sys/kern/subr_taskqueue.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2000 Doug Rabson
   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 __FBSDID("$FreeBSD$");
  31
  32 #include <sys/param.h>
  33 #include <sys/systm.h>
  34 #include <sys/bus.h>
  35 #include <sys/cpuset.h>
  36 #include <sys/interrupt.h>
  37 #include <sys/kernel.h>
  38 #include <sys/kthread.h>
  39 #include <sys/libkern.h>
  40 #include <sys/limits.h>
  41 #include <sys/lock.h>
  42 #include <sys/malloc.h>
  43 #include <sys/mutex.h>
  44 #include <sys/proc.h>
  45 #include <sys/epoch.h>
  46 #include <sys/sched.h>
  47 #include <sys/smp.h>
  48 #include <sys/taskqueue.h>
  49 #include <sys/unistd.h>
  50 #include <machine/stdarg.h>
  51
  52 static MALLOC_DEFINE(M_TASKQUEUE, "taskqueue", "Task Queues");
  53 static void     *taskqueue_giant_ih;
  54 static void     *taskqueue_ih;
  55 static void      taskqueue_fast_enqueue(void *);
  56 static void      taskqueue_swi_enqueue(void *);
  57 static void      taskqueue_swi_giant_enqueue(void *);
  58
  59 struct taskqueue_busy {
  60         struct task             *tb_running;
  61         u_int                    tb_seq;
  62         LIST_ENTRY(taskqueue_busy) tb_link;
  63 };
  64
  65 struct taskqueue {
  66         STAILQ_HEAD(, task)     tq_queue;
  67         LIST_HEAD(, taskqueue_busy) tq_active;
  68         struct task             *tq_hint;
  69         u_int                   tq_seq;
  70         int                     tq_callouts;
  71         struct mtx_padalign     tq_mutex;
  72         taskqueue_enqueue_fn    tq_enqueue;
  73         void                    *tq_context;
  74         char                    *tq_name;
  75         struct thread           **tq_threads;
  76         int                     tq_tcount;
  77         int                     tq_spin;
  78         int                     tq_flags;
  79         taskqueue_callback_fn   tq_callbacks[TASKQUEUE_NUM_CALLBACKS];
  80         void                    *tq_cb_contexts[TASKQUEUE_NUM_CALLBACKS];
  81 };
  82
  83 #define TQ_FLAGS_ACTIVE         (1 << 0)
  84 #define TQ_FLAGS_BLOCKED        (1 << 1)
  85 #define TQ_FLAGS_UNLOCKED_ENQUEUE       (1 << 2)
  86
  87 #define DT_CALLOUT_ARMED        (1 << 0)
  88 #define DT_DRAIN_IN_PROGRESS    (1 << 1)
  89
  90 #define TQ_LOCK(tq)                                                     \
  91         do {                                                            \
  92                 if ((tq)->tq_spin)                                      \
  93                         mtx_lock_spin(&(tq)->tq_mutex);                 \
  94                 else                                                    \
  95                         mtx_lock(&(tq)->tq_mutex);                      \
  96         } while (0)
  97 #define TQ_ASSERT_LOCKED(tq)    mtx_assert(&(tq)->tq_mutex, MA_OWNED)
  98
  99 #define TQ_UNLOCK(tq)                                                   \
 100         do {                                                            \
 101                 if ((tq)->tq_spin)                                      \
 102                         mtx_unlock_spin(&(tq)->tq_mutex);               \
 103                 else                                                    \
 104                         mtx_unlock(&(tq)->tq_mutex);                    \
 105         } while (0)
 106 #define TQ_ASSERT_UNLOCKED(tq)  mtx_assert(&(tq)->tq_mutex, MA_NOTOWNED)
 107
 108 void
 109 _timeout_task_init(struct taskqueue *queue, struct timeout_task *timeout_task,
 110     int priority, task_fn_t func, void *context)
 111 {
 112
 113         TASK_INIT(&timeout_task->t, priority, func, context);
 114         callout_init_mtx(&timeout_task->c, &queue->tq_mutex,
 115             CALLOUT_RETURNUNLOCKED);
 116         timeout_task->q = queue;
 117         timeout_task->f = 0;
 118 }
 119
 120 static __inline int
