Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
workqueue: Add stall detector sample module
Add a sample module under samples/workqueue/stall_detector/ that
reproduces a workqueue stall caused by PF_WQ_WORKER misuse. The
module queues two work items on the same per-CPU pool, then clears
PF_WQ_WORKER and sleeps in wait_event_idle(), hiding from the
concurrency manager and stalling the second work item indefinitely.
This is useful for testing the workqueue watchdog stall diagnostics.
Signed-off-by: Breno Leitao <leitao@debian.org>
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
authored by
Breno Leitao
and committed by
Tejun Heo
9e83d510
8823eaef
+99
+1
samples/workqueue/stall_detector/Makefile
+1
samples/workqueue/stall_detector/Makefile
···
1
+
obj-m += wq_stall.o
+98
samples/workqueue/stall_detector/wq_stall.c
+98
samples/workqueue/stall_detector/wq_stall.c
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
/*
3
+
* wq_stall - Test module for the workqueue stall detector.
4
+
*
5
+
* Deliberately creates a workqueue stall so the watchdog fires and
6
+
* prints diagnostic output. Useful for verifying that the stall
7
+
* detector correctly identifies stuck workers and produces useful
8
+
* backtraces.
9
+
*
10
+
* The stall is triggered by clearing PF_WQ_WORKER before sleeping,
11
+
* which hides the worker from the concurrency manager. A second
12
+
* work item queued on the same pool then sits in the worklist with
13
+
* no worker available to process it.
14
+
*
15
+
* After ~30s the workqueue watchdog fires:
16
+
* BUG: workqueue lockup - pool cpus=N ...
17
+
*
18
+
* Build:
19
+
* make -C <kernel tree> M=samples/workqueue/stall_detector modules
20
+
*
21
+
* Copyright (c) 2026 Meta Platforms, Inc. and affiliates.
22
+
* Copyright (c) 2026 Breno Leitao <leitao@debian.org>
23
+
*/
24
+
25
+
#include <linux/module.h>
26
+
#include <linux/workqueue.h>
27
+
#include <linux/wait.h>
28
+
#include <linux/atomic.h>
29
+
#include <linux/sched.h>
30
+
31
+
static DECLARE_WAIT_QUEUE_HEAD(stall_wq_head);
32
+
static atomic_t wake_condition = ATOMIC_INIT(0);
33
+
static struct work_struct stall_work1;
34
+
static struct work_struct stall_work2;
35
+
36
+
static void stall_work2_fn(struct work_struct *work)
37
+
{
38
+
pr_info("wq_stall: second work item finally ran\n");
39
+
}
40
+
41
+
static void stall_work1_fn(struct work_struct *work)
42
+
{
43
+
pr_info("wq_stall: first work item running on cpu %d\n",
44
+
raw_smp_processor_id());
45
+
46
+
/*
47
+
* Queue second item while we're still counted as running
48
+
* (pool->nr_running > 0). Since schedule_work() on a per-CPU
49
+
* workqueue targets raw_smp_processor_id(), item 2 lands on the
50
+
* same pool. __queue_work -> kick_pool -> need_more_worker()
51
+
* sees nr_running > 0 and does NOT wake a new worker.
52
+
*/
53
+
schedule_work(&stall_work2);
54
+
55
+
/*
56
+
* Hide from the workqueue concurrency manager. Without
57
+
* PF_WQ_WORKER, schedule() won't call wq_worker_sleeping(),
58
+
* so nr_running is never decremented and no replacement
59
+
* worker is created. Item 2 stays stuck in pool->worklist.
60
+
*/
61
+
current->flags &= ~PF_WQ_WORKER;
62
+
63
+
pr_info("wq_stall: entering wait_event_idle (PF_WQ_WORKER cleared)\n");
64
+
pr_info("wq_stall: expect 'BUG: workqueue lockup' in ~30-60s\n");
65
+
wait_event_idle(stall_wq_head, atomic_read(&wake_condition) != 0);
66
+
67
+
/* Restore so process_one_work() cleanup works correctly */
68
+
current->flags |= PF_WQ_WORKER;
69
+
pr_info("wq_stall: woke up, PF_WQ_WORKER restored\n");
70
+
}
71
+
72
+
static int __init wq_stall_init(void)
73
+
{
74
+
pr_info("wq_stall: loading\n");
75
+
76
+
INIT_WORK(&stall_work1, stall_work1_fn);
77
+
INIT_WORK(&stall_work2, stall_work2_fn);
78
+
schedule_work(&stall_work1);
79
+
80
+
return 0;
81
+
}
82
+
83
+
static void __exit wq_stall_exit(void)
84
+
{
85
+
pr_info("wq_stall: unloading\n");
86
+
atomic_set(&wake_condition, 1);
87
+
wake_up(&stall_wq_head);
88
+
flush_work(&stall_work1);
89
+
flush_work(&stall_work2);
90
+
pr_info("wq_stall: all work flushed, module unloaded\n");
91
+
}
92
+
93
+
module_init(wq_stall_init);
94
+
module_exit(wq_stall_exit);
95
+
96
+
MODULE_LICENSE("GPL");
97
+
MODULE_DESCRIPTION("Reproduce workqueue stall caused by PF_WQ_WORKER misuse");
98
+
MODULE_AUTHOR("Breno Leitao <leitao@debian.org>");