arch/x86/kernel/cpu/mcheck/mce-severity.c

   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 KERNEL .context = IN_KERNEL
  47 #define USER .context = IN_USER
  48 #define SER .ser = SER_REQUIRED
  49 #define NOSER .ser = NO_SER
  50 #define SEV(s) .sev = MCE_ ## s ## _SEVERITY
  51 #define BITCLR(x, s, m, r...) { .mask = x, .result = 0, SEV(s), .msg = m, ## r }
  52 #define BITSET(x, s, m, r...) { .mask = x, .result = x, SEV(s), .msg = m, ## r }
  53 #define MCGMASK(x, res, s, m, r...) \
  54         { .mcgmask = x, .mcgres = res, SEV(s), .msg = m, ## r }
  55 #define MASK(x, y, s, m, r...) \
  56         { .mask = x, .result = y, SEV(s), .msg = m, ## r }
  57 #define MCI_UC_S (MCI_STATUS_UC|MCI_STATUS_S)
  58 #define MCI_UC_SAR (MCI_STATUS_UC|MCI_STATUS_S|MCI_STATUS_AR)
  59 #define MCACOD 0xffff
  60
  61         BITCLR(MCI_STATUS_VAL, NO, "Invalid"),
  62         BITCLR(MCI_STATUS_EN, NO, "Not enabled"),
  63         BITSET(MCI_STATUS_PCC, PANIC, "Processor context corrupt"),
  64         /* When MCIP is not set something is very confused */
  65         MCGMASK(MCG_STATUS_MCIP, 0, PANIC, "MCIP not set in MCA handler"),
  66         /* Neither return not error IP -- no chance to recover -> PANIC */
  67         MCGMASK(MCG_STATUS_RIPV|MCG_STATUS_EIPV, 0, PANIC,
  68                 "Neither restart nor error IP"),
  69         MCGMASK(MCG_STATUS_RIPV, 0, PANIC, "In kernel and no restart IP",
  70                 KERNEL),
  71         BITCLR(MCI_STATUS_UC, KEEP, "Corrected error", NOSER),
  72         MASK(MCI_STATUS_OVER|MCI_STATUS_UC|MCI_STATUS_EN, MCI_STATUS_UC, SOME,
  73              "Spurious not enabled", SER),
  74
  75         /* ignore OVER for UCNA */
  76         MASK(MCI_UC_SAR, MCI_STATUS_UC, KEEP,
  77              "Uncorrected no action required", SER),
  78         MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_UC|MCI_STATUS_AR, PANIC,
  79              "Illegal combination (UCNA with AR=1)", SER),
  80         MASK(MCI_STATUS_S, 0, KEEP, "Non signalled machine check", SER),
  81
  82         /* AR add known MCACODs here */
  83         MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_OVER|MCI_UC_SAR, PANIC,
  84              "Action required with lost events", SER),
  85         MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCACOD, MCI_UC_SAR, PANIC,
  86              "Action required; unknown MCACOD", SER),
  87
  88         /* known AO MCACODs: */
  89         MASK(MCI_UC_SAR|MCI_STATUS_OVER|0xfff0, MCI_UC_S|0xc0, AO,
  90              "Action optional: memory scrubbing error", SER),
  91         MASK(MCI_UC_SAR|MCI_STATUS_OVER|MCACOD, MCI_UC_S|0x17a, AO,
  92              "Action optional: last level cache writeback error", SER),
  93
  94         MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_S, SOME,
  95              "Action optional unknown MCACOD", SER),
  96         MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_S|MCI_STATUS_OVER, SOME,
  97              "Action optional with lost events", SER),
  98         BITSET(MCI_STATUS_UC|MCI_STATUS_OVER, PANIC, "Overflowed uncorrected"),
  99         BITSET(MCI_STATUS_UC, UC, "Uncorrected"),
 100         BITSET(0, SOME, "No match")     /* always matches. keep at end */
 101 };
 102
 103 /*
 104  * If the EIPV bit is set, it means the saved IP is the
 105  * instruction which caused the MCE.
