EEH, hotplug, bash scripts, error reporting.

[linas at austin.ibm.com]

Greg,

Is there any sort of architecture or policy w.r.t. hot-plug scripts
and how they relate to syslog and to RAS?

On the pSeries platform, hardware errors are supposed to go through
RTAS into NVRAM, which then somehow gets into the syslog.  Then some
sort of "Error Log Analysis" (ELA) tool parses the syslog entries,
and reports the error up to the "Remote Management Control".  Which
then sends an event to the Service Focal Point (SFP) on the Hardware
Management Console (HMC).  This is the current offical pSeries Linux
RAS architecture.

But I'm sitting here thinking "gee, here's my EEH code, its detected
a permanant failure on some PCI slot.  It can/it should invoke some
hotplug script".  So it occured to me that the whole /sbin/hotplug
architecture is completely orthogonal to the whole syslog architecture.
And the syslog architecture is orhtogonal to the pSeries RAS architecture.

I beleive there's some work on a next-gen syslogd (in sourceforge,
I forgot the name) which will allow the sysadmin to write shell
scripts to deal with various system events, such as hotplug events
as well as demonic icmp redirect packets.  But those events seem
to be orthogonal to the subsystem of hotplug events.  So at this point
my mind boggled.

Any words of insight on this state of affairs?

--linas

** Sent via the linuxppc64-dev mail list. See http://lists.linuxppc.org/