1/*
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 <linux/cper.h>
13#include <linux/ratelimit.h>
14#include <linux/edac.h>
15#include <linux/ras.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
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 int old_edac_report_status;
47
48static u8 extlog_dsm_uuid[] __initdata = "663E35AF-CC10-41A4-88EA-5470AF055295";
49
50/* L1 table related physical address */
51static u64 elog_base;
52static size_t elog_size;
53static u64 l1_dirbase;
54static size_t l1_size;
55
56/* L1 table related virtual address */
57static void __iomem *extlog_l1_addr;
58static void __iomem *elog_addr;
59
60static void *elog_buf;
61
62static u64 *l1_entry_base;
63static u32 l1_percpu_entry;
64
65#define ELOG_IDX(cpu, bank) \
66 (cpu_physical_id(cpu) * l1_percpu_entry + (bank))
67
68#define ELOG_ENTRY_DATA(idx) \
69 (*(l1_entry_base + (idx)))
70
71#define ELOG_ENTRY_ADDR(phyaddr) \
72 (phyaddr - elog_base + (u8 *)elog_addr)
73
74static struct acpi_hest_generic_status *extlog_elog_entry_check(int cpu, int bank)
75{
76 int idx;
77 u64 data;
78 struct acpi_hest_generic_status *estatus;
79
80 WARN_ON(cpu < 0);
81 idx = ELOG_IDX(cpu, bank);
82 data = ELOG_ENTRY_DATA(idx);
83 if ((data & ELOG_ENTRY_VALID) == 0)
84 return NULL;
85
86 data &= EXT_ELOG_ENTRY_MASK;
87 estatus = (struct acpi_hest_generic_status *)ELOG_ENTRY_ADDR(data);
88
89 /* if no valid data in elog entry, just return */
90 if (estatus->block_status == 0)
91 return NULL;
92
93 return estatus;
94}
95
96static void __print_extlog_rcd(const char *pfx,
97 struct acpi_hest_generic_status *estatus, int cpu)
98{
99 static atomic_t seqno;
100 unsigned int curr_seqno;
101 char pfx_seq[64];
102
103 if (!pfx) {
104 if (estatus->error_severity <= CPER_SEV_CORRECTED)
105 pfx = KERN_INFO;
106 else
107 pfx = KERN_ERR;
108 }
109 curr_seqno = atomic_inc_return(&seqno);
110 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}", pfx, curr_seqno);
111 printk("%s""Hardware error detected on CPU%d\n", pfx_seq, cpu);
112 cper_estatus_print(pfx_seq, estatus);
113}
114
115static int print_extlog_rcd(const char *pfx,
116 struct acpi_hest_generic_status *estatus, int cpu)
117{
118 /* Not more than 2 messages every 5 seconds */
119 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
120 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
121 struct ratelimit_state *ratelimit;
122
123 if (estatus->error_severity == CPER_SEV_CORRECTED ||
124 (estatus->error_severity == CPER_SEV_INFORMATIONAL))
125 ratelimit = &ratelimit_corrected;
126 else
127 ratelimit = &ratelimit_uncorrected;
128 if (__ratelimit(ratelimit)) {
129 __print_extlog_rcd(pfx, estatus, cpu);
130 return 0;
131 }
132
133 return 1;
134}
135
136static int extlog_print(struct notifier_block *nb, unsigned long val,
137 void *data)
138{
139 struct mce *mce = (struct mce *)data;
140 int bank = mce->bank;
141 int cpu = mce->extcpu;
142 struct acpi_hest_generic_status *estatus, *tmp;
143 struct acpi_hest_generic_data *gdata;
144 const guid_t *fru_id = &guid_null;
145 char *fru_text = "";
146 guid_t *sec_type;
147 static u32 err_seq;
148
149 estatus = extlog_elog_entry_check(cpu, bank);
150 if (estatus == NULL)
151 return NOTIFY_DONE;
152
153 memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN);
154 /* clear record status to enable BIOS to update it again */
155 estatus->block_status = 0;
156
157 tmp = (struct acpi_hest_generic_status *)elog_buf;
158
159 if (!ras_userspace_consumers()) {
