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
35 static struct severity {
36 u64 mask;
37 u64 result;
38 unsigned char sev;
39 unsigned char mcgmask;
40 unsigned char mcgres;
41 unsigned char ser;
42 unsigned char context;
43 unsigned char covered;
44 char *msg;
45 } severities[] = {
46 #define MCESEV(s, m, c...) { .sev = MCE_ ## s ## _SEVERITY, .msg = m, ## c }
47 #define KERNEL .context = IN_KERNEL
48 #define USER .context = IN_USER
49 #define SER .ser = SER_REQUIRED
50 #define NOSER .ser = NO_SER
51 #define BITCLR(x) .mask = x, .result = 0
52 #define BITSET(x) .mask = x, .result = x
53 #define MCGMASK(x, y) .mcgmask = x, .mcgres = y
54 #define MASK(x, y) .mask = x, .result = y
55 #define MCI_UC_S (MCI_STATUS_UC|MCI_STATUS_S)
56 #define MCI_UC_SAR (MCI_STATUS_UC|MCI_STATUS_S|MCI_STATUS_AR)
57 #define MCACOD 0xffff
58
59 MCESEV(
60 NO, "Invalid",
61 BITCLR(MCI_STATUS_VAL)
62 ),
63 MCESEV(
64 NO, "Not enabled",
65 BITCLR(MCI_STATUS_EN)
66 ),
67 MCESEV(
68 PANIC, "Processor context corrupt",
69 BITSET(MCI_STATUS_PCC)
70 ),
71 /* When MCIP is not set something is very confused */
72 MCESEV(
73 PANIC, "MCIP not set in MCA handler",
74 MCGMASK(MCG_STATUS_MCIP, 0)
75 ),
76 /* Neither return not error IP -- no chance to recover -> PANIC */
77 MCESEV(
78 PANIC, "Neither restart nor error IP",
79 MCGMASK(MCG_STATUS_RIPV|MCG_STATUS_EIPV, 0)
80 ),
81 MCESEV(
82 PANIC, "In kernel and no restart IP",
83 KERNEL, MCGMASK(MCG_STATUS_RIPV, 0)
84 ),
85 MCESEV(
86 KEEP, "Corrected error",
87 NOSER, BITCLR(MCI_STATUS_UC)
88 ),
89
90 /* ignore OVER for UCNA */
91 MCESEV(
92 KEEP, "Uncorrected no action required",
93 SER, MASK(MCI_UC_SAR, MCI_STATUS_UC)
94 ),
95 MCESEV(
96 PANIC, "Illegal combination (UCNA with AR=1)",
97 SER,
98 MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_UC|MCI_STATUS_AR)
99 ),
100 MCESEV(
101 KEEP, "Non signalled machine check",
102 SER, BITCLR(MCI_STATUS_S)
103 ),
104
105 /* AR add known MCACODs here */
106 MCESEV(
107 PANIC, "Action required with lost events",
108 SER, BITSET(MCI_STATUS_OVER|MCI_UC_SAR)
109 ),
110 MCESEV(
111 PANIC, "Action required: unknown MCACOD",
112 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_SAR)
113 ),
114
115 /* known AO MCACODs: */
116 MCESEV(
117 AO, "Action optional: memory scrubbing error",
118 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|0xfff0, MCI_UC_S|0x00c0)
119 ),
120 MCESEV(
121 AO, "Action optional: last level cache writeback error",
122 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCACOD, MCI_UC_S|0x017a)
123 ),
124 MCESEV(
125 SOME, "Action optional: unknown MCACOD",
126 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_S)
127 ),
128 MCESEV(
129 SOME, "Action optional with lost events",
130 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_OVER|MCI_UC_S)
131 ),
132
133 MCESEV(
134 PANIC, "Overflowed uncorrected",
135 BITSET(MCI_STATUS_OVER|MCI_STATUS_UC)
136 ),
137 MCESEV(
138 UC, "Uncorrected",
139 BITSET(MCI_STATUS_UC)
140 ),
141 MCESEV(
142 SOME, "No match",
143 BITSET(0)
144 ) /* always matches. keep at end */
145 };
146
147 /*
148 * If the EIPV bit is set, it means the saved IP is the
149 * instruction which caused the MCE.
