2 * Copyright (c) 2017 Hans Petter Selasky
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions, and the following
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <linux/workqueue.h>
31 #include <linux/wait.h>
32 #include <linux/compat.h>
33 #include <linux/spinlock.h>
35 #include <sys/kernel.h>
38 * Define all work struct states
41 WORK_ST_IDLE, /* idle - not started */
42 WORK_ST_TIMER, /* timer is being started */
43 WORK_ST_TASK, /* taskqueue is being queued */
44 WORK_ST_EXEC, /* callback is being called */
45 WORK_ST_CANCEL, /* cancel is being requested */
50 * Define global workqueues
52 static struct workqueue_struct *linux_system_short_wq;
53 static struct workqueue_struct *linux_system_long_wq;
55 struct workqueue_struct *system_wq;
56 struct workqueue_struct *system_long_wq;
57 struct workqueue_struct *system_unbound_wq;
58 struct workqueue_struct *system_power_efficient_wq;
60 static int linux_default_wq_cpus = 4;
62 static void linux_delayed_work_timer_fn(void *);
65 * This function atomically updates the work state and returns the
66 * previous state at the time of update.
69 linux_update_state(atomic_t *v, const uint8_t *pstate)
75 while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
82 * A LinuxKPI task is allowed to free itself inside the callback function
83 * and cannot safely be referred after the callback function has
84 * completed. This function gives the linux_work_fn() function a hint,
85 * that the task is not going away and can have its state checked
86 * again. Without this extra hint LinuxKPI tasks cannot be serialized
87 * accross multiple worker threads.
90 linux_work_exec_unblock(struct work_struct *work)
92 struct workqueue_struct *wq;
93 struct work_exec *exec;
96 wq = work->work_queue;
97 if (unlikely(wq == NULL))
101 TAILQ_FOREACH(exec, &wq->exec_head, entry) {
102 if (exec->target == work) {
114 linux_delayed_work_enqueue(struct delayed_work *dwork)
116 struct taskqueue *tq;
118 tq = dwork->work.work_queue->taskqueue;
119 taskqueue_enqueue(tq, &dwork->work.work_task);
123 * This function queues the given work structure on the given
124 * workqueue. It returns non-zero if the work was successfully
125 * [re-]queued. Else the work is already pending for completion.
128 linux_queue_work_on(int cpu __unused, struct workqueue_struct *wq,
129 struct work_struct *work)
131 static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
132 [WORK_ST_IDLE] = WORK_ST_TASK, /* start queuing task */
133 [WORK_ST_TIMER] = WORK_ST_TIMER, /* NOP */
134 [WORK_ST_TASK] = WORK_ST_TASK, /* NOP */
135 [WORK_ST_EXEC] = WORK_ST_TASK, /* queue task another time */
136 [WORK_ST_CANCEL] = WORK_ST_TASK, /* start queuing task again */
139 if (atomic_read(&wq->draining) != 0)
140 return (!work_pending(work));
142 switch (linux_update_state(&work->state, states)) {
145 if (linux_work_exec_unblock(work) != 0)
149 work->work_queue = wq;
150 taskqueue_enqueue(wq->taskqueue, &work->work_task);
153 return (0); /* already on a queue */
158 * This function queues the given work structure on the given
159 * workqueue after a given delay in ticks. It returns non-zero if the
160 * work was successfully [re-]queued. Else the work is already pending
164 linux_queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
165 struct delayed_work *dwork, unsigned delay)
167 static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
168 [WORK_ST_IDLE] = WORK_ST_TIMER, /* start timeout */
169 [WORK_ST_TIMER] = WORK_ST_TIMER, /* NOP */
170 [WORK_ST_TASK] = WORK_ST_TASK, /* NOP */
171 [WORK_ST_EXEC] = WORK_ST_TIMER, /* start timeout */
172 [WORK_ST_CANCEL] = WORK_ST_TIMER, /* start timeout */
175 if (atomic_read(&wq->draining) != 0)
176 return (!work_pending(&dwork->work));
178 switch (linux_update_state(&dwork->work.state, states)) {
