RFC: Hugepage COW

[David Gibson]

On Thu, Jul 14, 2005 at 10:24:33AM -0700, Christoph Lameter wrote:
> On Thu, 7 Jul 2005, David Gibson wrote:
> 
> > Now that the hugepage code has been consolidated across the
> > architectures, it becomes much easier to implement copy-on-write.
> > Hugepage COW is of limited utility of itself, however, it is
> > essentially a prerequisite for any of a number of methods of allowing
> > userland programs to automatically use hugepages without code changes
> > e.g. hugepage malloc() libraries, implicit hugepage mmap(), hugepage
> > ELF segments.  For certain applications (particularly enormous HPC
> > FORTRAN programs), these can result in a large performance
> > improvement.
> > 
> > Thoughts?  Flames?
> 
> Great stuff. I am glad that you are cleaning up the hugepages and are 
> making progress improving them. What are your thoughts on implementing 
> fault handling for huge pages?

Well, the COW patch implements a fault handler, obviously.  What
specifically where you thinking about?

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/people/dgibson