Rusty Russell wrote:

>In message <400F5D71.7010702@cyberone.com.au> you write:
>
>>Hi Rusty,
>>Yes I do like it. It hass bothered me that the visible cpus_allowed
>>mask is changed in order to do the balancing. Thanks Rusty. Ingo?
>>
>
>Note that I should have used for_each_online_cpu, not for_each_cpu,
>since we explicitly don't want to look at offline cpus.
>
>New patch with one-line change below...
>Rusty.
>

I'm just reviewing this and porting it to -mm now. I have a few
comments to start with.

CPUs have (I think) always been int in sched.c, you're using unsigned
int. No big deal, but I'll change them to int.

My patches introduce a "move_tasks" function, so I'll rename yours
to migrate_task and __migrate_task.

Your version of move_task (now migrate_task) never actually used dest_cpu
unless I am seeing things. Fixed.

Hows that?