Peter Xu <pet...@redhat.com> writes:
> On Fri, Mar 07, 2025 at 05:15:03PM +0530, Prasad Pandit wrote:
>> diff --git a/migration/migration.c b/migration/migration.c
>> index 65fc4f5eed..da2c49c303 100644
>> --- a/migration/migration.c
>> +++ b/migration/migration.c
>> @@ -3401,9 +3401,10 @@ static MigIterateState
>> migration_iteration_run(MigrationState *s)
>> if (!in_postcopy && must_precopy <= s->threshold_size
>> && can_switchover && qatomic_read(&s->start_postcopy)) {
>> if (migrate_multifd()) {
>> - multifd_send_flush();
>> - multifd_send_sync_main(MULTIFD_SYNC_LOCAL);
>> - qemu_savevm_send_multifd_recv_sync(s->to_dst_file);
>> +/* multifd_send_flush();
>> + * multifd_send_sync_main(MULTIFD_SYNC_ALL);
>> + * qemu_savevm_send_multifd_recv_sync(s->to_dst_file);
>> + */
>> + qemu_savevm_state_complete_multifd(s->to_dst_file);
>> multifd_send_shutdown();
>
> Forgot to mention one thing:
>
> If you do flush and sync, IMHO we can keep the threads there and remove
> this shutdown, as long as we are sure it'll be properly shutdown when
> cleanup.
>
> With the assertion in dest threads, I think it should be OK.
Good point. Shutdown at random places makes it difficult to protect
against cleanup races.
