1 /* 2 * Copyright (c) 2009 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Copyright (c) 2012 Spectra Logic Corporation. 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 AUTHORS 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 AUTHORS 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/ck.h> 34 #include <sys/epoch.h> 35 #include <sys/kernel.h> 36 #include <sys/kmem.h> 37 #include <sys/lock.h> 38 #include <sys/mutex.h> 39 #include <sys/queue.h> 40 #include <sys/taskq.h> 41 #include <sys/taskqueue.h> 42 #include <sys/zfs_context.h> 43 44 #if defined(__i386__) || defined(__amd64__) || defined(__aarch64__) 45 #include <machine/pcb.h> 46 #endif 47 48 #include <vm/uma.h> 49 50 #if __FreeBSD_version < 1201522 51 #define taskqueue_start_threads_in_proc(tqp, count, pri, proc, name, ...) \ 52 taskqueue_start_threads(tqp, count, pri, name, __VA_ARGS__) 53 #endif 54 55 static uint_t taskq_tsd; 56 static uma_zone_t taskq_zone; 57 58 /* 59 * Global system-wide dynamic task queue available for all consumers. This 60 * taskq is not intended for long-running tasks; instead, a dedicated taskq 61 * should be created. 62 */ 63 taskq_t *system_taskq = NULL; 64 taskq_t *system_delay_taskq = NULL; 65 taskq_t *dynamic_taskq = NULL; 66 67 proc_t *system_proc; 68 69 extern int uma_align_cache; 70 71 static MALLOC_DEFINE(M_TASKQ, "taskq", "taskq structures"); 72 73 static CK_LIST_HEAD(tqenthashhead, taskq_ent) *tqenthashtbl; 74 static unsigned long tqenthash; 75 static unsigned long tqenthashlock; 76 static struct sx *tqenthashtbl_lock; 77 78 static taskqid_t tqidnext; 79 80 #define TQIDHASH(tqid) (&tqenthashtbl[(tqid) & tqenthash]) 81 #define TQIDHASHLOCK(tqid) (&tqenthashtbl_lock[((tqid) & tqenthashlock)]) 82 83 #define TIMEOUT_TASK 1 84 #define NORMAL_TASK 2 85 86 static void 87 system_taskq_init(void *arg) 88 { 89 int i; 90 91 tsd_create(&taskq_tsd, NULL); 92 tqenthashtbl = hashinit(mp_ncpus * 8, M_TASKQ, &tqenthash); 93 tqenthashlock = (tqenthash + 1) / 8; 94 if (tqenthashlock > 0) 95 tqenthashlock--; 96 tqenthashtbl_lock = 97 malloc(sizeof (*tqenthashtbl_lock) * (tqenthashlock + 1), 98 M_TASKQ, M_WAITOK | M_ZERO); 99 for (i = 0; i < tqenthashlock + 1; i++) 100 sx_init_flags(&tqenthashtbl_lock[i], "tqenthash", SX_DUPOK); 101 taskq_zone = uma_zcreate("taskq_zone", sizeof (taskq_ent_t), 102 NULL, NULL, NULL, NULL, 103 UMA_ALIGN_CACHE, 0); 104 system_taskq = taskq_create("system_taskq", mp_ncpus, minclsyspri, 105 0, 0, 0); 106 system_delay_taskq = taskq_create("system_delay_taskq", mp_ncpus, 107 minclsyspri, 0, 0, 0); 108 } 109 SYSINIT(system_taskq_init, SI_SUB_CONFIGURE, SI_ORDER_ANY, system_taskq_init, 110 NULL); 111 112 static void 113 system_taskq_fini(void *arg) 114 { 115 int i; 116 117 taskq_destroy(system_delay_taskq); 118 taskq_destroy(system_taskq); 119 uma_zdestroy(taskq_zone); 120 tsd_destroy(&taskq_tsd); 121 for (i = 0; i < tqenthashlock + 1; i++) 122 sx_destroy(&tqenthashtbl_lock[i]); 123 for (i = 0; i < tqenthash + 1; i++) 124 VERIFY(CK_LIST_EMPTY(&tqenthashtbl[i])); 125 free(tqenthashtbl_lock, M_TASKQ); 126 free(tqenthashtbl, M_TASKQ); 127 } 128 SYSUNINIT(system_taskq_fini, SI_SUB_CONFIGURE, SI_ORDER_ANY, system_taskq_fini, 129 NULL); 130 131 #ifdef __LP64__ 132 static taskqid_t 133 __taskq_genid(void) 134 { 135 taskqid_t tqid; 136 137 /* 138 * Assume a 64-bit counter will not wrap in practice. 139 */ 140 tqid = atomic_add_64_nv(&tqidnext, 1); 141 VERIFY(tqid); 142 return (tqid); 143 } 144 #else 145 static taskqid_t 146 __taskq_genid(void) 147 { 148 taskqid_t tqid; 149 150 for (;;) { 151 tqid = atomic_add_32_nv(&tqidnext, 1); 152 if (__predict_true(tqid != 0)) 153 break; 154 } 155 VERIFY(tqid); 156 return (tqid); 157 } 158 #endif 159 160 static taskq_ent_t * 161 taskq_lookup(taskqid_t tqid) 162 { 163 taskq_ent_t *ent = NULL; 164 165 sx_xlock(TQIDHASHLOCK(tqid)); 166 CK_LIST_FOREACH(ent, TQIDHASH(tqid), tqent_hash) { 167 if (ent->tqent_id == tqid) 168 break; 169 } 170 if (ent != NULL) 171 refcount_acquire(&ent->tqent_rc); 172 sx_xunlock(TQIDHASHLOCK(tqid)); 173 return (ent); 174 } 175 176 static taskqid_t 177 taskq_insert(taskq_ent_t *ent) 178 { 179 taskqid_t tqid; 180 181 tqid = __taskq_genid(); 182 ent->tqent_id = tqid; 183 ent->tqent_registered = B_TRUE; 184 sx_xlock(TQIDHASHLOCK(tqid)); 185 CK_LIST_INSERT_HEAD(TQIDHASH(tqid), ent, tqent_hash); 186 sx_xunlock(TQIDHASHLOCK(tqid)); 187 return (tqid); 188 } 189 190 static void 191 taskq_remove(taskq_ent_t *ent) 192 { 193 taskqid_t tqid = ent->tqent_id; 194 195 if (!ent->tqent_registered) 196 return; 197 198 sx_xlock(TQIDHASHLOCK(tqid)); 199 CK_LIST_REMOVE(ent, tqent_hash); 200 sx_xunlock(TQIDHASHLOCK(tqid)); 201 ent->tqent_registered = B_FALSE; 202 } 203 204 static void 205 taskq_tsd_set(void *context) 206 { 207 taskq_t *tq = context; 208 209 #if defined(__amd64__) || defined(__aarch64__) 210 if (context != NULL && tsd_get(taskq_tsd) == NULL) 211 fpu_kern_thread(FPU_KERN_NORMAL); 212 #endif 213 tsd_set(taskq_tsd, tq); 214 } 215 216 static taskq_t * 217 taskq_create_impl(const char *name, int nthreads, pri_t pri, 218 proc_t *proc __maybe_unused, uint_t flags) 219 { 220 taskq_t *tq; 221 222 if ((flags & TASKQ_THREADS_CPU_PCT) != 0) 223 nthreads = MAX((mp_ncpus * nthreads) / 100, 1); 224 225 tq = kmem_alloc(sizeof (*tq), KM_SLEEP); 226 tq->tq_queue = taskqueue_create(name, M_WAITOK, 227 taskqueue_thread_enqueue, &tq->tq_queue); 228 taskqueue_set_callback(tq->tq_queue, TASKQUEUE_CALLBACK_TYPE_INIT, 229 taskq_tsd_set, tq); 230 taskqueue_set_callback(tq->tq_queue, TASKQUEUE_CALLBACK_TYPE_SHUTDOWN, 231 taskq_tsd_set, NULL); 232 (void) taskqueue_start_threads_in_proc(&tq->tq_queue, nthreads, pri, 233 proc, "%s", name); 234 235 return ((taskq_t *)tq); 236 } 237 238 taskq_t * 239 taskq_create(const char *name, int nthreads, pri_t pri, int minalloc __unused, 240 int maxalloc __unused, uint_t flags) 241 { 242 return (taskq_create_impl(name, nthreads, pri, system_proc, flags)); 243 } 244 245 taskq_t * 246 taskq_create_proc(const char *name, int nthreads, pri_t pri, 247 int minalloc __unused, int maxalloc __unused, proc_t *proc, uint_t flags) 248 { 249 return (taskq_create_impl(name, nthreads, pri, proc, flags)); 250 } 251 252 void 253 taskq_destroy(taskq_t *tq) 254 { 255 256 taskqueue_free(tq->tq_queue); 257 kmem_free(tq, sizeof (*tq)); 258 } 259 260 int 261 taskq_member(taskq_t *tq, kthread_t *thread) 262 { 263 264 return (taskqueue_member(tq->tq_queue, thread)); 265 } 266 267 taskq_t * 268 taskq_of_curthread(void) 269 { 270 return (tsd_get(taskq_tsd)); 271 } 272 273 static void 274 taskq_free(taskq_ent_t *task) 275 { 276 taskq_remove(task); 277 if (refcount_release(&task->tqent_rc)) 278 uma_zfree(taskq_zone, task); 279 } 280 281 int 282 taskq_cancel_id(taskq_t *tq, taskqid_t tid) 283 { 284 uint32_t pend; 285 int rc; 286 taskq_ent_t *ent; 287 288 if (tid == 0) 289 return (0); 290 291 if ((ent = taskq_lookup(tid)) == NULL) 292 return (0); 293 294 ent->tqent_cancelled = B_TRUE; 295 if (ent->tqent_type == TIMEOUT_TASK) { 296 rc = taskqueue_cancel_timeout(tq->tq_queue, 297 &ent->tqent_timeout_task, &pend); 298 } else 299 rc = taskqueue_cancel(tq->tq_queue, &ent->tqent_task, &pend); 300 if (rc == EBUSY) { 301 taskqueue_drain(tq->tq_queue, &ent->tqent_task); 302 } else if (pend) { 303 /* 304 * Tasks normally free themselves when run, but here the task 305 * was cancelled so it did not free itself. 