On 14/06/2017 14:41, Roman Kagan wrote: > On Wed, Jun 14, 2017 at 02:11:56PM +0200, Paolo Bonzini wrote: >> On 14/06/2017 13:54, Roman Kagan wrote: >>>> Why not disable the zeroing for host-initiated MSR writes? This is >>>> pretty clearly a KVM bug, we can push it to stable kernels too. >>> >>> The only problem with this is that QEMU will have no reliable way to >>> know if the KVM it runs with has this bug fixed or not. Machines >>> without vmbus work and even migrate fine with the current KVM despite >>> this bug (the only user of those pages currently is synic timers which >>> re-arm themselves and post messages regardless of zeroing). Now >>> updating QEMU to a vmbus-enabled version without updating the kernel >>> will make the migrations cause guest hangs. >> >> Return 2 from KVM_CHECK_EXTENSION(KVM_CAP_HYPERV_SYNIC)? Then you can >> make new QEMU refuse to enable synic if a new kernel is not available. > > Indeed, that's a possibility. > > I'll probably make it in both directions then: on > KVM_ENABLE_CAP(KVM_CAP_HYPERV_SYNIC, 2) disable zeroing completely, > including on guest writes, to better match Hyper-V. Or does it deserve > a separate cap number?
Unfortunately kvm_vcpu_ioctl_enable_cap is not checking arguments and refusing nonzero values, so I'm a bit wary of doing that unconditionally---and I'm not sure it deserves a separate capability number. Maybe a flag KVM_ENABLE_CAP_STRICT_ARG_CHECK in cap->flags, but I don't want you to do too much unrelated work. Keeping the zeroing on guest writes doesn't seem too bad. Or the nuclear option, introduce KVM_CAP_HYPERV_SYNIC2 and drop the broken KVM_CAP_HYPERV_SYNIC altogether. Has its charm... Paolo