Re: ddb5477 fixes for 2.6

[Daniel Jacobowitz <dan@debian.org>]

On Thu, Sep 18, 2003 at 07:19:21PM +0200, Maciej W. Rozycki wrote:
> On Thu, 18 Sep 2003, Daniel Jacobowitz wrote:
> 
> > >  Is it OK for other PCI systems?
> > 
> > Yes, I think so.  Those two functions seem to have migrated in from
> > pci-hplj.c; they can't possibly compile, since they use constants only
> > defined in that file.
> 
>  Then the snippet should probably get removed altogether.
> 
> > >  Why do you want these suffixes?  They don't work for assembly sources.
> > 
> > Because otherwise uses of XKPHYS in a 32-bit kernel generate noisy
> > warnings.  I don't remember where it was offhand.  Wrap it in
> > __ASSEMBLY__ if you like.
> 
>  Hmm, but is there any use for XKPHYS, etc. in a 32-bit kernel at all? 
> The address cannot effectively be used anyway. 

That patch is a total hack/slash job.  You're probably right on both
counts.

Take a look at __ioremap_mode which probably needs some #ifdefs for
32-bit kernels.  It uses cpu_has_64bit_addresses, which means at
runtime it's OK, but we get a lot of warnings.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer