start of some pci cleanups

Dave Airlie airlied at gmail.com
Wed Jul 20 09:17:14 PDT 2005


> 
> On Wed, 20 Jul 2005, Benjamin Herrenschmidt wrote:
> 
> > On Tue, 2005-07-19 at 08:57 -0700, Andy Ritger wrote:
> > > When modifying existing structures and functions, please be conscious
> > > of binary compatibility.
> >
> > In that case, there is no real choice I'm afraid, that stuff need to be
> > cleaned up...
> >
> > It might be possible to get a first patch in that only removes the use
> > of the direct IO port access for PCI config on x86 and uses /proc
> > instead as it does on other architectures. That would be a big step
> > forward already without breaking backward compatibility. But we need to
> > get the PCI & VGA IO stuff cleaned up as well and that will
> > unfortunately cause some breakage.
> 
> Understood.  If interfaces need to change, then by all means please
> do so.  However, in many cases, there are alternate solutions like
> adding a separate entry point, rather than changing the arguments
> to an existing entry point.  Similarly, a new field like 'domain'
> could be added at the end of pciVideoRec rather than inside PciBusId.
> 
> If the ABI needs to be changed, then have good reason and do it
> consciously, and please give vendors a way to know which version
> of the ABI should be used, so users know which build of the driver
> to install.
> 
> I know there have been some plans in the works for other ABI changes,
> like Kevin Martin's DMX work to eliminate the MAXSCREENS compile
> time constant.  It might be good to time these things such that
> they can all be encompassed in a single rev of the ABI.
> 

Well the plan for X11R7 is to break some ABIs to clean things up,
hacking in domain support in a different place is not clean, and we
would have to live with it with little point, the current interfaces
are insufficent for a number of things, I'm about to fix up a few
more, note I'm not intending to add any sort of domain support beyond
what is there, but I would like at least to get the interfaces for R7
clean so that implementation can happen later without breaking them...

Dave.



More information about the xorg mailing list