mced is a very small daemon which monitors /dev/mcelog for machine check events, and then reports them to other applications.
Machine checks can indicate failing hardware, system overheats, bad DIMMs or other problems. Some MCEs are 'fatal' and can't generally be survived. Many, however are not fatal, and mced can help you detect and report them more accurately.
Notes:
-
Kernels prior to 2.6.22 do not support poll() on /dev/mcelog. In that case, the polling intervals of the kernel and of mced will limit the rate at which MCEs can be logged. To get better resolution, decrease these intervals.
-
Not every system tracks a "current boot number". For those that do, the -b (--bootnum) option can be used. Tracking a boot number can help to identify crashes that might be related to MCEs.
-
mced does not try to decode MCEs. If you want that, you should try Andi Kleen's mcelog tool (https://www.mcelog.org/) or the simple generic MCE decoder included in this package (mce_decode).
-
The latest code for this project can be found at the github site: https://github.com/thockin/mcedaemon The mced-devel mailing list is where development discussions and patches get sent: http://groups.google.com/group/mced-devel
-
mced supports sending events over D-Bus. See the mced man page for details or the mce_listen code for specifics on how to receive and decode these events. Building with D-Bus support requires development packages for glib, gobject, gtype, pcre, dbus and dbus-glib.
-
If there is something you need out of mced, or if something is not working right, please contact me - thockin@hockin.org.