If you come across the following LPM issue with IVM and JS22 blades,

IZ63357: LPM HUNG AT 100% ON BOTH SOURCE AND DESTINATION IVMS

Error description

The problem that these customers are seeing is the same

As described in this particular defect.

Migrating mobile partition from one IVM to another:

The destination IVM GUI shows the migration is complete,

but the percent complete is "unknown".

The source IVM GUI shows the "migration status" as

Migration Complete and "percent Complete" as 100%.

Once the recover operation is performed, the status shows

that the LPAR is now "Running" on the target and removed

from the source, as expected.

There are two PMRs with the same issue and the

Environment is the same on both the cases:

* Firmware level FSP: EA340_075

* IVM VIOS Level: 2.1.1.10-FP21

* Machine Type: JS22 blade

Local fix

Run the "Recover" operation as the Migration appears to

be "hung". Once the recover is complete, the status shows

that the LPAR is now "Running" on the target and removed

from the source, as expected.

Then I recommend you request the following ifix from IBM support:

IZ63357F21.091110.epkg.Z

From IBM support:

It appears that this problem is due to a timing issue.

The trace data shows that IVM is calling the migmover -m end command on the destination partition. That command in turn calls a system call. When everything is in the correct state, that system call will end the migration and cleanup. The problem is it is not being called because the mover has not yet been notified that VASI is in the END state as indicated by the mover's state being: MIG_STATE_COMPLETED. The trace shows this notification arriving a split second after the migmover -m end command was issued.

On a good migration, the mover is notified by VASI of the state change before IVM calls migmover -m. IVM runs migmover -m end once it gets a migration completed message from phyp. If this message is issued about the same time as the message that tells VASI that
the migration has ended, a timing issue seems a real possibility.

The problem was fixed in the VIOS code. A new call was added when the "end" state change comes in from IVM/HMC. If the state is complete we are good to go, otherwise we issue an error.

Installing this fix on my VIOS resolved the problem. My VIOS are running 2.1.1.10-FP21. If you are running FP22 for VIOS 2.1, then you must request a different ifix from IBM support.