On Tue, Aug 09, 2022, David Woodhouse wrote: > On Tue, 2022-08-09 at 14:59 +0200, Paolo Bonzini wrote: > > On 8/9/22 11:22, David Woodhouse wrote: > > > On Mon, 2022-08-08 at 14:06 -0500, Coleman Dietsch wrote: > > > > Stop Xen timer (if it's running) prior to changing the IRQ vector and > > > > potentially (re)starting the timer. Changing the IRQ vector while the > > > > timer is still running can result in KVM injecting a garbage event, e.g. > > > > vm_xen_inject_timer_irqs() could see a non-zero xen.timer_pending from > > > > a previous timer but inject the new xen.timer_virq. > > > > > > Hm, wasn't that already addressed in the first patch I saw, which just > > > called kvm_xen_stop_timer() unconditionally before (possibly) setting > > > it up again? > > > > Which patch is that? > > The one I acked in > https://lore.kernel.org/all/9bad724858b6a06c25ead865b2b3d9dfc216d01c.camel@xxxxxxxxxxxxx/ It's effectively the same patch. I had asked Coleman to split it into two separate patches: (1) fix the re-initialization of an active timer bug and (2) stop the active timer before changing the vector (this patch).
