lib/libzpool/taskq.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25 /*
  26  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  27  * Copyright 2012 Garrett D'Amore <garrett@damore.org>.  All rights reserved.
  28  */
  29
  30 #include <sys/zfs_context.h>
  31
  32 int taskq_now;
  33 taskq_t *system_taskq;
  34
  35 #define TASKQ_ACTIVE    0x00010000
  36
  37 struct taskq {
  38         kmutex_t        tq_lock;
  39         krwlock_t       tq_threadlock;
  40         kcondvar_t      tq_dispatch_cv;
  41         kcondvar_t      tq_wait_cv;
  42         kthread_t       **tq_threadlist;
  43         int             tq_flags;
  44         int             tq_active;
  45         int             tq_nthreads;
  46         int             tq_nalloc;
  47         int             tq_minalloc;
  48         int             tq_maxalloc;
  49         kcondvar_t      tq_maxalloc_cv;
  50         int             tq_maxalloc_wait;
  51         taskq_ent_t     *tq_freelist;
  52         taskq_ent_t     tq_task;
  53 };
  54
  55 static taskq_ent_t *
  56 task_alloc(taskq_t *tq, int tqflags)
  57 {
  58         taskq_ent_t *t;
  59         int rv;
  60
  61 again:  if ((t = tq->tq_freelist) != NULL && tq->tq_nalloc >= tq->tq_minalloc) {
  62                 ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC));
  63                 tq->tq_freelist = t->tqent_next;
  64         } else {
  65                 if (tq->tq_nalloc >= tq->tq_maxalloc) {
  66                         if (!(tqflags & KM_SLEEP))
  67                                 return (NULL);
  68
  69                         /*
  70                          * We don't want to exceed tq_maxalloc, but we can't
  71                          * wait for other tasks to complete (and thus free up
  72                          * task structures) without risking deadlock with
  73                          * the caller.  So, we just delay for one second
  74                          * to throttle the allocation rate. If we have tasks
  75                          * complete before one second timeout expires then
  76                          * taskq_ent_free will signal us and we will
  77                          * immediately retry the allocation.
  78                          */
  79                         tq->tq_maxalloc_wait++;
  80                         rv = cv_timedwait(&tq->tq_maxalloc_cv,
  81                             &tq->tq_lock, ddi_get_lbolt() + hz);
  82                         tq->tq_maxalloc_wait--;
  83                         if (rv > 0)
  84                                 goto again;             /* signaled */
  85                 }
  86                 mutex_exit(&tq->tq_lock);
  87
  88                 t = kmem_alloc(sizeof (taskq_ent_t), tqflags);
  89
  90                 mutex_enter(&tq->tq_lock);
  91                 if (t != NULL) {
  92                         /* Make sure we start without any flags */
  93                         t->tqent_flags = 0;
  94                         tq->tq_nalloc++;
  95                 }
  96         }
  97         return (t);
  98 }
  99
 100 static void
 101 task_free(taskq_t *tq, taskq_ent_t *t)
 102 {
 103         if (tq->tq_nalloc <= tq->tq_minalloc) {
 104                 t->tqent_next = tq->tq_freelist;
 105                 tq->tq_freelist = t;
 106         } else {
 107                 tq->tq_nalloc--;
 108                 mutex_exit(&tq->tq_lock);
 109                 kmem_free(t, sizeof (taskq_ent_t));
 110                 mutex_enter(&tq->tq_lock);
 111         }
 112
 113         if (tq->tq_maxalloc_wait)
 114                 cv_signal(&tq->tq_maxalloc_cv);
 115 }
 116
 117 taskqid_t
 118 taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags)
 119 {
 120         taskq_ent_t *t;
 121
 122         if (taskq_now) {
 123                 func(arg);
 124                 return (1);
 125         }
 126
 127         mutex_enter(&tq->tq_lock);
 128         ASSERT(tq->tq_flags & TASKQ_ACTIVE);
 129         if ((t = task_alloc(tq, tqflags)) == NULL) {
 130                 mutex_exit(&tq->tq_lock);
 131                 return (0);
 132         }
 133         if (tqflags & TQ_FRONT) {
 134                 t->tqent_next = tq->tq_task.tqent_next;
 135                 t->tqent_prev = &tq->tq_task;
 136         } else {
 137                 t->tqent_next = &tq->tq_task;
 138                 t->tqent_prev = tq->tq_task.tqent_prev;
 139         }
 140         t->tqent_next->tqent_prev = t;
 141         t->tqent_prev->tqent_next = t;
 142         t->tqent_func = func;
 143         t->tqent_arg = arg;
 144         t->tqent_flags = 0;
 145         cv_signal(&tq->tq_dispatch_cv);
 146         mutex_exit(&tq->tq_lock);
 147         return (1);
 148 }
 149
 150 int
 151 taskq_empty_ent(taskq_ent_t *t)
 152 {
 153         return t->tqent_next == NULL;
 154 }
 155
 156 void
 157 taskq_init_ent(taskq_ent_t *t)
 158 {
 159         t->tqent_next = NULL;
 160         t->tqent_prev = NULL;
 161         t->tqent_func = NULL;
 162         t->tqent_arg = NULL;
 163         t->tqent_flags = 0;
 164 }
 165
 166 void
 167 taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags,
 168     taskq_ent_t *t)
 169 {
 170         ASSERT(func != NULL);
 171         ASSERT(!(tq->tq_flags & TASKQ_DYNAMIC));
 172
 173         /*
 174          * Mark it as a prealloc'd task.  This is important
 175          * to ensure that we don't free it later.
