Home | History | Annotate | Line # | Download | only in kern
subr_workqueue.c revision 1.46
      1 /*	$NetBSD: subr_workqueue.c,v 1.46 2023/08/09 08:24:08 riastradh Exp $	*/
      2 
      3 /*-
      4  * Copyright (c)2002, 2005, 2006, 2007 YAMAMOTO Takashi,
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  *
     16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     26  * SUCH DAMAGE.
     27  */
     28 
     29 #include <sys/cdefs.h>
     30 __KERNEL_RCSID(0, "$NetBSD: subr_workqueue.c,v 1.46 2023/08/09 08:24:08 riastradh Exp $");
     31 
     32 #include <sys/param.h>
     33 
     34 #include <sys/condvar.h>
     35 #include <sys/cpu.h>
     36 #include <sys/kmem.h>
     37 #include <sys/kthread.h>
     38 #include <sys/mutex.h>
     39 #include <sys/proc.h>
     40 #include <sys/queue.h>
     41 #include <sys/sdt.h>
     42 #include <sys/systm.h>
     43 #include <sys/workqueue.h>
     44 
     45 typedef struct work_impl {
     46 	SIMPLEQ_ENTRY(work_impl) wk_entry;
     47 } work_impl_t;
     48 
     49 SIMPLEQ_HEAD(workqhead, work_impl);
     50 
     51 struct workqueue_queue {
     52 	kmutex_t q_mutex;
     53 	kcondvar_t q_cv;
     54 	struct workqhead q_queue_pending;
     55 	uint64_t q_gen;
     56 	lwp_t *q_worker;
     57 };
     58 
     59 struct workqueue {
     60 	void (*wq_func)(struct work *, void *);
     61 	void *wq_arg;
     62 	int wq_flags;
     63 
     64 	char wq_name[MAXCOMLEN];
     65 	pri_t wq_prio;
     66 	void *wq_ptr;
     67 };
     68 
     69 #define	WQ_SIZE		(roundup2(sizeof(struct workqueue), coherency_unit))
     70 #define	WQ_QUEUE_SIZE	(roundup2(sizeof(struct workqueue_queue), coherency_unit))
     71 
     72 #define	POISON	0xaabbccdd
     73 
     74 SDT_PROBE_DEFINE7(sdt, kernel, workqueue, create,
     75     "struct workqueue *"/*wq*/,
     76     "const char *"/*name*/,
     77     "void (*)(struct work *, void *)"/*func*/,
     78     "void *"/*arg*/,
     79     "pri_t"/*prio*/,
     80     "int"/*ipl*/,
     81     "int"/*flags*/);
     82 SDT_PROBE_DEFINE1(sdt, kernel, workqueue, destroy,
     83     "struct workqueue *"/*wq*/);
     84 
     85 SDT_PROBE_DEFINE3(sdt, kernel, workqueue, enqueue,
     86     "struct workqueue *"/*wq*/,
     87     "struct work *"/*wk*/,
     88     "struct cpu_info *"/*ci*/);
     89 SDT_PROBE_DEFINE4(sdt, kernel, workqueue, entry,
     90     "struct workqueue *"/*wq*/,
     91     "struct work *"/*wk*/,
     92     "void (*)(struct work *, void *)"/*func*/,
     93     "void *"/*arg*/);
     94 SDT_PROBE_DEFINE4(sdt, kernel, workqueue, return,
     95     "struct workqueue *"/*wq*/,
     96     "struct work *"/*wk*/,
     97     "void (*)(struct work *, void *)"/*func*/,
     98     "void *"/*arg*/);
     99 SDT_PROBE_DEFINE2(sdt, kernel, workqueue, wait__start,
    100     "struct workqueue *"/*wq*/,
    101     "struct work *"/*wk*/);
