Cyclic execution of BE processes in a long windows

[benedond]

I thought that BE processes are executed in a cyclic manner in such a way that if the last one is interrupted (out of budget) and there is still some time remaining within the window, the first one will continue again. But that does not seem to be the case.

Assume this configuration with two infinite processes - limited by a budget of 1 second executed within the 10-seconds long window. I thought that these processes should be repeated executed about 5 times each before the window finishes.

partitions:
  - name: be_a53
    processes:
      - {budget:  1000, frequency:  345.6, cmd: "yes > /dev/null"}
      - {budget:  1000, frequency:  499.2, cmd: "yes > /dev/null"}
windows:
  - length: 10000
    slices:
      - cpu: 0,3,4,5
        be_partition: be_a53

But the trace shows that after suspending the second process (13846), there is an 8-seconds long idle-time before the window starts for the second time.

>>> 13:07:13.809 [trace] Starting BE partitions
>>> 13:07:13.809 [trace] Running process '13845' for '1000 milliseconds'
>>> 13:07:13.809 [trace] Resuming process '13845' (partition 'be_a53')
>>> 13:07:13.809 [trace] Changing CPU frequency to '345.6 MHz' for 'policy0'
>>> 13:07:14.809 [trace] Suspended process '13845' (partition 'be_a53')
>>> 13:07:14.809 [trace] Running process '13846' for '1000 milliseconds'
>>> 13:07:14.810 [trace] Resuming process '13846' (partition 'be_a53')
>>> 13:07:14.810 [trace] Changing CPU frequency to '499.2 MHz' for 'policy0'
>>> 13:07:15.809 [trace] Suspended process '13846' (partition 'be_a53')     # <-- the second process suspended here
>>> 13:07:23.809 [trace] Starting window
>>> 13:07:23.809 [trace] Starting BE partitions
>>> 13:07:23.809 [trace] Running process '13845' for '1000 milliseconds'
>>> 13:07:23.809 [trace] Resuming process '13845' (partition 'be_a53')
>>> 13:07:23.809 [trace] Changing CPU frequency to '345.6 MHz' for 'policy0'
>>> 13:07:24.809 [trace] Suspended process '13845' (partition 'be_a53')
>>> 13:07:24.809 [trace] Running process '13846' for '1000 milliseconds'
>>> 13:07:24.809 [trace] Resuming process '13846' (partition 'be_a53')
>>> 13:07:24.810 [trace] Changing CPU frequency to '499.2 MHz' for 'policy0'
>>> 13:07:25.809 [trace] Suspended process '13846' (partition 'be_a53')

[MatejKafka]

Currently, BE processes are executed at most once. It shouldn't be too hard to cycle them instead, if it's necessary.

[wentasah]

@MatejKafka can you try to implement it? At least a quick draft to see how many changes it would require?

[MatejKafka]

@wentasah See the commit referencing this issue (above) for a quick mockup. This should be functional, but see the commit message.