1 ====================================
2 Concurrency Managed Workqueue (cmwq)
3 ====================================
6 :Author: Tejun Heo <tj@kernel.org>
7 :Author: Florian Mickler <florian@mickler.org>
13 There are many cases where an asynchronous process execution context
14 is needed and the workqueue (wq) API is the most commonly used
15 mechanism for such cases.
17 When such an asynchronous execution context is needed, a work item
18 describing which function to execute is put on a queue. An
19 independent thread serves as the asynchronous execution context. The
20 queue is called workqueue and the thread is called worker.
22 While there are work items on the workqueue the worker executes the
23 functions associated with the work items one after the other. When
24 there is no work item left on the workqueue the worker becomes idle.
25 When a new work item gets queued, the worker begins executing again.
31 In the original wq implementation, a multi threaded (MT) wq had one
32 worker thread per CPU and a single threaded (ST) wq had one worker
33 thread system-wide. A single MT wq needed to keep around the same
34 number of workers as the number of CPUs. The kernel grew a lot of MT
35 wq users over the years and with the number of CPU cores continuously
36 rising, some systems saturated the default 32k PID space just booting
39 Although MT wq wasted a lot of resource, the level of concurrency
40 provided was unsatisfactory. The limitation was common to both ST and
41 MT wq albeit less severe on MT. Each wq maintained its own separate
42 worker pool. An MT wq could provide only one execution context per CPU
43 while an ST wq one for the whole system. Work items had to compete for
44 those very limited execution contexts leading to various problems
45 including proneness to deadlocks around the single execution context.
47 The tension between the provided level of concurrency and resource
48 usage also forced its users to make unnecessary tradeoffs like libata
49 choosing to use ST wq for polling PIOs and accepting an unnecessary
50 limitation that no two polling PIOs can progress at the same time. As
51 MT wq don't provide much better concurrency, users which require
52 higher level of concurrency, like async or fscache, had to implement
53 their own thread pool.
55 Concurrency Managed Workqueue (cmwq) is a reimplementation of wq with
56 focus on the following goals.
58 * Maintain compatibility with the original workqueue API.
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
64 * Automatically regulate worker pool and level of concurrency so that
65 the API users don't need to worry about such details.
71 In order to ease the asynchronous execution of functions a new
72 abstraction, the work item, is introduced.
74 A work item is a simple struct that holds a pointer to the function
75 that is to be executed asynchronously. Whenever a driver or subsystem
76 wants a function to be executed asynchronously it has to set up a work
77 item pointing to that function and queue that work item on a
80 Special purpose threads, called worker threads, execute the functions
81 off of the queue, one after the other. If no work is queued, the
82 worker threads become idle. These worker threads are managed in so
85 The cmwq design differentiates between the user-facing workqueues that
86 subsystems and drivers queue work items on and the backend mechanism
87 which manages worker-pools and processes the queued work items.
89 There are two worker-pools, one for normal work items and the other
90 for high priority ones, for each possible CPU and some extra
91 worker-pools to serve work items queued on unbound workqueues - the
92 number of these backing pools is dynamic.
94 Subsystems and drivers can create and queue work items through special
95 workqueue API functions as they see fit. They can influence some
96 aspects of the way the work items are executed by setting flags on the
97 workqueue they are putting the work item on. These flags include
98 things like CPU locality, concurrency limits, priority and more. To
99 get a detailed overview refer to the API description of
100 ``alloc_workqueue()`` below.
102 When a work item is queued to a workqueue, the target worker-pool is
103 determined according to the queue parameters and workqueue attributes
104 and appended on the shared worklist of the worker-pool. For example,
105 unless specifically overridden, a work item of a bound workqueue will
106 be queued on the worklist of either normal or highpri worker-pool that
107 is associated to the CPU the issuer is running on.
109 For any worker pool implementation, managing the concurrency level
110 (how many execution contexts are active) is an important issue. cmwq
111 tries to keep the concurrency at a minimal but sufficient level.
112 Minimal to save resources and sufficient in that the system is used at
115 Each worker-pool bound to an actual CPU implements concurrency
116 management by hooking into the scheduler. The worker-pool is notified
117 whenever an active worker wakes up or sleeps and keeps track of the
118 number of the currently runnable workers. Generally, work items are
119 not expected to hog a CPU and consume many cycles. That means
120 maintaining just enough concurrency to prevent work processing from
121 stalling should be optimal. As long as there are one or more runnable
122 workers on the CPU, the worker-pool doesn't start execution of a new
123 work, but, when the last running worker goes to sleep, it immediately
124 schedules a new worker so that the CPU doesn't sit idle while there
125 are pending work items. This allows using a minimal number of workers
126 without losing execution bandwidth.
