blob: a6869e110ce5650a86b6f666bc28347031c16275 [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>
12#include <acpi/acpi_bus.h>
13#include <linux/cper.h>
14#include <linux/ratelimit.h>
15#include <asm/cpu.h>
16#include <asm/mce.h>
17
18#include "apei/apei-internal.h"
19
20#define EXT_ELOG_ENTRY_MASK GENMASK_ULL(51, 0) /* elog entry address mask */
21
22#define EXTLOG_DSM_REV 0x0
23#define EXTLOG_FN_QUERY 0x0
24#define EXTLOG_FN_ADDR 0x1
25
26#define FLAG_OS_OPTIN BIT(0)
27#define EXTLOG_QUERY_L1_EXIST BIT(1)
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
34struct 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
46static u8 extlog_dsm_uuid[] = "663E35AF-CC10-41A4-88EA-5470AF055295";
47
48/* L1 table related physical address */
49static u64 elog_base;
50static size_t elog_size;
51static u64 l1_dirbase;
52static size_t l1_size;
53
54/* L1 table related virtual address */
55static void __iomem *extlog_l1_addr;
56static void __iomem *elog_addr;
57
58static void *elog_buf;
59
60static u64 *l1_entry_base;
61static 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
72static struct acpi_generic_status *extlog_elog_entry_check(int cpu, int bank)
73{
74 int idx;
75 u64 data;
76 struct acpi_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_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
94static void __print_extlog_rcd(const char *pfx,
95 struct acpi_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
113static int print_extlog_rcd(const char *pfx,
114 struct acpi_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
134static 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_generic_status *estatus;
141 int rc;
142
143 estatus = extlog_elog_entry_check(cpu, bank);
144 if (estatus == NULL)
145 return NOTIFY_DONE;
146
147 memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN);
148 /* clear record status to enable BIOS to update it again */
149 estatus->block_status = 0;
150
151 rc = print_extlog_rcd(NULL, (struct acpi_generic_status *)elog_buf, cpu);
152
153 return NOTIFY_DONE;
154}
155
156static int extlog_get_dsm(acpi_handle handle, int rev, int func, u64 *ret)
157{
158 struct acpi_buffer buf = {ACPI_ALLOCATE_BUFFER, NULL};
159 struct acpi_object_list input;
160 union acpi_object params[4], *obj;
161 u8 uuid[16];
162 int i;
163
164 acpi_str_to_uuid(extlog_dsm_uuid, uuid);
165 input.count = 4;
166 input.pointer = params;
167 params[0].type = ACPI_TYPE_BUFFER;
168 params[0].buffer.length = 16;
169 params[0].buffer.pointer = uuid;
170 params[1].type = ACPI_TYPE_INTEGER;
171 params[1].integer.value = rev;
172 params[2].type = ACPI_TYPE_INTEGER;
173 params[2].integer.value = func;
174 params[3].type = ACPI_TYPE_PACKAGE;
175 params[3].package.count = 0;
176 params[3].package.elements = NULL;
177
178 if (ACPI_FAILURE(acpi_evaluate_object(handle, "_DSM", &input, &buf)))
179 return -1;
180
181 *ret = 0;
182 obj = (union acpi_object *)buf.pointer;
183 if (obj->type == ACPI_TYPE_INTEGER) {
184 *ret = obj->integer.value;
185 } else if (obj->type == ACPI_TYPE_BUFFER) {
186 if (obj->buffer.length <= 8) {
187 for (i = 0; i < obj->buffer.length; i++)
188 *ret |= (obj->buffer.pointer[i] << (i * 8));
189 }
190 }
191 kfree(buf.pointer);
192
193 return 0;
194}
195
196static bool extlog_get_l1addr(void)
197{
198 acpi_handle handle;
199 u64 ret;
200
201 if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle)))
202 return false;
203
204 if (extlog_get_dsm(handle, EXTLOG_DSM_REV, EXTLOG_FN_QUERY, &ret) ||
205 !(ret & EXTLOG_QUERY_L1_EXIST))
206 return false;
