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