rts: Accept data race in work-stealing implementation

This race is okay since the task is owned by the capability pushing it.
By Note [Ownership of Task] this means that the capability is free to
write to `task->cap` without taking `task->lock`.

Fixes #17276.

1 files changed, 3 insertions, 0 deletions

diff --git a/rts/Capability.c b/rts/Capability.c
index 2fe54f2889..0b5f009800 100644
--- a/rts/Capability.c
+++ b/rts/Capability.c
@@ -660,6 +660,9 @@ static Capability * waitForWorkerCapability (Task *task)
         ACQUIRE_LOCK(&task->lock);
         // task->lock held, cap->lock not held
         if (!task->wakeup) waitCondition(&task->cond, &task->lock);
+        // The happens-after matches the happens-before in
+        // schedulePushWork, which does owns 'task' when it sets 'task->cap'.
+        TSAN_ANNOTATE_HAPPENS_AFTER(&task->cap);
         cap = task->cap;
         task->wakeup = false;
         RELEASE_LOCK(&task->lock);