[PATCH v3 2/4] powerpc: Fixes for CONFIG_PTE_64BIT for SMP support

[Benjamin Herrenschmidt]

On Fri, 2008-09-05 at 14:44 -0500, Kumar Gala wrote:
> > I also wonder if you should first ensure that the PTE is invalid and
> > if not, clear it and flush the TLB page ... Or at least add a
> > WARN_ON(pte_valid()) in case we get that wrong ...
> 
> I believe that's an issue since kmap_atomic() will call set_pte_at and  
> have the valid bit set.

Hrm... on the other hand, it's safe because kmap_atomic() is per-CPU and
thus won't race with anything...

On the other hand, if others (mprotect, mremap, whoever...) does it,
it's not safe.

I'm keen on letting set_pte_at() to the job and maybe if we can remove
the explicit flush kmap_atomic does... Though it's not totally trivial
to "know" it's a flush that doesn't need global invalidations...

May be worth moving your current stuff into a __set_pte_at() that you
use from kmap atomic, and have set_pte_at() wrap that along with a
present warn. That would do for the time being.

Ben.