On 21/06/2016 11:36, David Vrabel wrote: > On 21/06/16 08:35, Paolo Bonzini wrote: >> >> >> On 21/06/2016 03:44, Jason Wang wrote: >>> >>> >>> On 2016年06月21日 01:53, David Vrabel wrote: >>>> Commit 9d29cdeaaca3a0383af764000b71492c4fc67c6e (rtl8139: port >>>> TallyCounters to vmstate) introduced in incompatibility in the v4 >>>> format as it omitted the RxOkMul counter. >>>> >>>> There are presumably no users that were impacted by the v4 to v4' >>>> breakage, so increase the save version to 5 and re-add the field, >>>> keeping backward compatibility with v4'. >>>> >>>> We can't have a field conditional on the section version in >>>> vmstate_tally_counters since this version checked would not be the >>>> section version (but the version defined in this structure). So, move >>>> all the fields into the main state structure. >>>> >>>> Signed-off-by: David Vrabel <david.vra...@citrix.com> >>> >>> Migration to old version is important for the user and this patch seems >>> to break this. How about something like: >>> >>> - introduce a subsection for RXOKMul >>> - only migrate it for new version (e.g >= 2.7) > > I don't see how this can work with snapshots where the QEMU version that > is going to restore the snapshot is not known in advance.
By "new version" he meant the versioned machine types, e.g. pc-i440fx-2.6 and older wouldn't migrate it. >> Introducing a subsection is not really necessary if the value is going >> to be migrated always, and upstream generally does not have "migrate it >> only in some version" checks. This is left for downstreams to implement >> if they care. We just don't have the manpower to ensure that migration >> to older versions works between all releases of QEMU. > > So is this patch acceptable as is? I think it is. Paolo