128 Keeping idle workers around doesn't cost other than the memory space
129 for kthreads, so cmwq holds onto idle ones for a while before killing
132 For unbound workqueues, the number of backing pools is dynamic.
133 Unbound workqueue can be assigned custom attributes using
134 ``apply_workqueue_attrs()`` and workqueue will automatically create
135 backing worker pools matching the attributes. The responsibility of
136 regulating concurrency level is on the users. There is also a flag to
137 mark a bound wq to ignore the concurrency management. Please refer to
138 the API section for details.
140 Forward progress guarantee relies on that workers can be created when
141 more execution contexts are necessary, which in turn is guaranteed
142 through the use of rescue workers. All work items which might be used
143 on code paths that handle memory reclaim are required to be queued on
144 wq's that have a rescue-worker reserved for execution under memory
145 pressure. Else it is possible that the worker-pool deadlocks waiting
146 for execution contexts to free up.
149 Application Programming Interface (API)
150 =======================================
152 ``alloc_workqueue()`` allocates a wq. The original
153 ``create_*workqueue()`` functions are deprecated and scheduled for
154 removal. ``alloc_workqueue()`` takes three arguments - ``@name``,
155 ``@flags`` and ``@max_active``. ``@name`` is the name of the wq and
156 also used as the name of the rescuer thread if there is one.
158 A wq no longer manages execution resources but serves as a domain for
159 forward progress guarantee, flush and work item attributes. ``@flags``
160 and ``@max_active`` control how work items are assigned execution
161 resources, scheduled and executed.
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
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.
181 * Long running CPU intensive workloads which can be better
182 managed by the system scheduler.
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.
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.
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.
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.
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.
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.
217 This flag is meaningless for unbound wq.
223 ``@max_active`` determines the maximum number of execution contexts per
224 CPU 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
226 at the same time per CPU. This is always a per-CPU attribute, even for
229 The maximum limit for ``@max_active`` is 512 and the default value used
230 when 0 is specified is 256. These values are chosen sufficiently high
231 such that they are not the limiting factor while providing protection in
234 The number of active work items of a wq is usually regulated by the
235 users of the wq, more specifically, by how many work items the users
236 may queue at the same time. Unless there is a specific need for
237 throttling the number of active work items, specifying '0' is
240 Some users depend on the strict execution ordering of ST wq. The
241 combination of ``@max_active`` of 1 and ``WQ_UNBOUND`` used to
242 achieve this behavior. Work items on such wq were always queued to the
243 unbound worker-pools and only one work item could be active at any given
244 time thus achieving the same ordering property as ST wq.
246 In the current implementation the above configuration only guarantees
247 ST behavior within a given NUMA node. Instead ``alloc_ordered_queue()`` should
248 be used to achieve system-wide ST behavior.
251 Example Execution Scenarios
252 ===========================
254 The following example execution scenarios try to illustrate how cmwq
255 behave under different configurations.
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
262 Ignoring all other tasks, works and processing overhead, and assuming
263 simple FIFO scheduling, the following is one highly simplified version
264 of possible sequences of events with the original wq. ::
267 0 w0 starts and burns CPU
269 15 w0 wakes up and burns CPU
271 20 w1 starts and burns CPU
273 35 w1 wakes up and finishes
274 35 w2 starts and burns CPU
276 50 w2 wakes up and finishes
278 And with cmwq with ``@max_active`` >= 3, ::
281 0 w0 starts and burns CPU
283 5 w1 starts and burns CPU
285 10 w2 starts and burns CPU
287 15 w0 wakes up and burns CPU
289 20 w1 wakes up and finishes
290 25 w2 wakes up and finishes
292 If ``@max_active`` == 2, ::
295 0 w0 starts and burns CPU
297 5 w1 starts and burns CPU
299 15 w0 wakes up and burns CPU
301 20 w1 wakes up and finishes
302 20 w2 starts and burns CPU
304 35 w2 wakes up and finishes
306 Now, let's assume w1 and w2 are queued to a different wq q1 which has
307 ``WQ_CPU_INTENSIVE`` set, ::
310 0 w0 starts and burns CPU
312 5 w1 and w2 start and burn CPU
315 15 w0 wakes up and burns CPU
317 20 w1 wakes up and finishes
318 25 w2 wakes up and finishes
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
331 * Unless strict ordering is required, there is no need to use ST wq.
