readl/writel: swap or not to swap (was: [PATCH] fbdev: Fix IO access in rivafb)

[Yasushi SHOJI <yashi@atmark-techno.com>]

At Sat, 13 Nov 2004 21:20:03 +0900,
yashi wrote:
[...]
> > Why not just use __raw_readl/__raw_writel?
> > 
> > That's what they exist for, and they still do any IO accesses correctly, 
> > which a direct store does not do (it would seriously break on older 
> > alphas, for example).
> 
> sorry for a dumb question but should readl/writel on big endian system
> swap like ppc does?

I guess everyone is busy hacking.  but can at least someone give me a
hit?

I'm worring about this issue because I'm about to use two deferent
linux arch on same board.  it's based on reconfigurable device so I
can configure to have deferent cpu on it.

if you are using two deferent arch of linux, it's natual to think you
want to share all device drivers.  but if one arch swap with readl and
the other doesn't, I have to abstruct these low level access
methods. (given that those to arch are same endian and connected with
same bus and to the same devices)

is there any rule we should follow?  Is the ppc way the right
direction to follow? I can ifdef anytime for my own use but I just
want to know what _should_ be done.

thanks,
--
             yashi