[PULL] Linux and bus subsystem cleanups

[Adam Jackson]

On Tue, 2009-11-10 at 12:55 -0800, keithp wrote:
> Excerpts from Adam Jackson's message of Tue Nov 10 11:43:21 -0800 2009:
> >       bus: Fix open-coded implementation of xf86MakeNewMapping
> 
> 	This leaves vp uninitialized. You need to keep calling
> 	getVidMapRec

D'oh.

> >       bus: Change xf86MakeNewMapping prototype to match reality
> 
> 	If no drivers use it, please remove it from the ABI.

Slightly easier said than done.  sdksyms.sh pretty much wants _every_
definition in sdk headers to be prefaced with _X_EXPORT, last time I
read it.  So it'd need to be in a new header, or we'd need to fix
sdksyms.sh.

I'm of the opinion that the shell script is far far worse than the
static list we had before, in that it actually grew the ABI.  The joke
is that it did so in the name of being able to use -fvisibility=hidden
to seal off more of the ABI!  Some kind of massive misunderstanding of
the problem.

> >       linux: PCI_DOM_MASK is 0xff everywhere.
> 
> 	The kernel allows for 16 bits here; do we need to fix this? Or
> 	can we live with only 8 (yes, I know that would require
> 	actually fixing the code)?

Fixing it here is straightforward, but the PCITAG type encodes this
mistake.  On ILP32, only the high 8 bits of PCITAG are available for
domain number.  PCITAG is a complete mistake anyway at this point, we
should be passing around struct pci_device * or nothing, but that's the
kind of ABI break I'd like merged drivers for.

- ajax
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part
Url : http://lists.x.org/archives/xorg-devel/attachments/20091110/0143ef87/attachment.pgp