On Aug 14, 2006, at 6:07 PM, Martin Husemann wrote:

> On Mon, Aug 14, 2006 at 10:12:18PM +0200, Martin Husemann wrote:
>> Besides wedges not autoconfiguring the dump device (which is pretty  
>> easy to
>> solve and testable w/o writing a dump all over your disk ;-})
>
> Here is a complete patch, including the autoconfig parts.
> I haven't tried writing a kernel dump with it yet.

I don't really like the autoconfig parts.  I would rather not export  
the notion of "finding swap" from the wedge code itself.  Honestly,  
I'm not sure I see the value in auto-configuring the dump device  
anymore.  Let it be configured during early system start-up.  I think  
we should remove all that code completely from the kernel.

As far as writing the dump is concerned:

1- The lockmgr stuff is going to be interesting -- what if the kernel  
panic'd while the lock was held, etc.  I guess this is a general  
problem with crash dumps anyway -- an awful lot of the kernel needs to  
work in order to be able to use them.

2- Yes, you do need to open the device, because you need to make an  
open call to the backing device (it might need to spin up the media,  
or whatever...)

-- thorpej