On Fri, Jun 21, 2013 at 12:02:02AM +0200, Alexander Graf wrote: > > On 20.06.2013, at 23:59, Peter Maydell wrote: > > > On 20 June 2013 22:55, Alexander Graf <agraf@xxxxxxx> wrote: > >> > >> On 20.06.2013, at 22:37, Christoffer Dall wrote: > >> > >>> On Thu, Jun 20, 2013 at 08:29:30PM +0100, Peter Maydell wrote: > >>>> On 20 June 2013 19:32, Christoffer Dall <christoffer.dall@xxxxxxxxxx> wrote: > >>>>> Marc wrote: > >>>>>> So there is just one thing we absolutely need to make sure here: no vcpu > >>>>>> can run before they've all had their timer restored, and hence a stable > >>>>>> cntvoff. Otherwise two vcpus will have a different view of time. > >>>>>> > >>>>>> Can we guarantee this? > >>>> > >>>>> Do we need to? User space is free to modify time and all sort of other > >>>>> registers at any point during VM execution - it will just break the > >>>>> guest that it's running. > >>>> > >>>> Note that QEMU will stop all CPUs before doing a migration or > >>>> similar operation. However there is a monitor command to query > >>>> the current CPU registers etc which won't try to stop the VM > >>>> first. So we might try to read vcpu registers (though I hope we > >>>> don't allow writing them). > >>>> > >>> Sounds like we need to add a -EBUSY return on SET_ONE_REG if the VM is > >>> running. > >> > >> The ONE_REG API should already be protected here, as it does > >> vcpu_load() in kvm_vcpu_ioctl(). So a separate thread can't possibly > >> do ONE_REG accesses while another thread has the same vcpu running. > > > > Doesn't protect you against confusion due to another thread running > > a different vcpu in the same vm, though. > > Ah, different ONE_REG API. Can't you just notify all vcpus to exit and refresh their timers? That's what kvm_make_request() is there for, no? > yes you can, but I don't think it's worth the trouble to add the code in the kernel to fix a case where user space does something completely broken, which does not muck with the hardware or host state, but can only break the guest. I didn't realize that ONE_REG does vcpu_load() (or, I probably did once, and forgot) so that means we're good. Conclusion on this patch: address Marc's comment to move the user space interface handling out of arch_timer.c and we should be good. Thanks, -Christoffer _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
