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
extlog_elog_entry_check(int cpu,int bank)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
__print_extlog_rcd(const char * pfx,struct acpi_hest_generic_status * estatus,int cpu)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(&seqno);
108 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}", pfx, curr_seqno);
109 printk("%s""Hardware error detected on CPU%d\n", pfx_seq, cpu);
110 cper_estatus_print(pfx_seq, estatus);
111 }
112
print_extlog_rcd(const char * pfx,struct acpi_hest_generic_status * estatus,int cpu)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
extlog_print(struct notifier_block * nb,unsigned long val,void * data)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)
149 return NOTIFY_DONE;
150
151 if (mce->kflags & MCE_HANDLED_CEC) {
152 estatus->block_status = 0;
153 return NOTIFY_DONE;
154 }
155
156 memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN);
157 /* clear record status to enable BIOS to update it again */
158 estatus->block_status = 0;
159
160 tmp = (struct acpi_hest_generic_status *)elog_buf;
161
162 if (!ras_userspace_consumers()) {
163 print_extlog_rcd(NULL, tmp, cpu);
164 goto out;
165 }
166
167 /* log event via trace */
168 err_seq++;
169 apei_estatus_for_each_section(tmp, gdata) {
170 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
171 fru_id = (guid_t *)gdata->fru_id;
172 else
173 fru_id = &guid_null;
174 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
175 fru_text = gdata->fru_text;
176 else
177 fru_text = "";
178 sec_type = (guid_t *)gdata->section_type;
179 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
180 struct cper_sec_mem_err *mem = acpi_hest_get_payload(gdata);
181
182 if (gdata->error_data_length >= sizeof(*mem))
183 trace_extlog_mem_event(mem, err_seq, fru_id, fru_text,
184 (u8)gdata->error_severity);
185 }
186 }
187
188 out:
189 mce->kflags |= MCE_HANDLED_EXTLOG;
190 return NOTIFY_OK;
191 }
192
extlog_get_l1addr(void)193 static bool __init extlog_get_l1addr(void)
194 {
195 guid_t guid;
196 acpi_handle handle;
197 union acpi_object *obj;
198
199 if (guid_parse(extlog_dsm_uuid, &guid))
200 return false;
201 if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle)))
202 return false;
203 if (!acpi_check_dsm(handle, &guid, EXTLOG_DSM_REV, 1 << EXTLOG_FN_ADDR))
204 return false;
205 obj = acpi_evaluate_dsm_typed(handle, &guid, EXTLOG_DSM_REV,
206 EXTLOG_FN_ADDR, NULL, ACPI_TYPE_INTEGER);
207 if (!obj) {
208 return false;
209 } else {
210 l1_dirbase = obj->integer.value;
211 ACPI_FREE(obj);
212 }
213
214 /* Spec says L1 directory must be 4K aligned, bail out if it isn't */
215 if (l1_dirbase & ((1 << 12) - 1)) {
216 pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n",
217 l1_dirbase);
218 return false;
219 }
220
221 return true;
222 }
223 static struct notifier_block extlog_mce_dec = {
224 .notifier_call = extlog_print,
225 .priority = MCE_PRIO_EXTLOG,
226 };
227
extlog_init(void)228 static int __init extlog_init(void)
229 {
230 struct extlog_l1_head *l1_head;
231 void __iomem *extlog_l1_hdr;
232 size_t l1_hdr_size;
233 struct resource *r;
234 u64 cap;
235 int rc;
236
237 if (rdmsrl_safe(MSR_IA32_MCG_CAP, &cap) ||
238 !(cap & MCG_ELOG_P) ||
239 !extlog_get_l1addr())
240 return -ENODEV;
241
242 rc = -EINVAL;
243 /* get L1 header to fetch necessary information */
244 l1_hdr_size = sizeof(struct extlog_l1_head);
245 r = request_mem_region(l1_dirbase, l1_hdr_size, "L1 DIR HDR");
246 if (!r) {
247 pr_warn(FW_BUG EMCA_BUG,
248 (unsigned long long)l1_dirbase,
249 (unsigned long long)l1_dirbase + l1_hdr_size);
250 goto err;
251 }
252
253 extlog_l1_hdr = acpi_os_map_iomem(l1_dirbase, l1_hdr_size);
254 l1_head = (struct extlog_l1_head *)extlog_l1_hdr;
255 l1_size = l1_head->total_len;
256 l1_percpu_entry = l1_head->entries;
257 elog_base = l1_head->elog_base;
258 elog_size = l1_head->elog_len;
259 acpi_os_unmap_iomem(extlog_l1_hdr, l1_hdr_size);
260 release_mem_region(l1_dirbase, l1_hdr_size);
261
262 /* remap L1 header again based on completed information */
263 r = request_mem_region(l1_dirbase, l1_size, "L1 Table");
264 if (!r) {
265 pr_warn(FW_BUG EMCA_BUG,
266 (unsigned long long)l1_dirbase,
267 (unsigned long long)l1_dirbase + l1_size);
268 goto err;
269 }
270 extlog_l1_addr = acpi_os_map_iomem(l1_dirbase, l1_size);
271 l1_entry_base = (u64 *)((u8 *)extlog_l1_addr + l1_hdr_size);
272
273 /* remap elog table */
274 r = request_mem_region(elog_base, elog_size, "Elog Table");
275 if (!r) {
276 pr_warn(FW_BUG EMCA_BUG,
277 (unsigned long long)elog_base,
278 (unsigned long long)elog_base + elog_size);
279 goto err_release_l1_dir;
280 }
281 elog_addr = acpi_os_map_iomem(elog_base, elog_size);
282
283 rc = -ENOMEM;
284 /* allocate buffer to save elog record */
285 elog_buf = kmalloc(ELOG_ENTRY_LEN, GFP_KERNEL);
286 if (elog_buf == NULL)
287 goto err_release_elog;
288
289 mce_register_decode_chain(&extlog_mce_dec);
290 /* enable OS to be involved to take over management from BIOS */
291 ((struct extlog_l1_head *)extlog_l1_addr)->flags |= FLAG_OS_OPTIN;
292
293 return 0;
294
295 err_release_elog:
296 if (elog_addr)
297 acpi_os_unmap_iomem(elog_addr, elog_size);
298 release_mem_region(elog_base, elog_size);
299 err_release_l1_dir:
300 if (extlog_l1_addr)
301 acpi_os_unmap_iomem(extlog_l1_addr, l1_size);
302 release_mem_region(l1_dirbase, l1_size);
303 err:
304 pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n");
305 return rc;
306 }
307
extlog_exit(void)308 static void __exit extlog_exit(void)
309 {
310 mce_unregister_decode_chain(&extlog_mce_dec);
311 if (extlog_l1_addr) {
312 ((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN;
313 acpi_os_unmap_iomem(extlog_l1_addr, l1_size);
314 }
315 if (elog_addr)
316 acpi_os_unmap_iomem(elog_addr, elog_size);
317 release_mem_region(elog_base, elog_size);
318 release_mem_region(l1_dirbase, l1_size);
319 kfree(elog_buf);
320 }
321
322 module_init(extlog_init);
323 module_exit(extlog_exit);
324
325 MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>");
326 MODULE_DESCRIPTION("Extended MCA Error Log Driver");
327 MODULE_LICENSE("GPL");
328