On Wed, Oct 20, 2004 at 08:29:59PM -0400, Jeff Garzik wrote:

I still merging stuff, so won't get around to it for another day or so :)

I certainly don't mind anyone stealing the task from me, but the effort 
is larger than the other iomap conversions.  The patch above hits all 
the easily-picked fruit, leaving the stuff that requires a modicum of 
effort:

* map/unmap N PCI bars (N >= 4, per controller)
* map/unmap 2 ISA I/O regions (0x170, 0x1f0)
* accurately handle the odd situation where IDE driver steals 0x170 
while libata steals 0x1f0 (or vice versa), a.k.a. the reason for 
quirk_intel_ide_combined() and the ____request_resource nastiness

Currently the code is set up to handle:
* N PIO ports
	or
* a single MMIO address that contains all the registers the driver needs
(mmio_base)

Hmm...  It misses a bunch of easy stuff, actually (tons of casts to void *
from what used to be unsigned long and is void __iomem * with your patch).

I don't see where you handle PIO stuff, though - no ioport_map() _or_
pci_iomap() in sight.  Note that ioport_map() is not equivalent to a cast -
we add a constant there.  How does ioread/iowrite work on the results?