Skip to contentSkip to conversation

workqueue: Use active mask for new worker when pool is DISASSOCIATED

PATCH Superseded Schspa Shi 2 participants 3 messages Jul 13, 2022 – Jul 14, 2022 Workqueue

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Schspa Shi

author
#0 lore raw

PATCH v2 1/1 workqueue: Use active mask for new worker when pool is DISASSOCIATED

When CPU-[un]hotplugs, all workers will be bound to active CPU via unbind_workers().

But the unbound worker still has a chance to create a new worker, which has bound the newly created task to pool->attrs->cpumask. But the CPU has been unplugged.

Please refer to the following scenarios.

CPU0 CPU1
------------------------------------------------------------------
sched_cpu_deactivate(cpu_active_mask clear) workqueue_offline_cpu(work pool POOL_DISASSOCIATED)
-- all worker will migrate to another cpu --
worker_thread
-- will create new worker if
pool->worklist is not empty create_worker()
-- new kworker will bound to CPU0
(pool->attrs->cpumask will be mask of CPU0).
kworker/0:x will running on rq

sched_cpu_dying
if (rq->nr_running != 1 || rq_has_pinned_tasks(rq))

    WARN(true, "Dying CPU not properly vacated!");
      ---------OOPS-------------

And the crash log as fellowing.

The crash log is as follows:

[ 1622.829074] ------------[ cut here ]------------
[ 1622.829081] Dying CPU not properly vacated!
[ 1622.829091] WARNING: CPU: 3 PID: 31 at kernel/sched/core.c:7756 sched_cpu_dying+0x74/0x204
[ 1622.829374] CPU: 3 PID: 31 Comm: migration/3 Tainted: P           O      5.10.59-rt52 #2
[ 1622.829386] Stopper: multi_cpu_stop+0x0/0x160 <- 0x0
[ 1622.829400] pstate: 60c00009 (nZCv daif +PAN +UAO -TCO BTYPE=--)
[ 1622.829408] pc : sched_cpu_dying+0x74/0x204
[ 1622.829415] lr : sched_cpu_dying+0x74/0x204
[ 1622.829421] sp : ffff800012933c70
[ 1622.829424] pmr_save: 00000060
[ 1622.829426] x29: ffff800012933c70 x28: 0000000000000000
[ 1622.829435] x27: 0000000000000000 x26: 0000000000000001
[ 1622.829444] x25: 0000000000000000 x24: ffff800018353c2c
[ 1622.829452] x23: 0000000000000003 x22: 0000000000000003
[ 1622.829460] x21: 0000000000000059 x20: 0000000000000000
[ 1622.829468] x19: ffff00027ee68a80 x18: 0000000000000000
[ 1622.829477] x17: 0000000000000000 x16: 0000000000000000
[ 1622.829485] x15: ffffffffffffffff x14: ffff80001169ae30
[ 1622.829493] x13: ffffffffffc38c07 x12: ffffffffffffffff
[ 1622.829501] x11: ffffffffffe00000 x10: ffff80001169ae58
[ 1622.829510] x9 : 000000000000001e x8 : ffff80001169ae30
[ 1622.829518] x7 : ffff800012933ab0 x6 : 00000000ffff0e20
[ 1622.829526] x5 : ffff00027ee62988 x4 : 00000000ffff0e20
[ 1622.829535] x3 : ffff800011e30180 x2 : 0000000100000002
[ 1622.829543] x1 : 0000000000000000 x0 : 0000000000000000
[ 1622.829552] Call trace:
[ 1622.829555]  sched_cpu_dying+0x74/0x204
[ 1622.829562]  cpuhp_invoke_callback+0xc0/0x1b0
[ 1622.829571]  take_cpu_down+0xbc/0xd4
[ 1622.829577]  multi_cpu_stop+0x138/0x160
[ 1622.829584]  cpu_stopper_thread+0x9c/0x118
[ 1622.829591]  smpboot_thread_fn+0x1e8/0x1ec
[ 1622.829598]  kthread+0x114/0x124
[ 1622.829606]  ret_from_fork+0x10/0x30
[ 1622.829615] ---[ end trace 0000000000000002 ]---
[ 1623.830273] CPU3 enqueued tasks (2 total):
[ 1623.830291] 	pid: 31, name: migration/3
[ 1623.830440] 	pid: 25654, name: kworker/3:0
[ 1623.830444] task:kworker/3:0     state:R  running task     stack:    0 pid:25654 ppid:     2 flags:0x00000028
[ 1623.830458] Call trace:
[ 1623.830460]  __switch_to+0x164/0x17c
[ 1623.830472]  __schedule+0x4cc/0x5c0
[ 1623.830483]  schedule+0x7c/0xcc
[ 1623.830491]  schedule_preempt_disabled+0x14/0x24
[ 1623.830500]  kthread+0xd8/0x124
[ 1623.830509]  ret_from_fork+0x10/0x30

