Re: [PATCH 00/33] hw/cpu/arm: Remove one use of qemu_get_cpu() in A7/A15 MPCore priv

[Peter Xu]

On Wed, Jan 10, 2024 at 09:09:51AM +0100, Markus Armbruster wrote:
> If an object has state that needs to be migrated only sometimes, and
> that part of the state is large enough to bother, we can put it in an
> optional subsection, can't we?
> 
> Destination: if present, take it.  If absent, initialize to default.
> 
> Source: send unless (known to be) in default state.

Hmm.. correct. I think I messed up VMSD's needed() hook with field_exists()
of the fields; my apologies.

The trick should be that VMSD's subsections is more flexible, due to the
fact that vmstate_subsection_load() has:

    while (qemu_peek_byte(f, 0) == QEMU_VM_SUBSECTION) {
        ...
        sub_vmsd = vmstate_get_subsection(vmsd->subsections, idstr);
        ...
    }

So it tolerates the case where the subsection doesn't exist, or out of
order subsections.

While field_exists() hook seems not that flexible, as it's implemented as
this in vmstate_load_state():

    while (field->name) {
        ...
        if (vmstate_field_exists(vmsd, field, opaque, version_id)) {
           ...
        }
        ...
        field++;
    }

So that vmstate_field_exists() needs to be known even on dest before the
main vmsd got loaded, and it should always return the same value as the
source.  Also, the field has ordering requirements.

Then yes, subsection should work for dynamic objects.

Thanks,

-- 
Peter Xu