On Fri, Oct 10, 2014 at 09:33:04AM +0200, Marcin Gibuła wrote: > >Does anybody know why the APIC state loaded by the first call to > >kvm_arch_get_registers() is wrong, in the first place? What exactly is > >different in the APIC state in the second kvm_arch_get_registers() call, > >and when/why does it change? > > > >If cpu_synchronize_state() does the wrong thing if it is called at the > >wrong moment, then we may have other hidden bugs, because the user can > >trigger cpu_synchronize_all_states() calls arbitrarily using monitor > >commands. > > My guess is, it's not wrong, it's just outdated when second call occures. > Maybe it's an ordering issue - could kvmclock state change handler be called > before other activity is suspended (?)
Changing the ordering could just hide the problem. If the APIC state in the kernel can change outside ioctl(KVM_RUN) calls, then it can't be cached the same way the other registers loaded by kvm_arch_get_registers() are. The current code invalidates the QEMU register data (i.e. sets kvm_vcpu_dirty=false) only on 3 cases: 1) VCPU init; 2) VCPU reset; 3) on kvm_cpu_exec(), immediately before calling ioctl(KVM_RUN). We may need a separate API to load APIC state? I don't know what are the actual users X86CPU.apic_state, today. -- Eduardo
