Alan,

Anton Blanchard reported a lock vs. interrupt-off ordering bug in
migration_thread.  The bug is already fixed in 2.5.  A patch is attached
to fix O(1) in your tree.

See the comments above double_rq_lock and double_rq_unlock (which,
sadly, I wrote)!  Without this patch there is a possibility to hang,
which Anton observed.

Patch is against 2.4.19-pre8-ac1, please apply.

	Robert Love