RFC on writel and writel_relaxed

[Jason Gunthorpe]

On Tue, Mar 27, 2018 at 08:22:55AM -0400, okaya at codeaurora.org wrote:
> On 2018-03-27 07:23, Benjamin Herrenschmidt wrote:
> >On Tue, 2018-03-27 at 11:44 +0200, Arnd Bergmann wrote:
> >>> The interesting thing is that we do seem to have a whole LOT of these
> >>> spurrious wmb before writel all over the tree, I suspect because of
> >>> that incorrect recommendation in memory-barriers.txt.
> >>>
> >>> We should fix that.
> >>
> >>Maybe the problem is just that it's so counter-intuitive that we don't
> >>need that barrier in Linux, when the hardware does need one on some
> >>architectures.
> >>
> >>How about we define a barrier type instruction specifically for this
> >>purpose, something like wmb_before_mmio() and have all architectures
> >>define that to an empty macro?
> >
> >This is exactly what wmb() is about and exactly what Linux rejected
> >back in the day (and in hindsight I agree with him).
> >
> >>That way, having correct code using wmb_before_mmio() will not
> >>trigger an incorrect review comment that leads to extra wmb(). ;-)
> >
> >Ah, you mean have an empty macro that will always be empty on all
> >architectures just to fool people ? :-)
> >
> >Not sure that will fly ... I think we just need to be documenting that
> >stuff better and not have incorrect examples. Also a sweep to remove
> >some useless ones like the one in e1000e would help.
> 
> I have been converting wmb+writel to wmb+writel_relaxed. (About 30 patches)
>
> I will have to just remove the wmb and keep writel, then repost.

Okay, but before you do that, can we get a statement how this works
for WC?

Some of these writels are to WC memory, do they need the wmb()?!?

Jason