Performance improvement to vga arbitration

[Mark Kettenis]

> Date: Mon, 14 Jun 2010 20:53:52 +0300
> From: "Vignatti Tiago (Nokia-D/Helsinki)" <tiago.vignatti at nokia.com>
> 
> Hi Henry,
> 
> On Sat, Jun 12, 2010 at 04:38:14AM +0200, ext Henry Zhao wrote:
> > On 06/11/10 04:56, Tiago Vignatti wrote:
> > >
> > > http://people.freedesktop.org/~vignatti/tmp/0001-xfree86-vgaarb-disable-VGA-decoding-after-POST.patch
> > >
> > > At this point on the X server, we already POSTed all cards and
> > > we could be optimistic, letting the drivers say when we actually
> > > need VGA legacy. Right now, as you said, we're assuming legacy
> > > access as default whenever the system has more than one card,
> > > not driven by DRI (DRI and VGA legacy conflict with the current
> > > design. Dave preferred to let this away).
> > 
> > The patch didn't go to git master yet ?
> 
> not yet. Do you mind to do a proper review and insert your review tag there? 

Of course your diff is not in yet.  It will kill systems with cards
that do need the legacy stuff.  And setting VGA_ARB_RSRC_NONE is
totally wrong.  The cards will still need access to their framebuffer
so minimally you want VGA_ARB_RSRC_NORMAL_MEM.

The proper way to handle this is to add to add the appropriate
xf86VGAarbiterDeviceDecodes() call to the drivers that are known not
to require VGA legacy access, and only bypass the arbiter once there
are no devices left in the system that require legacy access.