Re: open(2) says O_DIRECT works on 512 byte boundries?

[Andrea Arcangeli <aarcange@redhat.com>]

On Mon, Feb 02, 2009 at 07:50:12PM -0800, Greg KH wrote:
> On Mon, Feb 02, 2009 at 11:08:56PM +0100, Andrea Arcangeli wrote:
> > Hi Greg!
> > 
> > > Thanks for the pointers, I'll go read the thread and follow up there.
> > 
> > If you also run into this final fix is attached below. Porting to
> > mainline is a bit hard because of gup-fast... Perhaps we can use mmu
> > notifiers to fix gup-fast... need to think more about it then I'll
> > post something.
> > 
> > Please help testing the below on pre-gup-fast kernels, thanks!
> 
> Thanks a lot for the patch, I'll try this out tomorrow.

before testing I recommend to make this change:

-       if ((flags & FOLL_WRITE) && PageAnon(page) && !PageGUP(page))
+	if (PageAnon(page) && !PageGUP(page))

If gup triggers cow, gup clears FOLL_WRITE from foll_flags before
calling follow_page again, I think that's why it didn't work right for
everyone, it just couldn't work right if FOLL_WRITE was going away
before the last follow_page run. I wanted to set PG_gup only for
gup(write=1) and not for gup(write=0) but that optimization that
clears FOLL_WRITE prevents the above to work right, so the simple
solution is to always set PG_gup, it won't make a big difference.

I'm looking at the hugetlb cow before posting an update...