 121 TQ_SLEEP(struct taskqueue *tq, void *p, const char *wm)
 122 {
 123         if (tq->tq_spin)
 124                 return (msleep_spin(p, (struct mtx *)&tq->tq_mutex, wm, 0));
 125         return (msleep(p, &tq->tq_mutex, 0, wm, 0));
 126 }
 127
 128 static struct taskqueue *
 129 _taskqueue_create(const char *name, int mflags,
 130                  taskqueue_enqueue_fn enqueue, void *context,
 131                  int mtxflags, const char *mtxname __unused)
 132 {
 133         struct taskqueue *queue;
 134         char *tq_name;
 135
 136         tq_name = malloc(TASKQUEUE_NAMELEN, M_TASKQUEUE, mflags | M_ZERO);
 137         if (tq_name == NULL)
 138                 return (NULL);
 139
 140         queue = malloc(sizeof(struct taskqueue), M_TASKQUEUE, mflags | M_ZERO);
 141         if (queue == NULL) {
 142                 free(tq_name, M_TASKQUEUE);
 143                 return (NULL);
 144         }
 145
 146         snprintf(tq_name, TASKQUEUE_NAMELEN, "%s", (name) ? name : "taskqueue");
 147
 148         STAILQ_INIT(&queue->tq_queue);
 149         LIST_INIT(&queue->tq_active);
 150         queue->tq_enqueue = enqueue;
 151         queue->tq_context = context;
 152         queue->tq_name = tq_name;
 153         queue->tq_spin = (mtxflags & MTX_SPIN) != 0;
 154         queue->tq_flags |= TQ_FLAGS_ACTIVE;
 155         if (enqueue == taskqueue_fast_enqueue ||
 156             enqueue == taskqueue_swi_enqueue ||
 157             enqueue == taskqueue_swi_giant_enqueue ||
 158             enqueue == taskqueue_thread_enqueue)
 159                 queue->tq_flags |= TQ_FLAGS_UNLOCKED_ENQUEUE;
 160         mtx_init(&queue->tq_mutex, tq_name, NULL, mtxflags);
 161
 162         return (queue);
 163 }
 164
 165 struct taskqueue *
 166 taskqueue_create(const char *name, int mflags,
 167                  taskqueue_enqueue_fn enqueue, void *context)
 168 {
 169
 170         return _taskqueue_create(name, mflags, enqueue, context,
 171                         MTX_DEF, name);
 172 }
 173
 174 void
 175 taskqueue_set_callback(struct taskqueue *queue,
 176     enum taskqueue_callback_type cb_type, taskqueue_callback_fn callback,
 177     void *context)
 178 {
 179
 180         KASSERT(((cb_type >= TASKQUEUE_CALLBACK_TYPE_MIN) &&
 181             (cb_type <= TASKQUEUE_CALLBACK_TYPE_MAX)),
 182             ("Callback type %d not valid, must be %d-%d", cb_type,
 183             TASKQUEUE_CALLBACK_TYPE_MIN, TASKQUEUE_CALLBACK_TYPE_MAX));
 184         KASSERT((queue->tq_callbacks[cb_type] == NULL),
 185             ("Re-initialization of taskqueue callback?"));
 186
 187         queue->tq_callbacks[cb_type] = callback;
 188         queue->tq_cb_contexts[cb_type] = context;
 189 }
 190
 191 /*
 192  * Signal a taskqueue thread to terminate.
 193  */
 194 static void
 195 taskqueue_terminate(struct thread **pp, struct taskqueue *tq)
 196 {
 197
 198         while (tq->tq_tcount > 0 || tq->tq_callouts > 0) {
 199                 wakeup(tq);
 200                 TQ_SLEEP(tq, pp, "tq_destroy");
 201         }
 202 }
 203
 204 void
 205 taskqueue_free(struct taskqueue *queue)
 206 {
 207
 208         TQ_LOCK(queue);
 209         queue->tq_flags &= ~TQ_FLAGS_ACTIVE;
 210         taskqueue_terminate(queue->tq_threads, queue);
 211         KASSERT(LIST_EMPTY(&queue->tq_active), ("Tasks still running?"));
 212         KASSERT(queue->tq_callouts == 0, ("Armed timeout tasks"));
 213         mtx_destroy(&queue->tq_mutex);
 214         free(queue->tq_threads, M_TASKQUEUE);
 215         free(queue->tq_name, M_TASKQUEUE);
 216         free(queue, M_TASKQUEUE);
 217 }
 218
 219 static int
 220 taskqueue_enqueue_locked(struct taskqueue *queue, struct task *task)
 221 {
 222         struct task *ins;
 223         struct task *prev;
 224
 225         KASSERT(task->ta_func != NULL, ("enqueueing task with NULL func"));
 226         /*
 227          * Count multiple enqueues.
