Commit | Line | Data |
---|---|---|
4b3db708 CG |
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 | ||
10 | #include <linux/module.h> | |
11 | #include <linux/acpi.h> | |
4b3db708 CG |
12 | #include <linux/cper.h> |
13 | #include <linux/ratelimit.h> | |
42139eb3 | 14 | #include <linux/edac.h> |
4b3db708 CG |
15 | #include <asm/cpu.h> |
16 | #include <asm/mce.h> | |
17 | ||
18 | #include "apei/apei-internal.h" | |
2dfb7d51 | 19 | #include <ras/ras_event.h> |
4b3db708 CG |
20 | |
21 | #define EXT_ELOG_ENTRY_MASK GENMASK_ULL(51, 0) /* elog entry address mask */ | |
22 | ||
23 | #define EXTLOG_DSM_REV 0x0 | |
4b3db708 CG |
24 | #define EXTLOG_FN_ADDR 0x1 |
25 | ||
26 | #define FLAG_OS_OPTIN BIT(0) | |
4b3db708 CG |
27 | #define ELOG_ENTRY_VALID (1ULL<<63) |
28 | #define ELOG_ENTRY_LEN 0x1000 | |
29 | ||
30 | #define EMCA_BUG \ | |
31 | "Can not request iomem region <0x%016llx-0x%016llx> - eMCA disabled\n" | |
32 | ||
33 | struct extlog_l1_head { | |
34 | u32 ver; /* Header Version */ | |
35 | u32 hdr_len; /* Header Length */ | |
36 | u64 total_len; /* entire L1 Directory length including this header */ | |
37 | u64 elog_base; /* MCA Error Log Directory base address */ | |
38 | u64 elog_len; /* MCA Error Log Directory length */ | |
39 | u32 flags; /* bit 0 - OS/VMM Opt-in */ | |
40 | u8 rev0[12]; | |
41 | u32 entries; /* Valid L1 Directory entries per logical processor */ | |
42 | u8 rev1[12]; | |
43 | }; | |
44 | ||
42139eb3 CG |
45 | static int old_edac_report_status; |
46 | ||
7ede9f8a | 47 | static u8 extlog_dsm_uuid[] __initdata = "663E35AF-CC10-41A4-88EA-5470AF055295"; |
4b3db708 CG |
48 | |
49 | /* L1 table related physical address */ | |
50 | static u64 elog_base; | |
51 | static size_t elog_size; | |
52 | static u64 l1_dirbase; | |
53 | static size_t l1_size; | |
54 | ||
55 | /* L1 table related virtual address */ | |
56 | static void __iomem *extlog_l1_addr; | |
57 | static void __iomem *elog_addr; | |
58 | ||
59 | static void *elog_buf; | |
60 | ||
61 | static u64 *l1_entry_base; | |
62 | static u32 l1_percpu_entry; | |
63 | ||
64 | #define ELOG_IDX(cpu, bank) \ | |
65 | (cpu_physical_id(cpu) * l1_percpu_entry + (bank)) | |
66 | ||
67 | #define ELOG_ENTRY_DATA(idx) \ | |
68 | (*(l1_entry_base + (idx))) | |
69 | ||
70 | #define ELOG_ENTRY_ADDR(phyaddr) \ | |
71 | (phyaddr - elog_base + (u8 *)elog_addr) | |
72 | ||
73 | static struct acpi_generic_status *extlog_elog_entry_check(int cpu, int bank) | |
74 | { | |
75 | int idx; | |
76 | u64 data; | |
77 | struct acpi_generic_status *estatus; | |
78 | ||
79 | WARN_ON(cpu < 0); | |
80 | idx = ELOG_IDX(cpu, bank); | |
81 | data = ELOG_ENTRY_DATA(idx); | |
82 | if ((data & ELOG_ENTRY_VALID) == 0) | |
83 | return NULL; | |
84 | ||
85 | data &= EXT_ELOG_ENTRY_MASK; | |
86 | estatus = (struct acpi_generic_status *)ELOG_ENTRY_ADDR(data); | |
87 | ||
88 | /* if no valid data in elog entry, just return */ | |
89 | if (estatus->block_status == 0) | |
90 | return NULL; | |
91 | ||
92 | return estatus; | |
