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 #include <sys/zfs_context.h> 27 28 int taskq_now; 29 taskq_t *system_taskq; 30 31 typedef struct task { 32 struct task *task_next; 33 struct task *task_prev; 34 task_func_t *task_func; 35 void *task_arg; 36 } task_t; 37 38 #define TASKQ_ACTIVE 0x00010000 39 40 struct taskq { 41 kmutex_t tq_lock; 42 krwlock_t tq_threadlock; 43 kcondvar_t tq_dispatch_cv; 44 kcondvar_t tq_wait_cv; 45 thread_t *tq_threadlist; 46 int tq_flags; 47 int tq_active; 48 int tq_nthreads; 49 int tq_nalloc; 50 int tq_minalloc; 51 int tq_maxalloc; 52 kcondvar_t tq_maxalloc_cv; 53 int tq_maxalloc_wait; 54 task_t *tq_freelist; 55 task_t tq_task; 56 }; 57 58 static task_t * 59 task_alloc(taskq_t *tq, int tqflags) 60 { 61 task_t *t; 62 int rv; 63 64 again: if ((t = tq->tq_freelist) != NULL && tq->tq_nalloc >= tq->tq_minalloc) { 65 tq->tq_freelist = t->task_next; 66 } else { 67 if (tq->tq_nalloc >= tq->tq_maxalloc) { 68 if (!(tqflags & KM_SLEEP)) 69 return (NULL); 70 71 /* 72 * We don't want to exceed tq_maxalloc, but we can't 73 * wait for other tasks to complete (and thus free up 74 * task structures) without risking deadlock with 75 * the caller. So, we just delay for one second 76 * to throttle the allocation rate. If we have tasks 77 * complete before one second timeout expires then 78 * taskq_ent_free will signal us and we will 79 * immediately retry the allocation. 80 */ 81 tq->tq_maxalloc_wait++; 82 rv = cv_timedwait(&tq->tq_maxalloc_cv, 83 &tq->tq_lock, ddi_get_lbolt() + hz); 84 tq->tq_maxalloc_wait--; 85 if (rv > 0) 86 goto again; /* signaled */ 87 } 88 mutex_exit(&tq->tq_lock); 89 90 t = kmem_alloc(sizeof (task_t), tqflags); 91 92 mutex_enter(&tq->tq_lock); 93 if (t != NULL) 94 tq->tq_nalloc++; 95 } 96 return (t); 97 } 98 99 static void 100 task_free(taskq_t *tq, task_t *t) 101 { 102 if (tq->tq_nalloc <= tq->tq_minalloc) { 103 t->task_next = tq->tq_freelist; 104 tq->tq_freelist = t; 105 } else { 106 tq->tq_nalloc--; 107 mutex_exit(&tq->tq_lock); 108 kmem_free(t, sizeof (task_t)); 109 mutex_enter(&tq->tq_lock); 110 } 111 112 if (tq->tq_maxalloc_wait) 113 cv_signal(&tq->tq_maxalloc_cv); 114 } 115 116 taskqid_t 117 taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags) 118 { 119 task_t *t; 120 121 if (taskq_now) { 122 func(arg); 123 return (1); 124 } 125 126 mutex_enter(&tq->tq_lock); 127 ASSERT(tq->tq_flags & TASKQ_ACTIVE); 128 if ((t = task_alloc(tq, tqflags)) == NULL) { 129 mutex_exit(&tq->tq_lock); 130 return (0); 131 } 132 if (tqflags & TQ_FRONT) { 133 t->task_next = tq->tq_task.task_next; 134 t->task_prev = &tq->tq_task; 135 } else { 136 t->task_next = &tq->tq_task; 137 t->task_prev = tq->tq_task.task_prev; 138 } 139 t->task_next->task_prev = t; 140 t->task_prev->task_next = t; 141 t->task_func = func; 142 t->task_arg = arg; 143 cv_signal(&tq->tq_dispatch_cv); 144 mutex_exit(&tq->tq_lock); 145 return (1); 146 } 147 148 void 149 taskq_wait(taskq_t *tq) 150 { 151 mutex_enter(&tq->tq_lock); 152 while (tq->tq_task.task_next != &tq->tq_task || tq->tq_active != 0) 153 cv_wait(&tq->tq_wait_cv, &tq->tq_lock); 154 mutex_exit(&tq->tq_lock); 155 } 156 157 static void * 158 taskq_thread(void *arg) 159 { 160 taskq_t *tq = arg; 161 task_t *t; 162 163 mutex_enter(&tq->tq_lock); 164 while (tq->tq_flags & TASKQ_ACTIVE) { 165 if ((t = tq->tq_task.task_next) == &tq->tq_task) { 166 if (--tq->tq_active == 0) 167 cv_broadcast(&tq->tq_wait_cv); 168 