On Mon, 2021-06-14 at 14:19 -0700, Jim Mattson wrote: > On Fri, Jun 11, 2021 at 3:31 AM Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote: > > On 10/06/21 22:39, Jim Mattson wrote: > > > But, even worse, it can modify guest memory, > > > even while all vCPU threads are stopped! > > > > To some extent this is a userspace issue---they could declare vCPU > > threads stopped only after KVM_GET_MPSTATE is done, and only start the > > downtime phase of migration after that. But it is nevertheless a pretty > > bad excuse. > > I agree that this could be fixed by documenting the behavior. Since I > don't think there's any existing documentation that says which ioctls > can modify guest memory, such a documentation change wouldn't actually > constitute an API breakage. > > BTW, which ioctls can modify guest memory? > > And, while we're at it, can we document the required orderings of the > various _GET_ and _SET_ ioctls for save and restore? > I strongly vote to make KVM_GET_MP_STATE not change guest state. It will backfire one day. Best regards, Maxim Levitsky
