Re: [PATCH] sched_ext: Use IRQ_WORK_INIT_HARD() to initialize sch->disable_irq_work

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] The patch forces scx_disable_irq_workfn to run in hardirq context, causing unbounded execution time with interrupts disabled, which severely breaks PREEMPT_RT latency guarantees.
--

commit 3f0e90ad5f7a3132810aa08d6137cba68235020c
Author: Zqiang <qiang.zhang@linux.dev>

sched_ext: Use IRQ_WORK_INIT_HARD() to initialize sch->disable_irq_work

This commit changes the initialization of the disable_irq_work in sched_ext
to use IRQ_WORK_INIT_HARD(). The intention is to force
scx_disable_irq_workfn() to run in hardirq context rather than a threaded
kthread on PREEMPT_RT, so that scx_dump_state() captures the expected
comm and pid.

> diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
> index 345aa11b84b28..c6b6038a97010 100644
> --- a/kernel/sched/ext.c
> +++ b/kernel/sched/ext.c
[ ... ]
> @@ -6559,7 +6559,7 @@ static struct scx_sched *scx_alloc_and_add_sched(struct sched_ext_ops *ops,
>  
>  	sch->slice_dfl = SCX_SLICE_DFL;
>  	atomic_set(&sch->exit_kind, SCX_EXIT_NONE);
> -	init_irq_work(&sch->disable_irq_work, scx_disable_irq_workfn);
> +	sch->disable_irq_work = IRQ_WORK_INIT_HARD(scx_disable_irq_workfn);

Will executing this in hardirq context cause severe latency spikes or hard
lockups on PREEMPT_RT?

scx_disable_irq_workfn() calls scx_dump_state(), which performs a heavy
workload. It iterates over all possible CPUs, takes their raw_spinlock_t
runqueue locks, formats sequence buffers, saves stack traces, and executes
BPF dump callbacks.

Performing this work with interrupts entirely disabled can break PREEMPT_RT
latency guarantees.

Instead of moving the entire dump to hardirq context, could the original
task's comm and pid be captured and stored in scx_exit_info during
scx_vexit()?

This would allow the heavy dump workload to remain safely deferred to the
threaded kthread execution.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260508115045.23739-1-qiang.zhang@linux.dev?part=1