1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * Extended Error Log driver |
4 | * |
5 | * Copyright (C) 2013 Intel Corp. |
6 | * Author: Chen, Gong <gong.chen@intel.com> |
7 | */ |
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 <linux/ras.h> |
15 | #include <acpi/ghes.h> |
16 | #include <asm/cpu.h> |
17 | #include <asm/mce.h> |
18 | |
19 | #include "apei/apei-internal.h" |
20 | #include <ras/ras_event.h> |
21 | |
22 | #define EXT_ELOG_ENTRY_MASK GENMASK_ULL(51, 0) /* elog entry address mask */ |
23 | |
24 | #define EXTLOG_DSM_REV 0x0 |
25 | #define EXTLOG_FN_ADDR 0x1 |
26 | |
27 | #define FLAG_OS_OPTIN BIT(0) |
28 | #define ELOG_ENTRY_VALID (1ULL<<63) |
29 | #define ELOG_ENTRY_LEN 0x1000 |
30 | |
31 | #define EMCA_BUG \ |
32 | "Can not request iomem region <0x%016llx-0x%016llx> - eMCA disabled\n" |
33 | |
34 | struct extlog_l1_head { |
35 | u32 ver; /* Header Version */ |
36 | u32 hdr_len; /* Header Length */ |
37 | u64 total_len; /* entire L1 Directory length including this header */ |
38 | u64 elog_base; /* MCA Error Log Directory base address */ |
39 | u64 elog_len; /* MCA Error Log Directory length */ |
40 | u32 flags; /* bit 0 - OS/VMM Opt-in */ |
41 | u8 rev0[12]; |
42 | u32 entries; /* Valid L1 Directory entries per logical processor */ |
43 | u8 rev1[12]; |
44 | }; |
45 | |
46 | static u8 extlog_dsm_uuid[] __initdata = "663E35AF-CC10-41A4-88EA-5470AF055295" ; |
47 | |
48 | /* L1 table related physical address */ |
49 | static u64 elog_base; |
50 | static size_t elog_size; |
51 | static u64 l1_dirbase; |
52 | static size_t l1_size; |
53 | |
54 | /* L1 table related virtual address */ |
55 | static void __iomem *extlog_l1_addr; |
56 | static void __iomem *elog_addr; |
57 | |
58 | static void *elog_buf; |
59 | |
60 | static u64 *l1_entry_base; |
61 | static u32 l1_percpu_entry; |
62 | |
63 | #define ELOG_IDX(cpu, bank) \ |
64 | (cpu_physical_id(cpu) * l1_percpu_entry + (bank)) |
65 | |
66 | #define ELOG_ENTRY_DATA(idx) \ |
67 | (*(l1_entry_base + (idx))) |
68 | |
69 | #define ELOG_ENTRY_ADDR(phyaddr) \ |
70 | (phyaddr - elog_base + (u8 *)elog_addr) |
71 | |
72 | static struct acpi_hest_generic_status *extlog_elog_entry_check(int cpu, int bank) |
73 | { |
74 | int idx; |
75 | u64 data; |
76 | struct acpi_hest_generic_status *estatus; |
77 | |
78 | WARN_ON(cpu < 0); |
79 | idx = ELOG_IDX(cpu, bank); |
80 | data = ELOG_ENTRY_DATA(idx); |
81 | if ((data & ELOG_ENTRY_VALID) == 0) |
82 | return NULL; |
83 | |
84 | data &= EXT_ELOG_ENTRY_MASK; |
85 | estatus = (struct acpi_hest_generic_status *)ELOG_ENTRY_ADDR(data); |
86 | |
87 | /* if no valid data in elog entry, just return */ |
88 | if (estatus->block_status == 0) |
89 | return NULL; |
90 | |
91 | return estatus; |
92 | } |
93 | |
94 | static void __print_extlog_rcd(const char *pfx, |
95 | struct acpi_hest_generic_status *estatus, int cpu) |
96 | { |
97 | static atomic_t seqno; |
98 | unsigned int curr_seqno; |
99 | char pfx_seq[64]; |
100 | |
101 | if (!pfx) { |
102 | if (estatus->error_severity <= CPER_SEV_CORRECTED) |
103 | pfx = KERN_INFO; |
104 | else |
105 | pfx = KERN_ERR; |
106 | } |
107 | curr_seqno = atomic_inc_return(v: &seqno); |
108 | snprintf(buf: pfx_seq, size: sizeof(pfx_seq), fmt: "%s{%u}" , pfx, curr_seqno); |
109 | printk("%s" "Hardware error detected on CPU%d\n" , pfx_seq, cpu); |
110 | cper_estatus_print(pfx: pfx_seq, estatus); |
111 | } |
112 | |
113 | static int print_extlog_rcd(const char *pfx, |
114 | struct acpi_hest_generic_status *estatus, int cpu) |
115 | { |
116 | /* Not more than 2 messages every 5 seconds */ |