    102 SDT_PROBE_DEFINE2(sdt, kernel, workqueue, wait__self,
    103     "struct workqueue *"/*wq*/,
    104     "struct work *"/*wk*/);
    105 SDT_PROBE_DEFINE2(sdt, kernel, workqueue, wait__hit,
    106     "struct workqueue *"/*wq*/,
    107     "struct work *"/*wk*/);
    108 SDT_PROBE_DEFINE2(sdt, kernel, workqueue, wait__done,
    109     "struct workqueue *"/*wq*/,
    110     "struct work *"/*wk*/);
    111 
    112 SDT_PROBE_DEFINE1(sdt, kernel, workqueue, exit__start,
    113     "struct workqueue *"/*wq*/);
    114 SDT_PROBE_DEFINE1(sdt, kernel, workqueue, exit__done,
    115     "struct workqueue *"/*wq*/);
    116 
    117 static size_t
    118 workqueue_size(int flags)
    119 {
    120 
    121 	return WQ_SIZE
    122 	    + ((flags & WQ_PERCPU) != 0 ? ncpu : 1) * WQ_QUEUE_SIZE
    123 	    + coherency_unit;
    124 }
    125 
    126 static struct workqueue_queue *
    127 workqueue_queue_lookup(struct workqueue *wq, struct cpu_info *ci)
    128 {
    129 	u_int idx = 0;
    130 
    131 	if (wq->wq_flags & WQ_PERCPU) {
    132 		idx = ci ? cpu_index(ci) : cpu_index(curcpu());
    133 	}
    134 
    135 	return (void *)((uintptr_t)(wq) + WQ_SIZE + (idx * WQ_QUEUE_SIZE));
    136 }
    137 
    138 static void
    139 workqueue_runlist(struct workqueue *wq, struct workqhead *list)
    140 {
    141 	work_impl_t *wk;
    142 	work_impl_t *next;
    143 
    144 	for (wk = SIMPLEQ_FIRST(list); wk != NULL; wk = next) {
    145 		next = SIMPLEQ_NEXT(wk, wk_entry);
    146 		SDT_PROBE4(sdt, kernel, workqueue, entry,
    147 		    wq, wk, wq->wq_func, wq->wq_arg);
    148 		(*wq->wq_func)((void *)wk, wq->wq_arg);
    149 		SDT_PROBE4(sdt, kernel, workqueue, return,
    150 		    wq, wk, wq->wq_func, wq->wq_arg);
    151 	}
    152 }
    153 
    154 static void
    155 workqueue_worker(void *cookie)
    156 {
    157 	struct workqueue *wq = cookie;
    158 	struct workqueue_queue *q;
    159 	int s;
    160 
    161 	/* find the workqueue of this kthread */
    162 	q = workqueue_queue_lookup(wq, curlwp->l_cpu);
    163 
    164 	if (wq->wq_flags & WQ_FPU)
    165 		s = kthread_fpu_enter();
    166 	mutex_enter(&q->q_mutex);
    167 	for (;;) {
    168 		struct workqhead tmp;
    169 
    170 		SIMPLEQ_INIT(&tmp);
    171 
    172 		while (SIMPLEQ_EMPTY(&q->q_queue_pending))
    173 			cv_wait(&q->q_cv, &q->q_mutex);
    174 		SIMPLEQ_CONCAT(&tmp, &q->q_queue_pending);
    175 		SIMPLEQ_INIT(&q->q_queue_pending);
    176 
    177 		/*
    178 		 * Mark the queue as actively running a batch of work
    179 		 * by setting the generation number odd.
    180 		 */
    181 		q->q_gen |= 1;
    182 		mutex_exit(&q->q_mutex);
    183 
    184 		workqueue_runlist(wq, &tmp);
    185 
    186 		/*
    187 		 * Notify workqueue_wait that we have completed a batch
    188 		 * of work by incrementing the generation number.
