On Thu, Oct 13, 2022, Sean Christopherson wrote: > On Mon, Oct 10, 2022, Sean Christopherson wrote: > > On Wed, Sep 21, 2022, Michal Luczaj wrote: > > If this fixes things on your end (I'll properly test tomorrow too), I'll post a > > v2 of the entire series. There are some cleanups that can be done on top, e.g. > > I think we should drop kvm_gpc_unmap() entirely until there's actually a user, > > because it's not at all obvious that it's (a) necessary and (b) has desirable > > behavior. > > Sorry for the delay, I initially missed that you included a selftest for the race > in the original RFC. The kernel is no longer exploding, but the test is intermittently > soft hanging waiting for the "IRQ". I'll debug and hopefully post tomorrow. Ended up being a test bug (technically). KVM drops the timer IRQ if the shared info page is invalid. As annoying as that is, there's isn't really a better option, and invalidating a shared page while vCPUs are running really is a VMM bug. To fix, I added an intermediate stage in the test that re-arms the timer if the IRQ doesn't arrive in a reasonable amount of time. Patches incoming...
