[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [vfio-users] Passing through an intregrated Nvidia mobile GPU (NVIDIA Quadro M1000M)

I hate to be that guy, but this really isn't anywhere near as complex as it's being presented. I've passed through the K2100M on my Precision M4800 before, I'll set it up again, but I had stopped work because it required an external monitor to work properly. I had then set up my desktop instead and been happy with the results.

Optimus doesn't really have any 'special sauce' making things happen. The OS and drivers do a lot of the work, basically Optimus is just a secondary GPU without directly connected I/Os (ideally). The firmware tricks are mainly to keep the OS from trying to bring up the secondary GPU as a primary one. You can see traits of this via lspci. The Intel GPU shows as a VGA controller, the Nvidia GPU shows as a "3D controller", which I had last seen in Linux in reference to a 3Dfx Voodoo2.

I'll try to get this running again on my M4800 and take notes/screenshots/etc to assist.

Someone suggested disabling Optimus. Do *not* do this. Your job will become massively more complex. Disabling Optimus on the Precision laptops disables the Intel GPU almost entirely and uses it only to connect the outputs. The Nvidia GPU will be fully functional and identified normally, and will be providing all of your graphical output. This can make it much more complicated to control the host OS.

I believe most of the difficulty I had was related to the lscpi issues I described above, the PCI IDs are changed when Optimus is enabled, and I believe this is the issue that is causing your problems. I can't recall, I'll need to check my laptop, but I believe I had spoofed the single-GPU-no-Optimus PCI IDs along with passing the device, but I'll have to check.

On Thu, Mar 31, 2016 at 1:35 PM, Alex Williamson <alex l williamson gmail com> wrote:
Nobody has successfully assigned an Optimus GPU to a guest yet, afaik.  These hybrid laptops are not simply an integrated and discrete GPU in the same case with the ability to switch between them, there are proprietary BIOS requirements that we cannot replicate in open source firmware to make that happen and it's not clear if there's any way to make the GPU work independently of those features.  If you can make it work, document what you've done, but don't expect that it should just work.  Good luck.

vfio-users mailing list
vfio-users redhat com

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]