Re: [PATCH v3 1/6] migration: Allow immutable device state to be migrated early (i.e., before RAM)

[Peter Xu <peterx@redhat.com>]

On Tue, Jan 10, 2023 at 12:52:32PM +0100, David Hildenbrand wrote:
> The following seems to work,

That looks much better at least from the diffstat pov (comparing to the
existing patch 1+5 and the framework changes), thanks.

> but makes analyze-migration.py angry:
> 
> $ ./scripts/analyze-migration.py -f STATEFILE
> Traceback (most recent call last):
>   File "/home/dhildenb/git/qemu/./scripts/analyze-migration.py", line 605, in <module>
>     dump.read(dump_memory = args.memory)
>   File "/home/dhildenb/git/qemu/./scripts/analyze-migration.py", line 539, in read
>     classdesc = self.section_classes[section_key]
>                 ~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
> KeyError: ('0000:00:03.0/virtio-mem-early', 0)
> 
> 
> We need the vmdesc to create info for the device.

Migration may ignore the save entry if save_state() not provided in the
"devices" section:

        if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
            continue;
        }

Could you try providing a shim save_state() for the new virtio-mem save
entry?

/*
 * Shim function to make sure the save entry will be dumped into "devices"
 * section, to make analyze-migration.py happy.
 */
static void virtio_mem_save_state_early(QEMUFile *file, void *opaque)
{
}

Then:

static const SaveVMHandlers vmstate_virtio_mem_device_early_ops = {
    .save_setup = virtio_mem_save_setup_early,
    .save_state = virtio_mem_save_state_early,
    .load_state = virtio_mem_load_state_early,
};

I'm not 100% sure it'll work yet, but maybe worth trying.

Thanks,

-- 
Peter Xu