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.
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
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.
353 An unbound workqueue groups CPUs according to its affinity scope to improve
354 cache locality. For example, if a workqueue is using the default affinity
355 scope of "cache", it will group CPUs according to last level cache
356 boundaries. A work item queued on the workqueue will be assigned to a worker
357 on one of the CPUs which share the last level cache with the issuing CPU.
358 Once started, the worker may or may not be allowed to move outside the scope
359 depending on the ``affinity_strict`` setting of the scope.
361 Workqueue currently supports the following five affinity scopes.
364 CPUs are not grouped. A work item issued on one CPU is processed by a
365 worker on the same CPU. This makes unbound workqueues behave as per-cpu
366 workqueues without concurrency management.
369 CPUs are grouped according to SMT boundaries. This usually means that the
370 logical threads of each physical CPU core are grouped together.
373 CPUs are grouped according to cache boundaries. Which specific cache
374 boundary is used is determined by the arch code. L3 is used in a lot of
375 cases. This is the default affinity scope.
378 CPUs are grouped according to NUMA bounaries.
381 All CPUs are put in the same group. Workqueue makes no effort to process a
382 work item on a CPU close to the issuing CPU.
384 The default affinity scope can be changed with the module parameter
385 ``workqueue.default_affinity_scope`` and a specific workqueue's affinity
386 scope can be changed using ``apply_workqueue_attrs()``.
388 If ``WQ_SYSFS`` is set, the workqueue will have the following affinity scope
389 related interface files under its ``/sys/devices/virtual/WQ_NAME/``
393 Read to see the current affinity scope. Write to change.
396 0 by default indicating that affinity scopes are not strict. When a work
397 item starts execution, workqueue makes a best-effort attempt to ensure
398 that the worker is inside its affinity scope, which is called
399 repatriation. Once started, the scheduler is free to move the worker
400 anywhere in the system as it sees fit. This enables benefiting from scope
401 locality while still being able to utilize other CPUs if necessary and
404 If set to 1, all workers of the scope are guaranteed always to be in the
405 scope. This may be useful when crossing affinity scopes has other
406 implications, for example, in terms of power consumption or workload
407 isolation. Strict NUMA scope can also be used to match the workqueue
408 behavior of older kernels.
411 Examining Configuration
412 =======================
414 Use tools/workqueue/wq_dump.py to examine unbound CPU affinity
415 configuration, worker pools and how workqueues map to the pools: ::
417 $ tools/workqueue/wq_dump.py
420 wq_unbound_cpumask=0000000f
424 pod_cpus [0]=00000001 [1]=00000002 [2]=00000004 [3]=00000008
425 pod_node [0]=0 [1]=0 [2]=1 [3]=1
426 cpu_pod [0]=0 [1]=1 [2]=2 [3]=3
430 pod_cpus [0]=00000001 [1]=00000002 [2]=00000004 [3]=00000008
431 pod_node [0]=0 [1]=0 [2]=1 [3]=1
432 cpu_pod [0]=0 [1]=1 [2]=2 [3]=3
436 pod_cpus [0]=00000003 [1]=0000000c
438 cpu_pod [0]=0 [1]=0 [2]=1 [3]=1
442 pod_cpus [0]=00000003 [1]=0000000c
444 cpu_pod [0]=0 [1]=0 [2]=1 [3]=1
448 pod_cpus [0]=0000000f
450 cpu_pod [0]=0 [1]=0 [2]=0 [3]=0
454 pool[00] ref= 1 nice= 0 idle/workers= 4/ 4 cpu= 0
455 pool[01] ref= 1 nice=-20 idle/workers= 2/ 2 cpu= 0
456 pool[02] ref= 1 nice= 0 idle/workers= 4/ 4 cpu= 1
457 pool[03] ref= 1 nice=-20 idle/workers= 2/ 2 cpu= 1
458 pool[04] ref= 1 nice= 0 idle/workers= 4/ 4 cpu= 2
459 pool[05] ref= 1 nice=-20 idle/workers= 2/ 2 cpu= 2
460 pool[06] ref= 1 nice= 0 idle/workers= 3/ 3 cpu= 3
461 pool[07] ref= 1 nice=-20 idle/workers= 2/ 2 cpu= 3
462 pool[08] ref=42 nice= 0 idle/workers= 6/ 6 cpus=0000000f
463 pool[09] ref=28 nice= 0 idle/workers= 3/ 3 cpus=00000003
464 pool[10] ref=28 nice= 0 idle/workers= 17/ 17 cpus=0000000c
465 pool[11] ref= 1 nice=-20 idle/workers= 1/ 1 cpus=0000000f
466 pool[12] ref= 2 nice=-20 idle/workers= 1/ 1 cpus=00000003
467 pool[13] ref= 2 nice=-20 idle/workers= 1/ 1 cpus=0000000c
469 Workqueue CPU -> pool
470 =====================
471 [ workqueue \ CPU 0 1 2 3 dfl]