 228          */
 229         if (task->ta_pending) {
 230                 if (task->ta_pending < USHRT_MAX)
 231                         task->ta_pending++;
 232                 TQ_UNLOCK(queue);
 233                 return (0);
 234         }
 235
 236         /*
 237          * Optimise cases when all tasks use small set of priorities.
 238          * In case of only one priority we always insert at the end.
 239          * In case of two tq_hint typically gives the insertion point.
 240          * In case of more then two tq_hint should halve the search.
 241          */
 242         prev = STAILQ_LAST(&queue->tq_queue, task, ta_link);
 243         if (!prev || prev->ta_priority >= task->ta_priority) {
 244                 STAILQ_INSERT_TAIL(&queue->tq_queue, task, ta_link);
 245         } else {
 246                 prev = queue->tq_hint;
 247                 if (prev && prev->ta_priority >= task->ta_priority) {
 248                         ins = STAILQ_NEXT(prev, ta_link);
 249                 } else {
 250                         prev = NULL;
 251                         ins = STAILQ_FIRST(&queue->tq_queue);
 252                 }
 253                 for (; ins; prev = ins, ins = STAILQ_NEXT(ins, ta_link))
 254                         if (ins->ta_priority < task->ta_priority)
 255                                 break;
 256
 257                 if (prev) {
 258                         STAILQ_INSERT_AFTER(&queue->tq_queue, prev, task, ta_link);
 259                         queue->tq_hint = task;
 260                 } else
 261                         STAILQ_INSERT_HEAD(&queue->tq_queue, task, ta_link);
 262         }
 263
 264         task->ta_pending = 1;
 265         if ((queue->tq_flags & TQ_FLAGS_UNLOCKED_ENQUEUE) != 0)
 266                 TQ_UNLOCK(queue);
 267         if ((queue->tq_flags & TQ_FLAGS_BLOCKED) == 0)
 268                 queue->tq_enqueue(queue->tq_context);
 269         if ((queue->tq_flags & TQ_FLAGS_UNLOCKED_ENQUEUE) == 0)
 270                 TQ_UNLOCK(queue);
 271
 272         /* Return with lock released. */
 273         return (0);
 274 }
 275
 276 int
 277 taskqueue_enqueue(struct taskqueue *queue, struct task *task)
 278 {
 279         int res;
 280
 281         TQ_LOCK(queue);
 282         res = taskqueue_enqueue_locked(queue, task);
 283         /* The lock is released inside. */
 284
 285         return (res);
 286 }
 287
 288 static void
 289 taskqueue_timeout_func(void *arg)
 290 {
 291         struct taskqueue *queue;
 292         struct timeout_task *timeout_task;
 293
 294         timeout_task = arg;
 295         queue = timeout_task->q;
 296         KASSERT((timeout_task->f & DT_CALLOUT_ARMED) != 0, ("Stray timeout"));
 297         timeout_task->f &= ~DT_CALLOUT_ARMED;
 298         queue->tq_callouts--;
 299         taskqueue_enqueue_locked(timeout_task->q, &timeout_task->t);
 300         /* The lock is released inside. */
 301 }
 302
 303 int
 304 taskqueue_enqueue_timeout_sbt(struct taskqueue *queue,
 305     struct timeout_task *timeout_task, sbintime_t sbt, sbintime_t pr, int flags)
 306 {
 307         int res;
 308
 309         TQ_LOCK(queue);