    189 		 */
    190 		mutex_enter(&q->q_mutex);
    191 		KASSERTMSG(q->q_gen & 1, "q=%p gen=%"PRIu64, q, q->q_gen);
    192 		q->q_gen++;
    193 		cv_broadcast(&q->q_cv);
    194 	}
    195 	mutex_exit(&q->q_mutex);
    196 	if (wq->wq_flags & WQ_FPU)
    197 		kthread_fpu_exit(s);
    198 }
    199 
    200 static void
    201 workqueue_init(struct workqueue *wq, const char *name,
    202     void (*callback_func)(struct work *, void *), void *callback_arg,
    203     pri_t prio, int ipl)
    204 {
    205 
    206 	KASSERT(sizeof(wq->wq_name) > strlen(name));
    207 	strncpy(wq->wq_name, name, sizeof(wq->wq_name));
    208 
    209 	wq->wq_prio = prio;
    210 	wq->wq_func = callback_func;
    211 	wq->wq_arg = callback_arg;
    212 }
    213 
    214 static int
    215 workqueue_initqueue(struct workqueue *wq, struct workqueue_queue *q,
    216     int ipl, struct cpu_info *ci)
    217 {
    218 	int error, ktf;
    219 
    220 	KASSERT(q->q_worker == NULL);
    221 
    222 	mutex_init(&q->q_mutex, MUTEX_DEFAULT, ipl);
    223 	cv_init(&q->q_cv, wq->wq_name);
    224 	SIMPLEQ_INIT(&q->q_queue_pending);
    225 	q->q_gen = 0;
    226 	ktf = ((wq->wq_flags & WQ_MPSAFE) != 0 ? KTHREAD_MPSAFE : 0);
    227 	if (wq->wq_prio < PRI_KERNEL)
    228 		ktf |= KTHREAD_TS;
    229 	if (ci) {
    230 		error = kthread_create(wq->wq_prio, ktf, ci, workqueue_worker,
    231 		    wq, &q->q_worker, "%s/%u", wq->wq_name, ci->ci_index);
    232 	} else {
    233 		error = kthread_create(wq->wq_prio, ktf, ci, workqueue_worker,
    234 		    wq, &q->q_worker, "%s", wq->wq_name);
    235 	}
    236 	if (error != 0) {
    237 		mutex_destroy(&q->q_mutex);
    238 		cv_destroy(&q->q_cv);
    239 		KASSERT(q->q_worker == NULL);
    240 	}
    241 	return error;
    242 }
    243 
    244 struct workqueue_exitargs {
    245 	work_impl_t wqe_wk;
    246 	struct workqueue_queue *wqe_q;
    247 };
    248 
    249 static void
    250 workqueue_exit(struct work *wk, void *arg)
    251 {
    252 	struct workqueue_exitargs *wqe = (void *)wk;
    253 	struct workqueue_queue *q = wqe->wqe_q;
    254 
    255 	/*
    256 	 * only competition at this point is workqueue_finiqueue.
