Re: [PATCH] ppc64: Fix possible race with set_pte on a present PTE

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

On Wed, 2004-05-26 at 14:54, Linus Torvalds wrote:
> On Wed, 26 May 2004, Benjamin Herrenschmidt wrote:
> >
> > Ok, your patch just missed the do_wp_page() case which needs the same
> > medication as break_cow(), which leaves us with only one caller of
> > ptep_establish... the one which just sets those 2 bits and shouldn't
> > be named ptep establish at all ;)
> 
> Actually, the do_wp_page() one doesn't change the page, but it potentially
> changes _three_ bits: accessed, dirty _and_ writable.

Hrm... that would work if we only ever set writable, so yes.

> But the fix on ppc64 should be to add the writable bit to the mask of 
> things, and rename the function to reflect that fact.
> 
> > I'd rather have ptep_establish do the ptep_clear_flush & set_pte, and
> > the later just do the bit flip, but then, the arch (and possibly
> > generic) implementation of that bit flip must also do the TLB flush
> > when necessary (it's not on ppc).
> 
> I agree on the renaming, but please don't use the ptep_clear_flush() in 
> do_wp_page(), because it really isn't changing the virtual mapping, it's 
> only a (slightly) extended case of the same old "set another bit".

Ok, I just wasn't sure what this function was all about.

I'll cook a new patch including all the changes we discussed, may take some
time, but hopefully you'll get it today.

Ben.