[PULL] use LRMI for real-mode calls

[Vignatti Tiago (Nokia-D/Helsinki)]

On Fri, May 14, 2010 at 06:03:43PM +0200, ext Julien Cristau wrote:
> On Wed, Apr 28, 2010 at 20:58:58 +0300, Tiago Vignatti wrote:
> 
> > Keith,
> > 
> > All int10 module went to libx86, which in turn is accessed by a shiny new
> > lrmi. It became very easy to create real mode contexts now; for instance all X
> > server code has ~60 lines only. And the cool thing is that the API for drivers
> > keeps the same: 
> > 
> >     git://people.freedesktop.org/~vignatti/xserver lrmi-for-children
> >     git://people.freedesktop.org/~vignatti/libx86
> > 
> I'm completely unable to review this code, but a couple comments anyway
> (I already said this to Tiago on irc, but repeating here for the
> record):
> - the xserver series is unbisectable, either some effort should be spent
>   on that or the patches should be merged.  Especially because comments
>   like "I removed some unspecified code because I couldn't test it or I
>   think it's broken" don't inspire confidence that there won't be
>   regressions from this change
> - this makes libx86 a required dependency for the xfree86 ddx, but
>   libx86 doesn't build on non-x86 archs afaict.  This should be fixed
>   either by making libx86 build everywhere, or by making it optional (I
>   don't know which is appropriate).

I do agree on both points, though I'm not sure we're able to make xserver
series easy bisectable. Anyway, thanks for checking it Julien.

I'd like to emphasize that this is one of the hugest components that currently
shouldn't belong inside the server at all. Should be a shame to keep it in one
more server release. At the same time, I understand why this is sitting there
for while, given no one doing serious driver development recently is relying
in real mode accesses. 

Anyway, I'd like to keep with this work and fix it properly to go upstream,
but unfortunately I don't have time right now. So I totally encourage anyone
taking over those branches.

        Tiago