On Fri, Apr 28, 2006 at 04:27:56AM -0400, der Mouse wrote:
> 
> Of course, this doesn't help the case where your rootdev isn't present
> in /dev (whether because you booted from somewhere unusual or because
> /dev is not fully populated), which is when you need it most.  I'll
> definitely have to look at fixing it.

Or when I've got around to modifying /boot so that you can load a kernel
out of one filesystem, and pass the 'root' device as a different filesystem.
(This is useful for test booting kernels from $(OBJ)/arch/compile/xxx/netbsd
when that isn't part of the root filesystem.)

The other problem is what 'savecore' should do when the kernel was loaded
over the network.  (Another test senario where you netboot a kernel but
use a local disk for the root filesystem.)

It seems that putting the address (and maybe some other parameters) for
the in-kernel symbol table into the dump header so that 'kmem grovellers'
can find the real symbol table is the best solution.

Although I'm not sure where the symbol table itself comes from ?
(Is it loaded as part of the kernel image by pxeboot etc?)

	David

-- 
David Laight: david@l8s.co.uk