Re: [PATCH] libata-sff: Don't call bmdma_stop on non DMA capable controllers

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

> On sparc64, for example, after I pointed this out to DaveM, he was able 
> to implement the new iomap interface without the 'if (pio-mem-area)' 
> branch present on x86.

Yup, we did that from day 1 on powerpc :-) However, I don't totally
agree with adding some other remapping layer here, I think if we want
PIO 0 to be illegal, then make it illegal at the HW level too.

The reason is in fact the same as Linus invoked for remapping it in the
first place -> make things look like an x86 :-)

That is, quite a few non-x86 machines do have some kind of superIO chip
or other set of legacy devices around. They also commonly have VGA cards
hard decoding VGA PIO addresses.

So here's a very simplified version on how most non-x86 platforms do
PIO :

at boot:

 pci_io_base = ioremap(MAGIC_PIO_REGION);

and then

 #define inb(port)	(readb(pci_io_base + (port)))

The nice thing with that approach is that all those legacy x86 drivers
for bits in your SuperIO chip or for VGA do actually still work when they
do

 inb(STUPID_HARD_CODED_IO_PORT);

While if we now add some magic remapping to make 0 illegal, that will break
and all those legacy drivers would have to be fixed, which nobody wants
to do.

So I do stand firm there. I don't necessarily mind deciding that 0 is an
illegal PIO address, but if we're going to do that, we should make it
illegal as a HW PIO address, not by adding a remapping trick to something
that really doesn't want more than it already has :-)

Ben.