1=========
2Workqueue
3=========
4
5:Date: September, 2010
6:Author: Tejun Heo <tj@kernel.org>
7:Author: Florian Mickler <florian@mickler.org>
8
9
10Introduction
11============
12
13There are many cases where an asynchronous process execution context
14is needed and the workqueue (wq) API is the most commonly used
15mechanism for such cases.
16
17When such an asynchronous execution context is needed, a work item
18describing which function to execute is put on a queue.  An
19independent thread serves as the asynchronous execution context.  The
20queue is called workqueue and the thread is called worker.
21
22While there are work items on the workqueue the worker executes the
23functions associated with the work items one after the other.  When
24there is no work item left on the workqueue the worker becomes idle.
25When a new work item gets queued, the worker begins executing again.
26
27
28Why Concurrency Managed Workqueue?
29==================================
30
31In the original wq implementation, a multi threaded (MT) wq had one
32worker thread per CPU and a single threaded (ST) wq had one worker
33thread system-wide.  A single MT wq needed to keep around the same
34number of workers as the number of CPUs.  The kernel grew a lot of MT
35wq users over the years and with the number of CPU cores continuously
36rising, some systems saturated the default 32k PID space just booting
37up.
38
39Although MT wq wasted a lot of resource, the level of concurrency
40provided was unsatisfactory.  The limitation was common to both ST and
41MT wq albeit less severe on MT.  Each wq maintained its own separate
42worker pool.  An MT wq could provide only one execution context per CPU
43while an ST wq one for the whole system.  Work items had to compete for
44those very limited execution contexts leading to various problems
45including proneness to deadlocks around the single execution context.
46
47The tension between the provided level of concurrency and resource
48usage also forced its users to make unnecessary tradeoffs like libata
49choosing to use ST wq for polling PIOs and accepting an unnecessary
50limitation that no two polling PIOs can progress at the same time.  As
51MT wq don't provide much better concurrency, users which require
52higher level of concurrency, like async or fscache, had to implement
53their own thread pool.
54
55Concurrency Managed Workqueue (cmwq) is a reimplementation of wq with
56focus on the following goals.
57
58* Maintain compatibility with the original workqueue API.
59
60* Use per-CPU unified worker pools shared by all wq to provide
61  flexible level of concurrency on demand without wasting a lot of
62  resource.
63
64* Automatically regulate worker pool and level of concurrency so that
65  the API users don't need to worry about such details.
66
67
68The Design
69==========
70
71In order to ease the asynchronous execution of functions a new
72abstraction, the work item, is introduced.
73
74A work item is a simple struct that holds a pointer to the function
75that is to be executed asynchronously.  Whenever a driver or subsystem
76wants a function to be executed asynchronously it has to set up a work
77item pointing to that function and queue that work item on a
78workqueue.
79
80Special purpose threads, called worker threads, execute the functions
81off of the queue, one after the other.  If no work is queued, the
82worker threads become idle.  These worker threads are managed in so
83called worker-pools.
84
85The cmwq design differentiates between the user-facing workqueues that
86subsystems and drivers queue work items on and the backend mechanism
87which manages worker-pools and processes the queued work items.
88
89There are two worker-pools, one for normal work items and the other
90for high priority ones, for each possible CPU and some extra
91worker-pools to serve work items queued on unbound workqueues - the
92number of these backing pools is dynamic.
93
94Subsystems and drivers can create and queue work items through special
95workqueue API functions as they see fit. They can influence some
96aspects of the way the work items are executed by setting flags on the
97workqueue they are putting the work item on. These flags include
98things like CPU locality, concurrency limits, priority and more.  To
99get a detailed overview refer to the API description of
100``alloc_workqueue()`` below.
101
102When a work item is queued to a workqueue, the target worker-pool is
103determined according to the queue parameters and workqueue attributes
104and appended on the shared worklist of the worker-pool.  For example,
105unless specifically overridden, a work item of a bound workqueue will
106be queued on the worklist of either normal or highpri worker-pool that
107is associated to the CPU the issuer is running on.
108
109For any worker pool implementation, managing the concurrency level
110(how many execution contexts are active) is an important issue.  cmwq
111tries to keep the concurrency at a minimal but sufficient level.
112Minimal to save resources and sufficient in that the system is used at
113its full capacity.
