[Intel-gfx] i915 KMS interferes with nVidia card

[Bruno Prémont]

On Wed, 12 March 2014 Ben Johnson <intel-gfx at geekamole.com> wrote:
> My hardware:
> Asrock Z77 Extreme4 with iGPU Multi-Monitor enabled. Primary graphics is 
> set to PCIe but that doesn't make a difference.
> i7 3770k / HD 4000
> nVidia GTX 660 running nVidia's binary drivers v334.21, although 
> multiple versions have the issue
> 
> My software:
> Stable Gentoo with gentoo-sources-3.12.13, although multiple kernels 
> have the issue
> 
> I am trying to use both the HD 4000 graphics and the nVidia card 
> simultaneously in a single-seat multiple-X-server Zaphod scenario. I 
> have another cheap nVidia card on order so I don't have any urgent need 
> and maybe not any need at all to get this working with i915.
> 
> The Zaphod setup works fine with the exception of two problems that 
> occur only when the i915 driver is loaded with modesetting enabled. 
> I.e., not loading i915 or setting "nomodeset" or "i915.modeset=0" at 
> boot resolves the issue, but does not allow the HD 4000 to be used. With 
> i915 loaded normally:
> 
> 1) The nVidia virtual terminals are messed up. They appear zoomed-in and 
> don't update.
> 2) The nVidia card does not resume correctly from suspend-to-ram. It 
> prints some errors to dmesg and either doesn't wake the monitor up, or 
> doesn't ever show the X session again until I reboot.
> 
> I am a C programmer and I don't mind helping to fix this if it turns out 
> to be relatively simple issue and my help is desired. Is this a known 
> limitation, and does anyone have a clear idea what the problem is or 
> where to look?

Wild guess, maybe you would have some more chances of success if the nvidia
was the GPU with active VGA bits (e.g. telling vga-arbitration to have
nvidia card active/primary).

That way the nvidia blob would have best changes of living in a tested
environment and i915 hopefully being able to do its work anyway.

Bruno