181 if (delay == 0 && linux_work_exec_unblock(&dwork->work) != 0) {
182 dwork->timer.expires = jiffies;
187 dwork->work.work_queue = wq;
188 dwork->timer.expires = jiffies + delay;
191 linux_delayed_work_enqueue(dwork);
192 } else if (unlikely(cpu != WORK_CPU_UNBOUND)) {
193 mtx_lock(&dwork->timer.mtx);
194 callout_reset_on(&dwork->timer.callout, delay,
195 &linux_delayed_work_timer_fn, dwork, cpu);
196 mtx_unlock(&dwork->timer.mtx);
198 mtx_lock(&dwork->timer.mtx);
199 callout_reset(&dwork->timer.callout, delay,
200 &linux_delayed_work_timer_fn, dwork);
201 mtx_unlock(&dwork->timer.mtx);
205 return (0); /* already on a queue */
210 linux_work_fn(void *context, int pending)
212 static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
213 [WORK_ST_IDLE] = WORK_ST_IDLE, /* NOP */
214 [WORK_ST_TIMER] = WORK_ST_EXEC, /* delayed work w/o timeout */
215 [WORK_ST_TASK] = WORK_ST_EXEC, /* call callback */
216 [WORK_ST_EXEC] = WORK_ST_IDLE, /* complete callback */
217 [WORK_ST_CANCEL] = WORK_ST_EXEC, /* failed to cancel */
219 struct work_struct *work;
220 struct workqueue_struct *wq;
221 struct work_exec exec;
223 linux_set_current(curthread);
225 /* setup local variables */
227 wq = work->work_queue;
229 /* store target pointer */
232 /* insert executor into list */
234 TAILQ_INSERT_TAIL(&wq->exec_head, &exec, entry);
236 switch (linux_update_state(&work->state, states)) {
242 /* call work function */
246 /* check if unblocked */
247 if (exec.target != work) {
258 /* remove executor from list */
259 TAILQ_REMOVE(&wq->exec_head, &exec, entry);
264 linux_delayed_work_fn(void *context, int pending)
266 struct delayed_work *dwork = context;
269 * Make sure the timer belonging to the delayed work gets
270 * drained before invoking the work function. Else the timer
271 * mutex may still be in use which can lead to use-after-free
272 * situations, because the work function might free the work
273 * structure before returning.
275 callout_drain(&dwork->timer.callout);
277 linux_work_fn(&dwork->work, pending);
281 linux_delayed_work_timer_fn(void *arg)
283 static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
284 [WORK_ST_IDLE] = WORK_ST_IDLE, /* NOP */
285 [WORK_ST_TIMER] = WORK_ST_TASK, /* start queueing task */
286 [WORK_ST_TASK] = WORK_ST_TASK, /* NOP */
287 [WORK_ST_EXEC] = WORK_ST_EXEC, /* NOP */
288 [WORK_ST_CANCEL] = WORK_ST_TASK, /* failed to cancel */
290 struct delayed_work *dwork = arg;
292 switch (linux_update_state(&dwork->work.state, states)) {
295 linux_delayed_work_enqueue(dwork);
303 * This function cancels the given work structure in a synchronous
304 * fashion. It returns non-zero if the work was successfully
305 * cancelled. Else the work was already cancelled.
308 linux_cancel_work_sync(struct work_struct *work)
310 static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
311 [WORK_ST_IDLE] = WORK_ST_IDLE, /* NOP */
312 [WORK_ST_TIMER] = WORK_ST_TIMER, /* can't happen */
313 [WORK_ST_TASK] = WORK_ST_IDLE, /* cancel and drain */
314 [WORK_ST_EXEC] = WORK_ST_IDLE, /* too late, drain */
315 [WORK_ST_CANCEL] = WORK_ST_IDLE, /* cancel and drain */
317 struct taskqueue *tq;
319 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
320 "linux_cancel_work_sync() might sleep");
322 switch (linux_update_state(&work->state, states)) {
327 tq = work->work_queue->taskqueue;
328 if (taskqueue_cancel(tq, &work->work_task, NULL) != 0)
329 taskqueue_drain(tq, &work->work_task);
332 tq = work->work_queue->taskqueue;
333 if (taskqueue_cancel(tq, &work->work_task, NULL) != 0)
334 taskqueue_drain(tq, &work->work_task);
340 * This function atomically stops the timer and callback. The timer
341 * callback will not be called after this function returns. This
342 * functions returns true when the timeout was cancelled. Else the
343 * timeout was not started or has already been called.
