[PATCH] Remove XAA

[Mark Kettenis]

> Date: Mon, 23 Sep 2013 09:11:26 -0700
> From: Alan Coopersmith <alan.coopersmith at oracle.com>
> 
> On 09/22/13 09:27 AM, Matthieu Herrb wrote:
> > On Sun, Sep 22, 2013 at 08:59:42AM -0700, Alan Coopersmith wrote:
> >> On 09/21/13 07:49 AM, devzero at web.de wrote:
> >>>> Deliberately breaking drivers without giving people a chance to fix them
> >>>> isn't fair.  Because of the current development model of Xorg people may
> >>>> not notice that stuff is broken for more than six months.
> >>>
> >>> Yes, this sucks big.
> >>
> >> You're replying to a year and a half old e-mail.   People have had plenty of
> >> chance to fix the drivers they still use now, and most of the ones people use
> >> have been fixed.
> >
> > Sorry, but this is only partially true.
> >
> > In OpenBSD we're still having a number of issues with drivers that are
> > used and have not been fully fixed. My limited time and knowledge of
> > EXA have not allowed to make enough progress to tell that old drivers
> > are performing as good now with EXA than they were performing with
> > XAA.
> >
> > Affected drivers are at least
> >
> > - xf86-video-cirrus (no EXA code at all in xf86-video-cirrus)
> > - xf86-video-mach64 (render acceleration broken in EXA)
> > - xf86-video-mga (render acceleration broken in EXA, need to be
> >    disabled to have a working server)
> > - xf86-video-nv shadowfb broken, no EXA acceleration for older
> >     chipsets. have to run it completely un-acceleratied.
> > - xf86-video-sis (mostly broken for quite a bit of time)
> > - xf86-video-sunffb only shadowfb "acceleration" after dropping XAA
> 
> I was referring to "runs without XAA" as being fixed, not necessarily adding
> EXA support.

But as Matthieu's list shows, quite a few drivers are simply broken
without XAA, so they *don't* "run without XAA".

> And since the maintainers of -nv advised people to stop using it and move
> to -vesa 3 or 4 years ago now, I'm not sure that one's worth the effort.

Pretty useless advice.  For two reasons:

1. Most VESA BIOSen only support the standard VESA modes.  These are a
   bad match for any modern LCD monitor.

2. The -vesa driver only works for platforms that actually have a VESA
   BIOS.  So it doesn't work on older Apple machines for example.

Anyway, you're probably right that adding EXA support isn't worth the
effort for any of these drivers.  For regular 2D stuff shadowfb often
seems to be faster, even on hardware that supposedly has working EXA
support.  At least if you're not using a fancy compositing window
manager...