acpi_extlog.c 8.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324
  1. /*
  2. * Extended Error Log driver
  3. *
  4. * Copyright (C) 2013 Intel Corp.
  5. * Author: Chen, Gong <gong.chen@intel.com>
  6. *
  7. * This file is licensed under GPLv2.
  8. */
  9. #include <linux/module.h>
  10. #include <linux/acpi.h>
  11. #include <linux/cper.h>
  12. #include <linux/ratelimit.h>
  13. #include <linux/edac.h>
  14. #include <asm/cpu.h>
  15. #include <asm/mce.h>
  16. #include "apei/apei-internal.h"
  17. #include <ras/ras_event.h>
  18. #define EXT_ELOG_ENTRY_MASK GENMASK_ULL(51, 0) /* elog entry address mask */
  19. #define EXTLOG_DSM_REV 0x0
  20. #define EXTLOG_FN_ADDR 0x1
  21. #define FLAG_OS_OPTIN BIT(0)
  22. #define ELOG_ENTRY_VALID (1ULL<<63)
  23. #define ELOG_ENTRY_LEN 0x1000
  24. #define EMCA_BUG \
  25. "Can not request iomem region <0x%016llx-0x%016llx> - eMCA disabled\n"
  26. struct extlog_l1_head {
  27. u32 ver; /* Header Version */
  28. u32 hdr_len; /* Header Length */
  29. u64 total_len; /* entire L1 Directory length including this header */
  30. u64 elog_base; /* MCA Error Log Directory base address */
  31. u64 elog_len; /* MCA Error Log Directory length */
  32. u32 flags; /* bit 0 - OS/VMM Opt-in */
  33. u8 rev0[12];
  34. u32 entries; /* Valid L1 Directory entries per logical processor */
  35. u8 rev1[12];
  36. };
  37. static int old_edac_report_status;
  38. static u8 extlog_dsm_uuid[] __initdata = "663E35AF-CC10-41A4-88EA-5470AF055295";
  39. /* L1 table related physical address */
  40. static u64 elog_base;
  41. static size_t elog_size;
  42. static u64 l1_dirbase;
  43. static size_t l1_size;
  44. /* L1 table related virtual address */
  45. static void __iomem *extlog_l1_addr;
  46. static void __iomem *elog_addr;
  47. static void *elog_buf;
  48. static u64 *l1_entry_base;
  49. static u32 l1_percpu_entry;
  50. #define ELOG_IDX(cpu, bank) \
  51. (cpu_physical_id(cpu) * l1_percpu_entry + (bank))
  52. #define ELOG_ENTRY_DATA(idx) \
  53. (*(l1_entry_base + (idx)))
  54. #define ELOG_ENTRY_ADDR(phyaddr) \
  55. (phyaddr - elog_base + (u8 *)elog_addr)
  56. static struct acpi_generic_status *extlog_elog_entry_check(int cpu, int bank)
  57. {
  58. int idx;
  59. u64 data;
  60. struct acpi_generic_status *estatus;
  61. WARN_ON(cpu < 0);
  62. idx = ELOG_IDX(cpu, bank);
  63. data = ELOG_ENTRY_DATA(idx);
  64. if ((data & ELOG_ENTRY_VALID) == 0)
  65. return NULL;
  66. data &= EXT_ELOG_ENTRY_MASK;
  67. estatus = (struct acpi_generic_status *)ELOG_ENTRY_ADDR(data);
  68. /* if no valid data in elog entry, just return */
  69. if (estatus->block_status == 0)
  70. return NULL;
  71. return estatus;
  72. }
  73. static void __print_extlog_rcd(const char *pfx,
  74. struct acpi_generic_status *estatus, int cpu)
  75. {
  76. static atomic_t seqno;
  77. unsigned int curr_seqno;
  78. char pfx_seq[64];
  79. if (!pfx) {
  80. if (estatus->error_severity <= CPER_SEV_CORRECTED)
  81. pfx = KERN_INFO;
  82. else
  83. pfx = KERN_ERR;
  84. }
  85. curr_seqno = atomic_inc_return(&seqno);
  86. snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}", pfx, curr_seqno);
  87. printk("%s""Hardware error detected on CPU%d\n", pfx_seq, cpu);
  88. cper_estatus_print(pfx_seq, estatus);
  89. }
  90. static int print_extlog_rcd(const char *pfx,
  91. struct acpi_generic_status *estatus, int cpu)
  92. {
  93. /* Not more than 2 messages every 5 seconds */