In the crash log, the error enqueued kworker(pid: 25654, name: kworker/3:0) is a newly created thread, and have bind to a unpluged CPU 3.

Can crash dump can also verify this.
crash> task -R nr_cpus_allowed,cpus_mask 25654
PID: 25654 TASK: ffff000181ff0000 CPU: 3 COMMAND: "kworker/3:0"
nr_cpus_allowed = 1,

  cpus_mask = {
    bits = {8, 0}

},

crash> struct worker_pool.cpu,nr_workers,attrs 0xffff00027ee68380
cpu = 3,
nr_workers = 3,
attrs = 0xffff000180004480,
crash> struct workqueue_attrs 0xffff000180004480
struct workqueue_attrs {
nice = 0,

  cpumask = {{
      bits = {8, 0}
    }},

no_numa = false
}

The stack trace of the bad running task was dumped via the following patch:

And I think this debug patch needs to be added to the mainline, it can help us to debug this kind of problem

To fix it, we can use cpu_active_mask when work pool is DISASSOCIATED.

--

Changelog:
v1 -> v2:
- Move worker task bind to worker_attach_to_pool, remove extra wq_pool_attach_mutex added.
- Add a timing diagram to make this question clearer.

---
 kernel/workqueue.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
kernel/workqueue.c +4 -5
index 1ea50f6be843..3f6e6d884793 100644
@@ -1867,13 +1867,13 @@ static void worker_attach_to_pool(struct worker *worker,
 	 * stable across this function.  See the comments above the flag
 	 * definition for details.
 	 */
-	if (pool->flags & POOL_DISASSOCIATED)
+	if (pool->flags & POOL_DISASSOCIATED) {
 		worker->flags |= WORKER_UNBOUND;
-	else
+		set_cpus_allowed_ptr(worker->task, cpu_active_mask);
+	} else {
 		kthread_set_per_cpu(worker->task, pool->cpu);
-
-	if (worker->rescue_wq)
 		set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask);
+	}
 
 	list_add_tail(&worker->node, &pool->workers);
 	worker->pool = pool;
@@ -1952,7 +1952,6 @@ static struct worker *create_worker(struct worker_pool *pool)
 		goto fail;
 
 	set_user_nice(worker->task, pool->attrs->nice);
-	kthread_bind_mask(worker->task, pool->attrs->cpumask);
 
 	/* successful, attach the worker to the pool */
 	worker_attach_to_pool(worker, pool);
Signature · 1 line
2.29.0
reviewer
#1 lore raw
Quoted 132 lines · Schspa Shi

When CPU-[un]hotplugs, all workers will be bound to active CPU via unbind_workers().

But the unbound worker still has a chance to create a new worker, which has bound the newly created task to pool->attrs->cpumask. But the CPU has been unplugged.

Please refer to the following scenarios.