93 | } | |
94 | ||
95 | static void __print_extlog_rcd(const char *pfx, | |
96 | struct acpi_generic_status *estatus, int cpu) | |
97 | { | |
98 | static atomic_t seqno; | |
99 | unsigned int curr_seqno; | |
100 | char pfx_seq[64]; | |
101 | ||
102 | if (!pfx) { | |
103 | if (estatus->error_severity <= CPER_SEV_CORRECTED) | |
104 | pfx = KERN_INFO; | |
105 | else | |
106 | pfx = KERN_ERR; | |
107 | } | |
108 | curr_seqno = atomic_inc_return(&seqno); | |
109 | snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}", pfx, curr_seqno); | |
110 | printk("%s""Hardware error detected on CPU%d\n", pfx_seq, cpu); | |
111 | cper_estatus_print(pfx_seq, estatus); | |
112 | } | |
113 | ||
114 | static int print_extlog_rcd(const char *pfx, | |
115 | struct acpi_generic_status *estatus, int cpu) | |
116 | { | |
117 | /* Not more than 2 messages every 5 seconds */ | |
118 | static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2); | |
119 | static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2); | |
120 | struct ratelimit_state *ratelimit; | |
121 | ||
122 | if (estatus->error_severity == CPER_SEV_CORRECTED || | |
123 | (estatus->error_severity == CPER_SEV_INFORMATIONAL)) | |
124 | ratelimit = &ratelimit_corrected; | |
125 | else | |
126 | ratelimit = &ratelimit_uncorrected; | |
127 | if (__ratelimit(ratelimit)) { | |
128 | __print_extlog_rcd(pfx, estatus, cpu); | |
129 | return 0; | |
130 | } | |
131 | ||
132 | return 1; | |
133 | } | |
134 | ||
135 | static int extlog_print(struct notifier_block *nb, unsigned long val, | |
136 | void *data) | |
137 | { | |
138 | struct mce *mce = (struct mce *)data; | |
139 | int bank = mce->bank; | |
140 | int cpu = mce->extcpu; | |
2dfb7d51 CG |
141 | struct acpi_generic_status *estatus, *tmp; |
142 | struct acpi_generic_data *gdata; | |
143 | const uuid_le *fru_id = &NULL_UUID_LE; | |
144 | char *fru_text = ""; | |
145 | uuid_le *sec_type; | |
146 | static u32 err_seq; | |
4b3db708 CG |
147 | |
148 | estatus = extlog_elog_entry_check(cpu, bank); | |
149 | if (estatus == NULL) | |
150 | return NOTIFY_DONE; | |
151 | ||
152 | memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN); | |
153 | /* clear record status to enable BIOS to update it again */ | |
154 | estatus->block_status = 0; | |
155 | ||
2dfb7d51 CG |
156 | tmp = (struct acpi_generic_status *)elog_buf; |
157 | print_extlog_rcd(NULL, tmp, cpu); | |
158 | ||
159 | /* log event via trace */ | |
160 | err_seq++; | |
161 | gdata = (struct acpi_generic_data *)(tmp + 1); | |
162 | if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID) | |
163 | fru_id = (uuid_le *)gdata->fru_id; | |
164 | if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT) | |
165 | fru_text = gdata->fru_text; | |
166 | sec_type = (uuid_le *)gdata->section_type; | |
167 | if (!uuid_le_cmp(*sec_type, CPER_SEC_PLATFORM_MEM)) { | |
168 | struct cper_sec_mem_err *mem = (void *)(gdata + 1); | |
169 | if (gdata->error_data_length >= sizeof(*mem)) | |
