1 /*-
2  * Copyright (c) 2017-2019 Hans Petter Selasky
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
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.
14  *
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.
25  */
26 
27 #include <sys/cdefs.h>
28 #include <linux/workqueue.h>
29 #include <linux/wait.h>
30 #include <linux/compat.h>
31 #include <linux/spinlock.h>
32 #include <linux/rcupdate.h>
33 #include <linux/irq_work.h>
34 
35 #include <sys/kernel.h>
36 
37 /*
38  * Define all work struct states
39  */
40 enum {
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 */
46 	WORK_ST_MAX,
47 };
48 
49 /*
50  * Define global workqueues
51  */
52 static struct workqueue_struct *linux_system_short_wq;
53 static struct workqueue_struct *linux_system_long_wq;
54 
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_highpri_wq;
59 struct workqueue_struct *system_power_efficient_wq;
60 
61 struct taskqueue *linux_irq_work_tq;
62 
63 static int linux_default_wq_cpus = 4;
64 
65 static void linux_delayed_work_timer_fn(void *);
66 
67 /*
68  * This function atomically updates the work state and returns the
69  * previous state at the time of update.
70  */
71 static uint8_t
linux_update_state(atomic_t * v,const uint8_t * pstate)72 linux_update_state(atomic_t *v, const uint8_t *pstate)
73 {
74 	int c, old;
75 
76 	c = v->counter;
77 
78 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
79 		c = old;
80 
81 	return (c);
82 }
83 
84 /*
85  * A LinuxKPI task is allowed to free itself inside the callback function
86  * and cannot safely be referred after the callback function has
87  * completed. This function gives the linux_work_fn() function a hint,
88  * that the task is not going away and can have its state checked
89  * again. Without this extra hint LinuxKPI tasks cannot be serialized
90  * across multiple worker threads.
91  */
92 static bool
linux_work_exec_unblock(struct work_struct * work)93 linux_work_exec_unblock(struct work_struct *work)
94 {
95 	struct workqueue_struct *wq;
96 	struct work_exec *exec;
97 	bool retval = false;
98 
99 	wq = work->work_queue;
100 	if (unlikely(wq == NULL))
101 		goto done;
102 
103 	WQ_EXEC_LOCK(wq);
104 	TAILQ_FOREACH(exec, &wq->exec_head, entry) {
105 		if (exec->target == work) {
106 			exec->target = NULL;
107 			retval = true;
108 			break;
109 		}
110 	}
111 	WQ_EXEC_UNLOCK(wq);
112 done:
113 	return (retval);
114 }
115 
116 static void
linux_delayed_work_enqueue(struct delayed_work * dwork)117 linux_delayed_work_enqueue(struct delayed_work *dwork)
118 {
119 	struct taskqueue *tq;
120 
121 	tq = dwork->work.work_queue->taskqueue;
122 	taskqueue_enqueue(tq, &dwork->work.work_task);
123 }
124 
125 /*
126  * This function queues the given work structure on the given
127  * workqueue. It returns non-zero if the work was successfully
128  * [re-]queued. Else the work is already pending for completion.
129  */
130 bool
linux_queue_work_on(int cpu __unused,struct workqueue_struct * wq,struct work_struct * work)131 linux_queue_work_on(int cpu __unused, struct workqueue_struct *wq,
132     struct work_struct *work)
133 {
134 	static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
135 		[WORK_ST_IDLE] = WORK_ST_TASK,		/* start queuing task */
136 		[WORK_ST_TIMER] = WORK_ST_TIMER,	/* NOP */
137 		[WORK_ST_TASK] = WORK_ST_TASK,		/* NOP */
138 		[WORK_ST_EXEC] = WORK_ST_TASK,		/* queue task another time */
139 		[WORK_ST_CANCEL] = WORK_ST_TASK,	/* start queuing task again */
140 	};
141 
142 	if (atomic_read(&wq->draining) != 0)
143 		return (!work_pending(work));
144 
145 	switch (linux_update_state(&work->state, states)) {
146 	case WORK_ST_EXEC:
147 	case WORK_ST_CANCEL:
148 		if (linux_work_exec_unblock(work) != 0)
149 			return (true);
150 		/* FALLTHROUGH */
151 	case WORK_ST_IDLE:
152 		work->work_queue = wq;
153 		taskqueue_enqueue(wq->taskqueue, &work->work_task);
154 		return (true);
155 	default:
156 		return (false);		/* already on a queue */
157 	}
158 }
159 
160 /*
161  * Callback func for linux_queue_rcu_work
162  */
163 static void
rcu_work_func(struct rcu_head * rcu)164 rcu_work_func(struct rcu_head *rcu)
165 {
166 	struct rcu_work *rwork;
167 
168 	rwork = container_of(rcu, struct rcu_work, rcu);
169 	linux_queue_work_on(WORK_CPU_UNBOUND, rwork->wq, &rwork->work);
170 }
171 
172 /*
173  * This function queue a work after a grace period
174  * If the work was already pending it returns false,
175  * if not it calls call_rcu and returns true.
