3 * Copyright 2008, 2009 Intel Corporation.
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
12 #include <linux/kernel.h>
13 #include <linux/seq_file.h>
14 #include <linux/init.h>
15 #include <linux/debugfs.h>
17 #include <asm/uaccess.h>
19 #include "mce-internal.h"
22 * Grade an mce by severity. In general the most severe ones are processed
23 * first. Since there are quite a lot of combinations test the bits in a
24 * table-driven way. The rules are simply processed in order, first
27 * Note this is only used for machine check exceptions, the corrected
28 * errors use much simpler rules. The exceptions still check for the corrected
29 * errors, but only to leave them alone for the CMCI handler (except for
33 enum context { IN_KERNEL = 1, IN_USER = 2, IN_KERNEL_RECOV = 3 };
34 enum ser { SER_REQUIRED = 1, NO_SER = 2 };
35 enum exception { EXCP_CONTEXT = 1, NO_EXCP = 2 };
37 static struct severity {
41 unsigned char mcgmask;
44 unsigned char context;
46 unsigned char covered;
49 #define MCESEV(s, m, c...) { .sev = MCE_ ## s ## _SEVERITY, .msg = m, ## c }
50 #define KERNEL .context = IN_KERNEL
51 #define USER .context = IN_USER
52 #define KERNEL_RECOV .context = IN_KERNEL_RECOV
53 #define SER .ser = SER_REQUIRED
54 #define NOSER .ser = NO_SER
55 #define EXCP .excp = EXCP_CONTEXT
56 #define NOEXCP .excp = NO_EXCP
57 #define BITCLR(x) .mask = x, .result = 0
58 #define BITSET(x) .mask = x, .result = x
59 #define MCGMASK(x, y) .mcgmask = x, .mcgres = y
60 #define MASK(x, y) .mask = x, .result = y
61 #define MCI_UC_S (MCI_STATUS_UC|MCI_STATUS_S)
62 #define MCI_UC_SAR (MCI_STATUS_UC|MCI_STATUS_S|MCI_STATUS_AR)
63 #define MCI_ADDR (MCI_STATUS_ADDRV|MCI_STATUS_MISCV)
67 BITCLR(MCI_STATUS_VAL)
71 EXCP, BITCLR(MCI_STATUS_EN)
74 PANIC, "Processor context corrupt",
75 BITSET(MCI_STATUS_PCC)
77 /* When MCIP is not set something is very confused */
79 PANIC, "MCIP not set in MCA handler",
80 EXCP, MCGMASK(MCG_STATUS_MCIP, 0)
82 /* Neither return not error IP -- no chance to recover -> PANIC */
84 PANIC, "Neither restart nor error IP",
85 EXCP, MCGMASK(MCG_STATUS_RIPV|MCG_STATUS_EIPV, 0)
88 PANIC, "In kernel and no restart IP",
89 EXCP, KERNEL, MCGMASK(MCG_STATUS_RIPV, 0)
92 PANIC, "In kernel and no restart IP",
93 EXCP, KERNEL_RECOV, MCGMASK(MCG_STATUS_RIPV, 0)
96 DEFERRED, "Deferred error",
97 NOSER, MASK(MCI_STATUS_UC|MCI_STATUS_DEFERRED|MCI_STATUS_POISON, MCI_STATUS_DEFERRED)
100 KEEP, "Corrected error",
101 NOSER, BITCLR(MCI_STATUS_UC)
104 /* ignore OVER for UCNA */
106 UCNA, "Uncorrected no action required",
107 SER, MASK(MCI_UC_SAR, MCI_STATUS_UC)
110 PANIC, "Illegal combination (UCNA with AR=1)",
112 MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_UC|MCI_STATUS_AR)
115 KEEP, "Non signalled machine check",
116 SER, BITCLR(MCI_STATUS_S)
120 PANIC, "Action required with lost events",
121 SER, BITSET(MCI_STATUS_OVER|MCI_UC_SAR)
124 /* known AR MCACODs: */
125 #ifdef CONFIG_MEMORY_FAILURE
127 KEEP, "Action required but unaffected thread is continuable",
128 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR, MCI_UC_SAR|MCI_ADDR),
129 MCGMASK(MCG_STATUS_RIPV|MCG_STATUS_EIPV, MCG_STATUS_RIPV)
132 AR, "Action required: data load in error recoverable area of kernel",
133 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_DATA),
137 AR, "Action required: data load error in a user process",
138 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_DATA),
142 AR, "Action required: instruction fetch error in a user process",
143 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_INSTR),
148 PANIC, "Action required: unknown MCACOD",
149 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_SAR)
152 /* known AO MCACODs: */
154 AO, "Action optional: memory scrubbing error",
155 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCACOD_SCRUBMSK, MCI_UC_S|MCACOD_SCRUB)
158 AO, "Action optional: last level cache writeback error",
159 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCACOD, MCI_UC_S|MCACOD_L3WB)
162 SOME, "Action optional: unknown MCACOD",
163 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_S)
166 SOME, "Action optional with lost events",
167 SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_OVER|MCI_UC_S)
171 PANIC, "Overflowed uncorrected",
172 BITSET(MCI_STATUS_OVER|MCI_STATUS_UC)
176 BITSET(MCI_STATUS_UC)
181 ) /* always matches. keep at end */
184 #define mc_recoverable(mcg) (((mcg) & (MCG_STATUS_RIPV|MCG_STATUS_EIPV)) == \
185 (MCG_STATUS_RIPV|MCG_STATUS_EIPV))
188 * If mcgstatus indicated that ip/cs on the stack were
189 * no good, then "m->cs" will be zero and we will have
190 * to assume the worst case (IN_KERNEL) as we actually
191 * have no idea what we were executing when the machine
193 * If we do have a good "m->cs" (or a faked one in the
194 * case we were executing in VM86 mode) we can use it to
195 * distinguish an exception taken in user from from one
196 * taken in the kernel.
