Skip to contentSkip to conversation

workqueue: Implement automatic CPU intensive detection and add monitoring

PATCH Superseded Tejun Heo 3 participants 11 messages Apr 18, 2023 – Apr 28, 2023 1 message not in the archive

Subsystems (3)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Tejun Heo

author maintainer
#0 lore raw

PATCH 1/1 workqueue: Implement automatic CPU intensive detection and add monitoring

Hello,

To reduce the number of concurrent worker threads, workqueue holds back starting per-cpu work items while the previous work item stays in the RUNNING state. As such a per-cpu work item which consumes a lot of CPU cycles, even if it has cond_resched()'s in the right places, can stall other per-cpu work items.

To support per-cpu work items that may occupy the CPU for a substantial period of time, workqueue has WQ_CPU_INTENSIVE flag which exempts work items issued through the marked workqueue from concurrency management - they're started immediately and don't block other work items. While this works, it's error-prone in that a workqueue user can easily forget to set the flag or set it unnecessarily. Furthermore, the impacts of the wrong flag setting can be rather indirect and challenging to root-cause.

This patchset makes workqueue auto-detect CPU intensive work items based on CPU consumption. If a work item consumes more than the threshold (5ms by default) of CPU time, it's automatically marked as CPU intensive when it gets scheduled out which unblocks starting of pending per-cpu work items.

The mechanism isn't foolproof in that the detection delays can add up if many CPU-hogging work items are queued at the same time. However, in such situations, the bigger problem likely is the CPU being saturated with per-cpu work items and the solution would be making them UNBOUND. Future changes will make UNBOUND workqueues more attractive by improving their locality behaviors and eventually remove the explicit WQ_CPU_INTENSIVE flag.

While at it, add statistics and a monitoring script. Lack of visibility has always been a bit of pain point when debugging workqueue related issues and with this change and more drastic ones planned for workqueue, this is a good time to address the shortcoming.

This patchset was born out of the discussion in the following thread:

https://lkml.kernel.org/r/CAHk-=wgE9kORADrDJ4nEsHHLirqPCZ1tGaEPAZejHdZ03qCOGg@mail.gmail.com

and contains the following five patches:

0001-workqueue-sched-Notify-workqueue-of-scheduling-of-RU.patch 0002-workqueue-Re-order-struct-worker-fields.patch 0003-workqueue-Move-worker_set-clr_flags-upwards.patch 0004-workqueue-Automatically-mark-CPU-hogging-work-items-.patch 0005-workqueue-Add-pwq-stats-and-a-monitoring-script.patch

and also available in the following git branch:

git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git auto-cpu-intensive

diffstat follows. Thanks.

 Documentation/core-api/workqueue.rst |   30 ++++++++
 kernel/sched/core.c                  |   18 +----
 kernel/workqueue.c                   |  221 +++++++++++++++++++++++++++++++++++++++-----------------------
 kernel/workqueue_internal.h          |   14 +--
 tools/workqueue/wq_monitor.py        |  148 +++++++++++++++++++++++++++++++++++++++++
 5 files changed, 333 insertions(+), 98 deletions(-)

--
tejun

author maintainer
#1 lore raw

PATCH 1/5 workqueue, sched: Notify workqueue of scheduling of RUNNING tasks

When a workqueue kworker goes to sleep, wq_worker_sleeping() is called so that workqueue can manage concurrency. This patch renames wq_worker_sleeping() to wq_worker_stopping() can calls it whenever a kworker is scheduled out whether it's going to sleep or not.

Workqueue will use the schedule-out event of running tasks to automatically detect CPU hogging work items and exclude them from concurrency management so that they can't stall other work items.

This patch just moves the task_is_running() test from sched_submit_work() to wq_worker_stopping(). No behavior change is intended. While at it, remove the already outdated comment which doesn't cover the io_wq case.

