Re: XFree86 4.1.0: call for help

[Kostas Gewrgiou <gewrgiou@imbc.gr>]

On Wed, 20 Jun 2001, Michel Dänzer wrote:

> Kostas Gewrgiou wrote:
> >
> > Since fbdev is handling the console there is no need to restore
> > vga fonts/colormap (i don't know if it is possible to run vgacon
> > under some of the ibm machines, if yes then it is needed there)
>
> That's my point, either we don't need anything anywhere (we don't need to save
> the VGA mode with fbcon) or we need nothing on some machines and everything on
> others. Saving and restoring only part of the VGA context doesn't make sense
> to me.

Hmm you have a point there, i have a feeling that the text/colormap
was not crasing in the isa i/o access but in the isa mem ones,
(we mmap 0xA0000, i wonder what lives there since isa mem definitely
isn't)

> > > So, what can we do? We can't put arch or OS specific code to handle this
> > > in vgaHW either. Maybe it's really best to handle it in the kernel?
> >
> > Everything that is os/arch specific more or less goes in xfree86/os-support
> > adding something like xf86isIOAvailable(device/screen/whatever) is what we
> > need.
>
> Which won't work without kernel assistance, will it? So we could as well just
> let the existing syscall handle everything.

We don't need any more kernel assistance, just to fix the syscall to
return an error for the AGP bus of Uni-N if we can't do any io there.

The existing syscall can't handle everything as we use it right now,
we only call it to get the ioBase for the first bus/device/function
and we use this ioBase for *all* the cards that XFree is handling!!!
The base that we get might have nothing to do with the base of the
card that we use...

> > We still need to workout a way to handle multiple ioBases btw, there are
> > other paths in the code that will do io in the wrong base and we need
> > to handle this. Anyone with a second card in the pci bus will end up
> > getting the isa io access for the agp card in the wrong card and thats not
> > what we want.
>
> Yup, but that's for 4.2 at the earliest...

Since 4.1 went out broken we can't do much now, the best solution will
be to disable the ioctl (since we don't get the right base anyway)

 Kostas