f20, anaconda, net install and video out of range ....

[Matthew Garrett]

On Fri, Feb 07, 2014 at 11:19:38AM -0500, Paul Wouters wrote:
> On Thu, 6 Feb 2014, Adam Williamson wrote:
> 
> >painstakingly hand-weeding something like M*a's ldetect-lst you can get
> >some minor benefits, like doing this kind of distinction where we want
> >to load the native driver for a real card but not qemu's emulated
> >cirrus.
> 
> You are telling me it is hard to detect the real physical card versus
> the emulated card? Come on! You can even make that decision by looking
> at the cpu type. If your cpu is QEMU Virtual CPU, how about using the
> virtual cirrus driver....

Oh, that's not difficult - it'd be possible to make the cirrus driver 
bail on virtualised hardware (easy to detect, just check whether there's 
a KMS driver bound). But that would still require someone to care about 
maintaining the cirrus driver, which nobody does.

> Taking out everyone who tries to run fedora or rhel7 using a physical
> cirrus card IMHO is just sloppy and lazy. Yes, people still run P-III
> servers with SCSI disks and cirrus cards. In fact, I think you will
> see it more within the enterprise then outside it.

As Adam points out, nobody's running RHEL7 on a physical cirrus card. 
However, we'd still expect the vesa driver to work, and the fact that it 
doesn't is a bug[1]. Nobody ever added exa support to Cirrus, so it's 
not like you'd see any meaningful difference in performance.

[1] I mean plausibly it's a bug in this particular Cirrus video BIOS, 
but it'd be nice to actually figure that out.

-- 
Matthew Garrett | mjg59 at srcf.ucam.org