1 /*
2 * MCE grading rules.
3 * Copyright 2008, 2009 Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; version 2
8 * of the License.
9 *
10 * Author: Andi Kleen
11 */
12 #include <linux/kernel.h>
13 #include <linux/seq_file.h>
14 #include <linux/init.h>
15 #include <linux/debugfs.h>
16 #include <asm/mce.h>
17
18 #include "mce-internal.h"
19
20 /*
21 * Grade an mce by severity. In general the most severe ones are processed
22 * first. Since there are quite a lot of combinations test the bits in a
23 * table-driven way. The rules are simply processed in order, first
24 * match wins.
25 *
26 * Note this is only used for machine check exceptions, the corrected
27 * errors use much simpler rules. The exceptions still check for the corrected
28 * errors, but only to leave them alone for the CMCI handler (except for
29 * panic situations)
30 */
31
32 enum context { IN_KERNEL = 1, IN_USER = 2 };
33 enum ser { SER_REQUIRED = 1, NO_SER = 2 };
34 enum exception { EXCP_CONTEXT = 1, NO_EXCP = 2 };
35
36 static struct severity {
37 u64 mask;
38 u64 result;
39 unsigned char sev;
40 unsigned char mcgmask;
41 unsigned char mcgres;
42 unsigned char ser;
43 unsigned char context;
44 unsigned char excp;
45 unsigned char covered;
46 char *msg;
47 } severities[] = {
48 #define MCESEV(s, m, c...) { .sev = MCE_ ## s ## _SEVERITY, .msg = m, ## c }
49 #define KERNEL .context = IN_KERNEL
50 #define USER .context = IN_USER
51 #define SER .ser = SER_REQUIRED
52 #define NOSER .ser = NO_SER
53 #define EXCP .excp = EXCP_CONTEXT
54 #define NOEXCP .excp = NO_EXCP
55 #define BITCLR(x) .mask = x, .result = 0
56 #define BITSET(x) .mask = x, .result = x
57 #define MCGMASK(x, y) .mcgmask = x, .mcgres = y
58 #define MASK(x, y) .mask = x, .result = y
59 #define MCI_UC_S (MCI_STATUS_UC|MCI_STATUS_S)
60 #define MCI_UC_SAR (MCI_STATUS_UC|MCI_STATUS_S|MCI_STATUS_AR)
61 #define MCI_ADDR (MCI_STATUS_ADDRV|MCI_STATUS_MISCV)
62
63 MCESEV(
64 NO, "Invalid",
65 BITCLR(MCI_STATUS_VAL)
66 ),
67 MCESEV(
68 NO, "Not enabled",
69 EXCP, BITCLR(MCI_STATUS_EN)
70 ),
71 MCESEV(
72 PANIC, "Processor context corrupt",
73 BITSET(MCI_STATUS_PCC)
74 ),
75 /* When MCIP is not set something is very confused */
76 MCESEV(
77 PANIC, "MCIP not set in MCA handler",
78 EXCP, MCGMASK(MCG_STATUS_MCIP, 0)
79 ),
80 /* Neither return not error IP -- no chance to recover -> PANIC */
81 MCESEV(
82 PANIC, "Neither restart nor error IP",
83 EXCP, MCGMASK(MCG_STATUS_RIPV|MCG_STATUS_EIPV, 0)
84 ),
85 MCESEV(
86 PANIC, "In kernel and no restart IP",
87 EXCP, KERNEL, MCGMASK(MCG_STATUS_RIPV, 0)
88 ),
89 MCESEV(
90 DEFERRED, "Deferred error",
91 NOSER, MASK(MCI_STATUS_UC|MCI_STATUS_DEFERRED|MCI_STATUS_POISON, MCI_STATUS_DEFERRED)
92 ),
93 MCESEV(
94 KEEP, "Corrected error",
95 NOSER, BITCLR(MCI_STATUS_UC)
96 ),
97
98 /* ignore OVER for UCNA */
99 MCESEV(
100 UCNA, "Uncorrected no action required",
101 SER, MASK(MCI_UC_SAR, MCI_STATUS_UC)
102 ),
103 MCESEV(
104 PANIC, "Illegal combination (UCNA with AR=1)",
105 SER,
106 MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_UC|MCI_STATUS_AR)
107 ),
108 MCESEV(
109 KEEP, "Non signalled machine check",
110 SER, BITCLR(MCI_STATUS_S)
111 ),
112
113 MCESEV(
114 PANIC, "Action required with lost events",
115 SER, BITSET(MCI_STATUS_OVER|MCI_UC_SAR)
116 ),
117
118 /* known AR MCACODs: */
119 #ifdef CONFIG_MEMORY_FAILURE
120 MCESEV(
121 KEEP, "Action required but unaffected thread is continuable",
122 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR, MCI_UC_SAR|MCI_ADDR),
123 MCGMASK(MCG_STATUS_RIPV|MCG_STATUS_EIPV, MCG_STATUS_RIPV)
