[Regression v8.2/v9.0+] stuck at SeaBIOS for >30s with 100% CPU (1T)
Host environment
- Operating system: Gentoo
- OS/kernel version: 6.6.32
- Architecture: amd64
- CPU: AMD Ryzen 7950X3D, AMD Ryzen 5950X, AMD EPYC 7502P (various different systems)
- QEMU flavor: qemu-system-x86_64
- QEMU version: v9.0.0 and v8.2.3 - v8.1.5 is OK
- QEMU command line:
ultra-minimal, still hangs with 100% CPU for ~30s, before going to failed boot due to no disk, etc.:
qemu-system-x86_64 -name template16 -smp cores=16,threads=1,sockets=1 -m 16G -vnc 0.0.0.0:1,lossy=on -enable-kvm -machine q35,accel=kvm -cpu host -display none -vga virtio -daemonize -runas kvm
ultra-ultra minimal, still hangs, so I think it must be a generic issue, I can't really remove anything else without it becoming very hard/impossible to verify if the bug still occurs:
qemu-system-x86_64 -smp cores=16,threads=1,sockets=1 -m 16G -vnc 0.0.0.0:1,lossy=on -enable-kvm -machine q35,accel=kvm -cpu host -display none
Description of problem
starting our Linux direct-kernel-boot VMs with same args on different hosts/hardware will get stuck at SeaBIOS for 30-60s with 100% 1T CPU load starting with v8.2 and also in v9.0. v9.0.0 and v8.2.3 - v8.1.5 is OK. To be clear, everything seems to be fine after that, though I did not do any benchmarks to compare performance. It just delays (re)booting by almost 1 minute, which is a shame, because before that update/regression it was instant and our VMs only take 4s to boot, which is now more like 60s. Downgrading to v8.1 instantly fixes it, upgrading to v8.2/v9.0 instantly breaks it.
Steps to reproduce
- start VM with same args on different versions
somehow if I save this bug with /label ~"kind::Bug"
it disappears, so I'm unable to add/keep the label