198 static int error_context(struct mce *m)
200 if ((m->cs & 3) == 3)
202 if (mc_recoverable(m->mcgstatus) && ex_has_fault_handler(m->ip))
203 return IN_KERNEL_RECOV;
208 * See AMD Error Scope Hierarchy table in a newer BKDG. For example
209 * 49125_15h_Models_30h-3Fh_BKDG.pdf, section "RAS Features"
211 static int mce_severity_amd(struct mce *m, int tolerant, char **msg, bool is_excp)
213 enum context ctx = error_context(m);
215 /* Processor Context Corrupt, no need to fumble too much, die! */
216 if (m->status & MCI_STATUS_PCC)
217 return MCE_PANIC_SEVERITY;
219 if (m->status & MCI_STATUS_UC) {
222 * On older systems where overflow_recov flag is not present, we
223 * should simply panic if an error overflow occurs. If
224 * overflow_recov flag is present and set, then software can try
225 * to at least kill process to prolong system operation.
227 if (mce_flags.overflow_recov) {
228 /* software can try to contain */
229 if (!(m->mcgstatus & MCG_STATUS_RIPV) && (ctx == IN_KERNEL))
230 return MCE_PANIC_SEVERITY;
232 /* kill current process */
233 return MCE_AR_SEVERITY;
235 /* at least one error was not logged */
236 if (m->status & MCI_STATUS_OVER)
237 return MCE_PANIC_SEVERITY;
241 * For any other case, return MCE_UC_SEVERITY so that we log the
242 * error and exit #MC handler.
244 return MCE_UC_SEVERITY;
248 * deferred error: poll handler catches these and adds to mce_ring so
249 * memory-failure can take recovery actions.
251 if (m->status & MCI_STATUS_DEFERRED)
252 return MCE_DEFERRED_SEVERITY;
255 * corrected error: poll handler catches these and passes responsibility
256 * of decoding the error to EDAC
258 return MCE_KEEP_SEVERITY;
261 static int mce_severity_intel(struct mce *m, int tolerant, char **msg, bool is_excp)
263 enum exception excp = (is_excp ? EXCP_CONTEXT : NO_EXCP);
264 enum context ctx = error_context(m);
267 for (s = severities;; s++) {
268 if ((m->status & s->mask) != s->result)
270 if ((m->mcgstatus & s->mcgmask) != s->mcgres)
272 if (s->ser == SER_REQUIRED && !mca_cfg.ser)
274 if (s->ser == NO_SER && mca_cfg.ser)
276 if (s->context && ctx != s->context)
278 if (s->excp && excp != s->excp)
283 if (s->sev >= MCE_UC_SEVERITY && ctx == IN_KERNEL) {
284 if (panic_on_oops || tolerant < 1)
285 return MCE_PANIC_SEVERITY;
291 /* Default to mce_severity_intel */
292 int (*mce_severity)(struct mce *m, int tolerant, char **msg, bool is_excp) =
295 void __init mcheck_vendor_init_severity(void)
297 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
298 mce_severity = mce_severity_amd;
301 #ifdef CONFIG_DEBUG_FS
302 static void *s_start(struct seq_file *f, loff_t *pos)
304 if (*pos >= ARRAY_SIZE(severities))
306 return &severities[*pos];
309 static void *s_next(struct seq_file *f, void *data, loff_t *pos)
311 if (++(*pos) >= ARRAY_SIZE(severities))
313 return &severities[*pos];
316 static void s_stop(struct seq_file *f, void *data)
320 static int s_show(struct seq_file *f, void *data)
322 struct severity *ser = data;
323 seq_printf(f, "%d\t%s\n", ser->covered, ser->msg);
327 static const struct seq_operations severities_seq_ops = {
334 static int severities_coverage_open(struct inode *inode, struct file *file)
336 return seq_open(file, &severities_seq_ops);
339 static ssize_t severities_coverage_write(struct file *file,
340 const char __user *ubuf,
341 size_t count, loff_t *ppos)
344 for (i = 0; i < ARRAY_SIZE(severities); i++)
345 severities[i].covered = 0;
349 static const struct file_operations severities_coverage_fops = {
350 .open = severities_coverage_open,
351 .release = seq_release,
353 .write = severities_coverage_write,
357 static int __init severities_debugfs_init(void)
359 struct dentry *dmce, *fsev;
361 dmce = mce_get_debugfs_dir();
365 fsev = debugfs_create_file("severities-coverage", 0444, dmce, NULL,
366 &severities_coverage_fops);
375 late_initcall(severities_debugfs_init);
376 #endif /* CONFIG_DEBUG_FS */