114
115Each worker-pool bound to an actual CPU implements concurrency
116management by hooking into the scheduler.  The worker-pool is notified
117whenever an active worker wakes up or sleeps and keeps track of the
118number of the currently runnable workers.  Generally, work items are
119not expected to hog a CPU and consume many cycles.  That means
120maintaining just enough concurrency to prevent work processing from
121stalling should be optimal.  As long as there are one or more runnable
122workers on the CPU, the worker-pool doesn't start execution of a new
123work, but, when the last running worker goes to sleep, it immediately
124schedules a new worker so that the CPU doesn't sit idle while there
125are pending work items.  This allows using a minimal number of workers
126without losing execution bandwidth.
127
128Keeping idle workers around doesn't cost other than the memory space
129for kthreads, so cmwq holds onto idle ones for a while before killing
130them.
131
132For unbound workqueues, the number of backing pools is dynamic.
133Unbound workqueue can be assigned custom attributes using
134``apply_workqueue_attrs()`` and workqueue will automatically create
135backing worker pools matching the attributes.  The responsibility of
136regulating concurrency level is on the users.  There is also a flag to
137mark a bound wq to ignore the concurrency management.  Please refer to
138the API section for details.
139
140Forward progress guarantee relies on that workers can be created when
141more execution contexts are necessary, which in turn is guaranteed
142through the use of rescue workers.  All work items which might be used
143on code paths that handle memory reclaim are required to be queued on
144wq's that have a rescue-worker reserved for execution under memory
145pressure.  Else it is possible that the worker-pool deadlocks waiting
146for execution contexts to free up.
147
148
149Application Programming Interface (API)
150=======================================
151
152``alloc_workqueue()`` allocates a wq.  The original
153``create_*workqueue()`` functions are deprecated and scheduled for
154removal.  ``alloc_workqueue()`` takes three arguments - ``@name``,
155``@flags`` and ``@max_active``.  ``@name`` is the name of the wq and
156also used as the name of the rescuer thread if there is one.
157
158A wq no longer manages execution resources but serves as a domain for
159forward progress guarantee, flush and work item attributes. ``@flags``
160and ``@max_active`` control how work items are assigned execution
161resources, scheduled and executed.
162
163
164``flags``
165---------
166
167``WQ_UNBOUND``
168  Work items queued to an unbound wq are served by the special
169  worker-pools which host workers which are not bound to any
170  specific CPU.  This makes the wq behave as a simple execution
171  context provider without concurrency management.  The unbound
172  worker-pools try to start execution of work items as soon as
173  possible.  Unbound wq sacrifices locality but is useful for
174  the following cases.
175
176  * Wide fluctuation in the concurrency level requirement is
177    expected and using bound wq may end up creating large number
178    of mostly unused workers across different CPUs as the issuer
179    hops through different CPUs.
180
181  * Long running CPU intensive workloads which can be better
182    managed by the system scheduler.
183
184``WQ_FREEZABLE``
185  A freezable wq participates in the freeze phase of the system
186  suspend operations.  Work items on the wq are drained and no
187  new work item starts execution until thawed.
188
189``WQ_MEM_RECLAIM``
190  All wq which might be used in the memory reclaim paths **MUST**
191  have this flag set.  The wq is guaranteed to have at least one
192  execution context regardless of memory pressure.
193
194``WQ_HIGHPRI``
195  Work items of a highpri wq are queued to the highpri
196  worker-pool of the target cpu.  Highpri worker-pools are
197  served by worker threads with elevated nice level.
198
199  Note that normal and highpri worker-pools don't interact with
200  each other.  Each maintains its separate pool of workers and
201  implements concurrency management among its workers.
202
203``WQ_CPU_INTENSIVE``
204  Work items of a CPU intensive wq do not contribute to the
205  concurrency level.  In other words, runnable CPU intensive
206  work items will not prevent other work items in the same
207  worker-pool from starting execution.  This is useful for bound
208  work items which are expected to hog CPU cycles so that their
209  execution is regulated by the system scheduler.
210
211  Although CPU intensive work items don't contribute to the
212  concurrency level, start of their executions is still
213  regulated by the concurrency management and runnable
214  non-CPU-intensive work items can delay execution of CPU
215  intensive work items.
