Re: [3/4] [PATCH]Diskdump - yet another crash dump function

[Keith Owens <kaos@sgi.com>]

On Thu, 17 Jun 2004 14:13:56 +0200, 
Ingo Molnar <mingo@elte.hu> wrote:
>but there's another possible method (suggested by Alan Cox) that
>requires no changes to the timer API hotpaths: 'clear' all timer lists
>upon a crash [once all CPUs have stopped and irqs are disabled] and just
>let the drivers use the normal timer APIs. Drive timer execution via a
>polling method.
>
>this basically approximates your polling based implementation but uses
>the existing kernel timer data structures and timer mechanism so should
>be robust and compatible. It doesnt rely on any previous state (because
>all currently pending timers are discarded) so it's as crash-safe as
>possible.

Don't forget live crash dumping.  The system is running and is behaving
strangely so you want to take a dump for investigation, but you do not
want to kill the system afterwards.  Live crash dumping is very useful
for problem diagnosis.

It is a little more complex than dumping after an oops because you must
not destroy any kernel data, including timer lists.