Hello.
I'm quite new with VirtualGL.
I'm trying to do build a virtual machine that will serve an old application over VNC. That application needs video acceleration.
I use a libvirt/qemu guest with PCI bypass to expose a Tesla T4 headless card to the virtual machine.
The virtual machine runs Debian 10, with the nvidia-tesla-470-driver package from the distribution.
I run a Xorg server on :0 which, according to nvidia-smi, actually uses the T4 card. A lightdm service runs on this X session.
I run a turbovnc session on :1 to use the virtual machine.
I followed the Headless nVidia Mini How-To, then the Documentation (chapter 6). For testing purposes, I didn't restrict virtualgl usage to a vglusers group. XTEST extension is enabled as I plan to use a vnc server.
According to the sanity test with glx info, the renderer string is "Tesla T4/PCIe/SSE2", which is fine. According to the same tool, there are numerous visuals with 24 bits and pbuffer support.
Unfortunately, I cannot run glxgears.
When running "vglrun glxgears" from a vnc client, a glxgears window flashes briefly, and I have the following error message:
[VGL] ERROR: in readPixels--
[VGL] 515: Could not map pixel buffer object
What is this 515 error?
I tried to run glreadtest, and it runs flawlessly without options. If I run glreadtest with -alpha options, I have a lot of errors:
X11 Error: BadMatch
I tried to run glxgears with the VLG_FORCEALPHA=1 option, but I still have my error.
Could you please help me on this? Any help would be appreciated.
I attach the vglrun trace (+tr option)