On Thu, Sep 25, 2025 at 6:33 AM Matt Fleming matt@readmodwrite.com wrote:
From: Matt Fleming mfleming@cloudflare.com
This reverts commit b7ca5743a2604156d6083b88cefacef983f3a3a6.
If we dequeue a task (task B) that was sched delayed then that task is definitely no longer on the rq and not tracked in the rbtree. Unfortunately, task_on_rq_queued(B) will still return true because dequeue_task() doesn't update p->on_rq.
Hey! Sorry again my patch has been causing you trouble. Thanks for your persistence in chasing this down!
It's confusing as this patch uses the similar logic as logic pick_next_entity() uses when a sched_delayed task is picked to run, as well as elsewhere in __sched_setscheduler() and in sched_ext, so I'd fret that similar
And my impression was that dequeue_task() on a sched_delayed task would update p->on_rq via calling __block_task() at the end of dequeue_entities().
However, there are two spots where we might exit dequeue_entities() early when cfs_rq_throttled(rq), so maybe that's what's catching us here?
Peter: Those cfs_rq_throttled() exits in dequeue_entities() seem a little odd, as the desired dequeue didn't really complete, but dequeue_task_fair() will still return true indicating success - not that too many places are checking the dequeue_task return. Is that right?
thanks -john