124 ),
125 MCESEV(
126 AR, "Action required: data load error in a user process",
127 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_DATA),
128 USER
129 ),
130 MCESEV(
131 AR, "Action required: instruction fetch error in a user process",
132 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_INSTR),
133 USER
134 ),
135 MCESEV(
136 PANIC, "Instruction fetch error in kernel",
137 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_INSTR),
138 KERNEL
139 ),
140 #endif
141 MCESEV(
142 PANIC, "Action required: unknown MCACOD",
143 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_SAR)
144 ),
145
146 /* known AO MCACODs: */
147 MCESEV(
148 AO, "Action optional: memory scrubbing error",
149 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCACOD_SCRUBMSK, MCI_UC_S|MCACOD_SCRUB)
150 ),
151 MCESEV(
152 AO, "Action optional: last level cache writeback error",
153 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCACOD, MCI_UC_S|MCACOD_L3WB)
154 ),
155 MCESEV(
156 SOME, "Action optional: unknown MCACOD",
157 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_S)
158 ),
159 MCESEV(
160 SOME, "Action optional with lost events",
161 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_OVER|MCI_UC_S)
162 ),
163
164 MCESEV(
165 PANIC, "Overflowed uncorrected",
166 BITSET(MCI_STATUS_OVER|MCI_STATUS_UC)
167 ),
168 MCESEV(
169 UC, "Uncorrected",
170 BITSET(MCI_STATUS_UC)
171 ),
172 MCESEV(
173 SOME, "No match",
174 BITSET(0)
175 ) /* always matches. keep at end */
176 };
177
178 /*
179 * If mcgstatus indicated that ip/cs on the stack were
180 * no good, then "m->cs" will be zero and we will have
181 * to assume the worst case (IN_KERNEL) as we actually
182 * have no idea what we were executing when the machine
183 * check hit.
184 * If we do have a good "m->cs" (or a faked one in the
185 * case we were executing in VM86 mode) we can use it to
186 * distinguish an exception taken in user from from one
187 * taken in the kernel.
188 */
error_context(struct mce * m)189 static int error_context(struct mce *m)
190 {
191 return ((m->cs & 3) == 3) ? IN_USER : IN_KERNEL;
192 }
193
194 /*
195 * See AMD Error Scope Hierarchy table in a newer BKDG. For example
196 * 49125_15h_Models_30h-3Fh_BKDG.pdf, section "RAS Features"
197 */
mce_severity_amd(struct mce * m,int tolerant,char ** msg,bool is_excp)198 static int mce_severity_amd(struct mce *m, int tolerant, char **msg, bool is_excp)
199 {
200 enum context ctx = error_context(m);
201
202 /* Processor Context Corrupt, no need to fumble too much, die! */
203 if (m->status & MCI_STATUS_PCC)
204 return MCE_PANIC_SEVERITY;
205
206 if (m->status & MCI_STATUS_UC) {
207
208 /*
209 * On older systems where overflow_recov flag is not present, we
210 * should simply panic if an error overflow occurs. If
211 * overflow_recov flag is present and set, then software can try
212 * to at least kill process to prolong system operation.
213 */
214 if (mce_flags.overflow_recov) {
215 /* software can try to contain */
216 if (!(m->mcgstatus & MCG_STATUS_RIPV) && (ctx == IN_KERNEL))
217 return MCE_PANIC_SEVERITY;
218
219 /* kill current process */
220 return MCE_AR_SEVERITY;
221 } else {
222 /* at least one error was not logged */
223 if (m->status & MCI_STATUS_OVER)
224 return MCE_PANIC_SEVERITY;
225 }
226
227 /*
228 * For any other case, return MCE_UC_SEVERITY so that we log the
229 * error and exit #MC handler.
230 */
231 return MCE_UC_SEVERITY;
232 }
233
234 /*
235 * deferred error: poll handler catches these and adds to mce_ring so
236 * memory-failure can take recovery actions.
