Control: tags -1 + moreinfo Hi Benjamin,
On Fri, Dec 15, 2023 at 01:21:06PM +0100, Benjamin Drung wrote: > The debvm 0.2.13 command1 autopkgtest fails on ppc64el on Ubuntu: > > ``` > qemu-system-ppc64el -netdev > user,id=net0,domainname=novalocal,hostfwd=tcp:127.0.0.1:2222-:22 -device > virtio-net-pci,netdev=net0 -append root=LABEL=debvm rw -nographic -device > virtio-rng-pci,rng=rng0 -object rng-random,filename=/dev/urandom,id=rng0 -cpu > host -machine accel=kvm:tcg -drive > media=disk,format=raw,discard=unmap,file=test.ext4,if=virtio,cache=unsafe > -no-user-config -name debvm-run test.ext4 -m 1G -kernel /proc/self/fd/3 > -initrd /proc/self/fd/4 > ioctl(KVM_CREATE_VM) failed: 22 Invalid argument > PPC KVM module is not loaded. Try modprobe kvm_hv. > qemu-system-ppc64el: failed to initialize kvm: Invalid argument > qemu-system-ppc64el: falling back to tcg > qemu-system-ppc64el: unable to find CPU model 'host' > ``` > > Full log: > https://autopkgtest.ubuntu.com/results/autopkgtest-noble/noble/ppc64el/d/debvm/20231210_104929_cc4af@/log.gz Thank you for the report. I note that the very same autopkgtest succeeds on Debian infrastructure. We should also consider a possible infrastructure issue here. > The environment where debvm is executed does not work with KVM. debvm > should check for that and not enable "accel=kvm:tcg" in that environment. As you can see, it does pass -machine accel=kvm:tcg to support this fallback here. What fails is not the machine selection but the cpu selection. As far as I understand it, we cannot use a similar fallback mechanism for the cpu selection. Therefore debvm tries to open /dev/kvm for writing and only passes -cpu host when that device is accessible. Comparing the log to Debian, we can see that on Debian no -cpu host is being passed. The difference thus is that Ubuntu provides a dysfunctional kvm device accessible to the user under test. Do you agree that this is a deficiency in your test environment? Please remove the moreinfo tag with your response. Helmut