207
208 if (extlog_get_dsm(handle, EXTLOG_DSM_REV, EXTLOG_FN_ADDR, &ret))
209 return false;
210
211 l1_dirbase = ret;
212 /* Spec says L1 directory must be 4K aligned, bail out if it isn't */
213 if (l1_dirbase & ((1 << 12) - 1)) {
214 pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n",
215 l1_dirbase);
216 return false;
217 }
218
219 return true;
220}
221static struct notifier_block extlog_mce_dec = {
222 .notifier_call = extlog_print,
223};
224
225static int __init extlog_init(void)
226{
227 struct extlog_l1_head *l1_head;
228 void __iomem *extlog_l1_hdr;
229 size_t l1_hdr_size;
230 struct resource *r;
231 u64 cap;
232 int rc;
233
234 rc = -ENODEV;
235
236 rdmsrl(MSR_IA32_MCG_CAP, cap);
237 if (!(cap & MCG_ELOG_P))
238 return rc;
239
240 if (!extlog_get_l1addr())
241 return rc;
242
243 rc = -EINVAL;
244 /* get L1 header to fetch necessary information */
245 l1_hdr_size = sizeof(struct extlog_l1_head);
246 r = request_mem_region(l1_dirbase, l1_hdr_size, "L1 DIR HDR");
247 if (!r) {
248 pr_warn(FW_BUG EMCA_BUG,
249 (unsigned long long)l1_dirbase,
250 (unsigned long long)l1_dirbase + l1_hdr_size);
251 goto err;
252 }
253
254 extlog_l1_hdr = acpi_os_map_memory(l1_dirbase, l1_hdr_size);
255 l1_head = (struct extlog_l1_head *)extlog_l1_hdr;
256 l1_size = l1_head->total_len;
257 l1_percpu_entry = l1_head->entries;
258 elog_base = l1_head->elog_base;
259 elog_size = l1_head->elog_len;
260 acpi_os_unmap_memory(extlog_l1_hdr, l1_hdr_size);
261 release_mem_region(l1_dirbase, l1_hdr_size);
262
263 /* remap L1 header again based on completed information */
264 r = request_mem_region(l1_dirbase, l1_size, "L1 Table");
265 if (!r) {
266 pr_warn(FW_BUG EMCA_BUG,
267 (unsigned long long)l1_dirbase,
268 (unsigned long long)l1_dirbase + l1_size);
269 goto err;
270 }
271 extlog_l1_addr = acpi_os_map_memory(l1_dirbase, l1_size);
272 l1_entry_base = (u64 *)((u8 *)extlog_l1_addr + l1_hdr_size);
273
274 /* remap elog table */
275 r = request_mem_region(elog_base, elog_size, "Elog Table");
276 if (!r) {
277 pr_warn(FW_BUG EMCA_BUG,
278 (unsigned long long)elog_base,
279 (unsigned long long)elog_base + elog_size);
280 goto err_release_l1_dir;
281 }
282 elog_addr = acpi_os_map_memory(elog_base, elog_size);
283
284 rc = -ENOMEM;
285 /* allocate buffer to save elog record */
286 elog_buf = kmalloc(ELOG_ENTRY_LEN, GFP_KERNEL);
287 if (elog_buf == NULL)
288 goto err_release_elog;
289
290 mce_register_decode_chain(&extlog_mce_dec);
291 /* enable OS to be involved to take over management from BIOS */
292 ((struct extlog_l1_head *)extlog_l1_addr)->flags |= FLAG_OS_OPTIN;
293
294 return 0;
295
296err_release_elog:
297 if (elog_addr)
298 acpi_os_unmap_memory(elog_addr, elog_size);
299 release_mem_region(elog_base, elog_size);
300err_release_l1_dir:
301 if (extlog_l1_addr)
302 acpi_os_unmap_memory(extlog_l1_addr, l1_size);
303 release_mem_region(l1_dirbase, l1_size);
304err:
305 pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n");
306 return rc;
307}
308
309static void __exit extlog_exit(void)
310{
311 mce_unregister_decode_chain(&extlog_mce_dec);
312 ((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN;
313 if (extlog_l1_addr)
314 acpi_os_unmap_memory(extlog_l1_addr, l1_size);
315 if (elog_addr)
316 acpi_os_unmap_memory(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
322module_init(extlog_init);
323module_exit(extlog_exit);
324
325MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>");
326MODULE_DESCRIPTION("Extended MCA Error Log Driver");
327MODULE_LICENSE("GPL");