CPU0 CPU1
------------------------------------------------------------------
sched_cpu_deactivate(cpu_active_mask clear) workqueue_offline_cpu(work pool POOL_DISASSOCIATED)
-- all worker will migrate to another cpu --
worker_thread
-- will create new worker if
pool->worklist is not empty create_worker()
-- new kworker will bound to CPU0
(pool->attrs->cpumask will be mask of CPU0).
kworker/0:x will running on rq

sched_cpu_dying
if (rq->nr_running != 1 || rq_has_pinned_tasks(rq))

    WARN(true, "Dying CPU not properly vacated!");
      ---------OOPS-------------

And the crash log as fellowing.

The crash log is as follows:

[ 1622.829074] ------------[ cut here ]------------
[ 1622.829081] Dying CPU not properly vacated!
[ 1622.829091] WARNING: CPU: 3 PID: 31 at kernel/sched/core.c:7756 sched_cpu_dying+0x74/0x204
[ 1622.829374] CPU: 3 PID: 31 Comm: migration/3 Tainted: P           O      5.10.59-rt52 #2
[ 1622.829386] Stopper: multi_cpu_stop+0x0/0x160 <- 0x0
[ 1622.829400] pstate: 60c00009 (nZCv daif +PAN +UAO -TCO BTYPE=--)
[ 1622.829408] pc : sched_cpu_dying+0x74/0x204
[ 1622.829415] lr : sched_cpu_dying+0x74/0x204
[ 1622.829421] sp : ffff800012933c70
[ 1622.829424] pmr_save: 00000060
[ 1622.829426] x29: ffff800012933c70 x28: 0000000000000000
[ 1622.829435] x27: 0000000000000000 x26: 0000000000000001
[ 1622.829444] x25: 0000000000000000 x24: ffff800018353c2c
[ 1622.829452] x23: 0000000000000003 x22: 0000000000000003
[ 1622.829460] x21: 0000000000000059 x20: 0000000000000000
[ 1622.829468] x19: ffff00027ee68a80 x18: 0000000000000000
[ 1622.829477] x17: 0000000000000000 x16: 0000000000000000
[ 1622.829485] x15: ffffffffffffffff x14: ffff80001169ae30
[ 1622.829493] x13: ffffffffffc38c07 x12: ffffffffffffffff
[ 1622.829501] x11: ffffffffffe00000 x10: ffff80001169ae58
[ 1622.829510] x9 : 000000000000001e x8 : ffff80001169ae30
[ 1622.829518] x7 : ffff800012933ab0 x6 : 00000000ffff0e20
[ 1622.829526] x5 : ffff00027ee62988 x4 : 00000000ffff0e20
[ 1622.829535] x3 : ffff800011e30180 x2 : 0000000100000002
[ 1622.829543] x1 : 0000000000000000 x0 : 0000000000000000
[ 1622.829552] Call trace:
[ 1622.829555]  sched_cpu_dying+0x74/0x204
[ 1622.829562]  cpuhp_invoke_callback+0xc0/0x1b0
[ 1622.829571]  take_cpu_down+0xbc/0xd4
[ 1622.829577]  multi_cpu_stop+0x138/0x160
[ 1622.829584]  cpu_stopper_thread+0x9c/0x118
[ 1622.829591]  smpboot_thread_fn+0x1e8/0x1ec
[ 1622.829598]  kthread+0x114/0x124
[ 1622.829606]  ret_from_fork+0x10/0x30
[ 1622.829615] ---[ end trace 0000000000000002 ]---
[ 1623.830273] CPU3 enqueued tasks (2 total):
[ 1623.830291]  pid: 31, name: migration/3
[ 1623.830440]  pid: 25654, name: kworker/3:0
[ 1623.830444] task:kworker/3:0     state:R  running task     stack:    0 pid:25654 ppid:     2 flags:0x00000028
[ 1623.830458] Call trace:
[ 1623.830460]  __switch_to+0x164/0x17c
[ 1623.830472]  __schedule+0x4cc/0x5c0
[ 1623.830483]  schedule+0x7c/0xcc
[ 1623.830491]  schedule_preempt_disabled+0x14/0x24
[ 1623.830500]  kthread+0xd8/0x124
[ 1623.830509]  ret_from_fork+0x10/0x30

In the crash log, the error enqueued kworker(pid: 25654, name: kworker/3:0) is a newly created thread, and have bind to a unpluged CPU 3.