472 events percpu 0 2 4 6
473 events_highpri percpu 1 3 5 7
474 events_long percpu 0 2 4 6
475 events_unbound unbound 9 9 10 10 8
476 events_freezable percpu 0 2 4 6
477 events_power_efficient percpu 0 2 4 6
478 events_freezable_power_ percpu 0 2 4 6
479 rcu_gp percpu 0 2 4 6
480 rcu_par_gp percpu 0 2 4 6
481 slub_flushwq percpu 0 2 4 6
482 netns ordered 8 8 8 8 8
485 See the command's help message for more info.
491 Use tools/workqueue/wq_monitor.py to monitor workqueue operations: ::
493 $ tools/workqueue/wq_monitor.py events
494 total infl CPUtime CPUhog CMW/RPR mayday rescued
495 events 18545 0 6.1 0 5 - -
496 events_highpri 8 0 0.0 0 0 - -
497 events_long 3 0 0.0 0 0 - -
498 events_unbound 38306 0 0.1 - 7 - -
499 events_freezable 0 0 0.0 0 0 - -
500 events_power_efficient 29598 0 0.2 0 0 - -
501 events_freezable_power_ 10 0 0.0 0 0 - -
502 sock_diag_events 0 0 0.0 0 0 - -
504 total infl CPUtime CPUhog CMW/RPR mayday rescued
505 events 18548 0 6.1 0 5 - -
506 events_highpri 8 0 0.0 0 0 - -
507 events_long 3 0 0.0 0 0 - -
508 events_unbound 38322 0 0.1 - 7 - -
509 events_freezable 0 0 0.0 0 0 - -
510 events_power_efficient 29603 0 0.2 0 0 - -
511 events_freezable_power_ 10 0 0.0 0 0 - -
512 sock_diag_events 0 0 0.0 0 0 - -
516 See the command's help message for more info.
522 Because the work functions are executed by generic worker threads
523 there are a few tricks needed to shed some light on misbehaving
526 Worker threads show up in the process list as: ::
528 root 5671 0.0 0.0 0 0 ? S 12:07 0:00 [kworker/0:1]
529 root 5672 0.0 0.0 0 0 ? S 12:07 0:00 [kworker/1:2]
530 root 5673 0.0 0.0 0 0 ? S 12:12 0:00 [kworker/0:0]
531 root 5674 0.0 0.0 0 0 ? S 12:13 0:00 [kworker/1:0]
533 If kworkers are going crazy (using too much cpu), there are two types
534 of possible problems:
536 1. Something being scheduled in rapid succession
537 2. A single work item that consumes lots of cpu cycles
539 The first one can be tracked using tracing: ::
541 $ echo workqueue:workqueue_queue_work > /sys/kernel/tracing/set_event
542 $ cat /sys/kernel/tracing/trace_pipe > out.txt
546 If something is busy looping on work queueing, it would be dominating
547 the output and the offender can be determined with the work item
550 For the second type of problems it should be possible to just check
551 the stack trace of the offending worker thread. ::
553 $ cat /proc/THE_OFFENDING_KWORKER/stack
555 The work item's function should be trivially visible in the stack
559 Non-reentrance Conditions
560 =========================
562 Workqueue guarantees that a work item cannot be re-entrant if the following
563 conditions hold after a work item gets queued:
565 1. The work function hasn't been changed.
566 2. No one queues the work item to another workqueue.
567 3. The work item hasn't been reinitiated.
569 In other words, if the above conditions hold, the work item is guaranteed to be
570 executed by at most one worker system-wide at any given time.
572 Note that requeuing the work item (to the same queue) in the self function
573 doesn't break these conditions, so it's safe to do. Otherwise, caution is
574 required when breaking the conditions inside a work function.
577 Kernel Inline Documentations Reference
578 ======================================
580 .. kernel-doc:: include/linux/workqueue.h
582 .. kernel-doc:: kernel/workqueue.c
projects / platform / kernel / linux-starfive.git / blob