Re: [PATCH 3/4] Fix remaining big endian issue of hfcmulti

[Sean MacLennan <smaclennan@pikatech.com>]

On Tue, 5 Aug 2008 23:02:39 +0200
"Karsten Keil" <kkeil@suse.de> wrote:

> On Tue, Aug 05, 2008 at 11:42:56AM -0700, Linus Torvalds wrote:
> > 
> > 
> > On Tue, 5 Aug 2008, Karsten Keil wrote:
> > > 
> > > Maybe we can use the trick from lib/iomap.c to detect which
> > > kind of IO is needed, but unfortunately PIO_OFFSET, PIO_MASK and
> > > PIO_RESERVED are not exported so it would need to copy the
> > > defines, which isn't a really clean solution.
> > 
> > Even if they were exported, you couldn't.
> > 
> > lib/iomap.c is _not_ generic code. It's a library function for 
> > architectures that don't do it some other way. But various
> > architectures can choose to not use lib/iomap.c at all - for
> > example, they may have MMIO and PIO in the same address space, so
> > they don't need the conditionals at all (because all the work was
> > done at mapping time, not at runtime).
> > 
> > So if you actually have different models of operation for PIO and
> > MMIO, then yes, you need to handle that in the driver itself.
> > 
> 
> One question here, what is the better approach to do such a different
> implementation, use one local function like
> 
> static void
> my_out32(struct card *c, u_int offset, u-int data) 
> {
> 	if (c->mode == MMIO) {
> 		...
> 	} else {
> 		...
> 	}
> }
> 
> or use 2 function, one for the MMIO and one for the PIO model and
> then use indirect calls (like c->my_out32(...)) ?

Why not select PIO or MMIO at config time?

Cheers,
   Sean