Cc Lai Jiangshan <[email protected]>
Cc Peter Zijlstra <[email protected]>
---
 kernel/sched/core.c         | 18 ++++++------------
 kernel/workqueue.c          | 13 ++++++++-----
 kernel/workqueue_internal.h |  2 +-
 3 files changed, 15 insertions(+), 18 deletions(-)
kernel/sched/core.c +6 -12
index 0d18c3969f90..1d83ff00d587 100644
@@ -6650,22 +6650,16 @@ void __noreturn do_task_dead(void)
 
 static inline void sched_submit_work(struct task_struct *tsk)
 {
-	unsigned int task_flags;
+	unsigned int task_flags = tsk->flags;
+
+	if (task_flags & PF_WQ_WORKER)
+		wq_worker_stopping(tsk);
 
 	if (task_is_running(tsk))
 		return;
 
-	task_flags = tsk->flags;
-	/*
-	 * If a worker goes to sleep, notify and ask workqueue whether it
-	 * wants to wake up a task to maintain concurrency.
-	 */
-	if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) {
-		if (task_flags & PF_WQ_WORKER)
-			wq_worker_sleeping(tsk);
-		else
-			io_wq_worker_sleeping(tsk);
-	}
+	if (task_flags & PF_IO_WORKER)
+		io_wq_worker_sleeping(tsk);
 
 	/*
 	 * spinlock and rwlock must not flush block requests.  This will
kernel/workqueue.c +8 -5
index b8b541caed48..6199fbf10cec 100644
@@ -886,17 +886,20 @@ void wq_worker_running(struct task_struct *task)
 }
 
 /**
- * wq_worker_sleeping - a worker is going to sleep
- * @task: task going to sleep
+ * wq_worker_stopping - a worker is stopping
+ * @task: task stopping
  *
- * This function is called from schedule() when a busy worker is
- * going to sleep.
+ * This function is called from schedule() when a busy worker is going off the
+ * CPU.
  */
-void wq_worker_sleeping(struct task_struct *task)
+void wq_worker_stopping(struct task_struct *task)
 {
 	struct worker *worker = kthread_data(task);
 	struct worker_pool *pool;
 
+	if (task_is_running(task))
+		return;
+
 	/*
 	 * Rescuers, which may not have all the fields set up like normal
 	 * workers, also reach here, let's not access anything before
kernel/workqueue_internal.h +1 -1
index e00b1204a8e9..b3b4b2b41d93 100644
@@ -75,7 +75,7 @@ static inline struct worker *current_wq_worker(void)
  * sched/ and workqueue.c.
  */
 void wq_worker_running(struct task_struct *task);
-void wq_worker_sleeping(struct task_struct *task);
+void wq_worker_stopping(struct task_struct *task);
 work_func_t wq_worker_last_func(struct task_struct *task);
 
 #endif /* _KERNEL_WORKQUEUE_INTERNAL_H */
Signature · 1 line
2.40.0
author maintainer
#2 lore raw

PATCH 2/5 workqueue: Re-order struct worker fields

In reply to Tejun Heo

struct worker was laid out with the intent that all fields that are modified for each work item execution are in the first cacheline. However, this hasn't been true for a while with the addition of ->last_func. Let's just collect hot fields together at the top.

Move ->sleeping in the hole after ->current_color and move ->lst_func right below. While at it, drop the cacheline comment which isn't useful anymore.

Cc Lai Jiangshan <[email protected]>
---
 kernel/workqueue_internal.h | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)
kernel/workqueue_internal.h +5 -6
index b3b4b2b41d93..902459b328de 100644
@@ -32,9 +32,12 @@ struct worker {
 	work_func_t		current_func;	/* L: current_work's fn */
 	struct pool_workqueue	*current_pwq;	/* L: current_work's pwq */
 	unsigned int		current_color;	/* L: current_work's color */
-	struct list_head	scheduled;	/* L: scheduled works */
+	int			sleeping;	/* None */
+
+	/* used by the scheduler to determine a worker's last known identity */
+	work_func_t		last_func;	/* L: last work's fn */
 
-	/* 64 bytes boundary on 64bit, 32 on 32bit */
+	struct list_head	scheduled;	/* L: scheduled works */
 
 	struct task_struct	*task;		/* I: worker task */
 	struct worker_pool	*pool;		/* A: the associated pool */
@@ -45,7 +48,6 @@ struct worker {
 	unsigned long		last_active;	/* L: last active timestamp */
 	unsigned int		flags;		/* X: flags */
 	int			id;		/* I: worker id */
-	int			sleeping;	/* None */
 
 	/*
 	 * Opaque string set with work_set_desc().  Printed out with task
@@ -55,9 +57,6 @@ struct worker {
 
 	/* used only by rescuers to point to the target workqueue */
 	struct workqueue_struct	*rescue_wq;	/* I: the workqueue to rescue */
-
-	/* used by the scheduler to determine a worker's last known identity */
-	work_func_t		last_func;
 };
 
 /**
Signature · 1 line
2.40.0
author maintainer
#3 lore raw

PATCH 3/5 workqueue: Move worker_set/clr_flags() upwards

In reply to Tejun Heo

They're gonna be used wq_worker_stopping(). Move them upwards.

