Re: [PATCH v3 2/5] migration: Create MigrationState active field

[Juan Quintela <quintela@redhat.com>]

"Dr. David Alan Gilbert" <dgilbert@redhat.com> wrote:
> * Juan Quintela (quintela@redhat.com) wrote:
>> Right now, there is no easy way to dectect if we have already
>> cancelled/finished/failed a migration.  This field is setup to true
>> when we start a migration, and it is set to false as soon as we stop
>> it.
>> 
>> It fixes a real bug, in ram_save_iterate() we call functions that
>> wrote to the channel even if we know that migration has stopped for
>> any reason.  This gives problems with multifd because we need to
>> synchronize various semoaphores that we don't want to take.
>> 
>> Signed-off-by: Juan Quintela <quintela@redhat.com>
>
> Why can't you use migration_is_active() in the ram.c case?
> My preference would be just to stick with something derived
> from the state rather than tacking another state bit on.

My plan was to go the other way around.
We need to use the state with atomics, I wanted a single way of deciding
if we are/or not in the middle of a migration.  Just now it is too
confusing on my opinion.

>> ---
>>  migration/migration.c | 5 +++++
>>  migration/migration.h | 5 +++++
>>  migration/ram.c       | 2 +-
>>  migration/savevm.c    | 2 ++
>>  4 files changed, 13 insertions(+), 1 deletion(-)
>> 
>> diff --git a/migration/migration.c b/migration/migration.c
>> index 990bff00c0..60bc8710b6 100644
>> --- a/migration/migration.c
>> +++ b/migration/migration.c
>> @@ -1583,6 +1583,8 @@ static void migrate_fd_cancel(MigrationState *s)
>>      QEMUFile *f = migrate_get_current()->to_dst_file;
>>      trace_migrate_fd_cancel();
>>  
>> +    s->active = false;
>> +
>>      if (s->rp_state.from_dst_file) {
>>          /* shutdown the rp socket, so causing the rp thread to shutdown */
>>          qemu_file_shutdown(s->rp_state.from_dst_file);
>> @@ -2834,6 +2836,7 @@ static void migration_completion(MigrationState *s)
>>      }
>>  
>>      if (!migrate_colo_enabled()) {
>> +        s->active = false;
>>          migrate_set_state(&s->state, current_active_state,
>>                            MIGRATION_STATUS_COMPLETED);
>
> You've not always got these two the same way around - i.e. do you change
> the state first or do you set the active state first?  I think it needs
> to be consistent.

Ok.

Thanks, Juan.