346 linux_cancel_timer(struct delayed_work *dwork, bool drain)
350 mtx_lock(&dwork->timer.mtx);
351 cancelled = (callout_stop(&dwork->timer.callout) == 1);
352 mtx_unlock(&dwork->timer.mtx);
354 /* check if we should drain */
356 callout_drain(&dwork->timer.callout);
361 * This function cancels the given delayed work structure in a
362 * non-blocking fashion. It returns non-zero if the work was
363 * successfully cancelled. Else the work may still be busy or already
367 linux_cancel_delayed_work(struct delayed_work *dwork)
369 static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
370 [WORK_ST_IDLE] = WORK_ST_IDLE, /* NOP */
371 [WORK_ST_TIMER] = WORK_ST_CANCEL, /* try to cancel */
372 [WORK_ST_TASK] = WORK_ST_CANCEL, /* try to cancel */
373 [WORK_ST_EXEC] = WORK_ST_EXEC, /* NOP */
374 [WORK_ST_CANCEL] = WORK_ST_CANCEL, /* NOP */
376 struct taskqueue *tq;
378 switch (linux_update_state(&dwork->work.state, states)) {
381 if (linux_cancel_timer(dwork, 0)) {
382 atomic_cmpxchg(&dwork->work.state,
383 WORK_ST_CANCEL, WORK_ST_IDLE);
388 tq = dwork->work.work_queue->taskqueue;
389 if (taskqueue_cancel(tq, &dwork->work.work_task, NULL) == 0) {
390 atomic_cmpxchg(&dwork->work.state,
391 WORK_ST_CANCEL, WORK_ST_IDLE);
401 * This function cancels the given work structure in a synchronous
402 * fashion. It returns non-zero if the work was successfully
403 * cancelled. Else the work was already cancelled.
406 linux_cancel_delayed_work_sync(struct delayed_work *dwork)
408 static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
409 [WORK_ST_IDLE] = WORK_ST_IDLE, /* NOP */
410 [WORK_ST_TIMER] = WORK_ST_IDLE, /* cancel and drain */
411 [WORK_ST_TASK] = WORK_ST_IDLE, /* cancel and drain */
412 [WORK_ST_EXEC] = WORK_ST_IDLE, /* too late, drain */
413 [WORK_ST_CANCEL] = WORK_ST_IDLE, /* cancel and drain */
415 struct taskqueue *tq;
417 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
418 "linux_cancel_delayed_work_sync() might sleep");
420 switch (linux_update_state(&dwork->work.state, states)) {
424 tq = dwork->work.work_queue->taskqueue;
425 if (taskqueue_cancel(tq, &dwork->work.work_task, NULL) != 0)
426 taskqueue_drain(tq, &dwork->work.work_task);
430 if (linux_cancel_timer(dwork, 1)) {
432 * Make sure taskqueue is also drained before
435 tq = dwork->work.work_queue->taskqueue;
436 taskqueue_drain(tq, &dwork->work.work_task);
441 tq = dwork->work.work_queue->taskqueue;
442 if (taskqueue_cancel(tq, &dwork->work.work_task, NULL) != 0)
443 taskqueue_drain(tq, &dwork->work.work_task);
449 * This function waits until the given work structure is completed.
450 * It returns non-zero if the work was successfully
451 * waited for. Else the work was not waited for.
454 linux_flush_work(struct work_struct *work)
456 struct taskqueue *tq;
459 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
460 "linux_flush_work() might sleep");
462 switch (atomic_read(&work->state)) {
466 tq = work->work_queue->taskqueue;
467 retval = taskqueue_poll_is_busy(tq, &work->work_task);
468 taskqueue_drain(tq, &work->work_task);
474 * This function waits until the given delayed work structure is
475 * completed. It returns non-zero if the work was successfully waited
476 * for. Else the work was not waited for.