Cc Lai Jiangshan <[email protected]>
---
 kernel/workqueue.c | 108 ++++++++++++++++++++++-----------------------
 1 file changed, 54 insertions(+), 54 deletions(-)
kernel/workqueue.c +54 -54
index 6199fbf10cec..b9e8dc54272d 100644
@@ -859,6 +859,60 @@ static void wake_up_worker(struct worker_pool *pool)
 		wake_up_process(worker->task);
 }
 
+/**
+ * worker_set_flags - set worker flags and adjust nr_running accordingly
+ * @worker: self
+ * @flags: flags to set
+ *
+ * Set @flags in @worker->flags and adjust nr_running accordingly.
+ *
+ * CONTEXT:
+ * raw_spin_lock_irq(pool->lock)
+ */
+static inline void worker_set_flags(struct worker *worker, unsigned int flags)
+{
+	struct worker_pool *pool = worker->pool;
+
+	WARN_ON_ONCE(worker->task != current);
+
+	/* If transitioning into NOT_RUNNING, adjust nr_running. */
+	if ((flags & WORKER_NOT_RUNNING) &&
+	    !(worker->flags & WORKER_NOT_RUNNING)) {
+		pool->nr_running--;
+	}
+
+	worker->flags |= flags;
+}
+
+/**
+ * worker_clr_flags - clear worker flags and adjust nr_running accordingly
+ * @worker: self
+ * @flags: flags to clear
+ *
+ * Clear @flags in @worker->flags and adjust nr_running accordingly.
+ *
+ * CONTEXT:
+ * raw_spin_lock_irq(pool->lock)
+ */
+static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
+{
+	struct worker_pool *pool = worker->pool;
+	unsigned int oflags = worker->flags;
+
+	WARN_ON_ONCE(worker->task != current);
+
+	worker->flags &= ~flags;
+
+	/*
+	 * If transitioning out of NOT_RUNNING, increment nr_running.  Note
+	 * that the nested NOT_RUNNING is not a noop.  NOT_RUNNING is mask
+	 * of multiple flags, not a single flag.
+	 */
+	if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING))
+		if (!(worker->flags & WORKER_NOT_RUNNING))
+			pool->nr_running++;
+}
+
 /**
  * wq_worker_running - a worker is running again
  * @task: task waking up
@@ -964,60 +1018,6 @@ work_func_t wq_worker_last_func(struct task_struct *task)
 	return worker->last_func;
 }
 
-/**
- * worker_set_flags - set worker flags and adjust nr_running accordingly
- * @worker: self
- * @flags: flags to set
- *
- * Set @flags in @worker->flags and adjust nr_running accordingly.
- *
- * CONTEXT:
- * raw_spin_lock_irq(pool->lock)
- */
-static inline void worker_set_flags(struct worker *worker, unsigned int flags)
-{
-	struct worker_pool *pool = worker->pool;
-
-	WARN_ON_ONCE(worker->task != current);
-
-	/* If transitioning into NOT_RUNNING, adjust nr_running. */
-	if ((flags & WORKER_NOT_RUNNING) &&
-	    !(worker->flags & WORKER_NOT_RUNNING)) {
-		pool->nr_running--;
-	}
-
-	worker->flags |= flags;
-}
-
-/**
- * worker_clr_flags - clear worker flags and adjust nr_running accordingly
- * @worker: self
- * @flags: flags to clear
- *
- * Clear @flags in @worker->flags and adjust nr_running accordingly.
- *
- * CONTEXT:
- * raw_spin_lock_irq(pool->lock)
- */
-static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
-{
-	struct worker_pool *pool = worker->pool;
-	unsigned int oflags = worker->flags;
-
-	WARN_ON_ONCE(worker->task != current);
-
-	worker->flags &= ~flags;
-
-	/*
-	 * If transitioning out of NOT_RUNNING, increment nr_running.  Note
-	 * that the nested NOT_RUNNING is not a noop.  NOT_RUNNING is mask
-	 * of multiple flags, not a single flag.
-	 */
-	if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING))
-		if (!(worker->flags & WORKER_NOT_RUNNING))
-			pool->nr_running++;
-}
-
 /**
  * find_worker_executing_work - find worker which is executing a work
  * @pool: pool of interest
Signature · 1 line
2.40.0
author maintainer
#4 lore raw

PATCH 4/5 workqueue: Automatically mark CPU-hogging work items CPU_INTENSIVE

In reply to Tejun Heo

If a per-cpu work item hogs the CPU, it can prevent other work items from starting through concurrency management. A per-cpu workqueue which intends to host such CPU-hogging work items can choose to not participate in concurrency management by setting %WQ_CPU_INTENSIVE; however, this can be error-prone and difficult to debug when missed.

