Re: [PATCH 16/18] x86: io: implement dummy relaxed accessor macros for writes

[Will Deacon <will.deacon@arm.com>]

On Wed, May 21, 2014 at 02:53:27AM +0100, Brian Norris wrote:
> Hi Will,

Hello,

> On Tue, Apr 22, 2014 at 05:08:21PM +0100, Will Deacon wrote:
> > On Thu, Apr 17, 2014 at 02:44:19PM +0100, Will Deacon wrote:
> > > write{b,w,l,q}_relaxed are implemented by some architectures in order to
> > > permit memory-mapped I/O accesses with weaker barrier semantics than the
> > > non-relaxed variants.
> > > 
> > > This patch adds dummy macros for the write accessors to x86, in the
> > > same vein as the dummy definitions for the relaxed read accessors.
> [...]
> > > --- a/arch/x86/include/asm/io.h
> > > +++ b/arch/x86/include/asm/io.h
> > > @@ -74,6 +74,9 @@ build_mmio_write(__writel, "l", unsigned int, "r", )
> > >  #define __raw_readw __readw
> > >  #define __raw_readl __readl
> > >  
> > > +#define writeb_relaxed(v, a) __writeb(v, a)
> > > +#define writew_relaxed(v, a) __writew(v, a)
> > > +#define writel_relaxed(v, a) __writel(v, a)
> > >  #define __raw_writeb __writeb
> > >  #define __raw_writew __writew
> > >  #define __raw_writel __writel
> [...]
> > 
> > Actually, I should be using the regular (i.e. without the double underscore
> > prefix) accessors for the relaxed variants, including the existing read
> > flavours here. The proposed semantics are that the accessors are ordered
> > with respect to each other, which necessitates a compiler barrier.
> 
> Are you planning on resubmitting this series? I've run into several
> situations in which I can't compile-test a driver on a different ARCH
> just because of this issue.

Yeah, I was just hoping for some input from Ben on the semantics I proposed.

I have a fix for the x86 patch, so I guess that justifies a v2. I'll post
something later on.

Cheers,

Will