176  */
177 bool
linux_queue_rcu_work(struct workqueue_struct * wq,struct rcu_work * rwork)178 linux_queue_rcu_work(struct workqueue_struct *wq, struct rcu_work *rwork)
179 {
180 
181 	if (!linux_work_pending(&rwork->work)) {
182 		rwork->wq = wq;
183 		linux_call_rcu(RCU_TYPE_REGULAR, &rwork->rcu, rcu_work_func);
184 		return (true);
185 	}
186 	return (false);
187 }
188 
189 /*
190  * This function waits for the last execution of a work and then
191  * flush the work.
192  * It returns true if the work was pending and we waited, it returns
193  * false otherwise.
194  */
195 bool
linux_flush_rcu_work(struct rcu_work * rwork)196 linux_flush_rcu_work(struct rcu_work *rwork)
197 {
198 
199 	if (linux_work_pending(&rwork->work)) {
200 		linux_rcu_barrier(RCU_TYPE_REGULAR);
201 		linux_flush_work(&rwork->work);
202 		return (true);
203 	}
204 	return (linux_flush_work(&rwork->work));
205 }
206 
207 /*
208  * This function queues the given work structure on the given
209  * workqueue after a given delay in ticks. It returns true if the
210  * work was successfully [re-]queued. Else the work is already pending
211  * for completion.
212  */
213 bool
linux_queue_delayed_work_on(int cpu,struct workqueue_struct * wq,struct delayed_work * dwork,unsigned delay)214 linux_queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
215     struct delayed_work *dwork, unsigned delay)
216 {
217 	static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
218 		[WORK_ST_IDLE] = WORK_ST_TIMER,		/* start timeout */
219 		[WORK_ST_TIMER] = WORK_ST_TIMER,	/* NOP */
220 		[WORK_ST_TASK] = WORK_ST_TASK,		/* NOP */
221 		[WORK_ST_EXEC] = WORK_ST_TIMER,		/* start timeout */
222 		[WORK_ST_CANCEL] = WORK_ST_TIMER,	/* start timeout */
223 	};
224 	bool res;
225 
226 	if (atomic_read(&wq->draining) != 0)
227 		return (!work_pending(&dwork->work));
228 
229 	mtx_lock(&dwork->timer.mtx);
230 	switch (linux_update_state(&dwork->work.state, states)) {
231 	case WORK_ST_EXEC:
232 	case WORK_ST_CANCEL:
233 		if (delay == 0 && linux_work_exec_unblock(&dwork->work)) {
234 			dwork->timer.expires = jiffies;
235 			res = true;
236 			goto out;
237 		}
238 		/* FALLTHROUGH */
239 	case WORK_ST_IDLE:
240 		dwork->work.work_queue = wq;
241 		dwork->timer.expires = jiffies + delay;
242 
243 		if (delay == 0) {
244 			linux_delayed_work_enqueue(dwork);
245 		} else if (unlikely(cpu != WORK_CPU_UNBOUND)) {
246 			callout_reset_on(&dwork->timer.callout, delay,
247 			    &linux_delayed_work_timer_fn, dwork, cpu);
248 		} else {
249 			callout_reset(&dwork->timer.callout, delay,
250 			    &linux_delayed_work_timer_fn, dwork);
251 		}
252 		res = true;
253 		break;
254 	default:
255 		res = false;
256 		break;
257 	}
258 out:
259 	mtx_unlock(&dwork->timer.mtx);
260 	return (res);
261 }
262 
263 void
linux_work_fn(void * context,int pending)264 linux_work_fn(void *context, int pending)
265 {
266 	static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