 310         KASSERT(timeout_task->q == NULL || timeout_task->q == queue,
 311             ("Migrated queue"));
 312         KASSERT(!queue->tq_spin, ("Timeout for spin-queue"));
 313         timeout_task->q = queue;
 314         res = timeout_task->t.ta_pending;
 315         if (timeout_task->f & DT_DRAIN_IN_PROGRESS) {
 316                 /* Do nothing */
 317                 TQ_UNLOCK(queue);
 318                 res = -1;
 319         } else if (sbt == 0) {
 320                 taskqueue_enqueue_locked(queue, &timeout_task->t);
 321                 /* The lock is released inside. */
 322         } else {
 323                 if ((timeout_task->f & DT_CALLOUT_ARMED) != 0) {
 324                         res++;
 325                 } else {
 326                         queue->tq_callouts++;
 327                         timeout_task->f |= DT_CALLOUT_ARMED;
 328                         if (sbt < 0)
 329                                 sbt = -sbt; /* Ignore overflow. */
 330                 }
 331                 if (sbt > 0) {
 332                         callout_reset_sbt(&timeout_task->c, sbt, pr,
 333                             taskqueue_timeout_func, timeout_task, flags);
 334                 }
 335                 TQ_UNLOCK(queue);
 336         }
 337         return (res);
 338 }
 339
 340 int
 341 taskqueue_enqueue_timeout(struct taskqueue *queue,
 342     struct timeout_task *ttask, int ticks)
 343 {
 344
 345         return (taskqueue_enqueue_timeout_sbt(queue, ttask, ticks * tick_sbt,
 346             0, 0));
 347 }
 348
 349 static void
 350 taskqueue_task_nop_fn(void *context, int pending)
 351 {
 352 }
 353
 354 /*
 355  * Block until all currently queued tasks in this taskqueue
 356  * have begun execution.  Tasks queued during execution of
 357  * this function are ignored.
 358  */
 359 static int
 360 taskqueue_drain_tq_queue(struct taskqueue *queue)
 361 {
 362         struct task t_barrier;
 363
 364         if (STAILQ_EMPTY(&queue->tq_queue))
 365                 return (0);
 366
 367         /*
 368          * Enqueue our barrier after all current tasks, but with
 369          * the highest priority so that newly queued tasks cannot
 370          * pass it.  Because of the high priority, we can not use
 371          * taskqueue_enqueue_locked directly (which drops the lock
 372          * anyway) so just insert it at tail while we have the
 373          * queue lock.
 374          */
 375         TASK_INIT(&t_barrier, UCHAR_MAX, taskqueue_task_nop_fn, &t_barrier);
 376         STAILQ_INSERT_TAIL(&queue->tq_queue, &t_barrier, ta_link);
 377         queue->tq_hint = &t_barrier;
 378         t_barrier.ta_pending = 1;
 379
 380         /*
 381          * Once the barrier has executed, all previously queued tasks
 382          * have completed or are currently executing.
 383          */
 384         while (t_barrier.ta_pending != 0)
 385                 TQ_SLEEP(queue, &t_barrier, "tq_qdrain");
 386         return (1);
 387 }
 388
 389 /*
 390  * Block until all currently executing tasks for this taskqueue
 391  * complete.  Tasks that begin execution during the execution
 392  * of this function are ignored.
