Re: [RFC] should VM_BUG_ON(cond) really evaluate cond

[Ben Hutchings <bhutchings@solarflare.com>]

On Thu, 2011-10-27 at 18:34 -0700, Linus Torvalds wrote:
> On Thu, Oct 27, 2011 at 6:25 PM, Andi Kleen <andi@firstfloor.org> wrote:
> >
> > Seems reasonable too. In fact we usually should have memory barriers
> > for this anyways which obsolete the volatile.
> 
> No we shouldn't. Memory barriers are insanely expensive, and pointless
> for atomics - that aren't ordered anyway.
> 
> You may mean compiler barriers.
> 
> That said, removing the volatile entirely might be a good idea, and
> never mind any barriers at all. The ordering for atomics really isn't
> well enough specified that we should care. So I wouldn't object to a
> patch that just removes the volatile entirely, but it would have to be
> accompanied with quite a bit of testing, in case some odd case ends up
> depending on it. But nothing *should* be looping on those things
> anyway.

Whether or not it needs to provide any ordering guarantee, atomic_read()
must never read more than once, and I think that requires the volatile
qualification.  It might be clearer to use the ACCESS_ONCE macro,
however.

Ben.

-- 
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.