> -----Original Message----- > From: Jan Beulich <[email protected]> > Sent: 01 October 2019 11:50 > To: Paul Durrant <[email protected]> > Cc: George Dunlap <[email protected]>; Andrew Cooper > <[email protected]>; Roger Pau > Monne <[email protected]>; [email protected]; Juergen Gross > <[email protected]>; Wei Liu > <[email protected]> > Subject: Re: [Xen-devel] [PATCH-for-4.13] x86/mm: don't needlessly veto > migration > > On 01.10.2019 12:40, Paul Durrant wrote: > >> From: George Dunlap <[email protected]> > >> Sent: 01 October 2019 11:34 > >> > >> One thing we could imagine is that when specific devices have an active > >> emulator (or whatever) propagating the dirty information, for that code > >> to tell Xen, "I am implementing dirty tracking for this device". Then > >> when the toolstack enables logdirty, the check can be, "Are there any > >> devices *that don't have external dirty tracking enabled* assigned to > >> the guest?" > > > > And what about existing emulators setting pages dirty at the moment? I > > don't see why Xen's internal dirty page logging is considered definitive > > because AFAICT that is really not the case even now. > > I don't think external emulators already setting pages dirty matter here. > All they want/need to do is advertise which device(s) they take care of. > These emulators actually _help_ Xen maintain a correct picture. What your > patch imo does though is (further) weaken the current model. >
Well that's where we disagree. I don't think the hypervisor currently is the authoritative source of information on the state of the domain. IMO that is what the toolstack is for and Xen should not be refusing to provide its input to the dirty page tracking information simply because it may not have the complete picture. Paul > Jan _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
