Linux Virtualization - Nov 2016 - [PATCH 0/3] virtio/vringh: kill off ACCESS_ONCE()

On Fri, Nov 25, 2016 at 09:52:50AM -0800, Linus Torvalds
wrote:
> READ/WRITE_ONCE() are atomic *WHEN*THAT*IS*POSSIBLE*.
> But sometimes it's not going to be atomic.
That's the problem.

Common code may rely on something being atomic when that's only true on
a subset of platforms. On others, it's silently "fixed" into
something
that isn't atomic, and we get no diagnostic. The bug lurks beneath the
surface.

Thanks,
Mark.

On Fri, Nov 25, 2016 at 10:07 AM, Mark Rutland <mark.rutland at arm.com>
wrote:
> On Fri, Nov 25, 2016 at 09:52:50AM -0800, Linus Torvalds wrote:
>> READ/WRITE_ONCE() are atomic *WHEN*THAT*IS*POSSIBLE*.
>
>> But sometimes it's not going to be atomic.
>
> That's the problem.
It has never really been much of a problem, and quite frankly, the
solution would never be to add _another_ crazy new function that will
just confuse everybody.

If you have code that depends on atomicity of READ_ONCE() and friends,
then you should add the appropriate built-time assert to *your* code.
Not to some random generic function that others care about and that
others do _not_ have problems with.

So if you have a data structure in virtio that is
architecture-dependent and might not be a word size, you add the

   BUILD_BUG_ON(sizeof(mytype) > sizeof(long));

or whatever. With a big comment saying "this needs to actually fit in
a single register so that we can do atomic accesses".

You do not screw it up for everybody else.

                Linus