Re: URGENT: [diagnostic] cache_lock: blocked on 0xda5265a8 ""

[YONETANI Tomokazu <qhwt+dfly@xxxxxxxxxx>]

On Sun, Aug 19, 2007 at 02:21:51PM +0200, Simon 'corecode' Schubert wrote:
> YONETANI Tomokazu wrote:
> >>I didn't look at the code, but you probably check for the dumpdev to be 
> >>used?  Overwriting swap with memory contents would be disastrous.
> >All it does is call dumpsys() and decrement the variable `dumping'
> >which gets incremented before calling dev_ddump().  dumpsys() already
> >checks to see if dumpdev is set.
> 
> yes, but usually it wouldn't care that dumpdev == swapdev.  when continuing 
> to run, this is rather bad.

A-ha, I see that.  And maybe the dump routine in some drivers is not
even written to resume from dumpsys()?

> but thinking about it... why wouldn't you just do a dd if=/dev/kmem 
> of=mycore?

Well, the system stops most of activities during the dump, so I thought
I can have more consistent dump than dd'ing from /dev/mem .

> maybe add some smp synchronisation, and that's it?  I usually go and debug 
> live machines with just kgdb, attaching to the live memory image.

I haven't used kgdb that way, does it take care of processes appearing
or disappearing during debug?

Cheers.