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