 176          */
 177         t->tqent_flags |= TQENT_FLAG_PREALLOC;
 178         /*
 179          * Enqueue the task to the underlying queue.
 180          */
 181         mutex_enter(&tq->tq_lock);
 182
 183         if (flags & TQ_FRONT) {
 184                 t->tqent_next = tq->tq_task.tqent_next;
 185                 t->tqent_prev = &tq->tq_task;
 186         } else {
 187                 t->tqent_next = &tq->tq_task;
 188                 t->tqent_prev = tq->tq_task.tqent_prev;
 189         }
 190         t->tqent_next->tqent_prev = t;
 191         t->tqent_prev->tqent_next = t;
 192         t->tqent_func = func;
 193         t->tqent_arg = arg;
 194         cv_signal(&tq->tq_dispatch_cv);
 195         mutex_exit(&tq->tq_lock);
 196 }
 197
 198 void
 199 taskq_wait(taskq_t *tq)
 200 {
 201         mutex_enter(&tq->tq_lock);
 202         while (tq->tq_task.tqent_next != &tq->tq_task || tq->tq_active != 0)
 203                 cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
 204         mutex_exit(&tq->tq_lock);
 205 }
 206
 207 static void
 208 taskq_thread(void *arg)
 209 {
 210         taskq_t *tq = arg;
 211         taskq_ent_t *t;
 212         boolean_t prealloc;
 213
 214         mutex_enter(&tq->tq_lock);
 215         while (tq->tq_flags & TASKQ_ACTIVE) {
 216                 if ((t = tq->tq_task.tqent_next) == &tq->tq_task) {
 217                         if (--tq->tq_active == 0)
 218                                 cv_broadcast(&tq->tq_wait_cv);
 219                         cv_wait(&tq->tq_dispatch_cv, &tq->tq_lock);
 220                         tq->tq_active++;
 221                         continue;
 222                 }
 223                 t->tqent_prev->tqent_next = t->tqent_next;
 224                 t->tqent_next->tqent_prev = t->tqent_prev;
 225                 t->tqent_next = NULL;
 226                 t->tqent_prev = NULL;
 227                 prealloc = t->tqent_flags & TQENT_FLAG_PREALLOC;
 228                 mutex_exit(&tq->tq_lock);
 229
 230                 rw_enter(&tq->tq_threadlock, RW_READER);
 231                 t->tqent_func(t->tqent_arg);
 232                 rw_exit(&tq->tq_threadlock);
 233
 234                 mutex_enter(&tq->tq_lock);
 235                 if (!prealloc)
 236                         task_free(tq, t);
 237         }
 238         tq->tq_nthreads--;
 239         cv_broadcast(&tq->tq_wait_cv);
 240         mutex_exit(&tq->tq_lock);
 241         thread_exit();
 242 }
 243
 244 /*ARGSUSED*/
 245 taskq_t *
 246 taskq_create(const char *name, int nthreads, pri_t pri,
 247         int minalloc, int maxalloc, uint_t flags)
 248 {
 249         taskq_t *tq = kmem_zalloc(sizeof (taskq_t), KM_SLEEP);
 250         int t;
 251
 252         if (flags & TASKQ_THREADS_CPU_PCT) {
 253                 int pct;
 254                 ASSERT3S(nthreads, >=, 0);
 255                 ASSERT3S(nthreads, <=, 100);
 256                 pct = MIN(nthreads, 100);
 257                 pct = MAX(pct, 0);
 258
 259                 nthreads = (sysconf(_SC_NPROCESSORS_ONLN) * pct) / 100;