216
217  This flag is meaningless for unbound wq.
218
219
220``max_active``
221--------------
222
223``@max_active`` determines the maximum number of execution contexts per
224CPU which can be assigned to the work items of a wq. For example, with
225``@max_active`` of 16, at most 16 work items of the wq can be executing
226at the same time per CPU. This is always a per-CPU attribute, even for
227unbound workqueues.
228
229The maximum limit for ``@max_active`` is 512 and the default value used
230when 0 is specified is 256. These values are chosen sufficiently high
231such that they are not the limiting factor while providing protection in
232runaway cases.
233
234The number of active work items of a wq is usually regulated by the
235users of the wq, more specifically, by how many work items the users
236may queue at the same time.  Unless there is a specific need for
237throttling the number of active work items, specifying '0' is
238recommended.
239
240Some users depend on the strict execution ordering of ST wq.  The
241combination of ``@max_active`` of 1 and ``WQ_UNBOUND`` used to
242achieve this behavior.  Work items on such wq were always queued to the
243unbound worker-pools and only one work item could be active at any given
244time thus achieving the same ordering property as ST wq.
245
246In the current implementation the above configuration only guarantees
247ST behavior within a given NUMA node. Instead ``alloc_ordered_workqueue()`` should
248be used to achieve system-wide ST behavior.
249
250
251Example Execution Scenarios
252===========================
253
254The following example execution scenarios try to illustrate how cmwq
255behave under different configurations.
256
257 Work items w0, w1, w2 are queued to a bound wq q0 on the same CPU.
258 w0 burns CPU for 5ms then sleeps for 10ms then burns CPU for 5ms
259 again before finishing.  w1 and w2 burn CPU for 5ms then sleep for
260 10ms.
261
262Ignoring all other tasks, works and processing overhead, and assuming
263simple FIFO scheduling, the following is one highly simplified version
264of possible sequences of events with the original wq. ::
265
266 TIME IN MSECS	EVENT
267 0		w0 starts and burns CPU
268 5		w0 sleeps
269 15		w0 wakes up and burns CPU
270 20		w0 finishes
271 20		w1 starts and burns CPU
272 25		w1 sleeps
273 35		w1 wakes up and finishes
274 35		w2 starts and burns CPU
275 40		w2 sleeps
276 50		w2 wakes up and finishes
277
278And with cmwq with ``@max_active`` >= 3, ::
279
280 TIME IN MSECS	EVENT
281 0		w0 starts and burns CPU
282 5		w0 sleeps
283 5		w1 starts and burns CPU
284 10		w1 sleeps
285 10		w2 starts and burns CPU
286 15		w2 sleeps
287 15		w0 wakes up and burns CPU
288 20		w0 finishes
289 20		w1 wakes up and finishes
290 25		w2 wakes up and finishes
291
292If ``@max_active`` == 2, ::
293
294 TIME IN MSECS	EVENT
295 0		w0 starts and burns CPU
296 5		w0 sleeps
297 5		w1 starts and burns CPU
298 10		w1 sleeps
299 15		w0 wakes up and burns CPU
300 20		w0 finishes
301 20		w1 wakes up and finishes
302 20		w2 starts and burns CPU
303 25		w2 sleeps
304 35		w2 wakes up and finishes
305
306Now, let's assume w1 and w2 are queued to a different wq q1 which has
307``WQ_CPU_INTENSIVE`` set, ::
308
309 TIME IN MSECS	EVENT
310 0		w0 starts and burns CPU
311 5		w0 sleeps
312 5		w1 and w2 start and burn CPU
313 10		w1 sleeps
314 15		w2 sleeps
315 15		w0 wakes up and burns CPU
316 20		w0 finishes
317 20		w1 wakes up and finishes
318 25		w2 wakes up and finishes
319
320
321Guidelines
322==========
323
324* Do not forget to use ``WQ_MEM_RECLAIM`` if a wq may process work
325  items which are used during memory reclaim.  Each wq with
326  ``WQ_MEM_RECLAIM`` set has an execution context reserved for it.  If
327  there is dependency among multiple work items used during memory
328  reclaim, they should be queued to separate wq each with
329  ``WQ_MEM_RECLAIM``.
330
331* Unless strict ordering is required, there is no need to use ST wq.
332
333* Unless there is a specific need, using 0 for @max_active is
334  recommended.  In most use cases, concurrency level usually stays
335  well under the default limit.
