> FWIW, there was a v3: Yes, this was last version.
The end of discussion was here: https://lists.gnu.org/archive/html/qemu-devel/2021-06/msg02128.html I wasn"t merged, because moslty, you can't have spectre/meltdown/... mitigation enabled cross amd/intel, and also, generally users don't mix intel/amd, so it's already possible to choose the lowest inte/amd cpu model of the cluster. Personnally, I think it could really be usefull for mixed intel/amd cluster. (as kvm64 will not work with newer distros). It's better than kvm64 anymay. (but no mitigation). a better way could be to auto choose the best compatible cpu model across cluster for user at vm creation. with some kind of new virtual "best" cpumodel at vm creation only (then the correct best compatible cpu model is compute and write in vm configuration). We could stream host cpumodel/revision/flags with pvestatd, then compare flags to choose the correct model. (mitigations autotuning could be great too) all flags are documented in libvirt here: https://gitlab.com/libvirt/libvirt/-/blob/master/src/cpu_map/ the simple algo is something like: - if nocluster/single node, set cpumodel to max/host - elsif cluster have only intel hosts, find best intel cpumodel - elsif cluster have only amd hosts, find best amd cpumodel - elsif cluster have mixed intel/amd hosts, find best x64-64-vX cpu model _______________________________________________ pve-devel mailing list pve-devel@lists.proxmox.com https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