306 */ 307 taskq_free(ent); 308 } 309 /* Free the extra reference we added with taskq_lookup. */ 310 taskq_free(ent); 311 return (rc); 312 } 313 314 static void 315 taskq_run(void *arg, int pending __unused) 316 { 317 taskq_ent_t *task = arg; 318 319 if (!task->tqent_cancelled) 320 task->tqent_func(task->tqent_arg); 321 taskq_free(task); 322 } 323 324 taskqid_t 325 taskq_dispatch_delay(taskq_t *tq, task_func_t func, void *arg, 326 uint_t flags, clock_t expire_time) 327 { 328 taskq_ent_t *task; 329 taskqid_t tqid; 330 clock_t timo; 331 int mflag; 332 333 timo = expire_time - ddi_get_lbolt(); 334 if (timo <= 0) 335 return (taskq_dispatch(tq, func, arg, flags)); 336 337 if ((flags & (TQ_SLEEP | TQ_NOQUEUE)) == TQ_SLEEP) 338 mflag = M_WAITOK; 339 else 340 mflag = M_NOWAIT; 341 342 task = uma_zalloc(taskq_zone, mflag); 343 if (task == NULL) 344 return (0); 345 task->tqent_func = func; 346 task->tqent_arg = arg; 347 task->tqent_type = TIMEOUT_TASK; 348 task->tqent_cancelled = B_FALSE; 349 refcount_init(&task->tqent_rc, 1); 350 tqid = taskq_insert(task); 351 TIMEOUT_TASK_INIT(tq->tq_queue, &task->tqent_timeout_task, 0, 352 taskq_run, task); 353 354 taskqueue_enqueue_timeout(tq->tq_queue, &task->tqent_timeout_task, 355 timo); 356 return (tqid); 357 } 358 359 taskqid_t 360 taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t flags) 361 { 362 taskq_ent_t *task; 363 int mflag, prio; 364 taskqid_t tqid; 365 366 if ((flags & (TQ_SLEEP | TQ_NOQUEUE)) == TQ_SLEEP) 367 mflag = M_WAITOK; 368 else 369 mflag = M_NOWAIT; 370 /* 371 * If TQ_FRONT is given, we want higher priority for this task, so it 372 * can go at the front of the queue. 373 */ 374 prio = !!(flags & TQ_FRONT); 375 376 task = uma_zalloc(taskq_zone, mflag); 377 if (task == NULL) 378 return (0); 379 refcount_init(&task->tqent_rc, 1); 380 task->tqent_func = func; 381 task->tqent_arg = arg; 382 task->tqent_cancelled = B_FALSE; 383 task->tqent_type = NORMAL_TASK; 384 tqid = taskq_insert(task); 385 TASK_INIT(&task->tqent_task, prio, taskq_run, task); 386 taskqueue_enqueue(tq->tq_queue, &task->tqent_task); 387 return (tqid); 388 } 389 390 static void 391 taskq_run_ent(void *arg, int pending __unused) 392 { 393 taskq_ent_t *task = arg; 394 395 task->tqent_func(task->tqent_arg); 396 } 397 398 void 399 taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint32_t flags, 400 taskq_ent_t *task) 401 { 402 int prio; 403 404 /* 405 * If TQ_FRONT is given, we want higher priority for this task, so it 406 * can go at the front of the queue. 407 */ 408 prio = !!(flags & TQ_FRONT); 409 task->tqent_cancelled = B_FALSE; 410 task->tqent_registered = B_FALSE; 411 task->tqent_id = 0; 412 task->tqent_func = func; 413 task->tqent_arg = arg; 414 415 TASK_INIT(&task->tqent_task, prio, taskq_run_ent, task); 416 taskqueue_enqueue(tq->tq_queue, &task->tqent_task); 417 } 418 419 void 420 taskq_wait(taskq_t *tq) 421 { 422 taskqueue_quiesce(tq->tq_queue); 423 } 424 425 void 426 taskq_wait_id(taskq_t *tq, taskqid_t tid) 427 { 428 taskq_ent_t *ent; 429 430 if (tid == 0) 431 return; 432 if ((ent = taskq_lookup(tid)) == NULL) 433 return; 434 435 taskqueue_drain(tq->tq_queue, &ent->tqent_task); 436 taskq_free(ent); 437 } 438 439 void 440 taskq_wait_outstanding(taskq_t *tq, taskqid_t id __unused) 441 { 442 taskqueue_drain_all(tq->tq_queue); 443 } 444 445 int 446 taskq_empty_ent(taskq_ent_t *t) 447 { 448 return (t->tqent_task.ta_pending == 0); 449 } 450