During investigation of RHEL-114506, an unexpected behavior was observed.

Observation

When only a FIFO bash process is running on a CPU (without any SCHED_NORMAL/fair task), the RCU kthreads (rcuc/, ktimers/, irq_work/*) are not being boosted by stalld, even though they would normally be detected as starving.

Test scenario

# Only run FIFO task on CPU 1
taskset -c 1 chrt -f 40 bash -c 'while :; do :; done' &

# Run stalld monitoring CPU 1
taskset -c 0 stalld -v -b queue_track -c 1 -a 0

In this scenario, the RCU kthreads are not boosted by stalld.

However, when a SCHED_NORMAL task is also present on the same CPU, stalld correctly detects and boosts the starving RCU kthreads.

Hypotheses

1. Without a normal task running, the RCU threads may not need explicit scheduling because their work might be done in the context of the running FIFO process or during scheduler switches.
2. There may simply be nothing for the RCU kthreads to do in this scenario.

Investigation needed

This may or may not be a stalld bug. If the test case behavior is erratic, this could be kernel-related rather than a stalld issue.

Verify whether this is expected behavior or requires changes in either stalld's detection/boosting logic or the kernel scheduler.

This was observed in the kernel version 6.12.0-183.el10.x86_64+rt.