On Mon, Jun 30, 2025 at 07:55:49PM +0200, Jason A. Donenfeld wrote:

Hi Yury,

quoted

quoted

quoted

quoted

quoted

diff --git a/drivers/net/wireguard/queueing.h b/drivers/net/wireguard/queueing.h
index 7eb76724b3ed..56314f98b6ba 100644
--- a/drivers/net/wireguard/queueing.h
+++ b/drivers/net/wireguard/queueing.h

@@ -104,16 +104,11 @@ static inline void wg_reset_packet(struct sk_buff *skb, bool encapsulating)

 static inline int wg_cpumask_choose_online(int *stored_cpu, unsigned int id)
 {
-       unsigned int cpu = *stored_cpu, cpu_index, i;
+       unsigned int cpu = *stored_cpu;
+
+       if (unlikely(cpu >= nr_cpu_ids || !cpu_online(cpu)))
+               cpu = *stored_cpu = cpumask_nth(id % num_online_cpus(), cpu_online_mask);

I was about to apply this but then it occurred to me: what happens if
cpu_online_mask changes (shrinks) after num_online_cpus() is evaluated?
cpumask_nth() will then return nr_cpu_ids?

It will return >= nd_cpu_ids. The original version based a for-loop
does the same, so I decided that the caller is safe against it.

Good point. I just checked... This goes into queue_work_on() which
eventually hits:

        /* pwq which will be used unless @work is executing elsewhere */
        if (req_cpu == WORK_CPU_UNBOUND) {

And it turns out WORK_CPU_UNBOUND is the same as nr_cpu_ids. So I guess
that's a fine failure mode.

Actually, cpumask_nth_cpu may return >= nr_cpu_ids because of
small_cpumask_nbits optimization. So it's safer to relax the
condition.

Can you consider applying the following patch for that?

Thanks,
Yury


From fbdce972342437fb12703cae0c3a4f8f9e218a1b Mon Sep 17 00:00:00 2001
From: Yury Norov (NVIDIA) <yury.norov@gmail.com>
Date: Mon, 30 Jun 2025 13:47:49 -0400
Subject: [PATCH] workqueue: relax condition in __queue_work()

Some cpumask search functions may return a number greater than
nr_cpu_ids when nothing is found. Adjust __queue_work() to it.

Signed-off-by: Yury Norov (NVIDIA) <yury.norov@gmail.com>
---
 kernel/workqueue.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 9f9148075828..abacfe157fe6 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c

@@ -2261,7 +2261,7 @@ static void __queue_work(int cpu, struct workqueue_struct *wq,
        rcu_read_lock();
 retry:
        /* pwq which will be used unless @work is executing elsewhere */
-       if (req_cpu == WORK_CPU_UNBOUND) {
+       if (req_cpu >= WORK_CPU_UNBOUND) {
                if (wq->flags & WQ_UNBOUND)
                        cpu = wq_select_unbound_cpu(raw_smp_processor_id());
                else

Seems reasonable to me... Maybe submit this to Tejun and CC me?

Sure, no problem.