long IOP cycle times during startup of front end in deb 10

Relevant TST alogs can be found here:

https://alog.ligo-la.caltech.edu/TST/index.php?callRep=12869 https://alog.ligo-la.caltech.edu/TST/index.php?callRep=12866 https://alog.ligo-la.caltech.edu/TST/index.php?callRep=12865 https://alog.ligo-la.caltech.edu/TST/index.php?callRep=12862

During start up, front ends running debian 10 in dts0 experienced multiple cycle times in the IOP model that were long enough (100 us +) to de-synch the model.

The first of these long cycles was isolated to the allocation of an mbuf shared memory buffer in the awgtpman instance pointed at the IOP model.

Testing showed that the mbuf module recompiled without printk did not cause a long cycle when awgtpman allocated a buffer.

Starting and stopping user models also caused long cycles in IOP. Elimination of printk() and cpu_up(), as cpu_up() also generates a kernel message, prevented long cycles while stopping. We don't know for sure if bringing up the cpu or the kernel message was the source of the problem in cpu_up. This could be tested by re-building the kernel with the message commented out.

The long cycles do not occur on older Xeon E5 machines.