_x32("vendor_flags", S_IRUSR | S_IWUSR,
1126 einj_debug_dir, &vendor_flags);
1127 }
1128
1129 if (vendor_errors.size)
1130 debugfs_create_blob("oem_error", 0600, einj_debug_dir,
1131 &vendor_errors);
1132
1133 pr_info("Error INJection is initialized.\n");
1134
1135 return 0;
1136
1137 err_release:
1138 apei_resources_release(&einj_resources);
1139 err_fini:
1140 apei_resources_fini(&einj_resources);
1141 debugfs_remove_recursive(einj_debug_dir);
1142 err_put_table:
1143 acpi_put_table((struct acpi_table_header *)einj_tab);
1144
1145 return rc;
1146 }
1147
einj_remove(struct faux_device * fdev)1148 static void einj_remove(struct faux_device *fdev)
1149 {
1150 struct apei_exec_context ctx;
1151
1152 if (einj_param) {
1153 acpi_size size;
1154
1155 if (v66param_size)
1156 size = v66param_size;
1157 else if (acpi5)
1158 size = v5param_size;
1159 else
1160 size = sizeof(struct einj_parameter);
1161
1162 acpi_os_unmap_iomem(einj_param, size);
1163 if (vendor_errors.size)
1164 acpi_os_unmap_memory(vendor_errors.data, vendor_errors.size);
1165 }
1166 einj_exec_ctx_init(&ctx);
1167 apei_exec_post_unmap_gars(&ctx);
1168 apei_resources_release(&einj_resources);
1169 apei_resources_fini(&einj_resources);
1170 debugfs_remove_recursive(einj_debug_dir);
1171 kfree(syndrome_data);
1172 acpi_put_table((struct acpi_table_header *)einj_tab);
1173 }
1174
1175 static struct faux_device *einj_dev;
1176 static struct faux_device_ops einj_device_ops = {
1177 .probe = einj_probe,
1178 .remove = einj_remove,
1179 };
1180
einj_init(void)1181 static int __init einj_init(void)
1182 {
1183 if (acpi_disabled) {
1184 pr_debug("ACPI disabled.\n");
1185 return -ENODEV;
1186 }
1187
1188 einj_dev = faux_device_create("acpi-einj", NULL, &einj_device_ops);
1189
1190 if (einj_dev)
1191 einj_initialized = true;
1192
1193 return 0;
1194 }
1195
einj_exit(void)1196 static void __exit einj_exit(void)
1197 {
1198 faux_device_destroy(einj_dev);
1199 }
1200
1201 module_init(einj_init);
1202 module_exit(einj_exit);
1203
1204 MODULE_AUTHOR("Huang Ying");
1205 MODULE_DESCRIPTION("APEI Error INJection support");
1206 MODULE_LICENSE("GPL");
1207