117 | static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2); |
118 | static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2); |
119 | struct ratelimit_state *ratelimit; |
120 | |
121 | if (estatus->error_severity == CPER_SEV_CORRECTED || |
122 | (estatus->error_severity == CPER_SEV_INFORMATIONAL)) |
123 | ratelimit = &ratelimit_corrected; |
124 | else |
125 | ratelimit = &ratelimit_uncorrected; |
126 | if (__ratelimit(ratelimit)) { |
127 | __print_extlog_rcd(pfx, estatus, cpu); |
128 | return 0; |
129 | } |
130 | |
131 | return 1; |
132 | } |
133 | |
134 | static int extlog_print(struct notifier_block *nb, unsigned long val, |
135 | void *data) |
136 | { |
137 | struct mce *mce = (struct mce *)data; |
138 | int bank = mce->bank; |
139 | int cpu = mce->extcpu; |
140 | struct acpi_hest_generic_status *estatus, *tmp; |
141 | struct acpi_hest_generic_data *gdata; |
142 | const guid_t *fru_id; |
143 | char *fru_text; |
144 | guid_t *sec_type; |
145 | static u32 err_seq; |
146 | |
147 | estatus = extlog_elog_entry_check(cpu, bank); |
148 | if (estatus == NULL || (mce->kflags & MCE_HANDLED_CEC)) |
149 | return NOTIFY_DONE; |
150 | |
151 | memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN); |
152 | /* clear record status to enable BIOS to update it again */ |
153 | estatus->block_status = 0; |
154 | |
155 | tmp = (struct acpi_hest_generic_status *)elog_buf; |
156 | |
157 | if (!ras_userspace_consumers()) { |
158 | print_extlog_rcd(NULL, estatus: tmp, cpu); |
159 | goto out; |
160 | } |
161 | |
162 | /* log event via trace */ |
163 | err_seq++; |
164 | apei_estatus_for_each_section(tmp, gdata) { |
165 | if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID) |
166 | fru_id = (guid_t *)gdata->fru_id; |
167 | else |
168 | fru_id = &guid_null; |
169 | if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT) |
170 | fru_text = gdata->fru_text; |
171 | else |
172 | fru_text = "" ; |
173 | sec_type = (guid_t *)gdata->section_type; |
174 | if (guid_equal(u1: sec_type, u2: &CPER_SEC_PLATFORM_MEM)) { |
175 | struct cper_sec_mem_err *mem = acpi_hest_get_payload(gdata); |
176 | |
177 | if (gdata->error_data_length >= sizeof(*mem)) |
178 | trace_extlog_mem_event(mem, err_seq, fru_id, fru_text, |
179 | sev: (u8)gdata->error_severity); |
180 | } |
181 | } |
182 | |
183 | out: |
184 | mce->kflags |= MCE_HANDLED_EXTLOG; |
185 | return NOTIFY_OK; |
186 | } |
187 | |
188 | static bool __init extlog_get_l1addr(void) |
189 | { |
190 | guid_t guid; |
191 | acpi_handle handle; |
192 | union acpi_object *obj; |
193 | |
194 | if (guid_parse(uuid: extlog_dsm_uuid, u: &guid)) |
195 | return false; |
196 | if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB" , &handle))) |
197 | return false; |
198 | if (!acpi_check_dsm(handle, guid: &guid, EXTLOG_DSM_REV, funcs: 1 << EXTLOG_FN_ADDR)) |
199 | return false; |
200 | obj = acpi_evaluate_dsm_typed(handle, guid: &guid, EXTLOG_DSM_REV, |
201 | EXTLOG_FN_ADDR, NULL, ACPI_TYPE_INTEGER); |
202 | if (!obj) { |
203 | return false; |
204 | } else { |
205 | l1_dirbase = obj->integer.value; |
206 | ACPI_FREE(obj); |
207 | } |
208 | |
209 | /* Spec says L1 directory must be 4K aligned, bail out if it isn't */ |
210 | if (l1_dirbase & ((1 << 12) - 1)) { |
211 | pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n" , |
212 | l1_dirbase); |
213 | return false; |
214 | } |
215 | |
216 | return true; |
217 | } |
218 | static struct notifier_block extlog_mce_dec = { |
219 | .notifier_call = extlog_print, |
220 | .priority = MCE_PRIO_EXTLOG, |
221 | }; |
222 | |
223 | static int __init extlog_init(void) |
224 | { |