267 		[WORK_ST_IDLE] = WORK_ST_IDLE,		/* NOP */
268 		[WORK_ST_TIMER] = WORK_ST_EXEC,		/* delayed work w/o timeout */
269 		[WORK_ST_TASK] = WORK_ST_EXEC,		/* call callback */
270 		[WORK_ST_EXEC] = WORK_ST_IDLE,		/* complete callback */
271 		[WORK_ST_CANCEL] = WORK_ST_EXEC,	/* failed to cancel */
272 	};
273 	struct work_struct *work;
274 	struct workqueue_struct *wq;
275 	struct work_exec exec;
276 	struct task_struct *task;
277 
278 	task = current;
279 
280 	/* setup local variables */
281 	work = context;
282 	wq = work->work_queue;
283 
284 	/* store target pointer */
285 	exec.target = work;
286 
287 	/* insert executor into list */
288 	WQ_EXEC_LOCK(wq);
289 	TAILQ_INSERT_TAIL(&wq->exec_head, &exec, entry);
290 	while (1) {
291 		switch (linux_update_state(&work->state, states)) {
292 		case WORK_ST_TIMER:
293 		case WORK_ST_TASK:
294 		case WORK_ST_CANCEL:
295 			WQ_EXEC_UNLOCK(wq);
296 
297 			/* set current work structure */
298 			task->work = work;
299 
300 			/* call work function */
301 			work->func(work);
302 
303 			/* set current work structure */
304 			task->work = NULL;
305 
306 			WQ_EXEC_LOCK(wq);
307 			/* check if unblocked */
308 			if (exec.target != work) {
309 				/* reapply block */
310 				exec.target = work;
311 				break;
312 			}
313 			/* FALLTHROUGH */
314 		default:
315 			goto done;
316 		}
317 	}
318 done:
319 	/* remove executor from list */
320 	TAILQ_REMOVE(&wq->exec_head, &exec, entry);
321 	WQ_EXEC_UNLOCK(wq);
322 }
323 
324 void
linux_delayed_work_fn(void * context,int pending)325 linux_delayed_work_fn(void *context, int pending)
326 {
327 	struct delayed_work *dwork = context;
328 
329 	/*
330 	 * Make sure the timer belonging to the delayed work gets
331 	 * drained before invoking the work function. Else the timer
332 	 * mutex may still be in use which can lead to use-after-free
333 	 * situations, because the work function might free the work
334 	 * structure before returning.
335 	 */
336 	callout_drain(&dwork->timer.callout);
337 
338 	linux_work_fn(&dwork->work, pending);
339 }
340 
341 static void
linux_delayed_work_timer_fn(void * arg)342 linux_delayed_work_timer_fn(void *arg)
343 {
344 	static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
345 		[WORK_ST_IDLE] = WORK_ST_IDLE,		/* NOP */
346 		[WORK_ST_TIMER] = WORK_ST_TASK,		/* start queueing task */
347 		[WORK_ST_TASK] = WORK_ST_TASK,		/* NOP */
348 		[WORK_ST_EXEC] = WORK_ST_EXEC,		/* NOP */
349 		[WORK_ST_CANCEL] = WORK_ST_TASK,	/* failed to cancel */
350 	};
351 	struct delayed_work *dwork = arg;
352 
353 	switch (linux_update_state(&dwork->work.state, states)) {
354 	case WORK_ST_TIMER:
355 	case WORK_ST_CANCEL:
356 		linux_delayed_work_enqueue(dwork);
357 		break;
358 	default:
359 		break;
360 	}
361 }
362 
363 /*
364  * This function cancels the given work structure in a
365  * non-blocking fashion. It returns non-zero if the work was
366  * successfully cancelled. Else the work may still be busy or already
367  * cancelled.