    257 	 */
    258 
    259 	KASSERT(q->q_worker == curlwp);
    260 	KASSERT(SIMPLEQ_EMPTY(&q->q_queue_pending));
    261 	mutex_enter(&q->q_mutex);
    262 	q->q_worker = NULL;
    263 	cv_broadcast(&q->q_cv);
    264 	mutex_exit(&q->q_mutex);
    265 	kthread_exit(0);
    266 }
    267 
    268 static void
    269 workqueue_finiqueue(struct workqueue *wq, struct workqueue_queue *q)
    270 {
    271 	struct workqueue_exitargs wqe;
    272 
    273 	KASSERT(wq->wq_func == workqueue_exit);
    274 
    275 	wqe.wqe_q = q;
    276 	KASSERT(SIMPLEQ_EMPTY(&q->q_queue_pending));
    277 	KASSERT(q->q_worker != NULL);
    278 	mutex_enter(&q->q_mutex);
    279 	SIMPLEQ_INSERT_TAIL(&q->q_queue_pending, &wqe.wqe_wk, wk_entry);
    280 	cv_broadcast(&q->q_cv);
    281 	while (q->q_worker != NULL) {
    282 		cv_wait(&q->q_cv, &q->q_mutex);
    283 	}
    284 	mutex_exit(&q->q_mutex);
    285 	mutex_destroy(&q->q_mutex);
    286 	cv_destroy(&q->q_cv);
    287 }
    288 
    289 /* --- */
    290 
    291 int
    292 workqueue_create(struct workqueue **wqp, const char *name,
    293     void (*callback_func)(struct work *, void *), void *callback_arg,
    294     pri_t prio, int ipl, int flags)
    295 {
    296 	struct workqueue *wq;
    297 	struct workqueue_queue *q;
    298 	void *ptr;
    299 	int error = 0;
    300 
    301 	CTASSERT(sizeof(work_impl_t) <= sizeof(struct work));
    302 
    303 	ptr = kmem_zalloc(workqueue_size(flags), KM_SLEEP);
    304 	wq = (void *)roundup2((uintptr_t)ptr, coherency_unit);
    305 	wq->wq_ptr = ptr;
    306 	wq->wq_flags = flags;
    307 
    308 	workqueue_init(wq, name, callback_func, callback_arg, prio, ipl);
    309 
    310 	if (flags & WQ_PERCPU) {
    311 		struct cpu_info *ci;
    312 		CPU_INFO_ITERATOR cii;
    313 
    314 		/* create the work-queue for each CPU */
    315 		for (CPU_INFO_FOREACH(cii, ci)) {
    316 			q = workqueue_queue_lookup(wq, ci);
    317 			error = workqueue_initqueue(wq, q, ipl, ci);
    318 			if (error) {
    319 				break;
    320 			}
    321 		}
    322 	} else {
    323 		/* initialize a work-queue */
    324 		q = workqueue_queue_lookup(wq, NULL);
    325 		error = workqueue_initqueue(wq, q, ipl, NULL);
    326 	}
    327 
    328 	if (error != 0) {
    329 		workqueue_destroy(wq);
    330 	} else {
    331 		*wqp = wq;
    332 	}
    333 
    334 	return error;
    335 }
    336 
    337 static bool
    338 workqueue_q_wait(struct workqueue *wq, struct workqueue_queue *q,
    339     work_impl_t *wk_target)
    340 {
    341 	work_impl_t *wk;
    342 	bool found = false;
    343 	uint64_t gen;
    344 
    345 	mutex_enter(&q->q_mutex);
    346 
    347 	/*
    348 	 * Avoid a deadlock scenario.  We can't guarantee that
    349 	 * wk_target has completed at this point, but we can't wait for
    350 	 * it either, so do nothing.
    351 	 *
    352 	 * XXX Are there use-cases that require this semantics?
    353 	 */
    354 	if (q->q_worker == curlwp) {
    355 		SDT_PROBE2(sdt, kernel, workqueue, wait__self,  wq, wk_target);
    356 		goto out;
    357 	}
    358 
    359 	/*
    360 	 * Wait until the target is no longer pending.  If we find it
    361 	 * on this queue, the caller can stop looking in other queues.
    362 	 * If we don't find it in this queue, however, we can't skip
    363 	 * waiting -- it may be hidden in the running queue which we
    364 	 * have no access to.
    365 	 */
    366     again:
    367 	SIMPLEQ_FOREACH(wk, &q->q_queue_pending, wk_entry) {
    368 		if (wk == wk_target) {
    369 			SDT_PROBE2(sdt, kernel, workqueue, wait__hit,  wq, wk);
    370 			found = true;
    371 			cv_wait(&q->q_cv, &q->q_mutex);
    372 			goto again;
    373 		}
    374 	}
    375 
    376 	/*
    377 	 * The target may be in the batch of work currently running,
    378 	 * but we can't touch that queue.  So if there's anything
    379 	 * running, wait until the generation changes.
