Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
From:   richard clark <richard.xnu.clark@gmail.com>
Date:   Mon, 5 Dec 2022 14:18:35 +0800
Message-ID: <CAJNi4rOgYmmtOaXVqYB9sAxDmRhGhS_vVXmZbCbMjvFCQsdjCw@mail.gmail.com>
Subject: work item still be scheduled to execute after destroy_workqueue?
To:     Lai Jiangshan <jiangshanlai@gmail.com>
Cc:     tj@kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Lai and Tejun,

Why can the work still be queued to and executed when queueing it to a
wq has been destroyed, for instance, the below code snippet in a
kernel module:
---------------------->8---------------------

struct workqueue_struct *wq0;
#define MAX_ACTIVE_WORKS        (3)

static void work0_func(struct work_struct *work);
static void work1_func(struct work_struct *work);
static void work2_func(struct work_struct *work);

static DECLARE_WORK(w0, work0_func);
static DECLARE_WORK(w1, work1_func);
static DECLARE_WORK(w2, work2_func);

/* work->func */
static void work0_func(struct work_struct *work)
{
        pr_info("+%s begins to sleep\n", __func__);
        /* sleep for 10s */
        schedule_timeout_interruptible(msecs_to_jiffies(10000));
        pr_info("+%s after sleep, begin to queue another work\n", __func__);
        queue_work_on(1, wq0, &w1);
}

/* work->func */
static void work1_func(struct work_struct *work)
{
        pr_info("+%s scheduled\n", __func__);
}

/* work->func */
static void work2_func(struct work_struct *work)
{
        pr_info("+%s scheduled\n", __func__);
}

static int destroy_init(void)
{
        wq0 = alloc_workqueue("percpu_wq0", 0, MAX_ACTIVE_WORKS);
        if (!wq0) {
                pr_err("alloc_workqueue failed\n");
                return -1;
        }
        queue_work_on(1, wq0, &w0);
        pr_info("Begin to destroy wq0...\n");
        destroy_workqueue(wq0);
        pr_info("queue w2 to the wq0 after destroyed...\n");
        queue_work_on(1, wq0, &w2);

        return 0;
}

The output on my x86_64 box is:

[344702.734480] +destroy_init+
[344702.734499] Begin to destroy wq0...
[344702.734516] +work0_func begins to sleep
[344712.791607] +work0_func after sleep, begin to queue another work
[344712.791620] +work1_func scheduled
[344712.791649] queue w2 to the wq0 after destroyed...
[344712.791663] +work2_func scheduled  <------------- work 2 still be scheduled?