237 */
238 if (m->status & MCI_STATUS_DEFERRED)
239 return MCE_DEFERRED_SEVERITY;
240
241 /*
242 * corrected error: poll handler catches these and passes responsibility
243 * of decoding the error to EDAC
244 */
245 return MCE_KEEP_SEVERITY;
246 }
247
mce_severity_intel(struct mce * m,int tolerant,char ** msg,bool is_excp)248 static int mce_severity_intel(struct mce *m, int tolerant, char **msg, bool is_excp)
249 {
250 enum exception excp = (is_excp ? EXCP_CONTEXT : NO_EXCP);
251 enum context ctx = error_context(m);
252 struct severity *s;
253
254 for (s = severities;; s++) {
255 if ((m->status & s->mask) != s->result)
256 continue;
257 if ((m->mcgstatus & s->mcgmask) != s->mcgres)
258 continue;
259 if (s->ser == SER_REQUIRED && !mca_cfg.ser)
260 continue;
261 if (s->ser == NO_SER && mca_cfg.ser)
262 continue;
263 if (s->context && ctx != s->context)
264 continue;
265 if (s->excp && excp != s->excp)
266 continue;
267 if (msg)
268 *msg = s->msg;
269 s->covered = 1;
270 if (s->sev >= MCE_UC_SEVERITY && ctx == IN_KERNEL) {
271 if (panic_on_oops || tolerant < 1)
272 return MCE_PANIC_SEVERITY;
273 }
274 return s->sev;
275 }
276 }
277
278 /* Default to mce_severity_intel */
279 int (*mce_severity)(struct mce *m, int tolerant, char **msg, bool is_excp) =
280 mce_severity_intel;
281
mcheck_vendor_init_severity(void)282 void __init mcheck_vendor_init_severity(void)
283 {
284 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
285 mce_severity = mce_severity_amd;
286 }
287
288 #ifdef CONFIG_DEBUG_FS
s_start(struct seq_file * f,loff_t * pos)289 static void *s_start(struct seq_file *f, loff_t *pos)
290 {
291 if (*pos >= ARRAY_SIZE(severities))
292 return NULL;
293 return &severities[*pos];
294 }
295
s_next(struct seq_file * f,void * data,loff_t * pos)296 static void *s_next(struct seq_file *f, void *data, loff_t *pos)
297 {
298 if (++(*pos) >= ARRAY_SIZE(severities))
299 return NULL;
300 return &severities[*pos];
301 }
302
s_stop(struct seq_file * f,void * data)303 static void s_stop(struct seq_file *f, void *data)
304 {
305 }
306
s_show(struct seq_file * f,void * data)307 static int s_show(struct seq_file *f, void *data)
308 {
309 struct severity *ser = data;
310 seq_printf(f, "%d\t%s\n", ser->covered, ser->msg);
311 return 0;
312 }
313
314 static const struct seq_operations severities_seq_ops = {
315 .start = s_start,
316 .next = s_next,
317 .stop = s_stop,
318 .show = s_show,
319 };
320
severities_coverage_open(struct inode * inode,struct file * file)321 static int severities_coverage_open(struct inode *inode, struct file *file)
322 {
323 return seq_open(file, &severities_seq_ops);
324 }
325
severities_coverage_write(struct file * file,const char __user * ubuf,size_t count,loff_t * ppos)326 static ssize_t severities_coverage_write(struct file *file,
327 const char __user *ubuf,
328 size_t count, loff_t *ppos)
329 {
330 int i;
331 for (i = 0; i < ARRAY_SIZE(severities); i++)
332 severities[i].covered = 0;
333 return count;
334 }
335
336 static const struct file_operations severities_coverage_fops = {
337 .open = severities_coverage_open,
338 .release = seq_release,
339 .read = seq_read,
340 .write = severities_coverage_write,
341 .llseek = seq_lseek,
342 };
343
severities_debugfs_init(void)344 static int __init severities_debugfs_init(void)
345 {
346 struct dentry *dmce, *fsev;
347
348 dmce = mce_get_debugfs_dir();
349 if (!dmce)
350 goto err_out;
351
352 fsev = debugfs_create_file("severities-coverage", 0444, dmce, NULL,
353 &severities_coverage_fops);
354 if (!fsev)
355 goto err_out;
356
357 return 0;
358
359 err_out:
360 return -ENOMEM;
361 }
362 late_initcall(severities_debugfs_init);
363 #endif /* CONFIG_DEBUG_FS */
364