336
337* A wq serves as a domain for forward progress guarantee
338  (``WQ_MEM_RECLAIM``, flush and work item attributes.  Work items
339  which are not involved in memory reclaim and don't need to be
340  flushed as a part of a group of work items, and don't require any
341  special attribute, can use one of the system wq.  There is no
342  difference in execution characteristics between using a dedicated wq
343  and a system wq.
344
345* Unless work items are expected to consume a huge amount of CPU
346  cycles, using a bound wq is usually beneficial due to the increased
347  level of locality in wq operations and work item execution.
348
349
350Affinity Scopes
351===============
352
353An unbound workqueue groups CPUs according to its affinity scope to improve
354cache locality. For example, if a workqueue is using the default affinity
355scope of "cache", it will group CPUs according to last level cache
356boundaries. A work item queued on the workqueue will be assigned to a worker
357on one of the CPUs which share the last level cache with the issuing CPU.
358Once started, the worker may or may not be allowed to move outside the scope
359depending on the ``affinity_strict`` setting of the scope.
360
361Workqueue currently supports the following affinity scopes.
362
363``default``
364  Use the scope in module parameter ``workqueue.default_affinity_scope``
365  which is always set to one of the scopes below.
366
367``cpu``
368  CPUs are not grouped. A work item issued on one CPU is processed by a
369  worker on the same CPU. This makes unbound workqueues behave as per-cpu
370  workqueues without concurrency management.
371
372``smt``
373  CPUs are grouped according to SMT boundaries. This usually means that the
374  logical threads of each physical CPU core are grouped together.
375
376``cache``
377  CPUs are grouped according to cache boundaries. Which specific cache
378  boundary is used is determined by the arch code. L3 is used in a lot of
379  cases. This is the default affinity scope.
380
381``numa``
382  CPUs are grouped according to NUMA bounaries.
383
384``system``
385  All CPUs are put in the same group. Workqueue makes no effort to process a
386  work item on a CPU close to the issuing CPU.
387
388The default affinity scope can be changed with the module parameter
389``workqueue.default_affinity_scope`` and a specific workqueue's affinity
390scope can be changed using ``apply_workqueue_attrs()``.
391
392If ``WQ_SYSFS`` is set, the workqueue will have the following affinity scope
393related interface files under its ``/sys/devices/virtual/workqueue/WQ_NAME/``
394directory.
395
396``affinity_scope``
397  Read to see the current affinity scope. Write to change.
398
399  When default is the current scope, reading this file will also show the
400  current effective scope in parentheses, for example, ``default (cache)``.
401
402``affinity_strict``
403  0 by default indicating that affinity scopes are not strict. When a work
404  item starts execution, workqueue makes a best-effort attempt to ensure
405  that the worker is inside its affinity scope, which is called
406  repatriation. Once started, the scheduler is free to move the worker
407  anywhere in the system as it sees fit. This enables benefiting from scope
408  locality while still being able to utilize other CPUs if necessary and
409  available.
410
411  If set to 1, all workers of the scope are guaranteed always to be in the
412  scope. This may be useful when crossing affinity scopes has other
413  implications, for example, in terms of power consumption or workload
414  isolation. Strict NUMA scope can also be used to match the workqueue
415  behavior of older kernels.
416
417
418Affinity Scopes and Performance
419===============================
420
421It'd be ideal if an unbound workqueue's behavior is optimal for vast
422majority of use cases without further tuning. Unfortunately, in the current
423kernel, there exists a pronounced trade-off between locality and utilization
424necessitating explicit configurations when workqueues are heavily used.
425
426Higher locality leads to higher efficiency where more work is performed for
427the same number of consumed CPU cycles. However, higher locality may also
428cause lower overall system utilization if the work items are not spread
429enough across the affinity scopes by the issuers. The following performance
430testing with dm-crypt clearly illustrates this trade-off.
431
432The tests are run on a CPU with 12-cores/24-threads split across four L3
433caches (AMD Ryzen 9 3900x). CPU clock boost is turned off for consistency.
434``/dev/dm-0`` is a dm-crypt device created on NVME SSD (Samsung 990 PRO) and
435opened with ``cryptsetup`` with default settings.