 393  */
 394 static int
 395 taskqueue_drain_tq_active(struct taskqueue *queue)
 396 {
 397         struct taskqueue_busy *tb;
 398         u_int seq;
 399
 400         if (LIST_EMPTY(&queue->tq_active))
 401                 return (0);
 402
 403         /* Block taskq_terminate().*/
 404         queue->tq_callouts++;
 405
 406         /* Wait for any active task with sequence from the past. */
 407         seq = queue->tq_seq;
 408 restart:
 409         LIST_FOREACH(tb, &queue->tq_active, tb_link) {
 410                 if ((int)(tb->tb_seq - seq) <= 0) {
 411                         TQ_SLEEP(queue, tb->tb_running, "tq_adrain");
 412                         goto restart;
 413                 }
 414         }
 415
 416         /* Release taskqueue_terminate(). */
 417         queue->tq_callouts--;
 418         if ((queue->tq_flags & TQ_FLAGS_ACTIVE) == 0)
 419                 wakeup_one(queue->tq_threads);
 420         return (1);
 421 }
 422
 423 void
 424 taskqueue_block(struct taskqueue *queue)
 425 {
 426
 427         TQ_LOCK(queue);
 428         queue->tq_flags |= TQ_FLAGS_BLOCKED;
 429         TQ_UNLOCK(queue);
 430 }
 431
 432 void
 433 taskqueue_unblock(struct taskqueue *queue)
 434 {
 435
 436         TQ_LOCK(queue);
 437         queue->tq_flags &= ~TQ_FLAGS_BLOCKED;
 438         if (!STAILQ_EMPTY(&queue->tq_queue))
 439                 queue->tq_enqueue(queue->tq_context);
 440         TQ_UNLOCK(queue);
 441 }
 442
 443 static void
 444 taskqueue_run_locked(struct taskqueue *queue)
 445 {
 446         struct epoch_tracker et;
 447         struct taskqueue_busy tb;
 448         struct task *task;
 449         bool in_net_epoch;
 450         int pending;
 451
 452         KASSERT(queue != NULL, ("tq is NULL"));
 453         TQ_ASSERT_LOCKED(queue);
 454         tb.tb_running = NULL;
 455         LIST_INSERT_HEAD(&queue->tq_active, &tb, tb_link);
 456         in_net_epoch = false;
 457
 458         while ((task = STAILQ_FIRST(&queue->tq_queue)) != NULL) {
 459                 STAILQ_REMOVE_HEAD(&queue->tq_queue, ta_link);
 460                 if (queue->tq_hint == task)
 461                         queue->tq_hint = NULL;
 462                 pending = task->ta_pending;
 463                 task->ta_pending = 0;
 464                 tb.tb_running = task;
 465                 tb.tb_seq = ++queue->tq_seq;
 466                 TQ_UNLOCK(queue);
 467
 468                 KASSERT(task->ta_func != NULL, ("task->ta_func is NULL"));
 469                 if (!in_net_epoch && TASK_IS_NET(task)) {
 470                         in_net_epoch = true;
 471                         NET_EPOCH_ENTER(et);
 472                 } else if (in_net_epoch && !TASK_IS_NET(task)) {
 473                         NET_EPOCH_EXIT(et);
 474                         in_net_epoch = false;
 475                 }
 476                 task->ta_func(task->ta_context, pending);
 477
 478                 TQ_LOCK(queue);
 479                 wakeup(task);
 480         }
 481         if (in_net_epoch)
 482                 NET_EPOCH_EXIT(et);
 483         LIST_REMOVE(&tb, tb_link);
 484 }
 485
 486 void
 487 taskqueue_run(struct taskqueue *queue)
 488 {
 489
 490         TQ_LOCK(queue);
 491         taskqueue_run_locked(queue);
 492         TQ_UNLOCK(queue);
 493 }
 494
 495 static int
 496 task_is_running(struct taskqueue *queue, struct task *task)
 497 {
 498         struct taskqueue_busy *tb;
 499
 500         TQ_ASSERT_LOCKED(queue);
 501         LIST_FOREACH(tb, &queue->tq_active, tb_link) {
 502                 if (tb->tb_running == task)
 503                         return (1);
 504         }
 505         return (0);
 506 }
 507
 508 /*
 509  * Only use this function in single threaded contexts. It returns
 510  * non-zero if the given task is either pending or running. Else the
 511  * task is idle and can be queued again or freed.
