To that end, it would be interesting to see if after the first
do-while
loop in timer_interrupt(), you can end up with sched_time >
(processed_system_time + delta).
If so, we could end up running per-cpu processed_system_time ahead of
current system time and so get -ve delta the *next* time we run
timer_interrupt().
In my observations, delta is always positive. It is cpu_delta
which goes negative. I'll go count the cpu specific deltas and
will check to see if they are larger than the system wide delta.