cv_wait(&tq->tq_dispatch_cv, &tq->tq_lock); 169 tq->tq_active++; 170 continue; 171 } 172 t->task_prev->task_next = t->task_next; 173 t->task_next->task_prev = t->task_prev; 174 mutex_exit(&tq->tq_lock); 175 176 rw_enter(&tq->tq_threadlock, RW_READER); 177 t->task_func(t->task_arg); 178 rw_exit(&tq->tq_threadlock); 179 180 mutex_enter(&tq->tq_lock); 181 task_free(tq, t); 182 } 183 tq->tq_nthreads--; 184 cv_broadcast(&tq->tq_wait_cv); 185 mutex_exit(&tq->tq_lock); 186 return (NULL); 187 } 188 189 /*ARGSUSED*/ 190 taskq_t * 191 taskq_create(const char *name, int nthreads, pri_t pri, 192 int minalloc, int maxalloc, uint_t flags) 193 { 194 taskq_t *tq = kmem_zalloc(sizeof (taskq_t), KM_SLEEP); 195 int t; 196 197 if (flags & TASKQ_THREADS_CPU_PCT) { 198 int pct; 199 ASSERT3S(nthreads, >=, 0); 200 ASSERT3S(nthreads, <=, 100); 201 pct = MIN(nthreads, 100); 202 pct = MAX(pct, 0); 203 204 nthreads = (sysconf(_SC_NPROCESSORS_ONLN) * pct) / 100; 205 nthreads = MAX(nthreads, 1); /* need at least 1 thread */ 206 } else { 207 ASSERT3S(nthreads, >=, 1); 208 } 209 210 rw_init(&tq->tq_threadlock, NULL, RW_DEFAULT, NULL); 211 mutex_init(&tq->tq_lock, NULL, MUTEX_DEFAULT, NULL); 212 cv_init(&tq->tq_dispatch_cv, NULL, CV_DEFAULT, NULL); 213 cv_init(&tq->tq_wait_cv, NULL, CV_DEFAULT, NULL); 214 cv_init(&tq->tq_maxalloc_cv, NULL, CV_DEFAULT, NULL); 215 tq->tq_flags = flags | TASKQ_ACTIVE; 216 tq->tq_active = nthreads; 217 tq->tq_nthreads = nthreads; 218 tq->tq_minalloc = minalloc; 219 tq->tq_maxalloc = maxalloc; 220 tq->tq_task.task_next = &tq->tq_task; 221 tq->tq_task.task_prev = &tq->tq_task; 222 tq->tq_threadlist = kmem_alloc(nthreads * sizeof (thread_t), KM_SLEEP); 223 224 if (flags & TASKQ_PREPOPULATE) { 225 mutex_enter(&tq->tq_lock); 226 while (minalloc-- > 0) 227 task_free(tq, task_alloc(tq, KM_SLEEP)); 228 mutex_exit(&tq->tq_lock); 229 } 230 231 for (t = 0; t < nthreads; t++) 232 (void) thr_create(0, 0, taskq_thread, 233 tq, THR_BOUND, &tq->tq_threadlist[t]); 234 235 return (tq); 236 } 237 238 void 239 taskq_destroy(taskq_t *tq) 240 { 241 int t; 242 int nthreads = tq->tq_nthreads; 243 244 taskq_wait(tq); 245 246 mutex_enter(&tq->tq_lock); 247 248 tq->tq_flags &= ~TASKQ_ACTIVE; 249 cv_broadcast(&tq->tq_dispatch_cv); 250 251 while (tq->tq_nthreads != 0) 252 cv_wait(&tq->tq_wait_cv, &tq->tq_lock); 253 254 tq->tq_minalloc = 0; 255 while (tq->tq_nalloc != 0) { 256 ASSERT(tq->tq_freelist != NULL); 257 task_free(tq, task_alloc(tq, KM_SLEEP)); 258 } 259 260 mutex_exit(&tq->tq_lock); 261 262 for (t = 0; t < nthreads; t++) 263 (void) thr_join(tq->tq_threadlist[t], NULL, NULL); 264 265 kmem_free(tq->tq_threadlist, nthreads * sizeof (thread_t)); 266 267 rw_destroy(&tq->tq_threadlock); 268 mutex_destroy(&tq->tq_lock); 269 cv_destroy(&tq->tq_dispatch_cv); 270 cv_destroy(&tq->tq_wait_cv); 271 cv_destroy(&tq->tq_maxalloc_cv); 272 273 kmem_free(tq, sizeof (taskq_t)); 274 } 275 276 int 277 taskq_member(taskq_t *tq, void *t) 278 { 279 int i; 280 281 if (taskq_now) 282 return (1); 283 284 for (i = 0; i < tq->tq_nthreads; i++) 285 if (tq->tq_threadlist[i] == (thread_t)(uintptr_t)t) 286 return (1); 287 288 return (0); 289 } 290 291 void 292 system_taskq_init(void) 293 { 294 system_taskq = taskq_create("system_taskq", 64, minclsyspri, 4, 512, 295 TASKQ_DYNAMIC | TASKQ_PREPOPULATE); 296 } 297 298 void 299 system_taskq_fini(void) 300 { 301 taskq_destroy(system_taskq); 302 system_taskq = NULL; /* defensive */ 303 } 304