Can crash dump can also verify this.
crash> task -R nr_cpus_allowed,cpus_mask 25654
PID: 25654 TASK: ffff000181ff0000 CPU: 3 COMMAND: "kworker/3:0"
nr_cpus_allowed = 1,

  cpus_mask = {
    bits = {8, 0}

},

crash> struct worker_pool.cpu,nr_workers,attrs 0xffff00027ee68380
cpu = 3,
nr_workers = 3,
attrs = 0xffff000180004480,
crash> struct workqueue_attrs 0xffff000180004480
struct workqueue_attrs {
nice = 0,

  cpumask = {{
      bits = {8, 0}
    }},

no_numa = false
}

The stack trace of the bad running task was dumped via the following patch:

And I think this debug patch needs to be added to the mainline, it can help us to debug this kind of problem

To fix it, we can use cpu_active_mask when work pool is DISASSOCIATED.

--

Changelog:
v1 -> v2:
- Move worker task bind to worker_attach_to_pool, remove extra wq_pool_attach_mutex added.
- Add a timing diagram to make this question clearer.

---
 kernel/workqueue.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
kernel/workqueue.c +1 -1
index 1ea50f6be843..3f6e6d884793 100644
@@ -1867,13 +1867,13 @@ static void worker_attach_to_pool(struct worker *worker,
         * stable across this function.  See the comments above the flag
         * definition for details.
         */
-       if (pool->flags & POOL_DISASSOCIATED)
+       if (pool->flags & POOL_DISASSOCIATED) {

kernel/workqueue.c static void worker_attach_to_pool(struct worker *worker, · line 1877

Quoted 10 lines
kernel/workqueue.c +3 -3
                worker->flags |= WORKER_UNBOUND;
-       else
+               set_cpus_allowed_ptr(worker->task, cpu_active_mask);
 
+       } else {
                kthread_set_per_cpu(worker->task, pool->cpu);
-
-       if (worker->rescue_wq)
                set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask);
+       }

For unbound pools (get_unbound_pool(), which also has POOL_DISASSOCIATED), pool->attrs->cpumask should be used if pool->attrs->cpumask has active cpu.

         if (cpumask_intersects(pool->attrs->cpumask, cpu_active_mask))
                 set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask);
         else
                 set_cpus_allowed_ptr(worker->task, wq_unbound_cpumask);

And put it above the "if (pool->flags & POOL_DISASSOCIATED)" code block.

kernel/workqueue.c static struct worker *create_worker(struct worker_pool *pool) · line 1955

Quoted 8 lines
kernel/workqueue.c +0 -1
        list_add_tail(&worker->node, &pool->workers);
        worker->pool = pool;
@@ -1952,7 +1952,6 @@ static struct worker *create_worker(struct worker_pool *pool)
                goto fail;
 
        set_user_nice(worker->task, pool->attrs->nice);
-       kthread_bind_mask(worker->task, pool->attrs->cpumask);

worker->flags |= PF_NO_SETAFFINITY;
PF_NO_SETAFFINITY needs to be set for workers.

Quoted 6 lines
kernel/workqueue.c +0 -1
        /* successful, attach the worker to the pool */
        worker_attach_to_pool(worker, pool);
--

2.29.0

author
#2 lore raw
Quoted 149 lines · Lai Jiangshan
Quoted 132 lines · Schspa Shi

When CPU-[un]hotplugs, all workers will be bound to active CPU via unbind_workers().

But the unbound worker still has a chance to create a new worker, which has bound the newly created task to pool->attrs->cpumask. But the CPU has been unplugged.

Please refer to the following scenarios.

