IRQs

worker and worker_thread

###Summary
The struct worker is the really scheudle unit in workqueue.
Each struct worker has a corresponding thread(task) by worker->task.
A struct worker is linked to struct worker_pool->idle_list when work is idle.
and moved to struct worker_pool->busy_hash.

worker_thread

  1. move worker from pool->idle_list and clear worker ‘s WORKER_IDLE flag.
  2. check the pool and manage the workers(create/destory)
  3. Iterate all the `struct work_struct *work` in the `struct worker_pool->worklist`,
    and run them in sequence with process_one_work(worker, work);.
  4. move worker into idle list again.
  5. schedule();
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
2270 /**
2271  * worker_thread - the worker thread function
2272  * @__worker: self
2273  *
2274  * The worker thread function.  All workers belong to a worker_pool -
2275  * either a per-cpu one or dynamic unbound one.  These workers process all
2276  * work items regardless of their specific target workqueue.  The only
2277  * exception is work items which belong to workqueues with a rescuer which
2278  * will be explained in rescuer_thread().
2279  *
2280  * Return: 0
2281  */
2282 static int worker_thread(void *__worker)
2283 {
2284         struct worker *worker = __worker;
2285         struct worker_pool *pool = worker->pool;
2286
2287         /* tell the scheduler that this is a workqueue worker */
2288         worker->task->flags |= PF_WQ_WORKER;
2289 woke_up:
2290         spin_lock_irq(&pool->lock);
2291
2292         /* am I supposed to die? */
2293         if (unlikely(worker->flags & WORKER_DIE)) {
2294                 spin_unlock_irq(&pool->lock);
2295                 WARN_ON_ONCE(!list_empty(&worker->entry));
2296                 worker->task->flags &= ~PF_WQ_WORKER;
2297                 return 0;
2298         }
2299
2300         worker_leave_idle(worker);
2301 recheck:
2302         /* no more worker necessary? */
2303         if (!need_more_worker(pool))
2304                 goto sleep;
2305
2306         /* do we need to manage? */
2307         if (unlikely(!may_start_working(pool)) && manage_workers(worker))
2308                 goto recheck;
2309
2310         /*
2311          * ->scheduled list can only be filled while a worker is
2312          * preparing to process a work or actually processing it.
2313          * Make sure nobody diddled with it while I was sleeping.
2314          */
2315         WARN_ON_ONCE(!list_empty(&worker->scheduled));
2316
2317         /*
2318          * Finish PREP stage.  We're guaranteed to have at least one idle
2319          * worker or that someone else has already assumed the manager
2320          * role.  This is where @worker starts participating in concurrency
2321          * management if applicable and concurrency management is restored
2322          * after being rebound.  See rebind_workers() for details.
2323          */
2324         worker_clr_flags(worker, WORKER_PREP | WORKER_REBOUND);
2325
2326         do {
2327                 struct work_struct *work =
2328                         list_first_entry(&pool->worklist,
2329                                          struct work_struct, entry);
2330
2331                 if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) {
2332                         /* optimization path, not strictly necessary */
2333                         process_one_work(worker, work);
2334                         if (unlikely(!list_empty(&worker->scheduled)))
2335                                 process_scheduled_works(worker);
2336                 } else {
2337                         move_linked_works(work, &worker->scheduled, NULL);
2338                         process_scheduled_works(worker);
2339                 }
2340         } while (keep_working(pool));
2341
2342         worker_set_flags(worker, WORKER_PREP, false);
2343 sleep:
2344         if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker))
2345                 goto recheck;
2346
2347         /*
2348          * pool->lock is held and there's no work to process and no need to
2349          * manage, sleep.  Workers are woken up only while holding
2350          * pool->lock or from local cpu, so setting the current state
2351          * before releasing pool->lock is enough to prevent losing any
2352          * event.
2353          */
2354         worker_enter_idle(worker);
2355         __set_current_state(TASK_INTERRUPTIBLE);
2356         spin_unlock_irq(&pool->lock);
2357         schedule();
2358         goto woke_up;
2359 }