 106  */
 107 static int error_context(struct mce *m)
 108 {
 109         if (m->mcgstatus & MCG_STATUS_EIPV)
 110                 return (m->ip && (m->cs & 3) == 3) ? IN_USER : IN_KERNEL;
 111         /* Unknown, assume kernel */
 112         return IN_KERNEL;
 113 }
 114
 115 int mce_severity(struct mce *a, int tolerant, char **msg)
 116 {
 117         enum context ctx = error_context(a);
 118         struct severity *s;
 119
 120         for (s = severities;; s++) {
 121                 if ((a->status & s->mask) != s->result)
 122                         continue;
 123                 if ((a->mcgstatus & s->mcgmask) != s->mcgres)
 124                         continue;
 125                 if (s->ser == SER_REQUIRED && !mce_ser)
 126                         continue;
 127                 if (s->ser == NO_SER && mce_ser)
 128                         continue;
 129                 if (s->context && ctx != s->context)
 130                         continue;
 131                 if (msg)
 132                         *msg = s->msg;
 133                 s->covered = 1;
 134                 if (s->sev >= MCE_UC_SEVERITY && ctx == IN_KERNEL) {
 135                         if (panic_on_oops || tolerant < 1)
 136                                 return MCE_PANIC_SEVERITY;
 137                 }
 138                 return s->sev;
 139         }
 140 }
 141
 142 #ifdef CONFIG_DEBUG_FS
 143 static void *s_start(struct seq_file *f, loff_t *pos)
 144 {
 145         if (*pos >= ARRAY_SIZE(severities))
 146                 return NULL;
 147         return &severities[*pos];
 148 }
 149
 150 static void *s_next(struct seq_file *f, void *data, loff_t *pos)
 151 {
 152         if (++(*pos) >= ARRAY_SIZE(severities))
 153                 return NULL;
 154         return &severities[*pos];
 155 }
 156
 157 static void s_stop(struct seq_file *f, void *data)
 158 {
 159 }
 160
 161 static int s_show(struct seq_file *f, void *data)
 162 {
 163         struct severity *ser = data;
 164         seq_printf(f, "%d\t%s\n", ser->covered, ser->msg);
 165         return 0;
 166 }
 167
 168 static const struct seq_operations severities_seq_ops = {
 169         .start  = s_start,
 170         .next   = s_next,
 171         .stop   = s_stop,
 172         .show   = s_show,
 173 };
 174
 175 static int severities_coverage_open(struct inode *inode, struct file *file)
 176 {
 177         return seq_open(file, &severities_seq_ops);
 178 }
 179
 180 static ssize_t severities_coverage_write(struct file *file,
 181                                          const char __user *ubuf,
 182                                          size_t count, loff_t *ppos)
 183 {
 184         int i;
 185         for (i = 0; i < ARRAY_SIZE(severities); i++)
 186                 severities[i].covered = 0;
 187         return count;
 188 }
 189
 190 static const struct file_operations severities_coverage_fops = {
 191         .open           = severities_coverage_open,
 192         .release        = seq_release,
 193         .read           = seq_read,
 194         .write          = severities_coverage_write,
 195 };
 196
 197 static int __init severities_debugfs_init(void)
 198 {
 199         struct dentry *dmce = NULL, *fseverities_coverage = NULL;
 200
 201         dmce = mce_get_debugfs_dir();
 202         if (dmce == NULL)
 203                 goto err_out;
 204         fseverities_coverage = debugfs_create_file("severities-coverage",
 205                                                    0444, dmce, NULL,
 206                                                    &severities_coverage_fops);
 207         if (fseverities_coverage == NULL)
 208                 goto err_out;
 209
 210         return 0;
 211
 212 err_out:
 213         return -ENOMEM;
 214 }
 215 late_initcall(severities_debugfs_init);
 216 #endif