Re: [PATCH 8/8] i386: bitops: smp_mb__{before, after}_clear_bit() definitions

[Satyam Sharma <ssatyam@cse.iitk.ac.in>]

On Tue, 24 Jul 2007, Nick Piggin wrote:

> Satyam Sharma wrote:
> > On Tue, 24 Jul 2007, Nick Piggin wrote:
> > > Satyam Sharma wrote:
> > > [...]
> > > > So let's make these proper no-ops, because that's exactly what we
> > > > require
> > > > these to be on the i386 platform.
> > > 
> > > No. clear_bit is not a compiler barrier on i386,
> > 
> > Obvious.
> > 
> > > thus smp_mb__before/after
> > > must be.

> > Not so obvious. Why do we require these to be a full compiler barrier
> > is precisely the question I raised here.

> > Consider this (the above two functions exist only for clear_bit(),
> > the atomic variant, as you already know), the _only_ memory reference
> > we care about is that of the address of the passed bit-string:
> 
> No. Memory barriers explicitly extend to all memory references.

[ Compiler barrier, you mean, that's not true of CPU barriers. ]

In any case, I know that, obviously. I asked "why" not "what" :-) i.e.
why should we care about other addresses / why do we want to extend
the compiler barrier to all memory references -- but Jeremy seems to
have answered that ...

> > (1) The compiler must not optimize / elid it (i.e. we need to disallow
> >     compiler optimization for that reference) -- but we've already taken
> >     care of that with the __asm__ __volatile__ and the constraints on
> >     the memory "addr" operand there, and,
> > (2) For the i386, it also includes an implicit memory (CPU) barrier
> >     already.
> 
> Repeating what has been said before: A CPU memory barrier is not a
> compiler barrier or vice versa. Seeing as we are talking about
> the compiler barrier, it is irrelevant as to whether or not the
> assembly includes a CPU barrier.

I think it is quite relevant, in fact. From Documentation/atomic_ops.txt,
smp_mb__{before,after}_clear_bit(), as the name itself suggests, must
be _CPU barriers_ for those arch's that don't have an implicit
_CPU barrier_ in the clear_bit() itself [ which i386 does have already ].

As for a compiler barrier, the asm there already guarantees the compiler
will not optimize references to _that_ address, but there could still be
the memset()/set{clear}_bit() interspersing pitfalls for example, so yeah
the memory clobber would probably protect us there.


Satyam
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/