 512  */
 513 int
 514 taskqueue_poll_is_busy(struct taskqueue *queue, struct task *task)
 515 {
 516         int retval;
 517
 518         TQ_LOCK(queue);
 519         retval = task->ta_pending > 0 || task_is_running(queue, task);
 520         TQ_UNLOCK(queue);
 521
 522         return (retval);
 523 }
 524
 525 static int
 526 taskqueue_cancel_locked(struct taskqueue *queue, struct task *task,
 527     u_int *pendp)
 528 {
 529
 530         if (task->ta_pending > 0) {
 531                 STAILQ_REMOVE(&queue->tq_queue, task, task, ta_link);
 532                 if (queue->tq_hint == task)
 533                         queue->tq_hint = NULL;
 534         }
 535         if (pendp != NULL)
 536                 *pendp = task->ta_pending;
 537         task->ta_pending = 0;
 538         return (task_is_running(queue, task) ? EBUSY : 0);
 539 }
 540
 541 int
 542 taskqueue_cancel(struct taskqueue *queue, struct task *task, u_int *pendp)
 543 {
 544         int error;
 545
 546         TQ_LOCK(queue);
 547         error = taskqueue_cancel_locked(queue, task, pendp);
 548         TQ_UNLOCK(queue);
 549
 550         return (error);
 551 }
 552
 553 int
 554 taskqueue_cancel_timeout(struct taskqueue *queue,
 555     struct timeout_task *timeout_task, u_int *pendp)
 556 {
 557         u_int pending, pending1;
 558         int error;
 559
 560         TQ_LOCK(queue);
 561         pending = !!(callout_stop(&timeout_task->c) > 0);
 562         error = taskqueue_cancel_locked(queue, &timeout_task->t, &pending1);
 563         if ((timeout_task->f & DT_CALLOUT_ARMED) != 0) {
 564                 timeout_task->f &= ~DT_CALLOUT_ARMED;
 565                 queue->tq_callouts--;
 566         }
 567         TQ_UNLOCK(queue);
 568
 569         if (pendp != NULL)
 570                 *pendp = pending + pending1;
 571         return (error);
 572 }
 573
 574 void
 575 taskqueue_drain(struct taskqueue *queue, struct task *task)
 576 {
 577
 578         if (!queue->tq_spin)
 579                 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__);
 580
 581         TQ_LOCK(queue);
 582         while (task->ta_pending != 0 || task_is_running(queue, task))
 583                 TQ_SLEEP(queue, task, "tq_drain");
 584         TQ_UNLOCK(queue);
 585 }
 586
 587 void
 588 taskqueue_drain_all(struct taskqueue *queue)
 589 {
 590
 591         if (!queue->tq_spin)
 592                 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__);
 593
 594         TQ_LOCK(queue);
 595         (void)taskqueue_drain_tq_queue(queue);
 596         (void)taskqueue_drain_tq_active(queue);
 597         TQ_UNLOCK(queue);
 598 }
 599
 600 void
 601 taskqueue_drain_timeout(struct taskqueue *queue,
 602     struct timeout_task *timeout_task)
 603 {
 604
 605         /*
 606          * Set flag to prevent timer from re-starting during drain:
 607          */
 608         TQ_LOCK(queue);
 609         KASSERT((timeout_task->f & DT_DRAIN_IN_PROGRESS) == 0,
 610             ("Drain already in progress"));
 611         timeout_task->f |= DT_DRAIN_IN_PROGRESS;
 612         TQ_UNLOCK(queue);
 613
 614         callout_drain(&timeout_task->c);
 615         taskqueue_drain(queue, &timeout_task->t);
 616
 617         /*
 618          * Clear flag to allow timer to re-start:
 619          */
 620         TQ_LOCK(queue);
 621         timeout_task->f &= ~DT_DRAIN_IN_PROGRESS;
 622         TQ_UNLOCK(queue);
 623 }
 624
 625 void
 626 taskqueue_quiesce(struct taskqueue *queue)
 627 {
 628         int ret;
 629
 630         TQ_LOCK(queue);
 631         do {
 632                 ret = taskqueue_drain_tq_queue(queue);
 633                 if (ret == 0)
 634                         ret = taskqueue_drain_tq_active(queue);
 635         } while (ret != 0);
 636         TQ_UNLOCK(queue);
 637 }
 638
 639 static void
 640 taskqueue_swi_enqueue(void *context)
 641 {
 642         swi_sched(taskqueue_ih, 0);
 643 }
 644
 645 static void
 646 taskqueue_swi_run(void *dummy)
 647 {
 648         taskqueue_run(taskqueue_swi);
 649 }
 650