 260                 nthreads = MAX(nthreads, 1);    /* need at least 1 thread */
 261         } else {
 262                 ASSERT3S(nthreads, >=, 1);
 263         }
 264
 265         rw_init(&tq->tq_threadlock, NULL, RW_DEFAULT, NULL);
 266         mutex_init(&tq->tq_lock, NULL, MUTEX_DEFAULT, NULL);
 267         cv_init(&tq->tq_dispatch_cv, NULL, CV_DEFAULT, NULL);
 268         cv_init(&tq->tq_wait_cv, NULL, CV_DEFAULT, NULL);
 269         cv_init(&tq->tq_maxalloc_cv, NULL, CV_DEFAULT, NULL);
 270         tq->tq_flags = flags | TASKQ_ACTIVE;
 271         tq->tq_active = nthreads;
 272         tq->tq_nthreads = nthreads;
 273         tq->tq_minalloc = minalloc;
 274         tq->tq_maxalloc = maxalloc;
 275         tq->tq_task.tqent_next = &tq->tq_task;
 276         tq->tq_task.tqent_prev = &tq->tq_task;
 277         tq->tq_threadlist = kmem_alloc(nthreads*sizeof(kthread_t *), KM_SLEEP);
 278
 279         if (flags & TASKQ_PREPOPULATE) {
 280                 mutex_enter(&tq->tq_lock);
 281                 while (minalloc-- > 0)
 282                         task_free(tq, task_alloc(tq, KM_SLEEP));
 283                 mutex_exit(&tq->tq_lock);
 284         }
 285
 286         for (t = 0; t < nthreads; t++)
 287                 VERIFY((tq->tq_threadlist[t] = thread_create(NULL, 0,
 288                     taskq_thread, tq, TS_RUN, NULL, 0, 0)) != NULL);
 289
 290         return (tq);
 291 }
 292
 293 void
 294 taskq_destroy(taskq_t *tq)
 295 {
 296         int nthreads = tq->tq_nthreads;
 297
 298         taskq_wait(tq);
 299
 300         mutex_enter(&tq->tq_lock);
 301
 302         tq->tq_flags &= ~TASKQ_ACTIVE;
 303         cv_broadcast(&tq->tq_dispatch_cv);
 304
 305         while (tq->tq_nthreads != 0)
 306                 cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
 307
 308         tq->tq_minalloc = 0;
 309         while (tq->tq_nalloc != 0) {
 310                 ASSERT(tq->tq_freelist != NULL);
 311                 task_free(tq, task_alloc(tq, KM_SLEEP));
 312         }
 313
 314         mutex_exit(&tq->tq_lock);
 315
 316         kmem_free(tq->tq_threadlist, nthreads * sizeof (kthread_t *));
 317
 318         rw_destroy(&tq->tq_threadlock);
 319         mutex_destroy(&tq->tq_lock);
 320         cv_destroy(&tq->tq_dispatch_cv);
 321         cv_destroy(&tq->tq_wait_cv);
 322         cv_destroy(&tq->tq_maxalloc_cv);
 323
 324         kmem_free(tq, sizeof (taskq_t));
 325 }
 326
 327 int
 328 taskq_member(taskq_t *tq, kthread_t *t)
 329 {
 330         int i;
 331
 332         if (taskq_now)
 333                 return (1);
 334
 335         for (i = 0; i < tq->tq_nthreads; i++)
 336                 if (tq->tq_threadlist[i] == t)
 337                         return (1);
 338
 339         return (0);
 340 }
 341
 342 void
 343 system_taskq_init(void)
 344 {
 345         system_taskq = taskq_create("system_taskq", 64, minclsyspri, 4, 512,
 346             TASKQ_DYNAMIC | TASKQ_PREPOPULATE);
 347 }
 348
 349 void
 350 system_taskq_fini(void)
 351 {
 352         taskq_destroy(system_taskq);
 353         system_taskq = NULL; /* defensive */
 354 }