2 GPUs have the same ID

Software & Operating Systems Linux
,
21

That happens sometimes.

That script looks at all pci devices. This includes PCI bridges, USB controllers, SATA controllers, etc… That means that not all the devices it finds are going to have the endpoint “boot_vga” so it will output that error a few times. As long as you’ve got your GPU bound properly, it’s doing its job.

1 Like
22

I tried around a little the last couple of hours. It turned out, that the GPUs device path is not as expected:

GPU 1:
/sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0
/sys/devices/pci0000:00/0000:00:01.0/0000:01:00.1

GPU 2:
/sys/devices/pci0000:00/0000:00:01.1/0000:02:00.0
/sys/devices/pci0000:00/0000:00:01.1/0000:02:00.1

I also tried the script where the GPU is directly selected. It seems to work but when I execute lspci -nnv it still says that nouveau is still in use for both GPUs

I also tried to switch completely to my onboard GPU as boot GPU which didn’t have an effect.

And finally blacklisting nouveau doesn’t help either (apart from the Nvidia GPUs having no driver at all now)

And I am still not sure whether that script is getting executed at boot up or not. When I check the driver_override files, they return null.

23

Now that’s a problem. If vfio isn’t able to be bound to a gpu with no driver, there’s something seriously wrong.

Ive never seen this before. I’ll have to see if I can reproduce the issue. What distro are you on?

24

I am currently running ubuntu. I’ll try to provide some more detailed information:

           bits: 64 gcc: 7.2.0
           Desktop: Cinnamon 3.6.7 (Gtk 3.22.30-1ubuntu1)
           Distro: Ubuntu 18.04.1 LTS
Graphics:  Card-1: Intel Device 3e92 bus-ID: 00:02.0
           Card-2: NVIDIA GP107 [GeForce GTX 1050 Ti] bus-ID: 01:00.0
           Card-3: NVIDIA GP107 [GeForce GTX 1050 Ti] bus-ID: 02:00.0
           Display Server: X.Org 1.19.6
           drivers: modesetting (unloaded: fbdev,vesa)
           Resolution: [email protected], [email protected], [email protected]
           OpenGL: renderer: Mesa DRI Intel HD Graphics (Coffeelake 3x8 GT2)
           version: 4.5 Mesa 18.0.5 Direct Render: Yes

/etc/initramfs-tools/modules:

#softdep nouveau pre: vfio vfio_pci

vfio
vfio_iommu_type1
vfio_virqfd
#options vfio_pci ids=10de:1c82,10de:0fb9
#vfio_pci ids=10de:1c82,10de:0fb9
vfio-pci
#nouveau

as mentioned before: nouveau is blacklisted, I have a file in the modprobe.d directory called vfio.conf containing the following line:

install vfio-pci /usr/bin/vfio-pci-override.sh

The boot GPU is currently set to Onboard (i can switch to PCIe but cannot select an individual GPU. But when I select PCIe only one of the GPUs gets the boot_vga flag set to 1).