Jason Thorpe <thorpej@nas.nasa.gov> writes:

> On Wed, 03 Nov 1999 11:06:45 +0100 
>  Lennart Augustsson <lennart@augustsson.net> wrote:
> 
>  > Just about every PCI device driver that is converted to work
>  > on big-endian machines will need to do byte swapping.
>  > 
>  > Should we define pcitoh() and htopci() macros in pci.h?
>  > Currently each driver has its own homegrown definitions
>  > (possibly with different names).
> 
> No, because the names are wrong.
> 
> What we need is:
> 
> 	hto{le,be}{16,32}()
> 	{le,be}toh{16,32}()

Actually, I'd argue that you still want #defines to "pci-ish" names.

The reason is that there are very few places where people actually
need to worry about PCI bus endianness; e.g. the bus_space routines
paper over it, etc.  using appropriately-named functions make the
logical intent clear: "use the endianness of the PCI (or somet other)
bus," rather than "use little endianness (or big endianness)."


cgd
-- 
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.