170 | trace_extlog_mem_event(mem, err_seq, fru_id, fru_text, | |
171 | (u8)gdata->error_severity); | |
172 | } | |
4b3db708 | 173 | |
42139eb3 | 174 | return NOTIFY_STOP; |
4b3db708 CG |
175 | } |
176 | ||
7ede9f8a | 177 | static bool __init extlog_get_l1addr(void) |
4b3db708 | 178 | { |
4b3db708 | 179 | u8 uuid[16]; |
7ede9f8a JL |
180 | acpi_handle handle; |
181 | union acpi_object *obj; | |
4b3db708 CG |
182 | |
183 | acpi_str_to_uuid(extlog_dsm_uuid, uuid); | |
4b3db708 CG |
184 | |
185 | if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle))) | |
186 | return false; | |
7ede9f8a | 187 | if (!acpi_check_dsm(handle, uuid, EXTLOG_DSM_REV, 1 << EXTLOG_FN_ADDR)) |
4b3db708 | 188 | return false; |
7ede9f8a JL |
189 | obj = acpi_evaluate_dsm_typed(handle, uuid, EXTLOG_DSM_REV, |
190 | EXTLOG_FN_ADDR, NULL, ACPI_TYPE_INTEGER); | |
191 | if (!obj) { | |
4b3db708 | 192 | return false; |
7ede9f8a JL |
193 | } else { |
194 | l1_dirbase = obj->integer.value; | |
195 | ACPI_FREE(obj); | |
196 | } | |
4b3db708 | 197 | |
4b3db708 CG |
198 | /* Spec says L1 directory must be 4K aligned, bail out if it isn't */ |
199 | if (l1_dirbase & ((1 << 12) - 1)) { | |
200 | pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n", | |
201 | l1_dirbase); | |
202 | return false; | |
203 | } | |
204 | ||
205 | return true; | |
206 | } | |
207 | static struct notifier_block extlog_mce_dec = { | |
208 | .notifier_call = extlog_print, | |
209 | }; | |
210 | ||
211 | static int __init extlog_init(void) | |
212 | { | |
213 | struct extlog_l1_head *l1_head; | |
214 | void __iomem *extlog_l1_hdr; | |
215 | size_t l1_hdr_size; | |
216 | struct resource *r; | |
217 | u64 cap; | |
218 | int rc; | |
219 | ||
42139eb3 CG |
220 | if (get_edac_report_status() == EDAC_REPORTING_FORCE) { |
221 | pr_warn("Not loading eMCA, error reporting force-enabled through EDAC.\n"); | |
222 | return -EPERM; | |
223 | } | |
4b3db708 | 224 | |
42139eb3 | 225 | rc = -ENODEV; |
4b3db708 CG |
226 | rdmsrl(MSR_IA32_MCG_CAP, cap); |
227 | if (!(cap & MCG_ELOG_P)) | |
228 | return rc; | |
229 | ||
230 | if (!extlog_get_l1addr()) | |
231 | return rc; | |
232 | ||
233 | rc = -EINVAL; | |
234 | /* get L1 header to fetch necessary information */ | |
235 | l1_hdr_size = sizeof(struct extlog_l1_head); | |
236 | r = request_mem_region(l1_dirbase, l1_hdr_size, "L1 DIR HDR"); | |
237 | if (!r) { | |
238 | pr_warn(FW_BUG EMCA_BUG, | |
239 | (unsigned long long)l1_dirbase, | |
240 | (unsigned long long)l1_dirbase + l1_hdr_size); | |
241 | goto err; | |
242 | } | |
243 | ||
a238317c | 244 | extlog_l1_hdr = acpi_os_map_iomem(l1_dirbase, l1_hdr_size); |
4b3db708 CG |
245 | l1_head = (struct extlog_l1_head *)extlog_l1_hdr; |
246 | l1_size = l1_head->total_len; | |
247 | l1_percpu_entry = l1_head->entries; | |
248 | elog_base = l1_head->elog_base; | |
249 | elog_size = l1_head->elog_len; | |
a238317c | 250 | acpi_os_unmap_iomem(extlog_l1_hdr, l1_hdr_size); |
4b3db708 CG |
251 | release_mem_region(l1_dirbase, l1_hdr_size); |