368  */
369 bool
linux_cancel_work(struct work_struct * work)370 linux_cancel_work(struct work_struct *work)
371 {
372 	static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
373 		[WORK_ST_IDLE] = WORK_ST_IDLE,		/* NOP */
374 		[WORK_ST_TIMER] = WORK_ST_TIMER,	/* can't happen */
375 		[WORK_ST_TASK] = WORK_ST_IDLE,		/* cancel */
376 		[WORK_ST_EXEC] = WORK_ST_EXEC,		/* NOP */
377 		[WORK_ST_CANCEL] = WORK_ST_IDLE,	/* can't happen */
378 	};
379 	struct taskqueue *tq;
380 
381 	MPASS(atomic_read(&work->state) != WORK_ST_TIMER);
382 	MPASS(atomic_read(&work->state) != WORK_ST_CANCEL);
383 
384 	switch (linux_update_state(&work->state, states)) {
385 	case WORK_ST_TASK:
386 		tq = work->work_queue->taskqueue;
387 		if (taskqueue_cancel(tq, &work->work_task, NULL) == 0)
388 			return (true);
389 		/* FALLTHROUGH */
390 	default:
391 		return (false);
392 	}
393 }
394 
395 /*
396  * This function cancels the given work structure in a synchronous
397  * fashion. It returns non-zero if the work was successfully
398  * cancelled. Else the work was already cancelled.
399  */
400 bool
linux_cancel_work_sync(struct work_struct * work)401 linux_cancel_work_sync(struct work_struct *work)
402 {
403 	static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
404 		[WORK_ST_IDLE] = WORK_ST_IDLE,		/* NOP */
405 		[WORK_ST_TIMER] = WORK_ST_TIMER,	/* can't happen */
406 		[WORK_ST_TASK] = WORK_ST_IDLE,		/* cancel and drain */
407 		[WORK_ST_EXEC] = WORK_ST_IDLE,		/* too late, drain */
408 		[WORK_ST_CANCEL] = WORK_ST_IDLE,	/* cancel and drain */
409 	};
410 	struct taskqueue *tq;
411 	bool retval = false;
412 
413 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
414 	    "linux_cancel_work_sync() might sleep");
415 retry:
416 	switch (linux_update_state(&work->state, states)) {
417 	case WORK_ST_IDLE:
418 	case WORK_ST_TIMER:
419 		return (retval);
420 	case WORK_ST_EXEC:
421 		tq = work->work_queue->taskqueue;
422 		if (taskqueue_cancel(tq, &work->work_task, NULL) != 0)
423 			taskqueue_drain(tq, &work->work_task);
424 		goto retry;	/* work may have restarted itself */
425 	default:
426 		tq = work->work_queue->taskqueue;
427 		if (taskqueue_cancel(tq, &work->work_task, NULL) != 0)
428 			taskqueue_drain(tq, &work->work_task);
429 		retval = true;
430 		goto retry;
431 	}
432 }
433 
434 /*
435  * This function atomically stops the timer and callback. The timer
436  * callback will not be called after this function returns. This
437  * functions returns true when the timeout was cancelled. Else the
438  * timeout was not started or has already been called.
439  */
440 static inline bool
linux_cancel_timer(struct delayed_work * dwork,bool drain)441 linux_cancel_timer(struct delayed_work *dwork, bool drain)
442 {
443 	bool cancelled;
444 
445 	mtx_lock(&dwork->timer.mtx);
446 	cancelled = (callout_stop(&dwork->timer.callout) == 1);
447 	mtx_unlock(&dwork->timer.mtx);
448 
449 	/* check if we should drain */
450 	if (drain)
451 		callout_drain(&dwork->timer.callout);
452 	return (cancelled);
453 }
454 
455 /*
456  * This function cancels the given delayed work structure in a
457  * non-blocking fashion. It returns non-zero if the work was
458  * successfully cancelled. Else the work may still be busy or already
459  * cancelled.
