The number of timers to allocate time shares to is calculated by
disregarding system-UID execution. However, the actual allocation
of time *was* considering system-UID locks as well. This wound up
trying to share time across more timers than had been counted in the
first pass, leading to divide-by-zero exceptions.
This patch changes the logic to account the time slice only to
non-system-UID entities, in keeping with the commentary.
Change-Id: I7384a3c269aea44e52bc983927f9417e11613ad5
898c4dc59f