Re: [RFC] page fault retry with NOPAGE_RETRY

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Fri, 2006-09-15 at 14:30 +0100, Hugh Dickins wrote:
> On Fri, 15 Sep 2006, Andrew Morton wrote:
> > 
> > This assumes that no other heavyweight process will try to modify this
> > single-threaded process's mm.  I don't _think_ that happens anywhere, does
> > it?  access_process_vm() is the only case I can think of,
> 
> "Modify" in the sense of fault into.
> Yes, access_process_vm() is all I can think of too.
> 
> > and it does down_read(other process's mmap_sem).
> 
> If there were anything else, it'd have to do so too (if not down_write).
> 
> I too like NOPAGE_RETRY: as you've both observed, it can help to solve
> several different problems.

Yes, I don't need any of the safeguards that Andrew mentioned in my case
though. I want to return all the way to userland because I want signals
to be handled (which might also be a good thing in your case in fact, so
that a process being starved by that new mecanism can still be
interrupted).

I would ask that if you decide that the more complex approach is not
2.6.19 material, that the simple addition of NOPAGE_RETRY as I've
defined could be included in a first step so I can solve my problem (and
possibly other drivers wanting to do funky things with no_page() and
still take signals), and the google patch be rebased on top of that for
additional simmering :)

Cheers,
Ben.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/