This patch adds an automatic CPU usage based detection. If a concurrency-managed work item consumes more CPU time than the threshold (5ms by default), it's marked CPU_INTENSIVE automatically on schedule-out.

The mechanism isn't foolproof in that the 5ms detection delays can add up if many CPU-hogging work items are queued at the same time. However, in such situations, the bigger problem likely is the CPU being saturated with per-cpu work items and the solution would be making them UNBOUND.

For occasional CPU hogging, the new automatic mechanism should provide reasonable protection with minimal increase in code complexity.

Cc Linus Torvalds <[email protected]>
Cc Lai Jiangshan <[email protected]>
---
 kernel/workqueue.c          | 77 ++++++++++++++++++++++++++-----------
 kernel/workqueue_internal.h |  1 +
 2 files changed, 56 insertions(+), 22 deletions(-)
kernel/workqueue.c +55 -22
index b9e8dc54272d..d24b887ddd86 100644
@@ -306,6 +306,14 @@ static struct kmem_cache *pwq_cache;
 static cpumask_var_t *wq_numa_possible_cpumask;
 					/* possible CPUs of each node */
 
+/*
+ * Per-cpu work items which run for longer than the following threshold are
+ * automatically considered CPU intensive and excluded from concurrency
+ * management to prevent them from noticeably delaying other per-cpu work items.
+ */
+static unsigned long wq_cpu_intensive_thresh_us = 5000;
+module_param_named(cpu_intensive_thresh_us, wq_cpu_intensive_thresh_us, ulong, 0644);
+
 static bool wq_disable_numa;
 module_param_named(disable_numa, wq_disable_numa, bool, 0444);
 
@@ -951,9 +959,6 @@ void wq_worker_stopping(struct task_struct *task)
 	struct worker *worker = kthread_data(task);
 	struct worker_pool *pool;
 
-	if (task_is_running(task))
-		return;
-
 	/*
 	 * Rescuers, which may not have all the fields set up like normal
 	 * workers, also reach here, let's not access anything before
@@ -964,24 +969,49 @@ void wq_worker_stopping(struct task_struct *task)
 
 	pool = worker->pool;
 
-	/* Return if preempted before wq_worker_running() was reached */
-	if (worker->sleeping)
-		return;
+	if (task_is_running(task)) {
+		/*
+		 * Concurrency-managed @worker is still RUNNING. See if the
+		 * current work is hogging CPU stalling other per-cpu work
+		 * items. If so, mark @worker CPU_INTENSIVE to exclude it from
+		 * concurrency management. @worker->current_* are stable as they
+		 * can only be modified by @task which is %current.
+		 */
+		if (!worker->current_work ||
+		    task->se.sum_exec_runtime - worker->current_at <
+		    wq_cpu_intensive_thresh_us * NSEC_PER_USEC)
+			return;
 
-	worker->sleeping = 1;
-	raw_spin_lock_irq(&pool->lock);
+		raw_spin_lock_irq(&pool->lock);
+		worker_set_flags(worker, WORKER_CPU_INTENSIVE);
+	} else {
+		/*
+		 * Concurrency-managed @worker is sleeping. Decrement the
+		 * associated pool's nr_running accordingly.
+		 */
 
-	/*
-	 * Recheck in case unbind_workers() preempted us. We don't
-	 * want to decrement nr_running after the worker is unbound
-	 * and nr_running has been reset.
-	 */
-	if (worker->flags & WORKER_NOT_RUNNING) {
-		raw_spin_unlock_irq(&pool->lock);
-		return;
+		/* Return if preempted before wq_worker_running() was reached */
+		if (worker->sleeping)
+			return;
+
+		worker->sleeping = 1;
+		raw_spin_lock_irq(&pool->lock);
+
+		/*
+		 * Recheck in case unbind_workers() preempted us. We don't want
+		 * to decrement nr_running after the worker is unbound and
+		 * nr_running has been reset. As a running worker never sleeps
+		 * inbetween work items, we know that @worker->current_* are
+		 * valid after the following check.
+		 */
+		if (worker->flags & WORKER_NOT_RUNNING) {
+			raw_spin_unlock_irq(&pool->lock);
+			return;
+		}
+
+		pool->nr_running--;
 	}
 
