blob: 2635a01c5b3edc077eb80a704db6e316b2395857 [file] [log] [blame]
Chen, Gong4b3db702013-10-21 14:29:25 -07001/*
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>
Chen, Gong4b3db702013-10-21 14:29:25 -070012#include <linux/cper.h>
13#include <linux/ratelimit.h>
14#include <asm/cpu.h>
15#include <asm/mce.h>
16
17#include "apei/apei-internal.h"
18
19#define EXT_ELOG_ENTRY_MASK GENMASK_ULL(51, 0) /* elog entry address mask */
20
21#define EXTLOG_DSM_REV 0x0
22#define EXTLOG_FN_QUERY 0x0
23#define EXTLOG_FN_ADDR 0x1
24
25#define FLAG_OS_OPTIN BIT(0)
26#define EXTLOG_QUERY_L1_EXIST BIT(1)
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
33struct 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
45static u8 extlog_dsm_uuid[] = "663E35AF-CC10-41A4-88EA-5470AF055295";
46
47/* L1 table related physical address */
48static u64 elog_base;
49static size_t elog_size;
50static u64 l1_dirbase;
51static size_t l1_size;
52
53/* L1 table related virtual address */
54static void __iomem *extlog_l1_addr;
55static void __iomem *elog_addr;
56
57static void *elog_buf;
58
59static u64 *l1_entry_base;
60static 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
71static struct acpi_generic_status *extlog_elog_entry_check(int cpu, int bank)
72{
73 int idx;
74 u64 data;
75 struct acpi_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_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
93static void __print_extlog_rcd(const char *pfx,
94 struct acpi_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
112static int print_extlog_rcd(const char *pfx,
113 struct acpi_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
133static 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_generic_status *estatus;
140 int rc;
141
142 estatus = extlog_elog_entry_check(cpu, bank);
143 if (estatus == NULL)
144 return NOTIFY_DONE;
145
146 memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN);
147 /* clear record status to enable BIOS to update it again */
148 estatus->block_status = 0;
149
150 rc = print_extlog_rcd(NULL, (struct acpi_generic_status *)elog_buf, cpu);
151
152 return NOTIFY_DONE;
153}
154
155static int extlog_get_dsm(acpi_handle handle, int rev, int func, u64 *ret)
156{
157 struct acpi_buffer buf = {ACPI_ALLOCATE_BUFFER, NULL};
158 struct acpi_object_list input;
159 union acpi_object params[4], *obj;
160 u8 uuid[16];
161 int i;
162
163 acpi_str_to_uuid(extlog_dsm_uuid, uuid);
164 input.count = 4;
165 input.pointer = params;
166 params[0].type = ACPI_TYPE_BUFFER;
167 params[0].buffer.length = 16;
168 params[0].buffer.pointer = uuid;
169 params[1].type = ACPI_TYPE_INTEGER;
170 params[1].integer.value = rev;
171 params[2].type = ACPI_TYPE_INTEGER;
172 params[2].integer.value = func;
173 params[3].type = ACPI_TYPE_PACKAGE;
174 params[3].package.count = 0;
175 params[3].package.elements = NULL;
176
177 if (ACPI_FAILURE(acpi_evaluate_object(handle, "_DSM", &input, &buf)))
178 return -1;
179
180 *ret = 0;
181 obj = (union acpi_object *)buf.pointer;
182 if (obj->type == ACPI_TYPE_INTEGER) {
183 *ret = obj->integer.value;
184 } else if (obj->type == ACPI_TYPE_BUFFER) {
185 if (obj->buffer.length <= 8) {
186 for (i = 0; i < obj->buffer.length; i++)
187 *ret |= (obj->buffer.pointer[i] << (i * 8));
188 }
189 }
190 kfree(buf.pointer);
191
192 return 0;
193}
194
195static bool extlog_get_l1addr(void)
196{
197 acpi_handle handle;
198 u64 ret;
199
200 if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle)))
201 return false;
202
203 if (extlog_get_dsm(handle, EXTLOG_DSM_REV, EXTLOG_FN_QUERY, &ret) ||
204 !(ret & EXTLOG_QUERY_L1_EXIST))
205 return false;
206
207 if (extlog_get_dsm(handle, EXTLOG_DSM_REV, EXTLOG_FN_ADDR, &ret))
208 return false;
209
210 l1_dirbase = ret;
211 /* Spec says L1 directory must be 4K aligned, bail out if it isn't */
212 if (l1_dirbase & ((1 << 12) - 1)) {
213 pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n",
214 l1_dirbase);
215 return false;
216 }
217
218 return true;
219}
220static struct notifier_block extlog_mce_dec = {
221 .notifier_call = extlog_print,
222};
223
224static int __init extlog_init(void)
225{
226 struct extlog_l1_head *l1_head;
227 void __iomem *extlog_l1_hdr;
228 size_t l1_hdr_size;
229 struct resource *r;
230 u64 cap;
231 int rc;
232
233 rc = -ENODEV;
234
235 rdmsrl(MSR_IA32_MCG_CAP, cap);
236 if (!(cap & MCG_ELOG_P))
237 return rc;
238
239 if (!extlog_get_l1addr())
240 return rc;
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_memory(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_memory(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_memory(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_memory(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
295err_release_elog:
296 if (elog_addr)
297 acpi_os_unmap_memory(elog_addr, elog_size);
298 release_mem_region(elog_base, elog_size);
299err_release_l1_dir:
300 if (extlog_l1_addr)
301 acpi_os_unmap_memory(extlog_l1_addr, l1_size);
302 release_mem_region(l1_dirbase, l1_size);
303err:
304 pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n");
305 return rc;
306}
307
308static void __exit extlog_exit(void)
309{
310 mce_unregister_decode_chain(&extlog_mce_dec);
311 ((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN;
312 if (extlog_l1_addr)
313 acpi_os_unmap_memory(extlog_l1_addr, l1_size);
314 if (elog_addr)
315 acpi_os_unmap_memory(elog_addr, elog_size);
316 release_mem_region(elog_base, elog_size);
317 release_mem_region(l1_dirbase, l1_size);
318 kfree(elog_buf);
319}
320
321module_init(extlog_init);
322module_exit(extlog_exit);
323
324MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>");
325MODULE_DESCRIPTION("Extended MCA Error Log Driver");
326MODULE_LICENSE("GPL");