460  */
461 bool
linux_cancel_delayed_work(struct delayed_work * dwork)462 linux_cancel_delayed_work(struct delayed_work *dwork)
463 {
464 	static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
465 		[WORK_ST_IDLE] = WORK_ST_IDLE,		/* NOP */
466 		[WORK_ST_TIMER] = WORK_ST_CANCEL,	/* try to cancel */
467 		[WORK_ST_TASK] = WORK_ST_CANCEL,	/* try to cancel */
468 		[WORK_ST_EXEC] = WORK_ST_EXEC,		/* NOP */
469 		[WORK_ST_CANCEL] = WORK_ST_CANCEL,	/* NOP */
470 	};
471 	struct taskqueue *tq;
472 	bool cancelled;
473 
474 	mtx_lock(&dwork->timer.mtx);
475 	switch (linux_update_state(&dwork->work.state, states)) {
476 	case WORK_ST_TIMER:
477 	case WORK_ST_CANCEL:
478 		cancelled = (callout_stop(&dwork->timer.callout) == 1);
479 		if (cancelled) {
480 			atomic_cmpxchg(&dwork->work.state,
481 			    WORK_ST_CANCEL, WORK_ST_IDLE);
482 			mtx_unlock(&dwork->timer.mtx);
483 			return (true);
484 		}
485 		/* FALLTHROUGH */
486 	case WORK_ST_TASK:
487 		tq = dwork->work.work_queue->taskqueue;
488 		if (taskqueue_cancel(tq, &dwork->work.work_task, NULL) == 0) {
489 			atomic_cmpxchg(&dwork->work.state,
490 			    WORK_ST_CANCEL, WORK_ST_IDLE);
491 			mtx_unlock(&dwork->timer.mtx);
492 			return (true);
493 		}
494 		/* FALLTHROUGH */
495 	default:
496 		mtx_unlock(&dwork->timer.mtx);
497 		return (false);
498 	}
499 }
500 
501 /*
502  * This function cancels the given work structure in a synchronous
503  * fashion. It returns true if the work was successfully
504  * cancelled. Else the work was already cancelled.
505  */
506 static bool
linux_cancel_delayed_work_sync_int(struct delayed_work * dwork)507 linux_cancel_delayed_work_sync_int(struct delayed_work *dwork)
508 {
509 	static const uint8_t states[WORK_ST_MAX] __aligned(8) = {
510 		[WORK_ST_IDLE] = WORK_ST_IDLE,		/* NOP */
511 		[WORK_ST_TIMER] = WORK_ST_IDLE,		/* cancel and drain */
512 		[WORK_ST_TASK] = WORK_ST_IDLE,		/* cancel and drain */
513 		[WORK_ST_EXEC] = WORK_ST_IDLE,		/* too late, drain */
514 		[WORK_ST_CANCEL] = WORK_ST_IDLE,	/* cancel and drain */
515 	};
516 	struct taskqueue *tq;
517 	int ret, state;
518 	bool cancelled;
519 
520 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
521 	    "linux_cancel_delayed_work_sync() might sleep");
522 	mtx_lock(&dwork->timer.mtx);
523 
524 	state = linux_update_state(&dwork->work.state, states);
525 	switch (state) {
526 	case WORK_ST_IDLE:
527 		mtx_unlock(&dwork->timer.mtx);
528 		return (false);
529 	case WORK_ST_TIMER:
530 	case WORK_ST_CANCEL:
531 		cancelled = (callout_stop(&dwork->timer.callout) == 1);
532 
533 		tq = dwork->work.work_queue->taskqueue;
534 		ret = taskqueue_cancel(tq, &dwork->work.work_task, NULL);
535 		mtx_unlock(&dwork->timer.mtx);
536 
537 		callout_drain(&dwork->timer.callout);
538 		taskqueue_drain(tq, &dwork->work.work_task);
539 		return (cancelled || (ret != 0));
540 	default:
541 		tq = dwork->work.work_queue->taskqueue;
542 		ret = taskqueue_cancel(tq, &dwork->work.work_task, NULL);
543 		mtx_unlock(&dwork->timer.mtx);
544 		if (ret != 0)
545 			taskqueue_drain(tq, &dwork->work.work_task);
546 		return (ret != 0);
547 	}
548 }
549 
550 bool
linux_cancel_delayed_work_sync(struct delayed_work * dwork)551 linux_cancel_delayed_work_sync(struct delayed_work *dwork)
552 {
553 	bool res;
554 
555 	res = false;
556 	while (linux_cancel_delayed_work_sync_int(dwork))
557 		res = true;
558 	return (res);
559 }
560 
561 /*
562  * This function waits until the given work structure is completed.
563  * It returns non-zero if the work was successfully
564  * waited for. Else the work was not waited for.