-	pool->nr_running--;
 	if (need_more_worker(pool))
 		wake_up_worker(pool);
 	raw_spin_unlock_irq(&pool->lock);
@@ -2288,7 +2318,6 @@ __acquires(&pool->lock)
 {
 	struct pool_workqueue *pwq = get_work_pwq(work);
 	struct worker_pool *pool = worker->pool;
-	bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE;
 	unsigned long work_data;
 	struct worker *collision;
 #ifdef CONFIG_LOCKDEP
@@ -2325,6 +2354,7 @@ __acquires(&pool->lock)
 	worker->current_work = work;
 	worker->current_func = work->func;
 	worker->current_pwq = pwq;
+	worker->current_at = worker->task->se.sum_exec_runtime;
 	work_data = *work_data_bits(work);
 	worker->current_color = get_work_color(work_data);
 
@@ -2342,7 +2372,7 @@ __acquires(&pool->lock)
 	 * of concurrency management and the next code block will chain
 	 * execution of the pending work items.
 	 */
-	if (unlikely(cpu_intensive))
+	if (unlikely(pwq->wq->flags & WQ_CPU_INTENSIVE))
 		worker_set_flags(worker, WORKER_CPU_INTENSIVE);
 
 	/*
@@ -2420,9 +2450,12 @@ __acquires(&pool->lock)
 
 	raw_spin_lock_irq(&pool->lock);
 
-	/* clear cpu intensive status */
-	if (unlikely(cpu_intensive))
-		worker_clr_flags(worker, WORKER_CPU_INTENSIVE);
+	/*
+	 * In addition to %WQ_CPU_INTENSIVE, @worker may also have been marked
+	 * CPU intensive by wq_worker_stopping() if @work consumed more than
+	 * wq_cpu_intensive_thresh_us. Clear it.
+	 */
+	worker_clr_flags(worker, WORKER_CPU_INTENSIVE);
 
 	/* tag the worker for identification in schedule() */
 	worker->last_func = worker->current_func;
kernel/workqueue_internal.h +1 -0
index 902459b328de..08ffed41f104 100644
@@ -31,6 +31,7 @@ struct worker {
 	struct work_struct	*current_work;	/* L: work being processed */
 	work_func_t		current_func;	/* L: current_work's fn */
 	struct pool_workqueue	*current_pwq;	/* L: current_work's pwq */
+	u64			current_at;	/* L: runtime when current_work started */
 	unsigned int		current_color;	/* L: current_work's color */
 	int			sleeping;	/* None */
 
Signature · 1 line
2.40.0
reviewer
#5 lore raw

Re: [PATCH 4/5] workqueue: Automatically mark CPU-hogging work items CPU_INTENSIVE

Quoted 60 lines · Tejun Heo

If a per-cpu work item hogs the CPU, it can prevent other work items from starting through concurrency management. A per-cpu workqueue which intends to host such CPU-hogging work items can choose to not participate in concurrency management by setting %WQ_CPU_INTENSIVE; however, this can be error-prone and difficult to debug when missed.

This patch adds an automatic CPU usage based detection. If a concurrency-managed work item consumes more CPU time than the threshold (5ms by default), it's marked CPU_INTENSIVE automatically on schedule-out.

The mechanism isn't foolproof in that the 5ms detection delays can add up if many CPU-hogging work items are queued at the same time. However, in such situations, the bigger problem likely is the CPU being saturated with per-cpu work items and the solution would be making them UNBOUND.

For occasional CPU hogging, the new automatic mechanism should provide reasonable protection with minimal increase in code complexity.

