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 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 #endif
 136         MCESEV(
 137                 PANIC, "Action required: unknown MCACOD",
 138                 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_SAR)
 139                 ),
 140
 141         /* known AO MCACODs: */
 142         MCESEV(
 143                 AO, "Action optional: memory scrubbing error",
 144                 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCACOD_SCRUBMSK, MCI_UC_S|MCACOD_SCRUB)
 145                 ),
 146         MCESEV(
 147                 AO, "Action optional: last level cache writeback error",
 148                 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCACOD, MCI_UC_S|MCACOD_L3WB)
 149                 ),
 150         MCESEV(
 151                 SOME, "Action optional: unknown MCACOD",
 152                 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_S)
 153                 ),
 154         MCESEV(
 155                 SOME, "Action optional with lost events",
 156                 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_OVER|MCI_UC_S)
 157                 ),
 158
 159         MCESEV(
 160                 PANIC, "Overflowed uncorrected",
 161                 BITSET(MCI_STATUS_OVER|MCI_STATUS_UC)
 162                 ),
 163         MCESEV(
 164                 UC, "Uncorrected",
 165                 BITSET(MCI_STATUS_UC)
 166                 ),
 167         MCESEV(
 168                 SOME, "No match",
 169                 BITSET(0)
 170                 )       /* always matches. keep at end */
 171 };
 172
 173 /*
 174  * If mcgstatus indicated that ip/cs on the stack were
 175  * no good, then "m->cs" will be zero and we will have
 176  * to assume the worst case (IN_KERNEL) as we actually
 177  * have no idea what we were executing when the machine
 178  * check hit.
 179  * If we do have a good "m->cs" (or a faked one in the
 180  * case we were executing in VM86 mode) we can use it to
 181  * distinguish an exception taken in user from from one
 182  * taken in the kernel.
 183  */
 184 static int error_context(struct mce *m)
 185 {
 186         return ((m->cs & 3) == 3) ? IN_USER : IN_KERNEL;
 187 }
 188
 189 int mce_severity(struct mce *m, int tolerant, char **msg, bool is_excp)
 190 {
 191         enum exception excp = (is_excp ? EXCP_CONTEXT : NO_EXCP);
 192         enum context ctx = error_context(m);
 193         struct severity *s;
 194
 195         for (s = severities;; s++) {
 196                 if ((m->status & s->mask) != s->result)
 197                         continue;
 198                 if ((m->mcgstatus & s->mcgmask) != s->mcgres)
 199                         continue;
 200                 if (s->ser == SER_REQUIRED && !mca_cfg.ser)
 201                         continue;
 202                 if (s->ser == NO_SER && mca_cfg.ser)
 203                         continue;
 204                 if (s->context && ctx != s->context)
 205                         continue;
 206                 if (s->excp && excp != s->excp)
 207                         continue;
 208                 if (msg)
 209                         *msg = s->msg;
 210                 s->covered = 1;
 211                 if (s->sev >= MCE_UC_SEVERITY && ctx == IN_KERNEL) {
 212                         if (panic_on_oops || tolerant < 1)
 213                                 return MCE_PANIC_SEVERITY;
 214                 }
 215                 return s->sev;
 216         }
 217 }
 218
 219 #ifdef CONFIG_DEBUG_FS
 220 static void *s_start(struct seq_file *f, loff_t *pos)
 221 {
 222         if (*pos >= ARRAY_SIZE(severities))
 223                 return NULL;
 224         return &severities[*pos];
 225 }
 226
 227 static void *s_next(struct seq_file *f, void *data, loff_t *pos)
 228 {
 229         if (++(*pos) >= ARRAY_SIZE(severities))
 230                 return NULL;
 231         return &severities[*pos];
 232 }
 233
 234 static void s_stop(struct seq_file *f, void *data)
 235 {
 236 }
 237
 238 static int s_show(struct seq_file *f, void *data)
 239 {
 240         struct severity *ser = data;
 241         seq_printf(f, "%d\t%s\n", ser->covered, ser->msg);
 242         return 0;
 243 }
 244
 245 static const struct seq_operations severities_seq_ops = {
 246         .start  = s_start,
 247         .next   = s_next,
 248         .stop   = s_stop,
 249         .show   = s_show,
 250 };
 251
 252 static int severities_coverage_open(struct inode *inode, struct file *file)
 253 {
 254         return seq_open(file, &severities_seq_ops);
 255 }
 256
 257 static ssize_t severities_coverage_write(struct file *file,
 258                                          const char __user *ubuf,
 259                                          size_t count, loff_t *ppos)
 260 {
 261         int i;
 262         for (i = 0; i < ARRAY_SIZE(severities); i++)
 263                 severities[i].covered = 0;
 264         return count;
 265 }
 266
 267 static const struct file_operations severities_coverage_fops = {
 268         .open           = severities_coverage_open,
 269         .release        = seq_release,
 270         .read           = seq_read,
 271         .write          = severities_coverage_write,
 272         .llseek         = seq_lseek,
 273 };
 274
 275 static int __init severities_debugfs_init(void)
 276 {
 277         struct dentry *dmce, *fsev;
 278
 279         dmce = mce_get_debugfs_dir();
 280         if (!dmce)
 281                 goto err_out;
 282
 283         fsev = debugfs_create_file("severities-coverage", 0444, dmce, NULL,
 284                                    &severities_coverage_fops);
 285         if (!fsev)
 286                 goto err_out;
 287
 288         return 0;
 289
 290 err_out:
 291         return -ENOMEM;
 292 }
 293 late_initcall(severities_debugfs_init);
 294 #endif /* CONFIG_DEBUG_FS */