565  */
566 bool
linux_flush_work(struct work_struct * work)567 linux_flush_work(struct work_struct *work)
568 {
569 	struct taskqueue *tq;
570 	bool retval;
571 
572 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
573 	    "linux_flush_work() might sleep");
574 
575 	switch (atomic_read(&work->state)) {
576 	case WORK_ST_IDLE:
577 		return (false);
578 	default:
579 		tq = work->work_queue->taskqueue;
580 		retval = taskqueue_poll_is_busy(tq, &work->work_task);
581 		taskqueue_drain(tq, &work->work_task);
582 		return (retval);
583 	}
584 }
585 
586 /*
587  * This function waits until the given delayed work structure is
588  * completed. It returns non-zero if the work was successfully waited
589  * for. Else the work was not waited for.
590  */
591 bool
linux_flush_delayed_work(struct delayed_work * dwork)592 linux_flush_delayed_work(struct delayed_work *dwork)
593 {
594 	struct taskqueue *tq;
595 	bool retval;
596 
597 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
598 	    "linux_flush_delayed_work() might sleep");
599 
600 	switch (atomic_read(&dwork->work.state)) {
601 	case WORK_ST_IDLE:
602 		return (false);
603 	case WORK_ST_TIMER:
604 		if (linux_cancel_timer(dwork, 1))
605 			linux_delayed_work_enqueue(dwork);
606 		/* FALLTHROUGH */
607 	default:
608 		tq = dwork->work.work_queue->taskqueue;
609 		retval = taskqueue_poll_is_busy(tq, &dwork->work.work_task);
610 		taskqueue_drain(tq, &dwork->work.work_task);
611 		return (retval);
612 	}
613 }
614 
615 /*
616  * This function returns true if the given work is pending, and not
617  * yet executing:
618  */
619 bool
linux_work_pending(struct work_struct * work)620 linux_work_pending(struct work_struct *work)
621 {
622 	switch (atomic_read(&work->state)) {
623 	case WORK_ST_TIMER:
624 	case WORK_ST_TASK:
625 	case WORK_ST_CANCEL:
626 		return (true);
627 	default:
628 		return (false);
629 	}
630 }
631 
632 /*
633  * This function returns true if the given work is busy.
634  */
635 bool
linux_work_busy(struct work_struct * work)636 linux_work_busy(struct work_struct *work)
637 {
638 	struct taskqueue *tq;
639 
640 	switch (atomic_read(&work->state)) {
641 	case WORK_ST_IDLE:
642 		return (false);
643 	case WORK_ST_EXEC:
644 		tq = work->work_queue->taskqueue;
645 		return (taskqueue_poll_is_busy(tq, &work->work_task));
646 	default:
647 		return (true);
648 	}
649 }
650 
651 struct workqueue_struct *
linux_create_workqueue_common(const char * name,int cpus)652 linux_create_workqueue_common(const char *name, int cpus)
653 {
654 	struct workqueue_struct *wq;
655 
656 	/*
657 	 * If zero CPUs are specified use the default number of CPUs:
658 	 */
659 	if (cpus == 0)
660 		cpus = linux_default_wq_cpus;
661 
662 	wq = kmalloc(sizeof(*wq), M_WAITOK | M_ZERO);
663 	wq->taskqueue = taskqueue_create(name, M_WAITOK,
664 	    taskqueue_thread_enqueue, &wq->taskqueue);
665 	atomic_set(&wq->draining, 0);
666 	taskqueue_start_threads(&wq->taskqueue, cpus, PWAIT, "%s", name);
667 	TAILQ_INIT(&wq->exec_head);
668 	mtx_init(&wq->exec_mtx, "linux_wq_exec", NULL, MTX_DEF);
669 
670 	return (wq);
671 }
672 
673 void
linux_destroy_workqueue(struct workqueue_struct * wq)674 linux_destroy_workqueue(struct workqueue_struct *wq)
675 {
676 	atomic_inc(&wq->draining);
677 	drain_workqueue(wq);
678 	taskqueue_free(wq->taskqueue);
679 	mtx_destroy(&wq->exec_mtx);
680 	kfree(wq);
681 }
682 
683 void
linux_init_delayed_work(struct delayed_work * dwork,work_func_t func)684 linux_init_delayed_work(struct delayed_work *dwork, work_func_t func)
