fix(scheduler): entry-side check_cancelled in park_current — stop against a QUEUED actor was lossy

request_stop sets the flag and issues a wildcard unpark, but unpark
no-ops on a Queued actor (the pending run "is" the wake). If the
actor's first action on resume is a blocking park — no allocation, no
check!() on the way — the wake-side-only check in park_current is
unreachable: the actor parks with the stop flag already set and no
wake is ever coming. The runtime then idles forever.

Fix: check_cancelled at the ENTRY of park_current, before the switch.
The remaining window (flag lands after the entry check, before the
park) is covered by the existing protocol: the stop's unpark finds
Running / the prep-to-park window, sets Notified, and the park-return
re-queues into the wake-side check. Unwinding from the entry is the
same unwind path as the wake side; leftover wait registrations are
stale-epoch / dead-generation and self-clean at their wakers' failed
CAS.

This is bug #1 of the urus chunk-2 session (bug #2, the terminal
wake, is the preceding commit). Repro: tests/cancel.rs
stop_flagged_while_queued_lands_at_first_park — deadlocks without the
fix (watchdog-bounded), passes in <50ms with it.
This commit is contained in:
Claude
2026-06-11 21:15:03 +00:00
parent eddf3fe929
commit 7bab4d23ea
2 changed files with 70 additions and 0 deletions
+12
View File
@@ -212,6 +212,18 @@ pub fn yield_now() {
} }
pub fn park_current() { pub fn park_current() {
// Entry-side observation point: a stop flagged while we were QUEUED is
// otherwise lost — the stop's wildcard unpark no-ops on a Queued actor
// (the pending run "is" the wake), so if our first action on resume is
// this park, no wake is ever coming and the wake-side check below is
// unreachable. Checking here closes that hole; a flag that lands after
// this check is covered by the existing protocol (the stop's unpark
// finds Running / the prep-to-park window, sets Notified, and the
// park-return re-queues us into the wake-side check). Unwinding from
// here is the same unwind path as the wake side: leftover wait
// registrations are stale-epoch / dead-generation and self-clean at
// their wakers' failed CAS.
crate::preempt::check_cancelled();
runtime::set_yield_intent(YieldIntent::Park); runtime::set_yield_intent(YieldIntent::Park);
unsafe { crate::context::switch_to_scheduler() }; unsafe { crate::context::switch_to_scheduler() };
// Observation point on the wakeup side of every blocking primitive // Observation point on the wakeup side of every blocking primitive
+58
View File
@@ -130,3 +130,61 @@ fn join_on_stopped_actor_returns_ok() {
assert!(h.join().is_ok(), "join on a stopped actor returns Ok(())"); assert!(h.join().is_ok(), "join on a stopped actor returns Ok(())");
}); });
} }
/// Regression: `request_stop` against a QUEUED actor must not be lossy.
///
/// The stop flag is set, but the wildcard unpark no-ops on a Queued actor
/// (the pending run "is" the wake). If the actor's first action on resume
/// is a blocking park — no allocation, no `check!()` on the way — a
/// wake-side-only check in `park_current` never runs: the actor parks with
/// the stop flag already set, and nothing will ever wake it. The runtime
/// then idles forever (the root is parked on the monitor channel).
///
/// Fix: an entry-side `check_cancelled` in `park_current`. The remaining
/// window (flag set after the entry check, before the park lands) is closed
/// by the existing protocol: the stop's unpark then finds Running /
/// the prep-to-park window, sets Notified, and the park-return re-queues
/// into the wake-side check.
///
/// Watchdog harness: without the fix this deadlocks, so the runtime runs on
/// a side thread and the test fails on a timeout instead of hanging cargo.
#[test]
fn stop_flagged_while_queued_lands_at_first_park() {
use std::sync::mpsc;
use std::time::Duration;
let dropped = Arc::new(AtomicBool::new(false));
let saw_stopped = Arc::new(AtomicBool::new(false));
let (d, s) = (dropped.clone(), saw_stopped.clone());
let (done_tx, done_rx) = mpsc::channel::<()>();
std::thread::spawn(move || {
let rt = smarm::init(smarm::Config::exact(1));
rt.run(move || {
let h = spawn(move || {
let _g = DropFlag(d);
let (tx, rx) = channel::<u8>();
let _keep = tx; // keep the channel open: recv() parks
let _ = rx.recv(); // first observation point is this park
});
let pid = h.pid();
let down = monitor(pid);
// Stop while the child is still QUEUED — before it ever runs.
// The unpark no-ops; only the flag is left behind.
request_stop(pid);
let dn = down.rx.recv().expect("monitor channel closed before Down");
assert_eq!(dn.pid, pid);
if matches!(dn.reason, DownReason::Stopped) {
s.store(true, Ordering::SeqCst);
}
let _ = h.join();
});
let _ = done_tx.send(());
});
done_rx
.recv_timeout(Duration::from_secs(10))
.expect("runtime deadlocked: stop against a QUEUED actor was lost at its first park");
assert!(saw_stopped.load(Ordering::SeqCst), "expected DownReason::Stopped");
assert!(dropped.load(Ordering::SeqCst), "Drop guard must run during the cancellation unwind");
}