pub fn stall_ejection_skip(result: &VmResult) -> Result<()>Expand description
Environmental skip for a zero-scheduler-activity reading (e.g.
nr_dispatched / nr_yielded read 0 across every periodic sample)
that the guest’s OWN sched_ext runnable-stall watchdog explains: when
the console carries a watchdog ScxExitKind::Stall exit AND
capture_starvation_witness proves the host was contended, the BPF
scheduler was ejected because a descheduling host dilated the 5s
GUEST-time stall window past the timeout — not because the scheduler
failed to dispatch. Returns Err(post_vm_skip(..)) in that case; Ok
otherwise, so a Stall on a quiet host (no witness) or any non-Stall
zero still reaches the caller’s own regression ensure!.
Callers gate their zero-activity failure on it:
ⓘ
if !any_progress {
stall_ejection_skip(result)?;
}
anyhow::ensure!(any_progress, "... never dispatched ...");