436
437
438Scenario 1: Enough issuers and work spread across the machine
439-------------------------------------------------------------
440
441The command used: ::
442
443  $ fio --filename=/dev/dm-0 --direct=1 --rw=randrw --bs=32k --ioengine=libaio \
444    --iodepth=64 --runtime=60 --numjobs=24 --time_based --group_reporting \
445    --name=iops-test-job --verify=sha512
446
447There are 24 issuers, each issuing 64 IOs concurrently. ``--verify=sha512``
448makes ``fio`` generate and read back the content each time which makes
449execution locality matter between the issuer and ``kcryptd``. The followings
450are the read bandwidths and CPU utilizations depending on different affinity
451scope settings on ``kcryptd`` measured over five runs. Bandwidths are in
452MiBps, and CPU util in percents.
453
454.. list-table::
455   :widths: 16 20 20
456   :header-rows: 1
457
458   * - Affinity
459     - Bandwidth (MiBps)
460     - CPU util (%)
461
462   * - system
463     - 1159.40 ±1.34
464     - 99.31 ±0.02
465
466   * - cache
467     - 1166.40 ±0.89
468     - 99.34 ±0.01
469
470   * - cache (strict)
471     - 1166.00 ±0.71
472     - 99.35 ±0.01
473
474With enough issuers spread across the system, there is no downside to
475"cache", strict or otherwise. All three configurations saturate the whole
476machine but the cache-affine ones outperform by 0.6% thanks to improved
477locality.
478
479
480Scenario 2: Fewer issuers, enough work for saturation
481-----------------------------------------------------
482
483The command used: ::
484
485  $ fio --filename=/dev/dm-0 --direct=1 --rw=randrw --bs=32k \
486    --ioengine=libaio --iodepth=64 --runtime=60 --numjobs=8 \
487    --time_based --group_reporting --name=iops-test-job --verify=sha512
488
489The only difference from the previous scenario is ``--numjobs=8``. There are
490a third of the issuers but is still enough total work to saturate the
491system.
492
493.. list-table::
494   :widths: 16 20 20
495   :header-rows: 1
496
497   * - Affinity
498     - Bandwidth (MiBps)
499     - CPU util (%)
500
501   * - system
502     - 1155.40 ±0.89
503     - 97.41 ±0.05
504
505   * - cache
506     - 1154.40 ±1.14
507     - 96.15 ±0.09
508
509   * - cache (strict)
510     - 1112.00 ±4.64
511     - 93.26 ±0.35
512
513This is more than enough work to saturate the system. Both "system" and
514"cache" are nearly saturating the machine but not fully. "cache" is using
515less CPU but the better efficiency puts it at the same bandwidth as
516"system".
517
518Eight issuers moving around over four L3 cache scope still allow "cache
519(strict)" to mostly saturate the machine but the loss of work conservation
520is now starting to hurt with 3.7% bandwidth loss.
521
522
523Scenario 3: Even fewer issuers, not enough work to saturate
524-----------------------------------------------------------
525
526The command used: ::
527
528  $ fio --filename=/dev/dm-0 --direct=1 --rw=randrw --bs=32k \
529    --ioengine=libaio --iodepth=64 --runtime=60 --numjobs=4 \
530    --time_based --group_reporting --name=iops-test-job --verify=sha512
531
532Again, the only difference is ``--numjobs=4``. With the number of issuers
533reduced to four, there now isn't enough work to saturate the whole system
534and the bandwidth becomes dependent on completion latencies.
535
536.. list-table::
537   :widths: 16 20 20
538   :header-rows: 1
539
540   * - Affinity
541     - Bandwidth (MiBps)
542     - CPU util (%)
543
544   * - system
545     - 993.60 ±1.82
546     - 75.49 ±0.06
547
548   * - cache
549     - 973.40 ±1.52
550     - 74.90 ±0.07
551
552   * - cache (strict)
553     - 828.20 ±4.49
554     - 66.84 ±0.29
555
556Now, the tradeoff between locality and utilization is clearer. "cache" shows
5572% bandwidth loss compared to "system" and "cache (struct)" whopping 20%.
558
559
560Conclusion and Recommendations
561------------------------------
562
563In the above experiments, the efficiency advantage of the "cache" affinity
564scope over "system" is, while consistent and noticeable, small. However, the
565impact is dependent on the distances between the scopes and may be more
566pronounced in processors with more complex topologies.
