Struct VerifierVmResult 

pub struct VerifierVmResult {
    pub stats: Vec<ProgStats>,
    pub scheduler_log: String,
    pub attach: AttachOutcome,
    pub dispatched: bool,
    pub timed_out: bool,
}

Result of collecting verifier output from a VM run.

Fields

stats: Vec<ProgStats>

Per-program verifier statistics from host-side memory introspection (bpf_prog_aux->verified_insns).

scheduler_log: String

Scheduler log (stdout+stderr) from the VM. Contains libbpf’s verifier instruction traces when BPF load fails.

attach: AttachOutcome

Whether the scheduler positively confirmed attach. Derived from the guest’s lifecycle frames (AttachOutcome). Attach is necessary but NOT sufficient for a cell PASS: this must be AttachOutcome::Attached (the guest reached its post-attach dispatch phase) AND Self::dispatched must be true. Verification alone (non-empty stats) is not enough — a scheduler whose BPF loads but never reaches sched_ext enabled, or a guest that vanishes before the dispatch phase, is a real failure.

dispatched: bool

Whether the guest confirmed the injected verifier workload dispatched — a WorkloadDispatched lifecycle frame, emitted by ktstr_guest_init Phase 5 when the SpinWait probe recorded a worker with non-zero iterations under a confirmed SCHED_EXT policy after attach (so a fair-class fallback cannot false-confirm). A cell PASSes only when this is true AND Self::attach is AttachOutcome::Attached: a scheduler that turns on (sched_ext enabled) but never dispatches a runnable task is a real, distinct failure — worse than never attaching — that the attach verdict alone cannot catch. Derived from the run’s lifecycle frames; scheduler-agnostic — the probe runs as SCHED_EXT, so the BPF scheduler dispatches it under any switch mode (full or SCX_OPS_SWITCH_PARTIAL) and non-zero worker progress proves dispatch, unlike an scx-specific nr_dispatched counter.

timed_out: bool

The host watchdog fired (hard-deadline hang) before the guest exited. Orthogonal to Self::attach: the attach verdict already fails a guest that vanished BEFORE the dispatch phase — an early kernel panic reboots via panic=-1 (an i8042 reset → ExitAction::Shutdown, timed_out == false) and is AttachOutcome::Unconfirmed. This flag catches the remaining case: a guest that wedges AFTER attaching (during teardown), which leaves attach == Attached but never exits. A verifier cell FAILs on it too — but NOT on the guest exit code, which is 1 even on the verifier success path (no #[ktstr_test] body to dispatch).

Implementations

impl VerifierVmResult

pub fn cell_verdict(&self) -> Result<(), String>

The verifier cell PASS/FAIL verdict: Ok(()) when the scheduler verified its BPF, attached (sched_ext enabled), AND dispatched the injected workload; Err(reason) naming the first failing gate otherwise. Gate order — timed_out (a post-attach teardown hang), then attach (did it turn on?), then dispatch (did it schedule a task?) — so the root-cause failure is reported first: an attach failure is named before the dispatch gate it necessarily also trips. Does NOT key on the guest exit code, which is 1 even on the verifier success path (no #[ktstr_test] body to dispatch).