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

[Nobuhiro Tachino <ntachino@redhat.com>]

Jeff Moyer wrote:
> ==> Regarding Re: [3/4] [PATCH]Diskdump - yet another crash dump function; Ingo Molnar <mingo@elte.hu> adds:
> 
> mingo> * Christoph Hellwig <hch@infradead.org> wrote:
> 
> 
>>>Btw, now that we got you in the loop, any chance to see a forward-port
>>>of netdump to 2.6?  I think diskdump and netdump could share a lot of
>>>infrastructure, and given we already have the net polling hooks adding
>>>netdump shouldn't be that much work anymore.
> 
> 
> mingo> i think a forward port of netdump might already exist - Jeff, Dave?
> 
> Yes, I ported the code forward to 2.6.  The netpoll infrastructure needed a
> little tweaking to accommodate netdump, but nothing major.  Namely, we need
> to reset some locks, and I added an element to the netpoll data structure
> for the dump function.  I also updated the zap_completion_queue function to
> touch the nmi watchdog.
> 
> mingo> i agree that netdump and diskdump should be merged. (Red Hat is
> mingo> involved in the diskdump project too so this is an ultimate goal
> mingo> even though the patches are divergent.) Basically diskdumping is
> mingo> another IO transport - the format, userspace tools and much of the
> mingo> non-IO kernel mechanism is shared. Diskdumping is more complex on
> mingo> the driver level and it also needs to be more careful because it
> mingo> writes to media so it verifies various assumptions by reading
> mingo> on-disk sectors before writing to the area.
> 
> I'm not quite sure what infrastructure would be shared between the two.
> Page selection, perhaps?  Anything else?

I think freezing other CPUs code can be shared, but I can't find more.