CPU0 CPU1
------------------------------------------------------------------
sched_cpu_deactivate(cpu_active_mask clear) workqueue_offline_cpu(work pool POOL_DISASSOCIATED)
-- all worker will migrate to another cpu --
worker_thread
-- will create new worker if
pool->worklist is not empty create_worker()
-- new kworker will bound to CPU0
(pool->attrs->cpumask will be mask of CPU0).
kworker/0:x will running on rq

sched_cpu_dying
if (rq->nr_running != 1 || rq_has_pinned_tasks(rq))

    WARN(true, "Dying CPU not properly vacated!");
      ---------OOPS-------------

And the crash log as fellowing.

The crash log is as follows:

[ 1622.829074] ------------[ cut here ]------------
[ 1622.829081] Dying CPU not properly vacated!
[ 1622.829091] WARNING: CPU: 3 PID: 31 at kernel/sched/core.c:7756 sched_cpu_dying+0x74/0x204
[ 1622.829374] CPU: 3 PID: 31 Comm: migration/3 Tainted: P           O      5.10.59-rt52 #2
[ 1622.829386] Stopper: multi_cpu_stop+0x0/0x160 <- 0x0
[ 1622.829400] pstate: 60c00009 (nZCv daif +PAN +UAO -TCO BTYPE=--)
[ 1622.829408] pc : sched_cpu_dying+0x74/0x204
[ 1622.829415] lr : sched_cpu_dying+0x74/0x204
[ 1622.829421] sp : ffff800012933c70
[ 1622.829424] pmr_save: 00000060
[ 1622.829426] x29: ffff800012933c70 x28: 0000000000000000
[ 1622.829435] x27: 0000000000000000 x26: 0000000000000001
[ 1622.829444] x25: 0000000000000000 x24: ffff800018353c2c
[ 1622.829452] x23: 0000000000000003 x22: 0000000000000003
[ 1622.829460] x21: 0000000000000059 x20: 0000000000000000
[ 1622.829468] x19: ffff00027ee68a80 x18: 0000000000000000
[ 1622.829477] x17: 0000000000000000 x16: 0000000000000000
[ 1622.829485] x15: ffffffffffffffff x14: ffff80001169ae30
[ 1622.829493] x13: ffffffffffc38c07 x12: ffffffffffffffff
[ 1622.829501] x11: ffffffffffe00000 x10: ffff80001169ae58
[ 1622.829510] x9 : 000000000000001e x8 : ffff80001169ae30
[ 1622.829518] x7 : ffff800012933ab0 x6 : 00000000ffff0e20
[ 1622.829526] x5 : ffff00027ee62988 x4 : 00000000ffff0e20
[ 1622.829535] x3 : ffff800011e30180 x2 : 0000000100000002
[ 1622.829543] x1 : 0000000000000000 x0 : 0000000000000000
[ 1622.829552] Call trace:
[ 1622.829555]  sched_cpu_dying+0x74/0x204
[ 1622.829562]  cpuhp_invoke_callback+0xc0/0x1b0
[ 1622.829571]  take_cpu_down+0xbc/0xd4
[ 1622.829577]  multi_cpu_stop+0x138/0x160
[ 1622.829584]  cpu_stopper_thread+0x9c/0x118
[ 1622.829591]  smpboot_thread_fn+0x1e8/0x1ec
[ 1622.829598]  kthread+0x114/0x124
[ 1622.829606]  ret_from_fork+0x10/0x30
[ 1622.829615] ---[ end trace 0000000000000002 ]---
[ 1623.830273] CPU3 enqueued tasks (2 total):
[ 1623.830291]  pid: 31, name: migration/3
[ 1623.830440]  pid: 25654, name: kworker/3:0
[ 1623.830444] task:kworker/3:0     state:R  running task     stack:    0 pid:25654 ppid:     2 flags:0x00000028
[ 1623.830458] Call trace:
[ 1623.830460]  __switch_to+0x164/0x17c
[ 1623.830472]  __schedule+0x4cc/0x5c0
[ 1623.830483]  schedule+0x7c/0xcc
[ 1623.830491]  schedule_preempt_disabled+0x14/0x24
[ 1623.830500]  kthread+0xd8/0x124
[ 1623.830509]  ret_from_fork+0x10/0x30

In the crash log, the error enqueued kworker(pid: 25654, name: kworker/3:0) is a newly created thread, and have bind to a unpluged CPU 3.