252 | ||
253 | /* remap L1 header again based on completed information */ | |
254 | r = request_mem_region(l1_dirbase, l1_size, "L1 Table"); | |
255 | if (!r) { | |
256 | pr_warn(FW_BUG EMCA_BUG, | |
257 | (unsigned long long)l1_dirbase, | |
258 | (unsigned long long)l1_dirbase + l1_size); | |
259 | goto err; | |
260 | } | |
a238317c | 261 | extlog_l1_addr = acpi_os_map_iomem(l1_dirbase, l1_size); |
4b3db708 CG |
262 | l1_entry_base = (u64 *)((u8 *)extlog_l1_addr + l1_hdr_size); |
263 | ||
264 | /* remap elog table */ | |
265 | r = request_mem_region(elog_base, elog_size, "Elog Table"); | |
266 | if (!r) { | |
267 | pr_warn(FW_BUG EMCA_BUG, | |
268 | (unsigned long long)elog_base, | |
269 | (unsigned long long)elog_base + elog_size); | |
270 | goto err_release_l1_dir; | |
271 | } | |
a238317c | 272 | elog_addr = acpi_os_map_iomem(elog_base, elog_size); |
4b3db708 CG |
273 | |
274 | rc = -ENOMEM; | |
275 | /* allocate buffer to save elog record */ | |
276 | elog_buf = kmalloc(ELOG_ENTRY_LEN, GFP_KERNEL); | |
277 | if (elog_buf == NULL) | |
278 | goto err_release_elog; | |
279 | ||
42139eb3 CG |
280 | /* |
281 | * eMCA event report method has higher priority than EDAC method, | |
282 | * unless EDAC event report method is mandatory. | |
283 | */ | |
284 | old_edac_report_status = get_edac_report_status(); | |
285 | set_edac_report_status(EDAC_REPORTING_DISABLED); | |
4b3db708 CG |
286 | mce_register_decode_chain(&extlog_mce_dec); |
287 | /* enable OS to be involved to take over management from BIOS */ | |
288 | ((struct extlog_l1_head *)extlog_l1_addr)->flags |= FLAG_OS_OPTIN; | |
289 | ||
290 | return 0; | |
291 | ||
292 | err_release_elog: | |
293 | if (elog_addr) | |
a238317c | 294 | acpi_os_unmap_iomem(elog_addr, elog_size); |
4b3db708 CG |
295 | release_mem_region(elog_base, elog_size); |
296 | err_release_l1_dir: | |
297 | if (extlog_l1_addr) | |
a238317c | 298 | acpi_os_unmap_iomem(extlog_l1_addr, l1_size); |
4b3db708 CG |
299 | release_mem_region(l1_dirbase, l1_size); |
300 | err: | |
301 | pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n"); | |
302 | return rc; | |
303 | } | |
304 | ||
305 | static void __exit extlog_exit(void) | |
306 | { | |
42139eb3 | 307 | set_edac_report_status(old_edac_report_status); |
4b3db708 CG |
308 | mce_unregister_decode_chain(&extlog_mce_dec); |
309 | ((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN; | |
310 | if (extlog_l1_addr) | |
a238317c | 311 | acpi_os_unmap_iomem(extlog_l1_addr, l1_size); |
4b3db708 | 312 | if (elog_addr) |
a238317c | 313 | acpi_os_unmap_iomem(elog_addr, elog_size); |
4b3db708 CG |
314 | release_mem_region(elog_base, elog_size); |
315 | release_mem_region(l1_dirbase, l1_size); | |
316 | kfree(elog_buf); | |
317 | } | |
318 | ||
319 | module_init(extlog_init); | |
320 | module_exit(extlog_exit); | |
321 | ||
322 | MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>"); | |
323 | MODULE_DESCRIPTION("Extended MCA Error Log Driver"); | |
324 | MODULE_LICENSE("GPL"); |