Cc Linus Torvalds <[email protected]>
Cc Lai Jiangshan <[email protected]>
---
 kernel/workqueue.c          | 77 ++++++++++++++++++++++++++-----------
 kernel/workqueue_internal.h |  1 +
 2 files changed, 56 insertions(+), 22 deletions(-)
kernel/workqueue.c +8 -3
index b9e8dc54272d..d24b887ddd86 100644
@@ -306,6 +306,14 @@ static struct kmem_cache *pwq_cache;
 static cpumask_var_t *wq_numa_possible_cpumask;
                                        /* possible CPUs of each node */
 
+/*
+ * Per-cpu work items which run for longer than the following threshold are
+ * automatically considered CPU intensive and excluded from concurrency
+ * management to prevent them from noticeably delaying other per-cpu work items.
+ */
+static unsigned long wq_cpu_intensive_thresh_us = 5000;
+module_param_named(cpu_intensive_thresh_us, wq_cpu_intensive_thresh_us, ulong, 0644);
+
 static bool wq_disable_numa;
 module_param_named(disable_numa, wq_disable_numa, bool, 0444);
 
@@ -951,9 +959,6 @@ void wq_worker_stopping(struct task_struct *task)
        struct worker *worker = kthread_data(task);
        struct worker_pool *pool;
 
-       if (task_is_running(task))
-               return;
-
        /*
         * Rescuers, which may not have all the fields set up like normal
         * workers, also reach here, let's not access anything before
@@ -964,24 +969,49 @@ void wq_worker_stopping(struct task_struct *task)
 
        pool = worker->pool;
 

kernel/workqueue.c void wq_worker_stopping(struct task_struct *task) · line 978

Quoted 10 lines
kernel/workqueue.c +7 -3
-       /* Return if preempted before wq_worker_running() was reached */
-       if (worker->sleeping)
-               return;
+       if (task_is_running(task)) {
+               /*
+                * Concurrency-managed @worker is still RUNNING. See if the
+                * current work is hogging CPU stalling other per-cpu work
+                * items. If so, mark @worker CPU_INTENSIVE to exclude it from
+                * concurrency management. @worker->current_* are stable as they
+                * can only be modified by @task which is %current.

Hello

wq_worker_stopping() and sched_submit_work() are only called from schedule() and are not called for other various kinds of scheduling, such as schedule_rtlock(), preempt_schedule_*(), __cond_resched().

A work item hogging CPU may not call the bare schedule(). To make the new wq_worker_stopping() works, it has to be added to other kinds of scheduling, IMO.

Thanks
Lai

author maintainer
#6 lore raw

Re: [PATCH 4/5] workqueue: Automatically mark CPU-hogging work items CPU_INTENSIVE

Hello, Lai.

Quoted 7 lines · Lai Jiangshan

wq_worker_stopping() and sched_submit_work() are only called from schedule() and are not called for other various kinds of scheduling, such as schedule_rtlock(), preempt_schedule_*(), __cond_resched().

A work item hogging CPU may not call the bare schedule(). To make the new wq_worker_stopping() works, it has to be added to other kinds of scheduling, IMO.

Yeah, you're right. The proposed code would work fine only on !preempt kernels. I guess the right thing to do is splitting out the hook for non-sleeping schedulers and adding them in those paths. I'll look into it.

Thanks for spotting the issue.

Signature · 1 line
tejun
maintainer
#7 lore raw

Re: [PATCH 4/5] workqueue: Automatically mark CPU-hogging work items CPU_INTENSIVE

In reply to Tejun Heo
Quoted 5 lines · Tejun Heo

If a per-cpu work item hogs the CPU, it can prevent other work items from starting through concurrency management. A per-cpu workqueue which intends to host such CPU-hogging work items can choose to not participate in concurrency management by setting %WQ_CPU_INTENSIVE; however, this can be error-prone and difficult to debug when missed.

Well; you could use this very heuristic, and instead of silently trying to fix up, complain about the missing CPU_INTENSIVE thing.

Quoted 11 lines

This patch adds an automatic CPU usage based detection. If a concurrency-managed work item consumes more CPU time than the threshold (5ms by default), it's marked CPU_INTENSIVE automatically on schedule-out.

The mechanism isn't foolproof in that the 5ms detection delays can add up if many CPU-hogging work items are queued at the same time. However, in such situations, the bigger problem likely is the CPU being saturated with per-cpu work items and the solution would be making them UNBOUND.

For occasional CPU hogging, the new automatic mechanism should provide reasonable protection with minimal increase in code complexity.

But why not keep it a debug mechanism? Now you're got a heuristic with all the down-sides that they bring.

author maintainer
#8 lore raw

Re: [PATCH 4/5] workqueue: Automatically mark CPU-hogging work items CPU_INTENSIVE

Hello, Peter.

Quoted 9 lines · Peter Zijlstra
Quoted 5 lines · Tejun Heo

If a per-cpu work item hogs the CPU, it can prevent other work items from starting through concurrency management. A per-cpu workqueue which intends to host such CPU-hogging work items can choose to not participate in concurrency management by setting %WQ_CPU_INTENSIVE; however, this can be error-prone and difficult to debug when missed.

