On Mon, 2006-05-01 at 15:34 -0400, Jimi Xenidis wrote:
> On May 1, 2006, at 3:26 PM, Hollis Blanchard wrote:
>
> > On Mon, 2006-05-01 at 15:02 -0400, Jimi Xenidis wrote:
> >> On May 1, 2006, at 2:39 PM, Hollis Blanchard wrote:
> >>>
> >>> So... you're now fine with isa_io_base?
> >>
> >> no, I'd like to get rid of it and deprecate the use of inb/outb
> >> interfaces on power.
> >
> > Our Maple systems *have* an ISA bus. They do. It contains serial,
> > NVRAM,
> > and the RTC. I don't understand why you want to pretend they don't.
> I don't care what bus the device is on, just its address, and code
> exists to access the device that way.
> I see no reason to simulate an instruction from another architecture
> (inb/outb).
> If the 16550 was on the system without an ISA then inb/outb would be
> useless.
> I'm proposing we use readb/writeb exclusively.
Sure, I would expect nothing less in PowerPC-specific code.
In code that is shared with other architectures, though, such as device
drivers, inb/outb must be used. The current ns16550 driver is already
over-engineered, since you can see that no current architecture even
uses the ioremap path. (Hmm, which begs the question of how will they
keep it from breaking... especially since they've broken the normal
ns16550 path for weeks without anybody noticing except us.)
Since we're mostly just talking about the ns16550 driver, changing it
would not be a big deal (until the driver is rewritten again). However I
remain mystified by your apparently religious objection to emulating inb
and outb. I guess you shouldn't go looking at Linux code...
However, if you do change it, I reserve the right to shake my head and
say "I told you so" if x86 people break MMIO code or add more IO
accessors to drivers in the future.
--
Hollis Blanchard
IBM Linux Technology Center
_______________________________________________
Xen-ppc-devel mailing list
Xen-ppc-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ppc-devel
|
Home