Re: [PATCH] arch: Introduce read_acquire()

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Nov 11, 2014 at 12:45:23PM -0800, Alexander Duyck wrote:
> 
> On 11/11/2014 11:40 AM, Linus Torvalds wrote:
> >On Tue, Nov 11, 2014 at 10:57 AM,  <alexander.duyck@gmail.com> wrote:
> >>
> >>
> >>On reviewing the documentation and code for smp_load_acquire() it occurred
> >>to me that implementing something similar for CPU <-> device interaction
> >>would be worth while.  This commit provides just the load/read side of this
> >>in the form of read_acquire().
> >
> >So I don't hate the concept, but. there's a couple of reasons to think
> >this is broken.
> >
> >One is just the name. Why do we have "smp_load_acquire()", but then
> >call the non-smp version "read_acquire()"? That makes very little
> >sense to me. Why did "load" become "read"?
> 
> The idea behind read vs load in the name was because smp_load_acquire is
> using a full smp_mb() whereas this just falls back to rmb() for the cases it
> is dealing with.  My main conern is that a full memory barrier would be more
> expensive so I didn't want to give the idea that this is as completed as
> smp_load_acquire().  The read_acquire() call is not strictly enforcing any
> limitations on writes/stores, although there are a few cases where the
> barriers used do leak that functionality over as a side-effect.

Then I object. We should not name it acquire if it does not in fact
provides acquire semantics.

Memory ordering is hard enough, we don't need random weird semantics
mixed in just because.