    380 	 */
    381 	gen = q->q_gen;
    382 	if (gen & 1) {
    383 		do
    384 			cv_wait(&q->q_cv, &q->q_mutex);
    385 		while (gen == q->q_gen);
    386 	}
    387 
    388     out:
    389 	mutex_exit(&q->q_mutex);
    390 
    391 	return found;
    392 }
    393 
    394 /*
    395  * Wait for a specified work to finish.  The caller must ensure that no new
    396  * work will be enqueued before calling workqueue_wait.  Note that if the
    397  * workqueue is WQ_PERCPU, the caller can enqueue a new work to another queue
    398  * other than the waiting queue.
    399  */
    400 void
    401 workqueue_wait(struct workqueue *wq, struct work *wk)
    402 {
    403 	struct workqueue_queue *q;
    404 	bool found;
    405 
    406 	ASSERT_SLEEPABLE();
    407 
    408 	SDT_PROBE2(sdt, kernel, workqueue, wait__start,  wq, wk);
    409 	if (ISSET(wq->wq_flags, WQ_PERCPU)) {
    410 		struct cpu_info *ci;
    411 		CPU_INFO_ITERATOR cii;
    412 		for (CPU_INFO_FOREACH(cii, ci)) {
    413 			q = workqueue_queue_lookup(wq, ci);
    414 			found = workqueue_q_wait(wq, q, (work_impl_t *)wk);
    415 			if (found)
    416 				break;
    417 		}
    418 	} else {
    419 		q = workqueue_queue_lookup(wq, NULL);
    420 		(void)workqueue_q_wait(wq, q, (work_impl_t *)wk);
    421 	}
    422 	SDT_PROBE2(sdt, kernel, workqueue, wait__done,  wq, wk);
    423 }
    424 
    425 void
    426 workqueue_destroy(struct workqueue *wq)
    427 {
    428 	struct workqueue_queue *q;
    429 	struct cpu_info *ci;
    430 	CPU_INFO_ITERATOR cii;
    431 
    432 	ASSERT_SLEEPABLE();
    433 
    434 	SDT_PROBE1(sdt, kernel, workqueue, exit__start,  wq);
    435 	wq->wq_func = workqueue_exit;
    436 	for (CPU_INFO_FOREACH(cii, ci)) {
    437 		q = workqueue_queue_lookup(wq, ci);
    438 		if (q->q_worker != NULL) {
    439 			workqueue_finiqueue(wq, q);
    440 		}
    441 	}
    442 	SDT_PROBE1(sdt, kernel, workqueue, exit__done,  wq);
    443 	kmem_free(wq->wq_ptr, workqueue_size(wq->wq_flags));
    444 }
    445 
    446 #ifdef DEBUG
    447 static void
    448 workqueue_check_duplication(struct workqueue_queue *q, work_impl_t *wk)
    449 {
    450 	work_impl_t *_wk;
    451 
    452 	SIMPLEQ_FOREACH(_wk, &q->q_queue_pending, wk_entry) {
    453 		if (_wk == wk)
    454 			panic("%s: tried to enqueue a queued work", __func__);
    455 	}
    456 }
    457 #endif
    458 
    459 void
    460 workqueue_enqueue(struct workqueue *wq, struct work *wk0, struct cpu_info *ci)
    461 {
    462 	struct workqueue_queue *q;
    463 	work_impl_t *wk = (void *)wk0;
    464 
    465 	SDT_PROBE3(sdt, kernel, workqueue, enqueue,  wq, wk0, ci);
    466 
    467 	KASSERT(wq->wq_flags & WQ_PERCPU || ci == NULL);
    468 	q = workqueue_queue_lookup(wq, ci);
    469 
    470 	mutex_enter(&q->q_mutex);
    471 #ifdef DEBUG
    472 	workqueue_check_duplication(q, wk);
    473 #endif
    474 	SIMPLEQ_INSERT_TAIL(&q->q_queue_pending, wk, wk_entry);
    475 	cv_broadcast(&q->q_cv);
    476 	mutex_exit(&q->q_mutex);
    477 }
    478