Re: [Gc-linux-cvs] linux/include/asm-ppc io.h,1.8,1.9

[<a.o...@bl...>]

On Sun, Sep 11, 2005 at 11:01:27AM +0200, Albert Herranz wrote:
>  --- Arthur Othieno <a.o...@bl...> escribi?:
> > > We still have to maintain io.h for the
> > > read{b,w,l}/write{b,w,l} tweak anyway...
> > 
> > No. We don't. These are accessors for PCI and other
> > inherently
> > little-endian busses, only. In any case, the ones
> > we're using
> > for gamecube don't provide proper memory barriers.
> 
> We use that accessors in the GameCube to access i/o
> mapped devices, and that spaces are already guarded
> and cache-inhibited so we _don't_ need memory barriers
> to guarantee ordered access to "memory" (see ppc cache
> model and memory coherence).

That change doesn't touch the accessors bit, so I don't see where the
problem is..

> And we use that accessors because of the current lack
> of the appropiate ones. So, we _do_ need them (or we
> need to get the necessary patch). All drivers
> performing i/o currently use them.

The appropriate accessors _do_ exist. None of the drivers doing
I/O actually use them, unfortunately.

> > > So currently, and until someone adds big endian
> > > ioread*/iowrite* support, this change doesn't make
> > too
> > > much sense.
> > 
> > These are still queued up in patchwork. They should
> > make it upstream
> > RSN whenever paulus pushes to Linus, like he just
> > did for ppc64.
> 
> So, what's the _real_ effect of your patch?
> Why not wait for the ioread/iowrite stuff and submit a
> patch that _actually_ does something?

I think I clarified it enough in the commit message. Nothing has been
broken, intentionally or unintentionally. Like I said, patch exists, and
until it makes it into mainline, include/asm-ppc/io.h remains in CVS.

	Arthur