Re: Behaviour of smp_mb__{before,after}_spin* and acquire/release

[Will Deacon <will.deacon@arm.com>]

On Tue, Jan 20, 2015 at 09:35:45PM +0000, Paul E. McKenney wrote:
> On Tue, Jan 20, 2015 at 10:34:43AM +0100, Peter Zijlstra wrote:
> > On Tue, Jan 13, 2015 at 04:33:54PM +0000, Will Deacon wrote:
> > > Hi Paul,
> > > 
> > > I started dusting off a series I've been working to implement a relaxed
> > > atomic API in Linux (i.e. things like atomic_read(v, ACQUIRE)) but I'm
> > > having trouble making sense of the ordering semantics we have in mainline
> > > today:
> > 
> > >   2. Does smp_mb__after_unlock_lock order smp_store_release against
> > >      smp_load_acquire? Again, Documentation/memory-barriers.txt puts
> > >      these operations into the RELEASE and ACQUIRE classes respectively,
> > >      but since smp_mb__after_unlock_lock is a NOP everywhere other than
> > >      PowerPC, I don't think this is enforced by the current code. 
> > 
> > Yeah, wasn't Paul going to talk to Ben about that? PPC is the only arch
> > that has the weak ACQUIRE/RELEASE for its spinlocks.
> 
> I thought that you guys were going to propose something and we would see
> what the reaction was.  ;-)
> 
> > >      Most
> > >      architectures follow the pattern used by asm-generic/barrier.h:
> > > 
> > >        release: smp_mb(); STORE
> > >        acquire: LOAD; smp_mb();
> > > 
> > >      which doesn't provide any release -> acquire ordering afaict.
> > 
> > Only when combined on the same address, if the LOAD observes the result
> > of the STORE we can guarantee the rest of the ordering. And if you
> > build a locking primitive with them (or circular lists or whatnot) you
> > have that extra condition.
> > 
> > But yes, I see your argument that this implementation is weak like the
> > PPC.
> 
> A more complete example would be as follows:
> 
>        STOREs followed by release: smp_mb(); STORE A
>        acquire: LOAD A; smp_mb(); preceding LOADs
> 
> If the LOAD A gets the value from the STORE A, then the LOADs following
> the acquire are guaranteed to see the STOREs preceding the release.
> 
> And yes, this really truly does work fine with weaker ordering.

I agree, but if we consider the case where the acquire and the release are
operating on *different* addresses, then the current Documentation would
tell us to use smp_mb__after_unlock_lock to ensure ordering, which won't
work on anything other than Power.

Will