 651 static void
 652 taskqueue_swi_giant_enqueue(void *context)
 653 {
 654         swi_sched(taskqueue_giant_ih, 0);
 655 }
 656
 657 static void
 658 taskqueue_swi_giant_run(void *dummy)
 659 {
 660         taskqueue_run(taskqueue_swi_giant);
 661 }
 662
 663 static int
 664 _taskqueue_start_threads(struct taskqueue **tqp, int count, int pri,
 665     cpuset_t *mask, struct proc *p, const char *name, va_list ap)
 666 {
 667         char ktname[MAXCOMLEN + 1];
 668         struct thread *td;
 669         struct taskqueue *tq;
 670         int i, error;
 671
 672         if (count <= 0)
 673                 return (EINVAL);
 674
 675         vsnprintf(ktname, sizeof(ktname), name, ap);
 676         tq = *tqp;
 677
 678         tq->tq_threads = malloc(sizeof(struct thread *) * count, M_TASKQUEUE,
 679             M_NOWAIT | M_ZERO);
 680         if (tq->tq_threads == NULL) {
 681                 printf("%s: no memory for %s threads\n", __func__, ktname);
 682                 return (ENOMEM);
 683         }
 684
 685         for (i = 0; i < count; i++) {
 686                 if (count == 1)
 687                         error = kthread_add(taskqueue_thread_loop, tqp, p,
 688                             &tq->tq_threads[i], RFSTOPPED, 0, "%s", ktname);
 689                 else
 690                         error = kthread_add(taskqueue_thread_loop, tqp, p,
 691                             &tq->tq_threads[i], RFSTOPPED, 0,
 692                             "%s_%d", ktname, i);
 693                 if (error) {
 694                         /* should be ok to continue, taskqueue_free will dtrt */
 695                         printf("%s: kthread_add(%s): error %d", __func__,
 696                             ktname, error);
 697                         tq->tq_threads[i] = NULL;               /* paranoid */
 698                 } else
 699                         tq->tq_tcount++;
 700         }
 701         if (tq->tq_tcount == 0) {
 702                 free(tq->tq_threads, M_TASKQUEUE);
 703                 tq->tq_threads = NULL;
 704                 return (ENOMEM);
 705         }
 706         for (i = 0; i < count; i++) {
 707                 if (tq->tq_threads[i] == NULL)
 708                         continue;
 709                 td = tq->tq_threads[i];
 710                 if (mask) {
 711                         error = cpuset_setthread(td->td_tid, mask);
 712                         /*
 713                          * Failing to pin is rarely an actual fatal error;
 714                          * it'll just affect performance.
 715                          */
 716                         if (error)
 717                                 printf("%s: curthread=%llu: can't pin; "
 718                                     "error=%d\n",
 719                                     __func__,
 720                                     (unsigned long long) td->td_tid,
 721                                     error);
 722                 }
 723                 thread_lock(td);
 724                 sched_prio(td, pri);
 725                 sched_add(td, SRQ_BORING);
 726         }
 727
 728         return (0);
 729 }
 730
 731 int
 732 taskqueue_start_threads(struct taskqueue **tqp, int count, int pri,
 733     const char *name, ...)
 734 {
 735         va_list ap;
 736         int error;
 737
 738         va_start(ap, name);
 739         error = _taskqueue_start_threads(tqp, count, pri, NULL, NULL, name, ap);
 740         va_end(ap);
 741         return (error);
 742 }
 743
 744 int
 745 taskqueue_start_threads_in_proc(struct taskqueue **tqp, int count, int pri,
 746     struct proc *proc, const char *name, ...)
 747 {
 748         va_list ap;
 749         int error;
 750
 751         va_start(ap, name);
 752         error = _taskqueue_start_threads(tqp, count, pri, NULL, proc, name, ap);
 753         va_end(ap);
 754         return (error);
 755 }
 756
 757 int
 758 taskqueue_start_threads_cpuset(struct taskqueue **tqp, int count, int pri,
 759     cpuset_t *mask, const char *name, ...)