479 linux_flush_delayed_work(struct delayed_work *dwork)
481 struct taskqueue *tq;
484 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
485 "linux_flush_delayed_work() might sleep");
487 switch (atomic_read(&dwork->work.state)) {
491 if (linux_cancel_timer(dwork, 1))
492 linux_delayed_work_enqueue(dwork);
495 tq = dwork->work.work_queue->taskqueue;
496 retval = taskqueue_poll_is_busy(tq, &dwork->work.work_task);
497 taskqueue_drain(tq, &dwork->work.work_task);
503 * This function returns true if the given work is pending, and not
507 linux_work_pending(struct work_struct *work)
509 switch (atomic_read(&work->state)) {
520 * This function returns true if the given work is busy.
523 linux_work_busy(struct work_struct *work)
525 struct taskqueue *tq;
527 switch (atomic_read(&work->state)) {
531 tq = work->work_queue->taskqueue;
532 return (taskqueue_poll_is_busy(tq, &work->work_task));
538 struct workqueue_struct *
539 linux_create_workqueue_common(const char *name, int cpus)
541 struct workqueue_struct *wq;
544 * If zero CPUs are specified use the default number of CPUs:
547 cpus = linux_default_wq_cpus;
549 wq = kmalloc(sizeof(*wq), M_WAITOK | M_ZERO);
550 wq->taskqueue = taskqueue_create(name, M_WAITOK,
551 taskqueue_thread_enqueue, &wq->taskqueue);
552 atomic_set(&wq->draining, 0);
553 taskqueue_start_threads(&wq->taskqueue, cpus, PWAIT, "%s", name);
554 TAILQ_INIT(&wq->exec_head);
555 mtx_init(&wq->exec_mtx, "linux_wq_exec", NULL, MTX_DEF);
561 linux_destroy_workqueue(struct workqueue_struct *wq)
563 atomic_inc(&wq->draining);
565 taskqueue_free(wq->taskqueue);
566 mtx_destroy(&wq->exec_mtx);
571 linux_init_delayed_work(struct delayed_work *dwork, work_func_t func)
573 memset(dwork, 0, sizeof(*dwork));
574 dwork->work.func = func;
575 TASK_INIT(&dwork->work.work_task, 0, linux_delayed_work_fn, dwork);
576 mtx_init(&dwork->timer.mtx, spin_lock_name("lkpi-dwork"), NULL,
577 MTX_DEF | MTX_NOWITNESS);
578 callout_init_mtx(&dwork->timer.callout, &dwork->timer.mtx, 0);
582 linux_work_init(void *arg)
584 int max_wq_cpus = mp_ncpus + 1;
586 /* avoid deadlock when there are too few threads */
590 /* set default number of CPUs */
591 linux_default_wq_cpus = max_wq_cpus;
593 linux_system_short_wq = alloc_workqueue("linuxkpi_short_wq", 0, max_wq_cpus);
594 linux_system_long_wq = alloc_workqueue("linuxkpi_long_wq", 0, max_wq_cpus);
596 /* populate the workqueue pointers */
597 system_long_wq = linux_system_long_wq;
598 system_wq = linux_system_short_wq;
599 system_power_efficient_wq = linux_system_short_wq;
600 system_unbound_wq = linux_system_short_wq;
602 SYSINIT(linux_work_init, SI_SUB_TASKQ, SI_ORDER_THIRD, linux_work_init, NULL);
605 linux_work_uninit(void *arg)
607 destroy_workqueue(linux_system_short_wq);
608 destroy_workqueue(linux_system_long_wq);
610 /* clear workqueue pointers */
611 system_long_wq = NULL;
613 system_power_efficient_wq = NULL;
614 system_unbound_wq = NULL;
616 SYSUNINIT(linux_work_uninit, SI_SUB_TASKQ, SI_ORDER_THIRD, linux_work_uninit, NULL);