  94. static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
  95. static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
  96. struct ratelimit_state *ratelimit;
  97. if (estatus->error_severity == CPER_SEV_CORRECTED ||
  98. (estatus->error_severity == CPER_SEV_INFORMATIONAL))
  99. ratelimit = &ratelimit_corrected;
  100. else
  101. ratelimit = &ratelimit_uncorrected;
  102. if (__ratelimit(ratelimit)) {
  103. __print_extlog_rcd(pfx, estatus, cpu);
  104. return 0;
  105. }
  106. return 1;
  107. }
  108. static int extlog_print(struct notifier_block *nb, unsigned long val,
  109. void *data)
  110. {
  111. struct mce *mce = (struct mce *)data;
  112. int bank = mce->bank;
  113. int cpu = mce->extcpu;
  114. struct acpi_generic_status *estatus, *tmp;
  115. struct acpi_generic_data *gdata;
  116. const uuid_le *fru_id = &NULL_UUID_LE;
  117. char *fru_text = "";
  118. uuid_le *sec_type;
  119. static u32 err_seq;
  120. estatus = extlog_elog_entry_check(cpu, bank);
  121. if (estatus == NULL)
  122. return NOTIFY_DONE;
  123. memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN);
  124. /* clear record status to enable BIOS to update it again */
  125. estatus->block_status = 0;
  126. tmp = (struct acpi_generic_status *)elog_buf;
  127. print_extlog_rcd(NULL, tmp, cpu);
  128. /* log event via trace */
  129. err_seq++;
  130. gdata = (struct acpi_generic_data *)(tmp + 1);
  131. if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
  132. fru_id = (uuid_le *)gdata->fru_id;
  133. if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
  134. fru_text = gdata->fru_text;
  135. sec_type = (uuid_le *)gdata->section_type;
  136. if (!uuid_le_cmp(*sec_type, CPER_SEC_PLATFORM_MEM)) {
  137. struct cper_sec_mem_err *mem = (void *)(gdata + 1);
  138. if (gdata->error_data_length >= sizeof(*mem))
  139. trace_extlog_mem_event(mem, err_seq, fru_id, fru_text,
  140. (u8)gdata->error_severity);
  141. }
  142. return NOTIFY_STOP;
  143. }
  144. static bool __init extlog_get_l1addr(void)
  145. {
  146. u8 uuid[16];
  147. acpi_handle handle;
  148. union acpi_object *obj;
  149. acpi_str_to_uuid(extlog_dsm_uuid, uuid);
  150. if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle)))
  151. return false;
  152. if (!acpi_check_dsm(handle, uuid, EXTLOG_DSM_REV, 1 << EXTLOG_FN_ADDR))
  153. return false;
  154. obj = acpi_evaluate_dsm_typed(handle, uuid, EXTLOG_DSM_REV,
  155. EXTLOG_FN_ADDR, NULL, ACPI_TYPE_INTEGER);
  156. if (!obj) {
  157. return false;
  158. } else {
  159. l1_dirbase = obj->integer.value;
  160. ACPI_FREE(obj);
  161. }
  162. /* Spec says L1 directory must be 4K aligned, bail out if it isn't */
  163. if (l1_dirbase & ((1 << 12) - 1)) {
  164. pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n",
  165. l1_dirbase);
  166. return false;
  167. }
  168. return true;
  169. }
  170. static struct notifier_block extlog_mce_dec = {
  171. .notifier_call = extlog_print,
  172. };
  173. static int __init extlog_init(void)
  174. {
  175. struct extlog_l1_head *l1_head;
  176. void __iomem *extlog_l1_hdr;
  177. size_t l1_hdr_size;
  178. struct resource *r;
  179. u64 cap;
  180. int rc;
  181. if (get_edac_report_status() == EDAC_REPORTING_FORCE) {
  182. pr_warn("Not loading eMCA, error reporting force-enabled through EDAC.\n");
  183. return -EPERM;
  184. }
  185. rc = -ENODEV;
  186. rdmsrl(MSR_IA32_MCG_CAP, cap);