Can crash dump can also verify this.
crash> task -R nr_cpus_allowed,cpus_mask 25654
PID: 25654 TASK: ffff000181ff0000 CPU: 3 COMMAND: "kworker/3:0"
nr_cpus_allowed = 1,

  cpus_mask = {
    bits = {8, 0}

},

crash> struct worker_pool.cpu,nr_workers,attrs 0xffff00027ee68380
cpu = 3,
nr_workers = 3,
attrs = 0xffff000180004480,
crash> struct workqueue_attrs 0xffff000180004480
struct workqueue_attrs {
nice = 0,

  cpumask = {{
      bits = {8, 0}
    }},

no_numa = false
}

The stack trace of the bad running task was dumped via the following patch:

And I think this debug patch needs to be added to the mainline, it can help us to debug this kind of problem

To fix it, we can use cpu_active_mask when work pool is DISASSOCIATED.

--

Changelog:
v1 -> v2:
- Move worker task bind to worker_attach_to_pool, remove extra wq_pool_attach_mutex added.
- Add a timing diagram to make this question clearer.

---
 kernel/workqueue.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
kernel/workqueue.c +1 -1
index 1ea50f6be843..3f6e6d884793 100644
@@ -1867,13 +1867,13 @@ static void worker_attach_to_pool(struct worker *worker,
         * stable across this function.  See the comments above the flag
         * definition for details.
         */
-       if (pool->flags & POOL_DISASSOCIATED)
+       if (pool->flags & POOL_DISASSOCIATED) {

kernel/workqueue.c static void worker_attach_to_pool(struct worker *worker, · line 1877

Quoted 10 lines
kernel/workqueue.c +3 -3
                worker->flags |= WORKER_UNBOUND;
-       else
+               set_cpus_allowed_ptr(worker->task, cpu_active_mask);
 
+       } else {
                kthread_set_per_cpu(worker->task, pool->cpu);
-
-       if (worker->rescue_wq)
                set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask);
+       }

For unbound pools (get_unbound_pool(), which also has POOL_DISASSOCIATED), pool->attrs->cpumask should be used if pool->attrs->cpumask has active cpu.

Quoted 7 lines
         if (cpumask_intersects(pool->attrs->cpumask, cpu_active_mask))
                 set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask);
         else
                 set_cpus_allowed_ptr(worker->task, wq_unbound_cpumask);

And put it above the "if (pool->flags & POOL_DISASSOCIATED)" code block.

Thanks for remainding me. will change to this.

Quoted 6 lines
Quoted 6 lines
kernel/workqueue.c +0 -0
        list_add_tail(&worker->node, &pool->workers);
        worker->pool = pool;
@@ -1952,7 +1952,6 @@ static struct worker *create_worker(struct worker_pool *pool)
                goto fail;
 
Quoted 6 lines

kernel/workqueue.c static struct worker *create_worker(struct worker_pool *pool) · line 1955

Quoted 2 lines
kernel/workqueue.c +0 -1
        set_user_nice(worker->task, pool->attrs->nice);
-       kthread_bind_mask(worker->task, pool->attrs->cpumask);

worker->flags |= PF_NO_SETAFFINITY;
PF_NO_SETAFFINITY needs to be set for workers.

Sorry for missing this step, it will be added in the next version of the patch.

Quoted 6 lines
Quoted 6 lines
kernel/workqueue.c +0 -1
        /* successful, attach the worker to the pool */
        worker_attach_to_pool(worker, pool);
--

2.29.0

--
BRs
Schspa Shi