Unable to passthrough NVIDIA GPU

[DataHearth]

Hello folks!

I've been playing around with my server this afternoon to put my msi gtx1070 inside my dear server.

But I'm encountering issues that are far over my knowledge 😓.

Also, all the issues have in common the CG. When it's plugged in, one of the issue is there.

 

First issue encountered:

A freeze at boot time. When it's time to select the CLI or the GUI (auto CLI), sometimes the boot just freeze right after the `/bzroot OK` check. No error.

I haven't noticed any pattern in this issue. At first I thought that would be a bad configuration of my ACS override. But once I pull the CG, it boots as usual even with an ACS override configuration (multifunction, downstream, both).

This issue is kinda random for now. I've seen it few times.

 

Second issue:

If I do not encounter the previous issue, with an ACS override at multifunction and both (multifonction,downstream), a panic happened inside the kernel

KERNEL PANIC - NOT SYNCING: VPF: UNABLE TO MOUNT ROOT FS ON UNKNOWN-BLOCK

.

It seems to work correctly with downstream (IOMMU at correctly splitted but not enough for me to passthrough the GPU unfortunately).

Also, I've noticed something weird inside the stack trace:

sd 0:0:0:0: [sda] 7892992 512-byte logical blocks: (4.04 GB/3.76 GiB)

The size 3.76 GiB is the size of my USB stick which unraid is installed on. I was wondering if the kernel or whatever task is crashing, is trying to write something bigger than the USB stick size ? It is just a newbie supposition as I really don't understand this line in the trace.

 

I haven't been able to go further in because (I guess) my CG and it's sound card are packed in a big IOMMU group with others devices and it can't be passthrough.

 

The VM error I'm getting if I successfully  boot up

internal error: qemu unexpectedly closed the monitor: 2021-06-07T15:04:55.263662Z qemu-system-x86_64: -device vfio-pci,host=0000:04:00.0,id=hostdev0,bus=pci.0,addr=0x5: vfio 0000:04:00.0: group 15 is not viable
Please ensure all devices within the iommu_group are bound to their vfio bus driver.

 

If at the end these issues aren't related to the VM passthrough problem, these behaviours are kinda weird. 

 

Hope someone already had this problem (well actually no, that's ain't fun 😅) and can help me to solve it  .