Re: [patch 2/2] fix SMP data race in pagetable setup vs walking

[Nick Piggin <npiggin@suse.de>]

On Tue, May 06, 2008 at 12:11:53PM -0700, Paul E. McKenney wrote:
> On Tue, May 06, 2008 at 07:53:23AM -0700, Linus Torvalds wrote:
> > 
> > 
> > On Tue, 6 May 2008, Nick Piggin wrote:
> > > 
> > > Right. As the comment says, the x86 stuff is kind of a "reference"
> > > implementation, although if you prefer it isn't there, then I I can
> > > easily just make it alpha only.
> > 
> > If there really was a point in teaching people about 
> > "read_barrier_depends()", I'd agree that it's probably good to have it as 
> > a reference in the x86 implementation.
> > 
> > But since alpha is the only one that needs it, and is likely to remain so, 
> > it's not like we ever want to copy that code to anything else, and it 
> > really is better to make it alpha-only if the code is so much uglier.
> > 
> > Maybe just a comment?
> > 
> > As to the ACCESS_ONCE() thing, thinking about it some more, I doubt it 
> > really matters. We're never going to change pgd anyway, so who cares if we 
> > access it once or a hundred times?
> 
> If we are never going to change mm->pgd, then why do we need the
> smp_read_barrier_depends()?  Is this handling the initialization
> case or some such?

No, I had another look and I think Linus is correct. We don't need it for
mm->pgd.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>