Re: [PATCH 4/4] locking: Introduce smp_cond_acquire()

From: Linus Torvalds <torvalds@linux-foundation.org>
To: David Howells <dhowells@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@kernel.org>, Oleg Nesterov <oleg@redhat.com>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Paul McKenney <paulmck@linux.vnet.ibm.com>,
	boqun.feng@gmail.com, Jonathan Corbet <corbet@lwn.net>,
	Michal Hocko <mhocko@kernel.org>,
	Will Deacon <will.deacon@arm.com>
Subject: Re: [PATCH 4/4] locking: Introduce smp_cond_acquire()
Date: Mon, 2 Nov 2015 13:11:22 -0800	[thread overview]
Message-ID: <CA+55aFzheg4-ahv-rRWqn0PqwjwyTToQZaYF4voQKDLACYeZ+A@mail.gmail.com> (raw)
In-Reply-To: <8722.1446496609@warthog.procyon.org.uk>

On Mon, Nov 2, 2015 at 12:36 PM, David Howells <dhowells@redhat.com> wrote:
> Linus Torvalds <torvalds@linux-foundation.org> wrote:
>
>> > +       smp_read_barrier_depends(); /* ctrl */  \
>> > +       smp_rmb(); /* ctrl + rmb := acquire */  \
>
> Doesn't smp_rmb() imply an smp_read_barrier_depends() anyway?

Yes, it does. But that "smp_read_barrier_depends()" is actually
mis-used as a "barrier against subsequent dependent writes, thanks to
the control flow". It's not protecting against subsequent reads -
which is what the smp_rmb() is about.

Which is completely bogus, but that's what the comment implies.

Of course, on alpha (which is where smp_read_barrier_depends() makes a
difference), both that and smp_rmb() are just full memory barriers,
because alpha is some crazy sh*t. So yes, a "smp_rmb()" is sufficient
everywhere, but that is actually not where the confusion comes from in
the first place.

                   Linus