150 */
error_context(struct mce * m)151 static int error_context(struct mce *m)
152 {
153 if (m->mcgstatus & MCG_STATUS_EIPV)
154 return (m->ip && (m->cs & 3) == 3) ? IN_USER : IN_KERNEL;
155 /* Unknown, assume kernel */
156 return IN_KERNEL;
157 }
158
mce_severity(struct mce * m,int tolerant,char ** msg)159 int mce_severity(struct mce *m, int tolerant, char **msg)
160 {
161 enum context ctx = error_context(m);
162 struct severity *s;
163
164 for (s = severities;; s++) {
165 if ((m->status & s->mask) != s->result)
166 continue;
167 if ((m->mcgstatus & s->mcgmask) != s->mcgres)
168 continue;
169 if (s->ser == SER_REQUIRED && !mce_ser)
170 continue;
171 if (s->ser == NO_SER && mce_ser)
172 continue;
173 if (s->context && ctx != s->context)
174 continue;
175 if (msg)
176 *msg = s->msg;
177 s->covered = 1;
178 if (s->sev >= MCE_UC_SEVERITY && ctx == IN_KERNEL) {
179 if (panic_on_oops || tolerant < 1)
180 return MCE_PANIC_SEVERITY;
181 }
182 return s->sev;
183 }
184 }
185
186 #ifdef CONFIG_DEBUG_FS
s_start(struct seq_file * f,loff_t * pos)187 static void *s_start(struct seq_file *f, loff_t *pos)
188 {
189 if (*pos >= ARRAY_SIZE(severities))
190 return NULL;
191 return &severities[*pos];
192 }
193
s_next(struct seq_file * f,void * data,loff_t * pos)194 static void *s_next(struct seq_file *f, void *data, loff_t *pos)
195 {
196 if (++(*pos) >= ARRAY_SIZE(severities))
197 return NULL;
198 return &severities[*pos];
199 }
200
s_stop(struct seq_file * f,void * data)201 static void s_stop(struct seq_file *f, void *data)
202 {
203 }
204
s_show(struct seq_file * f,void * data)205 static int s_show(struct seq_file *f, void *data)
206 {
207 struct severity *ser = data;
208 seq_printf(f, "%d\t%s\n", ser->covered, ser->msg);
209 return 0;
210 }
211
212 static const struct seq_operations severities_seq_ops = {
213 .start = s_start,
214 .next = s_next,
215 .stop = s_stop,
216 .show = s_show,
217 };
218
severities_coverage_open(struct inode * inode,struct file * file)219 static int severities_coverage_open(struct inode *inode, struct file *file)
220 {
221 return seq_open(file, &severities_seq_ops);
222 }
223
severities_coverage_write(struct file * file,const char __user * ubuf,size_t count,loff_t * ppos)224 static ssize_t severities_coverage_write(struct file *file,
225 const char __user *ubuf,
226 size_t count, loff_t *ppos)
227 {
228 int i;
229 for (i = 0; i < ARRAY_SIZE(severities); i++)
230 severities[i].covered = 0;
231 return count;
232 }
233
234 static const struct file_operations severities_coverage_fops = {
235 .open = severities_coverage_open,
236 .release = seq_release,
237 .read = seq_read,
238 .write = severities_coverage_write,
239 .llseek = seq_lseek,
240 };
241
severities_debugfs_init(void)242 static int __init severities_debugfs_init(void)
243 {
244 struct dentry *dmce, *fsev;
245
246 dmce = mce_get_debugfs_dir();
247 if (!dmce)
248 goto err_out;
249
250 fsev = debugfs_create_file("severities-coverage", 0444, dmce, NULL,
251 &severities_coverage_fops);
252 if (!fsev)
253 goto err_out;
254
255 return 0;
256
257 err_out:
258 return -ENOMEM;
259 }
260 late_initcall(severities_debugfs_init);
261 #endif /* CONFIG_DEBUG_FS */
262