2016-02-03 17:51+0100, Paolo Bonzini: > On 03/02/2016 17:23, Radim Krčmář wrote: >> Discard policy doesn't rely on information from notifiers, so we don't >> need to register notifiers unconditionally. >> >> Use of ps->lock doesn't make sense, but isn't any worse than before.
Oops, it is worse than before ... toggling KVM_REINJECT_CONTROL when the guest is running and reading reinject without locking is now far more complex. This patch should have also ignored KVM_REINJECT_CONTROL when PIT has been started. > Oh, it's perfectly okay. Too fine-grained locks are bad, and lock > contention on ps->lock is a non-issue. > > Can you however add a patch that says what fields of kvm_kpit_state are > protected by which locks? Ok. (I'll be careful to not rewrite the whole PIT while at it. :]) > Then this patch will just add > > /* Protected by kvm_kpit_state lock. */ > > above the reinject field. There was no need to lock reinject in the past and v2 will hopefully achieve it again. > Otherwise > > Reviewed-by: Paolo Bonzini <pbonz...@redhat.com> Thanks. (Might not be applicable to v2, though; sorry.)
