Heavily loaded host systems may crash Multics with a PRDS stack overflow.

[canthony]

When running CI-kit on some heavily host systems, occasional "Fault in idle process" or " Fault/interrupt while on prds" sys$trouble crashes were observed.

This was determined to be a race condition in Multics triggered by the inability of the simulator to fairly allocate host cycles to the different simulated CPUs.

Details of the analysis are attached as "PRDS stack overflow.pdf"

[canthony]

PRDS stack overflow analysis

[canthony]

Rationale behind the “We see that not all CPUs are responding to the connects in a timely manner.” observation:

CPU 1 FAULT:  Fault 8(010), sub 0(00), 'Connect'
CPU 4 FAULT:  Fault 8(010), sub 0(00), 'Connect'
CPU 1 FAULT:  Fault 8(010), sub 0(00), 'Connect'
CPU 1 FAULT:  Fault 8(010), sub 0(00), 'Connect'
CPU 3 FAULT:  Fault 8(010), sub 0(00), 'Connect'

The second CPU 1 fault on the third line would only happen if the cam_clear loop timed out and reissued the connects; the lack of CPUs 2 and 3 messages suggests that they did not get a chance to respond in time.

[canthony]

Edit: add justification of  “We see that not all CPUs are responding to the connects in a timely manner."

[canthony]

"sys_trouble: Fault while in masked environment." has also been reported.

[Eric Swenson]

​https://s3.amazonaws.com/eswenson-multics/public/mcrs/MCR10144-v1.3.pdf