Re: [RFC PATCH] docs/memory-barriers.txt: Rewrite "KERNEL I/O BARRIER EFFECTS" section

From: Will Deacon
Date: Wed Feb 13 2019 - 12:20:56 EST


[+Tony]

On Mon, Feb 11, 2019 at 02:34:31PM -0800, Linus Torvalds wrote:
> On Mon, Feb 11, 2019 at 9:30 AM Will Deacon <will.deacon@xxxxxxx> wrote:
> >
> > +
> > + 1. All readX() and writeX() accesses to the same peripheral are ordered
> > + with respect to each other. For example, this ensures that MMIO register
> > + writes by the CPU to a particular device will arrive in program order.
>
> Hmm. I'd like more people look at strengthening this one wrt across
> CPUs and locking.
>
> Right now we document mmiowb(), but that "documentation" is really
> just a fairy tale. Very *very* few drivers actually do mmiowb() on
> their own.
>
> IOW, we should seriously just consider making the rule be that locking
> will order mmio too. Because that's practically the rule anyway.

I would /love/ to get rid of mmiowb() because I think it's both extremely
difficult to use and also pretty much never needed. It reminds me a lot of
smp_read_barrier_depends(), which we finally pushed into READ_ONCE for
Alpha.

> Powerpc already does it. IO within a locked region will serialize with the
> lock.

I thought ia64 was the hold out here? Did they actually have machines that
needed this in practice? If so, I think we can either:

(a) Add an mmiowb() to their spin_unlock() code, or
(b) Remove ia64 altogether if nobody complains

I know that Peter has been in favour of (b) for a while...

Will