On Wed, Apr 05, 2023 at 11:29:51AM +0200, Jan Beulich wrote: > On 05.04.2023 10:59, Roger Pau Monné wrote: > > We should likely move the setting of vm_assist to the domain create > > hypercall, instead of doing it at vCPU initialization. > > Perhaps, the more that setting the assist is limited to vCPU 0 init. > Which in a way makes sense when considering domain creation, but it > is odd for the case of vCPU 0 being brought down, reset, and then > re-initialized; IOW I think arch_set_info_guest() should further > have constrained the setting by a !d->creation_finished check.
Maybe, but still the right fix IMO would be to move this into domain_create. We could add the !d->creation_finished check but that feels more like a bodge than a proper solution. Restoring the previous check is better than nothing, but it would be nice if long term we could get rid of the vpcu related conditions. Thanks, Roger.