  187. if (!(cap & MCG_ELOG_P))
  188. return rc;
  189. if (!extlog_get_l1addr())
  190. return rc;
  191. rc = -EINVAL;
  192. /* get L1 header to fetch necessary information */
  193. l1_hdr_size = sizeof(struct extlog_l1_head);
  194. r = request_mem_region(l1_dirbase, l1_hdr_size, "L1 DIR HDR");
  195. if (!r) {
  196. pr_warn(FW_BUG EMCA_BUG,
  197. (unsigned long long)l1_dirbase,
  198. (unsigned long long)l1_dirbase + l1_hdr_size);
  199. goto err;
  200. }
  201. extlog_l1_hdr = acpi_os_map_iomem(l1_dirbase, l1_hdr_size);
  202. l1_head = (struct extlog_l1_head *)extlog_l1_hdr;
  203. l1_size = l1_head->total_len;
  204. l1_percpu_entry = l1_head->entries;
  205. elog_base = l1_head->elog_base;
  206. elog_size = l1_head->elog_len;
  207. acpi_os_unmap_iomem(extlog_l1_hdr, l1_hdr_size);
  208. release_mem_region(l1_dirbase, l1_hdr_size);
  209. /* remap L1 header again based on completed information */
  210. r = request_mem_region(l1_dirbase, l1_size, "L1 Table");
  211. if (!r) {
  212. pr_warn(FW_BUG EMCA_BUG,
  213. (unsigned long long)l1_dirbase,
  214. (unsigned long long)l1_dirbase + l1_size);
  215. goto err;
  216. }
  217. extlog_l1_addr = acpi_os_map_iomem(l1_dirbase, l1_size);
  218. l1_entry_base = (u64 *)((u8 *)extlog_l1_addr + l1_hdr_size);
  219. /* remap elog table */
  220. r = request_mem_region(elog_base, elog_size, "Elog Table");
  221. if (!r) {
  222. pr_warn(FW_BUG EMCA_BUG,
  223. (unsigned long long)elog_base,
  224. (unsigned long long)elog_base + elog_size);
  225. goto err_release_l1_dir;
  226. }
  227. elog_addr = acpi_os_map_iomem(elog_base, elog_size);
  228. rc = -ENOMEM;
  229. /* allocate buffer to save elog record */
  230. elog_buf = kmalloc(ELOG_ENTRY_LEN, GFP_KERNEL);
  231. if (elog_buf == NULL)
  232. goto err_release_elog;
  233. /*
  234. * eMCA event report method has higher priority than EDAC method,
  235. * unless EDAC event report method is mandatory.
  236. */
  237. old_edac_report_status = get_edac_report_status();
  238. set_edac_report_status(EDAC_REPORTING_DISABLED);
  239. mce_register_decode_chain(&extlog_mce_dec);
  240. /* enable OS to be involved to take over management from BIOS */
  241. ((struct extlog_l1_head *)extlog_l1_addr)->flags |= FLAG_OS_OPTIN;
  242. return 0;
  243. err_release_elog:
  244. if (elog_addr)
  245. acpi_os_unmap_iomem(elog_addr, elog_size);
  246. release_mem_region(elog_base, elog_size);
  247. err_release_l1_dir:
  248. if (extlog_l1_addr)
  249. acpi_os_unmap_iomem(extlog_l1_addr, l1_size);
  250. release_mem_region(l1_dirbase, l1_size);
  251. err:
  252. pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n");
  253. return rc;
  254. }
  255. static void __exit extlog_exit(void)
  256. {
  257. set_edac_report_status(old_edac_report_status);
  258. mce_unregister_decode_chain(&extlog_mce_dec);
  259. ((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN;
  260. if (extlog_l1_addr)
  261. acpi_os_unmap_iomem(extlog_l1_addr, l1_size);
  262. if (elog_addr)
  263. acpi_os_unmap_iomem(elog_addr, elog_size);
  264. release_mem_region(elog_base, elog_size);
  265. release_mem_region(l1_dirbase, l1_size);
  266. kfree(elog_buf);
  267. }
  268. module_init(extlog_init);
  269. module_exit(extlog_exit);
  270. MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>");
  271. MODULE_DESCRIPTION("Extended MCA Error Log Driver");
  272. MODULE_LICENSE("GPL");