225 | struct extlog_l1_head *l1_head; |
226 | void __iomem *extlog_l1_hdr; |
227 | size_t l1_hdr_size; |
228 | struct resource *r; |
229 | u64 cap; |
230 | int rc; |
231 | |
232 | if (rdmsrl_safe(MSR_IA32_MCG_CAP, p: &cap) || |
233 | !(cap & MCG_ELOG_P) || |
234 | !extlog_get_l1addr()) |
235 | return -ENODEV; |
236 | |
237 | rc = -EINVAL; |
238 | /* get L1 header to fetch necessary information */ |
239 | l1_hdr_size = sizeof(struct extlog_l1_head); |
240 | r = request_mem_region(l1_dirbase, l1_hdr_size, "L1 DIR HDR" ); |
241 | if (!r) { |
242 | pr_warn(FW_BUG EMCA_BUG, |
243 | (unsigned long long)l1_dirbase, |
244 | (unsigned long long)l1_dirbase + l1_hdr_size); |
245 | goto err; |
246 | } |
247 | |
248 | extlog_l1_hdr = acpi_os_map_iomem(phys: l1_dirbase, size: l1_hdr_size); |
249 | l1_head = (struct extlog_l1_head *)extlog_l1_hdr; |
250 | l1_size = l1_head->total_len; |
251 | l1_percpu_entry = l1_head->entries; |
252 | elog_base = l1_head->elog_base; |
253 | elog_size = l1_head->elog_len; |
254 | acpi_os_unmap_iomem(virt: extlog_l1_hdr, size: l1_hdr_size); |
255 | release_mem_region(l1_dirbase, l1_hdr_size); |
256 | |
257 | /* remap L1 header again based on completed information */ |
258 | r = request_mem_region(l1_dirbase, l1_size, "L1 Table" ); |
259 | if (!r) { |
260 | pr_warn(FW_BUG EMCA_BUG, |
261 | (unsigned long long)l1_dirbase, |
262 | (unsigned long long)l1_dirbase + l1_size); |
263 | goto err; |
264 | } |
265 | extlog_l1_addr = acpi_os_map_iomem(phys: l1_dirbase, size: l1_size); |
266 | l1_entry_base = (u64 *)((u8 *)extlog_l1_addr + l1_hdr_size); |
267 | |
268 | /* remap elog table */ |
269 | r = request_mem_region(elog_base, elog_size, "Elog Table" ); |
270 | if (!r) { |
271 | pr_warn(FW_BUG EMCA_BUG, |
272 | (unsigned long long)elog_base, |
273 | (unsigned long long)elog_base + elog_size); |
274 | goto err_release_l1_dir; |
275 | } |
276 | elog_addr = acpi_os_map_iomem(phys: elog_base, size: elog_size); |
277 | |
278 | rc = -ENOMEM; |
279 | /* allocate buffer to save elog record */ |
280 | elog_buf = kmalloc(ELOG_ENTRY_LEN, GFP_KERNEL); |
281 | if (elog_buf == NULL) |
282 | goto err_release_elog; |
283 | |
284 | mce_register_decode_chain(nb: &extlog_mce_dec); |
285 | /* enable OS to be involved to take over management from BIOS */ |
286 | ((struct extlog_l1_head *)extlog_l1_addr)->flags |= FLAG_OS_OPTIN; |
287 | |
288 | return 0; |
289 | |
290 | err_release_elog: |
291 | if (elog_addr) |
292 | acpi_os_unmap_iomem(virt: elog_addr, size: elog_size); |
293 | release_mem_region(elog_base, elog_size); |
294 | err_release_l1_dir: |
295 | if (extlog_l1_addr) |
296 | acpi_os_unmap_iomem(virt: extlog_l1_addr, size: l1_size); |
297 | release_mem_region(l1_dirbase, l1_size); |
298 | err: |
299 | pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n" ); |
300 | return rc; |
301 | } |
302 | |
303 | static void __exit extlog_exit(void) |
304 | { |
305 | mce_unregister_decode_chain(nb: &extlog_mce_dec); |
306 | ((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN; |
307 | if (extlog_l1_addr) |
308 | acpi_os_unmap_iomem(virt: extlog_l1_addr, size: l1_size); |
309 | if (elog_addr) |
310 | acpi_os_unmap_iomem(virt: elog_addr, size: elog_size); |
311 | release_mem_region(elog_base, elog_size); |
312 | release_mem_region(l1_dirbase, l1_size); |
313 | kfree(objp: elog_buf); |
314 | } |
315 | |
316 | module_init(extlog_init); |
317 | module_exit(extlog_exit); |
318 | |
319 | MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>" ); |
320 | MODULE_DESCRIPTION("Extended MCA Error Log Driver" ); |
321 | MODULE_LICENSE("GPL" ); |
322 | |