160 print_extlog_rcd(NULL, tmp, cpu);
161 goto out;
162 }
163
164 /* log event via trace */
165 err_seq++;
166 gdata = (struct acpi_hest_generic_data *)(tmp + 1);
167 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
168 fru_id = (guid_t *)gdata->fru_id;
169 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
170 fru_text = gdata->fru_text;
171 sec_type = (guid_t *)gdata->section_type;
172 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
173 struct cper_sec_mem_err *mem = (void *)(gdata + 1);
174 if (gdata->error_data_length >= sizeof(*mem))
175 trace_extlog_mem_event(mem, err_seq, fru_id, fru_text,
176 (u8)gdata->error_severity);
177 }
178
179out:
180 return NOTIFY_STOP;
181}
182
183static bool __init extlog_get_l1addr(void)
184{
185 guid_t guid;
186 acpi_handle handle;
187 union acpi_object *obj;
188
189 if (guid_parse(extlog_dsm_uuid, &guid))
190 return false;
191 if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle)))
192 return false;
193 if (!acpi_check_dsm(handle, &guid, EXTLOG_DSM_REV, 1 << EXTLOG_FN_ADDR))
194 return false;
195 obj = acpi_evaluate_dsm_typed(handle, &guid, EXTLOG_DSM_REV,
196 EXTLOG_FN_ADDR, NULL, ACPI_TYPE_INTEGER);
197 if (!obj) {
198 return false;
199 } else {
200 l1_dirbase = obj->integer.value;
201 ACPI_FREE(obj);
202 }
203
204 /* Spec says L1 directory must be 4K aligned, bail out if it isn't */
205 if (l1_dirbase & ((1 << 12) - 1)) {
206 pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n",
207 l1_dirbase);
208 return false;
209 }
210
211 return true;
212}
213static struct notifier_block extlog_mce_dec = {
214 .notifier_call = extlog_print,
215 .priority = MCE_PRIO_EXTLOG,
216};
217
218static int __init extlog_init(void)
219{
220 struct extlog_l1_head *l1_head;
221 void __iomem *extlog_l1_hdr;
222 size_t l1_hdr_size;
223 struct resource *r;
224 u64 cap;
225 int rc;
226
227 rdmsrl(MSR_IA32_MCG_CAP, cap);
228
229 if (!(cap & MCG_ELOG_P) || !extlog_get_l1addr())
230 return -ENODEV;
231
232 if (edac_get_report_status() == EDAC_REPORTING_FORCE) {
233 pr_warn("Not loading eMCA, error reporting force-enabled through EDAC.\n");
234 return -EPERM;
235 }
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(l1_dirbase, 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(extlog_l1_hdr, 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(l1_dirbase, 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(elog_base, 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 /*
285 * eMCA event report method has higher priority than EDAC method,
286 * unless EDAC event report method is mandatory.
287 */
288 old_edac_report_status = edac_get_report_status();
289 edac_set_report_status(EDAC_REPORTING_DISABLED);
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_iomem(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_iomem(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 edac_set_report_status(old_edac_report_status);
312 mce_unregister_decode_chain(&extlog_mce_dec);
313 ((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN;
314 if (extlog_l1_addr)
315 acpi_os_unmap_iomem(extlog_l1_addr, l1_size);
316 if (elog_addr)
317 acpi_os_unmap_iomem(elog_addr, elog_size);
318 release_mem_region(elog_base, elog_size);
319 release_mem_region(l1_dirbase, l1_size);
320 kfree(elog_buf);
321}
322
323module_init(extlog_init);
324module_exit(extlog_exit);
325
326MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>");
327MODULE_DESCRIPTION("Extended MCA Error Log Driver");
328MODULE_LICENSE("GPL");
329