[Ksummit-2012-discuss] [ATTEND] kernel core dump and "dying breath"

Jason Wessel jason.wessel at windriver.com
Thu Jun 21 15:05:10 UTC 2012


On 06/21/2012 09:45 AM, Cong Wang wrote:
> 
> Yeah, netconsole requires the UDP network stack and network interface 
> function well, if the panic/oops just happened in the network transmit 
> path, it certainly doesn't work.


The scope of places you can end up with no dump is far wider.  You can
can crash in any arbitrary irq context or have certain locks held and
not be able to get a netconsole dump.

The issue is that the network devices have state and with the ability
to pre-empt irq handlers, you can end up often with the circumstance
where the HW cannot be used to transmit the packets.  We don't have
any kind of auxilary stack or pre-allocated SKBs for use with the
existing NET POLL implementation so if you happen to also need a
buffer or clear a buffer etc... it is just another case of no trace
for you. :-)


>>
>> For the really tricky sorts of problems like memory corruption
>> however, the addresses tend to move around so this is not likely to
>> help much.  I tend to fall back to kdb, and the "kdb death script" (a
>> toy of mine that is not in the mainline), where you can assign an
>> action to output all the commands you would have other wise typed, and
>> then reboot.
> 
> 
> I know nearly nothing about kdb, it is good to know kdb could help this 
> topic as well!
> 


KDB is has been around a long time, but not in the mainline for too
long.  Seems that people are still using it, and perhaps it can be
extended a bit more to make it work a bit more cohesively with the
proposed "dying breath".



More information about the Ksummit-2012-discuss mailing list