Well; you could use this very heuristic, and instead of silently trying to fix up, complain about the missing CPU_INTENSIVE thing.

...

Quoted 2 lines

But why not keep it a debug mechanism? Now you're got a heuristic with all the down-sides that they bring.

I'm working on improving the locality of unbound workqueues and it isn't clear whether there's enough space for per-cpu CPU_INTENSIVE work items - ie. if it's gonna saturate the CPU for extended periods of time to the point of requiring CPU_INTENSIVE, it might as well be an unbound work item if the baseline locality can be good enough. There aren't that many users of CPU_INTENSIVE in tree and we can decide based on how unbound workqueues actually work out for them.

As for warnings, yeah, that'd be great to have regardless of how CPU_INTENSIVE turns out. Just gotta make sure it doesn't fire spuriously and become a nuisance.

Thanks.

Signature · 1 line
tejun
author maintainer
#9 lore raw

PATCH 5/5 workqueue: Add pwq->stats[] and a monitoring script

In reply to Tejun Heo

Currently, the only way to peer into workqueue operations is through tracing. While possible, it isn't easy or convenient to monitor per-workqueue behaviors over time this way. Let's add pwq->stats[] that track relevant events and a drgn monitoring script - tools/workqueue/wq_monitor.py.

It's arguable whether this needs to be configurable. However, it currently only has seven counters and the runtime overhead shouldn't be noticeable given that they're on pwq's which are per-cpu on per-cpu workqueues and per-numa-node on unbound ones. Let's keep it simple for the time being.

Cc Lai Jiangshan <[email protected]>
---
 Documentation/core-api/workqueue.rst |  30 ++++++
 kernel/workqueue.c                   |  29 +++++-
 tools/workqueue/wq_monitor.py        | 148 +++++++++++++++++++++++++++
 3 files changed, 206 insertions(+), 1 deletion(-)
 create mode 100755 tools/workqueue/wq_monitor.py
Documentation/core-api/workqueue.rst +30 -0
index 8ec4d6270b24..a4bb2208100e 100644
@@ -348,6 +348,35 @@ Guidelines
   level of locality in wq operations and work item execution.
 
 
+Monitoring
+==========
+
+Use tools/workqueue/wq_monitor.py to monitor workqueue operations: ::
+
+  $ tools/workqueue/wq_monitor.py events
+                              total  infl  CPUtime  CPUhog  CMwake  mayday rescued
+  events                      18545     0      6.1       0       5       -       -
+  events_highpri                  8     0      0.0       0       0       -       -
+  events_long                     3     0      0.0       0       0       -       -
+  events_unbound              38306     0      0.1       -       -       -       -
+  events_freezable                0     0      0.0       0       0       -       -
+  events_power_efficient      29598     0      0.2       0       0       -       -
+  events_freezable_power_        10     0      0.0       0       0       -       -
+  sock_diag_events                0     0      0.0       0       0       -       -
+
+                              total  infl  CPUtime  CPUhog  CMwake  mayday rescued
+  events                      18548     0      6.1       0       5       -       -
+  events_highpri                  8     0      0.0       0       0       -       -
+  events_long                     3     0      0.0       0       0       -       -
+  events_unbound              38322     0      0.1       -       -       -       -
+  events_freezable                0     0      0.0       0       0       -       -
+  events_power_efficient      29603     0      0.2       0       0       -       -
+  events_freezable_power_        10     0      0.0       0       0       -       -
+  sock_diag_events                0     0      0.0       0       0       -       -
+
+  ...
+
+
 Debugging
 =========
 
@@ -387,6 +416,7 @@ For the second type of problems it should be possible to just check
 The work item's function should be trivially visible in the stack
 trace.
 
