On 31 March 2014 15:17, Michael S. Tsirkin <m...@redhat.com> wrote:
> savevm will segfault if version_id < vmsd->minimum_version_id &&
> version_id >= vmsd->minimum_version_id_old
...and the vmstate has no load_state_old handler.
> This calls through a NULL pointer. This is a bug (should
> exit not crash).
I'd previously assumed that this was a vmstate description
bug if it happened (ie that a vmstate with
minimum_version_id_old < minimum_version_id but no
load_state_old wasn't allowed).
Rather than failing migration here, wouldn't it be better
to either:
(a) diagnose the bug, by asserting at the earliest
opportunity
(b) define that the value of minimum_version_id_old is not
relevant unless load_state_old is set
I would strongly prefer (b) -- this would allow us to
remove the now-pointless setting of minimum_version_id_old
in huge numbers of vmstate structures. (Only five devices
make use of load_state_old: acpi, apic, i440fx, pit and
the ppc cpu).
thanks
-- PMM
