On Thu, Mar 08, 2007 at 11:07:20PM +0100, Manuel Bouyer wrote:
>On Thu, Mar 08, 2007 at 05:00:09PM -0500, Steven M. Bellovin wrote:
>> On Thu, 8 Mar 2007 22:45:37 +0100
>> Manuel Bouyer <bouyer@antioche.eu.org> wrote:
>> 
>> > On Thu, Mar 08, 2007 at 04:34:55PM -0500, Perry E. Metzger wrote:
>> > > 
>> > > Manuel Bouyer <bouyer@antioche.eu.org> writes:
>> > > >> I'll amend that further, and make that suspend to swap.
>> > > >
>> > > > I would love this. But it's a way to large projet for SOC.
>> > > 
>> > > A framework for hibernation is not too large a project.
>> > 
>> > Maybe I'm missing somthing, but to resume from hibernation you'll load
>> > the kernel as a normal boot, and then restore state from a large file
>> > (or swap, or whatever). Then you need to be able to restore state of
>> > various parts of the kernel (sockets, open files, etc ...), which may
>> > not be at the same address as at hibernation time, because the kernel
>> > has done some work and rebooted in between. This is the difficult
>> > part, and may require redesign of large parts of the kernel. But
>> > maybe there's other way to get at it that I didn't see (doing the
>> > work from the boot loader ?)
>> 
>> You could do it by writing out all of memory to the swap file,
>> including the kernel text section.  This would work better, I think,
>> with a different boot loader design, where the boot loader either
>> pulled in /netbsd or restored from the hibernate (swap?)
>> file.  Alternatively you could assert that kernel changes simply
>> weren't allowed, and compared some magic number in the booted kernel to
>> that written out in the swap area -- put 16 bytes from /dev/random into
>> vers.c at kernel build time, for example.
>
>I was not talking about kernel change itself, just restoring the
>kernel's state. If the kernel itself does it you can't just block-copy
>from disk to ram. If something else does it - well, what's this
>something else ?

I understand that some of the newer CPU have functions to write
or read in their register state etc. Focusing on CPU with that
feature would be a lot easier.

// George


-- 
George Georgalis, systems architect, administrator <IXOYE><