Re: [PATCH 3/3] selftests/sched_ext: Add dispatch_cookie test

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [Medium] Child processes spin indefinitely and will leak if the parent terminates unexpectedly.
- [Medium] The dispatch operation returns immediately without attempting to dispatch another task if `bpf_task_from_pid()` fails, potentially leaving runnable tasks stranded.
--

commit 91ea259621d3c40168f6eb29f9526e50de76f0b2
Author: Cheng-Yang Chou <yphbchou0911@gmail.com>

selftests/sched_ext: Add dispatch_cookie test

This commit adds a new sched_ext selftest to exercise the dispatch
transaction API (scx_bpf_dsq_insert_begin and scx_bpf_dsq_insert_commit).
It uses a BPF scheduler with a queue map and a userspace program that forks
multiple spinning children, flipping their CPU affinities to trigger dequeue
and re-enqueue cycles.

> diff --git a/tools/testing/selftests/sched_ext/dispatch_cookie.bpf.c b/tools/testing/selftests/sched_ext/dispatch_cookie.bpf.c
> new file mode 100644
> index 0000000000000..160a77ed9000f
> --- /dev/null
> +++ b/tools/testing/selftests/sched_ext/dispatch_cookie.bpf.c

[ ... ]

> +void BPF_STRUCT_OPS(dispatch_cookie_dispatch, s32 cpu,
> +		    struct task_struct *prev)
> +{
> +	s32 pid;
> +	struct task_struct *p;
> +	u64 *stored, token;
> +
> +	if (bpf_map_pop_elem(&queue, &pid))
> +		return;
> +
> +	p = bpf_task_from_pid(pid);
> +	if (!p)
> +		return;

If a task exits and bpf_task_from_pid(pid) returns NULL, does returning
early here prevent other valid tasks in the queue from being dispatched?

Since ops.dispatch() expects a task to be inserted into a DSQ, would it be
better to loop and try popping another PID instead of returning immediately
and leaving the CPU idle?

[ ... ]

> diff --git a/tools/testing/selftests/sched_ext/dispatch_cookie.c b/tools/testing/selftests/sched_ext/dispatch_cookie.c
> new file mode 100644
> index 0000000000000..047d9ea1822b9
> --- /dev/null
> +++ b/tools/testing/selftests/sched_ext/dispatch_cookie.c

[ ... ]

> +	for (i = 0; i < NUM_CHILDREN; i++) {
> +		pids[i] = fork();
> +		if (pids[i] == 0) {
> +			while (1)
> +				sched_yield();
> +		}
> +		if (pids[i] > 0)
> +			nforked++;
> +	}

If the test is interrupted (e.g., via SIGINT, SIGTERM, or a test framework
timeout) before reaching the cleanup code, will these spinning children be
leaked and consume 100% of the CPUs indefinitely?

Could this be prevented by calling prctl(PR_SET_PDEATHSIG, SIGKILL) in the
child processes immediately after fork()?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260515172807.254804-1-yphbchou0911@gmail.com?part=3