+
 Non-reentrance Conditions
 =========================
 
kernel/workqueue.c +28 -1
index d24b887ddd86..c95e80de5ec3 100644
@@ -195,6 +195,22 @@ struct worker_pool {
 	struct rcu_head		rcu;
 };
 
+/*
+ * Per-pool_workqueue statistics. These can be monitored using
+ * tools/workqueue/wq_monitor.py.
+ */
+enum pool_workqueue_stats {
+	PWQ_STAT_STARTED,	/* work items started execution */
+	PWQ_STAT_COMPLETED,	/* work items completed execution */
+	PWQ_STAT_CPU_TIME,	/* total CPU time consumed */
+	PWQ_STAT_CPU_INTENSIVE,	/* wq_cpu_intensive_thresh_us expirations */
+	PWQ_STAT_CM_WAKEUP,	/* concurrency-management worker wakeups */
+	PWQ_STAT_MAYDAY,	/* maydays to rescuer */
+	PWQ_STAT_RESCUED,	/* linked work items executed by rescuer */
+
+	PWQ_NR_STATS,
+};
+
 /*
  * The per-pool workqueue.  While queued, the lower WORK_STRUCT_FLAG_BITS
  * of work_struct->data are used for flags and the remaining high bits
@@ -232,6 +248,8 @@ struct pool_workqueue {
 	struct list_head	pwqs_node;	/* WR: node on wq->pwqs */
 	struct list_head	mayday_node;	/* MD: node on wq->maydays */
 
+	u64			stats[PWQ_NR_STATS];
+
 	/*
 	 * Release of unbound pwq is punted to system_wq.  See put_pwq()
 	 * and pwq_unbound_release_workfn() for details.  pool_workqueue
@@ -984,6 +1002,7 @@ void wq_worker_stopping(struct task_struct *task)
 
 		raw_spin_lock_irq(&pool->lock);
 		worker_set_flags(worker, WORKER_CPU_INTENSIVE);
+		worker->current_pwq->stats[PWQ_STAT_CPU_INTENSIVE]++;
 	} else {
 		/*
 		 * Concurrency-managed @worker is sleeping. Decrement the
@@ -1012,8 +1031,10 @@ void wq_worker_stopping(struct task_struct *task)
 		pool->nr_running--;
 	}
 
-	if (need_more_worker(pool))
+	if (need_more_worker(pool)) {
+		worker->current_pwq->stats[PWQ_STAT_CM_WAKEUP]++;
 		wake_up_worker(pool);
+	}
 	raw_spin_unlock_irq(&pool->lock);
 }
 
@@ -2181,6 +2202,7 @@ static void send_mayday(struct work_struct *work)
 		get_pwq(pwq);
 		list_add_tail(&pwq->mayday_node, &wq->maydays);
 		wake_up_process(wq->rescuer->task);
+		pwq->stats[PWQ_STAT_MAYDAY]++;
 	}
 }
 
@@ -2419,6 +2441,7 @@ __acquires(&pool->lock)
 	 * workqueues), so hiding them isn't a problem.
 	 */
 	lockdep_invariant_state(true);
+	pwq->stats[PWQ_STAT_STARTED]++;
 	trace_workqueue_execute_start(work);
 	worker->current_func(work);
 	/*
@@ -2426,6 +2449,9 @@ __acquires(&pool->lock)
 	 * point will only record its address.
 	 */
 	trace_workqueue_execute_end(work, worker->current_func);
+	pwq->stats[PWQ_STAT_COMPLETED]++;
+	pwq->stats[PWQ_STAT_CPU_TIME] +=
+		worker->task->se.sum_exec_runtime - worker->current_at;
 	lock_map_release(&lockdep_map);
 	lock_map_release(&pwq->wq->lockdep_map);
 
@@ -2672,6 +2698,7 @@ static int rescuer_thread(void *__rescuer)
 				if (first)
 					pool->watchdog_ts = jiffies;
 				move_linked_works(work, scheduled, &n);
+				pwq->stats[PWQ_STAT_RESCUED]++;
 			}
 			first = false;
 		}
tools/workqueue/wq_monitor.py new file +148 -0

148 lines not on this page · raw message

Signature · 1 line
2.40.0

A message that is not in the archive (sent off-list or not yet received): [email protected]

author maintainer
#11 lore raw

Re: [PATCH 4/5] workqueue: Automatically mark CPU-hogging work items CPU_INTENSIVE

Hello,

On Wed, Apr 19, 2023 at 09:45:52AM +0800, Hillf Danton wrote:
...

Quoted 2 lines

Need to wake up another worker in case the current one with WORKER_CPU_INTENSIVE set is going to sleep with works left behind.

Please see below.

Quoted 7 lines

Review

Quoted 7 lines
(quoted diff) +2 -1
+
+		pool->nr_running--;
 	}
 
-	pool->nr_running--;
 	if (need_more_worker(pool))
 		wake_up_worker(pool);

That's what this block is for.

Thanks.

Signature · 1 line
tejun