On Tue, Oct 26, 2021, David Woodhouse wrote: > On Tue, 2021-10-26 at 14:42 +0000, Sean Christopherson wrote: > > What about taking the lock well early on so that the tail doesn't need to juggle > > errors? Dropping the lock for the KVM_MP_STATE_UNINITIALIZED case is a little > > unfortunate, but that at least pairs with similar logic in x86's other call to > > kvm_vcpu_block(). Relocking if xfer_to_guest_mode_handle_work() triggers an exit > > to userspace is also unfortunate but it's not the end of the world. > > > > On the plus side, the complete_userspace_io() callback doesn't need to worry > > about taking the lock. > > Yeah, that seems sensible for master, but I suspect I'd err on the side > of caution for backporting to stable first? Agreed, dirty-but-simple for stable.