685 {
686 	memset(dwork, 0, sizeof(*dwork));
687 	dwork->work.func = func;
688 	TASK_INIT(&dwork->work.work_task, 0, linux_delayed_work_fn, dwork);
689 	mtx_init(&dwork->timer.mtx, spin_lock_name("lkpi-dwork"), NULL,
690 	    MTX_DEF | MTX_NOWITNESS);
691 	callout_init_mtx(&dwork->timer.callout, &dwork->timer.mtx, 0);
692 }
693 
694 struct work_struct *
linux_current_work(void)695 linux_current_work(void)
696 {
697 	return (current->work);
698 }
699 
700 static void
linux_work_init(void * arg)701 linux_work_init(void *arg)
702 {
703 	int max_wq_cpus = mp_ncpus + 1;
704 
705 	/* avoid deadlock when there are too few threads */
706 	if (max_wq_cpus < 4)
707 		max_wq_cpus = 4;
708 
709 	/* set default number of CPUs */
710 	linux_default_wq_cpus = max_wq_cpus;
711 
712 	linux_system_short_wq = alloc_workqueue("linuxkpi_short_wq", 0, max_wq_cpus);
713 	linux_system_long_wq = alloc_workqueue("linuxkpi_long_wq", 0, max_wq_cpus);
714 
715 	/* populate the workqueue pointers */
716 	system_long_wq = linux_system_long_wq;
717 	system_wq = linux_system_short_wq;
718 	system_power_efficient_wq = linux_system_short_wq;
719 	system_unbound_wq = linux_system_short_wq;
720 	system_highpri_wq = linux_system_short_wq;
721 }
722 SYSINIT(linux_work_init, SI_SUB_TASKQ, SI_ORDER_THIRD, linux_work_init, NULL);
723 
724 static void
linux_work_uninit(void * arg)725 linux_work_uninit(void *arg)
726 {
727 	destroy_workqueue(linux_system_short_wq);
728 	destroy_workqueue(linux_system_long_wq);
729 
730 	/* clear workqueue pointers */
731 	system_long_wq = NULL;
732 	system_wq = NULL;
733 	system_power_efficient_wq = NULL;
734 	system_unbound_wq = NULL;
735 	system_highpri_wq = NULL;
736 }
737 SYSUNINIT(linux_work_uninit, SI_SUB_TASKQ, SI_ORDER_THIRD, linux_work_uninit, NULL);
738 
739 void
linux_irq_work_fn(void * context,int pending)740 linux_irq_work_fn(void *context, int pending)
741 {
742 	struct irq_work *irqw = context;
743 
744 	irqw->func(irqw);
745 }
746 
747 static void
linux_irq_work_init_fn(void * context,int pending)748 linux_irq_work_init_fn(void *context, int pending)
749 {
750 	/*
751 	 * LinuxKPI performs lazy allocation of memory structures required by
752 	 * current on the first access to it.  As some irq_work clients read
753 	 * it with spinlock taken, we have to preallocate td_lkpi_task before
754 	 * first call to irq_work_queue().  As irq_work uses a single thread,
755 	 * it is enough to read current once at SYSINIT stage.
756 	 */
757 	if (current == NULL)
758 		panic("irq_work taskqueue is not initialized");
759 }
760 static struct task linux_irq_work_init_task =
761     TASK_INITIALIZER(0, linux_irq_work_init_fn, &linux_irq_work_init_task);
762 
763 static void
linux_irq_work_init(void * arg)764 linux_irq_work_init(void *arg)
765 {
766 	linux_irq_work_tq = taskqueue_create_fast("linuxkpi_irq_wq",
767 	    M_WAITOK, taskqueue_thread_enqueue, &linux_irq_work_tq);
768 	taskqueue_start_threads(&linux_irq_work_tq, 1, PWAIT,
769 	    "linuxkpi_irq_wq");
770 	taskqueue_enqueue(linux_irq_work_tq, &linux_irq_work_init_task);
771 }
772 SYSINIT(linux_irq_work_init, SI_SUB_TASKQ, SI_ORDER_SECOND,
773     linux_irq_work_init, NULL);
774 
775 static void
linux_irq_work_uninit(void * arg)776 linux_irq_work_uninit(void *arg)
777 {
778 	taskqueue_drain_all(linux_irq_work_tq);
779 	taskqueue_free(linux_irq_work_tq);
780 }
781 SYSUNINIT(linux_irq_work_uninit, SI_SUB_TASKQ, SI_ORDER_SECOND,
782     linux_irq_work_uninit, NULL);
783