Dave Voutila <d...@sisu.io> writes: > Just a heads up, if you use vmm_clock[1] and virtio_vmmci[2] kernel > modules that I maintain, you'll need to pull the latest. > > Folks in Linux Land decided to twiddle some of the API surface area > forcing me to add yet another block of #elif's and kernel version checks > to the vmmci(4) implementation. The pvclock(4) implementation just > needed a new include. /eyeroll > > Secondarily, if you have strange reboot issues using Alpine guests on > Intel-based hosts, try adding the `reboot=triple` kernel parameter to > your /boot/extlinux.conf. I'm still tracking down something fishy in how > it calls into the BIOS to attempt a reset. Triple-fault works fine. >
Thanks to some sleuthing from kirill@ in the Linux kernel, if you're also experiencing strange failures to reboot on Alpine kernels (you call reboot(1) but it just dumps you back into your userland process/shell), for now a workaround seems to be passing tsc=reliable as a kernel boot arg. So in summary, at least on Intel hosts: tsc=reliable reboot=triple This is disgusting and I hate it...but it works for now. The TSC issue I can look more into, but last I did it's gated I think by a few hardware emulation issues. The reboot=triple stuff still strikes me as a problem with how the SeaBIOS payload is laid out in memory, but I haven't figured it out yet and it only bugs us on Alpine. Sigh. > > [1] https://github.com/voutilad/vmm_clock > [2] https://github.com/voutilad/virtio_vmmci
