• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1.. SPDX-License-Identifier: GPL-2.0
2
3===============================================================
4Configurable sysfs parameters for the x86-64 machine check code
5===============================================================
6
7Machine checks report internal hardware error conditions detected
8by the CPU. Uncorrected errors typically cause a machine check
9(often with panic), corrected ones cause a machine check log entry.
10
11Machine checks are organized in banks (normally associated with
12a hardware subsystem) and subevents in a bank. The exact meaning
13of the banks and subevent is CPU specific.
14
15mcelog knows how to decode them.
16
17When you see the "Machine check errors logged" message in the system
18log then mcelog should run to collect and decode machine check entries
19from /dev/mcelog. Normally mcelog should be run regularly from a cronjob.
20
21Each CPU has a directory in /sys/devices/system/machinecheck/machinecheckN
22(N = CPU number).
23
24The directory contains some configurable entries:
25
26bankNctl
27	(N bank number)
28
29	64bit Hex bitmask enabling/disabling specific subevents for bank N
30	When a bit in the bitmask is zero then the respective
31	subevent will not be reported.
32	By default all events are enabled.
33	Note that BIOS maintain another mask to disable specific events
34	per bank.  This is not visible here
35
36The following entries appear for each CPU, but they are truly shared
37between all CPUs.
38
39check_interval
40	How often to poll for corrected machine check errors, in seconds
41	(Note output is hexadecimal). Default 5 minutes.  When the poller
42	finds MCEs it triggers an exponential speedup (poll more often) on
43	the polling interval.  When the poller stops finding MCEs, it
44	triggers an exponential backoff (poll less often) on the polling
45	interval. The check_interval variable is both the initial and
46	maximum polling interval. 0 means no polling for corrected machine
47	check errors (but some corrected errors might be still reported
48	in other ways)
49
50tolerant
51	Tolerance level. When a machine check exception occurs for a non
52	corrected machine check the kernel can take different actions.
53	Since machine check exceptions can happen any time it is sometimes
54	risky for the kernel to kill a process because it defies
55	normal kernel locking rules. The tolerance level configures
56	how hard the kernel tries to recover even at some risk of
57	deadlock.  Higher tolerant values trade potentially better uptime
58	with the risk of a crash or even corruption (for tolerant >= 3).
59
60	0: always panic on uncorrected errors, log corrected errors
61	1: panic or SIGBUS on uncorrected errors, log corrected errors
62	2: SIGBUS or log uncorrected errors, log corrected errors
63	3: never panic or SIGBUS, log all errors (for testing only)
64
65	Default: 1
66
67	Note this only makes a difference if the CPU allows recovery
68	from a machine check exception. Current x86 CPUs generally do not.
69
70trigger
71	Program to run when a machine check event is detected.
72	This is an alternative to running mcelog regularly from cron
73	and allows to detect events faster.
74monarch_timeout
75	How long to wait for the other CPUs to machine check too on a
76	exception. 0 to disable waiting for other CPUs.
77	Unit: us
78
79TBD document entries for AMD threshold interrupt configuration
80
81For more details about the x86 machine check architecture
82see the Intel and AMD architecture manuals from their developer websites.
83
84For more details about the architecture
85see http://one.firstfloor.org/~andi/mce.pdf
86