567
568While the loss of work-conservation in certain scenarios hurts, it is a lot
569better than "cache (strict)" and maximizing workqueue utilization is
570unlikely to be the common case anyway. As such, "cache" is the default
571affinity scope for unbound pools.
572
573* As there is no one option which is great for most cases, workqueue usages
574  that may consume a significant amount of CPU are recommended to configure
575  the workqueues using ``apply_workqueue_attrs()`` and/or enable
576  ``WQ_SYSFS``.
577
578* An unbound workqueue with strict "cpu" affinity scope behaves the same as
579  ``WQ_CPU_INTENSIVE`` per-cpu workqueue. There is no real advanage to the
580  latter and an unbound workqueue provides a lot more flexibility.
581
582* Affinity scopes are introduced in Linux v6.5. To emulate the previous
583  behavior, use strict "numa" affinity scope.
584
585* The loss of work-conservation in non-strict affinity scopes is likely
586  originating from the scheduler. There is no theoretical reason why the
587  kernel wouldn't be able to do the right thing and maintain
588  work-conservation in most cases. As such, it is possible that future
589  scheduler improvements may make most of these tunables unnecessary.
590
591
592Examining Configuration
593=======================
594
595Use tools/workqueue/wq_dump.py to examine unbound CPU affinity
596configuration, worker pools and how workqueues map to the pools: ::
597
598  $ tools/workqueue/wq_dump.py
599  Affinity Scopes
600  ===============
601  wq_unbound_cpumask=0000000f
602
603  CPU
604    nr_pods  4
605    pod_cpus [0]=00000001 [1]=00000002 [2]=00000004 [3]=00000008
606    pod_node [0]=0 [1]=0 [2]=1 [3]=1
607    cpu_pod  [0]=0 [1]=1 [2]=2 [3]=3
608
609  SMT
610    nr_pods  4
611    pod_cpus [0]=00000001 [1]=00000002 [2]=00000004 [3]=00000008
612    pod_node [0]=0 [1]=0 [2]=1 [3]=1
613    cpu_pod  [0]=0 [1]=1 [2]=2 [3]=3
614
615  CACHE (default)
616    nr_pods  2
617    pod_cpus [0]=00000003 [1]=0000000c
618    pod_node [0]=0 [1]=1
619    cpu_pod  [0]=0 [1]=0 [2]=1 [3]=1
620
621  NUMA
622    nr_pods  2
623    pod_cpus [0]=00000003 [1]=0000000c
624    pod_node [0]=0 [1]=1
625    cpu_pod  [0]=0 [1]=0 [2]=1 [3]=1
626
627  SYSTEM
628    nr_pods  1
629    pod_cpus [0]=0000000f
630    pod_node [0]=-1
631    cpu_pod  [0]=0 [1]=0 [2]=0 [3]=0
632
633  Worker Pools
634  ============
635  pool[00] ref= 1 nice=  0 idle/workers=  4/  4 cpu=  0
636  pool[01] ref= 1 nice=-20 idle/workers=  2/  2 cpu=  0
637  pool[02] ref= 1 nice=  0 idle/workers=  4/  4 cpu=  1
638  pool[03] ref= 1 nice=-20 idle/workers=  2/  2 cpu=  1
639  pool[04] ref= 1 nice=  0 idle/workers=  4/  4 cpu=  2
640  pool[05] ref= 1 nice=-20 idle/workers=  2/  2 cpu=  2
641  pool[06] ref= 1 nice=  0 idle/workers=  3/  3 cpu=  3
642  pool[07] ref= 1 nice=-20 idle/workers=  2/  2 cpu=  3
643  pool[08] ref=42 nice=  0 idle/workers=  6/  6 cpus=0000000f
644  pool[09] ref=28 nice=  0 idle/workers=  3/  3 cpus=00000003
645  pool[10] ref=28 nice=  0 idle/workers= 17/ 17 cpus=0000000c
646  pool[11] ref= 1 nice=-20 idle/workers=  1/  1 cpus=0000000f
647  pool[12] ref= 2 nice=-20 idle/workers=  1/  1 cpus=00000003
648  pool[13] ref= 2 nice=-20 idle/workers=  1/  1 cpus=0000000c
649
650  Workqueue CPU -> pool
651  =====================
652  [    workqueue \ CPU              0  1  2  3 dfl]
653  events                   percpu   0  2  4  6
654  events_highpri           percpu   1  3  5  7
655  events_long              percpu   0  2  4  6
656  events_unbound           unbound  9  9 10 10  8
657  events_freezable         percpu   0  2  4  6
658  events_power_efficient   percpu   0  2  4  6
659  events_freezable_power_  percpu   0  2  4  6
660  rcu_gp                   percpu   0  2  4  6
661  rcu_par_gp               percpu   0  2  4  6
662  slub_flushwq             percpu   0  2  4  6
663  netns                    ordered  8  8  8  8  8
664  ...