 760 {
 761         va_list ap;
 762         int error;
 763
 764         va_start(ap, name);
 765         error = _taskqueue_start_threads(tqp, count, pri, mask, NULL, name, ap);
 766         va_end(ap);
 767         return (error);
 768 }
 769
 770 static inline void
 771 taskqueue_run_callback(struct taskqueue *tq,
 772     enum taskqueue_callback_type cb_type)
 773 {
 774         taskqueue_callback_fn tq_callback;
 775
 776         TQ_ASSERT_UNLOCKED(tq);
 777         tq_callback = tq->tq_callbacks[cb_type];
 778         if (tq_callback != NULL)
 779                 tq_callback(tq->tq_cb_contexts[cb_type]);
 780 }
 781
 782 void
 783 taskqueue_thread_loop(void *arg)
 784 {
 785         struct taskqueue **tqp, *tq;
 786
 787         tqp = arg;
 788         tq = *tqp;
 789         taskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_INIT);
 790         TQ_LOCK(tq);
 791         while ((tq->tq_flags & TQ_FLAGS_ACTIVE) != 0) {
 792                 /* XXX ? */
 793                 taskqueue_run_locked(tq);
 794                 /*
 795                  * Because taskqueue_run() can drop tq_mutex, we need to
 796                  * check if the TQ_FLAGS_ACTIVE flag wasn't removed in the
 797                  * meantime, which means we missed a wakeup.
 798                  */
 799                 if ((tq->tq_flags & TQ_FLAGS_ACTIVE) == 0)
 800                         break;
 801                 TQ_SLEEP(tq, tq, "-");
 802         }
 803         taskqueue_run_locked(tq);
 804         /*
 805          * This thread is on its way out, so just drop the lock temporarily
 806          * in order to call the shutdown callback.  This allows the callback
 807          * to look at the taskqueue, even just before it dies.
 808          */
 809         TQ_UNLOCK(tq);
 810         taskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_SHUTDOWN);
 811         TQ_LOCK(tq);
 812
 813         /* rendezvous with thread that asked us to terminate */
 814         tq->tq_tcount--;
 815         wakeup_one(tq->tq_threads);
 816         TQ_UNLOCK(tq);
 817         kthread_exit();
 818 }
 819
 820 void
 821 taskqueue_thread_enqueue(void *context)
 822 {
 823         struct taskqueue **tqp, *tq;
 824
 825         tqp = context;
 826         tq = *tqp;
 827         wakeup_any(tq);
 828 }
 829
 830 TASKQUEUE_DEFINE(swi, taskqueue_swi_enqueue, NULL,
 831                  swi_add(NULL, "task queue", taskqueue_swi_run, NULL, SWI_TQ,
 832                      INTR_MPSAFE, &taskqueue_ih));
 833
 834 TASKQUEUE_DEFINE(swi_giant, taskqueue_swi_giant_enqueue, NULL,
 835                  swi_add(NULL, "Giant taskq", taskqueue_swi_giant_run,
 836                      NULL, SWI_TQ_GIANT, 0, &taskqueue_giant_ih));
 837
 838 TASKQUEUE_DEFINE_THREAD(thread);
 839
 840 struct taskqueue *
 841 taskqueue_create_fast(const char *name, int mflags,
 842                  taskqueue_enqueue_fn enqueue, void *context)
 843 {
 844         return _taskqueue_create(name, mflags, enqueue, context,
 845                         MTX_SPIN, "fast_taskqueue");
 846 }
 847
 848 static void     *taskqueue_fast_ih;
 849
 850 static void
 851 taskqueue_fast_enqueue(void *context)
 852 {
 853         swi_sched(taskqueue_fast_ih, 0);
 854 }
 855
 856 static void
 857 taskqueue_fast_run(void *dummy)
 858 {
 859         taskqueue_run(taskqueue_fast);
 860 }
 861
 862 TASKQUEUE_FAST_DEFINE(fast, taskqueue_fast_enqueue, NULL,
 863         swi_add(NULL, "fast taskq", taskqueue_fast_run, NULL,
 864         SWI_TQ_FAST, INTR_MPSAFE, &taskqueue_fast_ih));
 865
 866 int
 867 taskqueue_member(struct taskqueue *queue, struct thread *td)
 868 {
 869         int i, j, ret = 0;
 870
 871         for (i = 0, j = 0; ; i++) {
 872                 if (queue->tq_threads[i] == NULL)
 873                         continue;
 874                 if (queue->tq_threads[i] == td) {
 875                         ret = 1;
 876                         break;
 877                 }
 878                 if (++j >= queue->tq_tcount)
 879                         break;
 880         }
 881         return (ret);
 882 }