So I set up a Windows 10 vm with vfio on the GPU. Everything seems to be working fine for a while, but after about an hour of GPU usage it will hard lock the whole system. It does not have to be heavy GPU usage, just playing video will do it. As far as I’ve been able to tell it remains stable if I don’t have anything running.
It locks up so badly I have to do a reset button reset to get the machine back up. I cannot find any record of any error anywhere I’ve looked (journal, dmesg, libvirtdomain.log)
I’m at work so I can’t post any actual logs or output from the machine at the moment, what I’m looking for is any ideas on how to find out what is going wrong, so if there’s anything I can post for a follow up please let me know.
I’m not brand new to Linux but I’m still getting comfortable to a full desktop daily driver system not just SSHing in to update server deployments.
I’m running a 2950x on ASRock x399 taichi tridentZ neo 3600c18, passing through 1050ti with gt710 for the host and a 1070ti for another vm.
Need logs at least, XML for the domain, methods used to mask the GPU from the host, what kernel, what distro, how much overclock, etc.
My 2950x has PBO enabled but nothing past that. 128G of ram clocked at 3ghz, running on the x399 Designare. Rock solid thankfully. I only do two GPU passthrough, a GTX 1660 to windows, and a Quadro p400 to Ubuntu (for Plex), the host uses a Radeon something.
So far I’ve tried PBO enabled, and Ram speeds from 3000, 3200, 3333, 3400 at cl16 and cl18. Booting into windows everything is rock solid @ either PBO or straight 4.1GHz with 3600cl18 or 3400 cl16.
My Grub config is as follows
quiet apparmor=1 security=apparmor resume=UUID=b74a48da-1cdc-4ef9-99ac-be4093df318d udev.log_priority=3 amd_iommu=fullflush hardlockup_all_cpu_backtrace=1 idle=nomwait kvm.ignore_msrs=1 default_hugepagesz=1G hugepagesz=1G hugepages=26 vfio-pci.ids=10de:1c82,10de:0fb9,10de:1b82,10de:10f0
I have also tried commandline arguments
iommu=pt pcie_aspm=off amd_iommu.avic=1
My domain.xml is below
1050ti.txt (7.8 KB)
There is nothing that shines any light on the crashes in either dmesg or the journal.
Any ideas would really be appreciated.