665
666See the command's help message for more info.
667
668
669Monitoring
670==========
671
672Use tools/workqueue/wq_monitor.py to monitor workqueue operations: ::
673
674  $ tools/workqueue/wq_monitor.py events
675                              total  infl  CPUtime  CPUhog CMW/RPR  mayday rescued
676  events                      18545     0      6.1       0       5       -       -
677  events_highpri                  8     0      0.0       0       0       -       -
678  events_long                     3     0      0.0       0       0       -       -
679  events_unbound              38306     0      0.1       -       7       -       -
680  events_freezable                0     0      0.0       0       0       -       -
681  events_power_efficient      29598     0      0.2       0       0       -       -
682  events_freezable_power_        10     0      0.0       0       0       -       -
683  sock_diag_events                0     0      0.0       0       0       -       -
684
685                              total  infl  CPUtime  CPUhog CMW/RPR  mayday rescued
686  events                      18548     0      6.1       0       5       -       -
687  events_highpri                  8     0      0.0       0       0       -       -
688  events_long                     3     0      0.0       0       0       -       -
689  events_unbound              38322     0      0.1       -       7       -       -
690  events_freezable                0     0      0.0       0       0       -       -
691  events_power_efficient      29603     0      0.2       0       0       -       -
692  events_freezable_power_        10     0      0.0       0       0       -       -
693  sock_diag_events                0     0      0.0       0       0       -       -
694
695  ...
696
697See the command's help message for more info.
698
699
700Debugging
701=========
702
703Because the work functions are executed by generic worker threads
704there are a few tricks needed to shed some light on misbehaving
705workqueue users.
706
707Worker threads show up in the process list as: ::
708
709  root      5671  0.0  0.0      0     0 ?        S    12:07   0:00 [kworker/0:1]
710  root      5672  0.0  0.0      0     0 ?        S    12:07   0:00 [kworker/1:2]
711  root      5673  0.0  0.0      0     0 ?        S    12:12   0:00 [kworker/0:0]
712  root      5674  0.0  0.0      0     0 ?        S    12:13   0:00 [kworker/1:0]
713
714If kworkers are going crazy (using too much cpu), there are two types
715of possible problems:
716
717	1. Something being scheduled in rapid succession
718	2. A single work item that consumes lots of cpu cycles
719
720The first one can be tracked using tracing: ::
721
722	$ echo workqueue:workqueue_queue_work > /sys/kernel/tracing/set_event
723	$ cat /sys/kernel/tracing/trace_pipe > out.txt
724	(wait a few secs)
725	^C
726
727If something is busy looping on work queueing, it would be dominating
728the output and the offender can be determined with the work item
729function.
730
731For the second type of problems it should be possible to just check
732the stack trace of the offending worker thread. ::
733
734	$ cat /proc/THE_OFFENDING_KWORKER/stack
735
736The work item's function should be trivially visible in the stack
737trace.
738
739
740Non-reentrance Conditions
741=========================
742
743Workqueue guarantees that a work item cannot be re-entrant if the following
744conditions hold after a work item gets queued:
745
746        1. The work function hasn't been changed.
747        2. No one queues the work item to another workqueue.
748        3. The work item hasn't been reinitiated.
749
750In other words, if the above conditions hold, the work item is guaranteed to be
751executed by at most one worker system-wide at any given time.
752
753Note that requeuing the work item (to the same queue) in the self function
754doesn't break these conditions, so it's safe to do. Otherwise, caution is
755required when breaking the conditions inside a work function.
756
757
758Kernel Inline Documentations Reference
759======================================
760
761.. kernel-doc:: include/linux/workqueue.h
762
763.. kernel-doc:: kernel/workqueue.c
764