doczkal / rpms / abrt

Forked from rpms/abrt 4 years ago
Clone

Blame SOURCES/0024-doc-MCE_readme.txt-new-file-documentation-about-MCE-.patch

b404a9
From 95e9590bfee2df447c8f4c0fd799e8c514beca80 Mon Sep 17 00:00:00 2001
b404a9
From: Denys Vlasenko <dvlasenk@redhat.com>
b404a9
Date: Tue, 10 Dec 2013 13:07:35 +0100
b404a9
Subject: [ABRT PATCH 24/27] doc/MCE_readme.txt: new file - documentation about
b404a9
 MCE handling
b404a9
b404a9
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
b404a9
b404a9
Related to rhbz#1032077
b404a9
b404a9
Signed-off-by: Jakub Filak <jfilak@redhat.com>
b404a9
---
b404a9
 doc/MCE_readme.txt | 86 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
b404a9
 1 file changed, 86 insertions(+)
b404a9
 create mode 100644 doc/MCE_readme.txt
b404a9
b404a9
diff --git a/doc/MCE_readme.txt b/doc/MCE_readme.txt
b404a9
new file mode 100644
b404a9
index 0000000..ed5b627
b404a9
--- /dev/null
b404a9
+++ b/doc/MCE_readme.txt
b404a9
@@ -0,0 +1,86 @@
b404a9
+	Background
b404a9
+
b404a9
+MCEs can be fatal (they panic kernel) or not.
b404a9
+Fatal MCE are delivered as exception#18.
b404a9
+Non-fatal ones sometimes are delivered as exception#18; other times
b404a9
+they are silently recorded in magic MSRs, CPU is not alerted.
b404a9
+Linux kernel periodically (up to 5 mins interval) reads those MSRs
b404a9
+and if MCE is seen there, it is piped in binary form through
b404a9
+/dev/mcelog to whoever listens on it. (Such as mcelog tool in
b404a9
+--daemon mode; but cat 
b404a9
+
b404a9
+"Machine Check Exception:" message is printed *only* by fatal MCEs.
b404a9
+It will be caught as vmcore if kdump is configured.
b404a9
+
b404a9
+Non-fatal MCEs have "[Hardware Error]: Machine check events logged"
b404a9
+message in kernel log.
b404a9
+When /dev/mcelog is read, *no additional kernel log messages appear*.
b404a9
+
b404a9
+> Are those magic MSR registers cleared when read via /dev/mcelog?
b404a9
+
b404a9
+Yes.
b404a9
+
b404a9
+> Without mcelog utility, we can directly read only binary form, right?
b404a9
+> Not nice, but still useful, right?
b404a9
+> (could be transferred to nice text form on other machine).
b404a9
+
b404a9
+No, raw /dev/mcelog data is not easy to interpret on other machine.
b404a9
+In fact, it can't be used by mcelog tool even on the same machine.
b404a9
+Technical reason is that mcelog uses an obscure ioctl on /dev/mcelog
b404a9
+in order to know the size of binary blob with MCE information.
b404a9
+When run on a file, ioctl fails, and mcelog bombs out.
b404a9
+
b404a9
+Looks like without mcelog running and processing /dev/mcelog data,
b404a9
+non-fatal MCE's can't be easily decoded with currently existing tools.
b404a9
+
b404a9
+mcelog tool can be configured to write log to /var/log/mcelog
b404a9
+(RHEL6 does that) or to syslog (RHEL7 does that).
b404a9
+
b404a9
+
b404a9
+	How ABRT catches MCEs
b404a9
+
b404a9
+Fatal MCEs are caught as any fatal kernel panic is caught - as a vmcore.
b404a9
+The oops text, which goes to "backtrace" element, will be the decoded
b404a9
+MCE message from kernel log buffer.
b404a9
+
b404a9
+Non-fatal MCEs are caught as kernel oopses.
b404a9
+If "Machine check events logged" message is seen in "dmesg" element,
b404a9
+we assume it's a MCE, and create "not-reportable" element with suitable
b404a9
+explanation.
b404a9
+Then we check whether /var/log/mcelog exists,
b404a9
+or whether system log contains "mcelog: Hardware event",
b404a9
+and create a "comment" element with explanatory text, followed by
b404a9
+last 20 lines from either of those files.
b404a9
+
b404a9
+
b404a9
+	How to test MCEs
b404a9
+
b404a9
+There is an MCE injection tool and a kernel module, both named mce-inject.
b404a9
+(The tool comes from mce-test project, may be found in ras-utils RHEL7 package).
b404a9
+The script I used is:
b404a9
+
b404a9
+modprobe mce-inject
b404a9
+sync &
b404a9
+sleep 1
b404a9
+sync
b404a9
+# This can crash the machine:
b404a9
+echo "Injecting MCE from file $1"
b404a9
+mce-inject "$1"
b404a9
+echo "Exitcode:$?"
b404a9
+
b404a9
+It requires files which describe MCE to simulate. I grabbed a few examples
b404a9
+from mce-test.tar.gz (source tarball of mce-test project).
b404a9
+I used this this file to cause a non-fatal MCE:
b404a9
+
b404a9
+CPU 0 BANK 2
b404a9
+STATUS VAL OVER EN
b404a9
+
b404a9
+And this one to cause a fatal one:
b404a9
+
b404a9
+CPU 0 BANK 4
b404a9
+MCGSTATUS MCIP
b404a9
+STATUS FATAL S
b404a9
+RIP 12343434
b404a9
+MISC 11
b404a9
+
b404a9
+(Not sure what failures exactly they imitate, maybe there are better examples).
b404a9
-- 
b404a9
1.8.3.1
b404a9