Re: [PATCH] IDE 58

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

(resent, I had the date screwed up previously, sorry about the
inconvenience).

>I don't see why all IDE-interfaces in the world have to be I/O-mapped just 
>because the first PC implementations used that. Sure it was an extended 
>ISA-bus but the ISA bus is long gone and we don't all run PC's anymore 
>either.
>
>So the simple abstraction we need to hit IDE-bus registers is a macro or 
>inline, instead of a call of an I/O-primitive. It was too much work to 
>abstract this when I inserted the CRIS-arch IDE-driver in the first place 
>so I found a workaround but now seems like a better time..

No, not a macro. There are cases where you want different access methods
on the same machine. For example, pmacs can have the "mac-io" (ide-pmac)
controller, which is MMIO based, _and_ a PCI-based legacy IDE controller
using inx/outx like IOs. (A typical example is the Blue&White G3 who has
both on the motherboard).

Ultimately, you want the hwif (or what it becomes in 2.5) provide a set
of functions for accessing taskfile registers and doing the PIO data
stream read/writes (that is replace inb/outb and insw/outsw).