process_one_work

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
2114 /**
2115  * process_one_work - process single work
2116  * @worker: self
2117  * @work: work to process
2118  *
2119  * Process @work.  This function contains all the logics necessary to
2120  * process a single work including synchronization against and
2121  * interaction with other workers on the same cpu, queueing and
2122  * flushing.  As long as context requirement is met, any worker can
2123  * call this function to process a work.
2124  *
2125  * CONTEXT:
2126  * spin_lock_irq(pool->lock) which is released and regrabbed.
2127  */
2128 static void process_one_work(struct worker *worker, struct work_struct *work)
2129 __releases(&pool->lock)
2130 __acquires(&pool->lock)
2131 {
2132         struct pool_workqueue *pwq = get_work_pwq(work);
2133         struct worker_pool *pool = worker->pool;
2134         bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE;
2135         int work_color;
2136         struct worker *collision;
2137 #ifdef CONFIG_LOCKDEP
2138         /*
2139          * It is permissible to free the struct work_struct from
2140          * inside the function that is called from it, this we need to
2141          * take into account for lockdep too.  To avoid bogus "held
2142          * lock freed" warnings as well as problems when looking into
2143          * work->lockdep_map, make a copy and use that here.
2144          */
2145         struct lockdep_map lockdep_map;
2146
2147         lockdep_copy_map(&lockdep_map, &work->lockdep_map);
2148 #endif
2149         /*
2150          * Ensure we're on the correct CPU.  DISASSOCIATED test is
2151          * necessary to avoid spurious warnings from rescuers servicing the
2152          * unbound or a disassociated pool.
2153          */
2154         WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) &&
2155                      !(pool->flags & POOL_DISASSOCIATED) &&
2156                      raw_smp_processor_id() != pool->cpu);
2157
2158         /*
2159          * A single work shouldn't be executed concurrently by
2160          * multiple workers on a single cpu.  Check whether anyone is
2161          * already processing the work.  If so, defer the work to the
2162          * currently executing one.
2163          */
2164         collision = find_worker_executing_work(pool, work);
2165         if (unlikely(collision)) {
2166                 move_linked_works(work, &collision->scheduled, NULL);
2167                 return;
2168         }
2169
2170         /* claim and dequeue */
2171         debug_work_deactivate(work);
2172         hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work);
2173         worker->current_work = work;
2174         worker->current_func = work->func;
2174         worker->current_func = work->func;
2175         worker->current_pwq = pwq;
2176         work_color = get_work_color(work);
2177
2178         list_del_init(&work->entry);
2179
2180         /*
2181          * CPU intensive works don't participate in concurrency
2182          * management.  They're the scheduler's responsibility.
2183          */
2184         if (unlikely(cpu_intensive))
2185                 worker_set_flags(worker, WORKER_CPU_INTENSIVE, true);
2186
2187         /*
2188          * Unbound pool isn't concurrency managed and work items should be
2189          * executed ASAP.  Wake up another worker if necessary.
2190          */
2191         if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool))
2192                 wake_up_worker(pool);
2193
2194         /*
2195          * Record the last pool and clear PENDING which should be the last
2196          * update to @work.  Also, do this inside @pool->lock so that
2197          * PENDING and queued state changes happen together while IRQ is
2198          * disabled.
2199          */
2200         set_work_pool_and_clear_pending(work, pool->id);
2201
2202         spin_unlock_irq(&pool->lock);
2203
2204         lock_map_acquire_read(&pwq->wq->lockdep_map);
2205         lock_map_acquire(&lockdep_map);
2206         trace_workqueue_execute_start(work);
2207         worker->current_func(work);
2208         /*
2209          * While we must be careful to not use "work" after this, the trace
2210          * point will only record its address.
2211          */
2212         trace_workqueue_execute_end(work);
2213         lock_map_release(&lockdep_map);
2214         lock_map_release(&pwq->wq->lockdep_map);
2215
2216         if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
2217                 pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n"
2218                        "     last function: %pf\n",
2219                        current->comm, preempt_count(), task_pid_nr(current),
2220                        worker->current_func);
2221                 debug_show_held_locks(current);
2222                 dump_stack();
2223         }
2224
2225         /*
2226          * The following prevents a kworker from hogging CPU on !PREEMPT
2227          * kernels, where a requeueing work item waiting for something to
2228          * happen could deadlock with stop_machine as such work item could
2229          * indefinitely requeue itself while all other CPUs are trapped in
2230          * stop_machine.
2231          */
2232         cond_resched();
2233
2234         spin_lock_irq(&pool->lock);
2235
2236         /* clear cpu intensive status */
2237         if (unlikely(cpu_intensive))
2238                 worker_clr_flags(worker, WORKER_CPU_INTENSIVE);
2239
2240         /* we're done with it, release */
2241         hash_del(&worker->hentry);
2242         worker->current_work = NULL;
2243         worker->current_func = NULL;
2244         worker->current_pwq = NULL;
2245         worker->desc_valid = false;
2246         pwq_dec_nr_in_flight(pwq, work_color);
2247 }