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 /*
  23  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26 
  27 /*
  28  * Copyright (c) 2012 by Delphix. All rights reserved.
  29  */
  30 
  31 #include <sys/thread.h>
  32 #include <sys/proc.h>
  33 #include <sys/debug.h>
  34 #include <sys/cmn_err.h>
  35 #include <sys/systm.h>
  36 #include <sys/sobject.h>
  37 #include <sys/sleepq.h>
  38 #include <sys/cpuvar.h>
  39 #include <sys/condvar.h>
  40 #include <sys/condvar_impl.h>
  41 #include <sys/schedctl.h>
  42 #include <sys/procfs.h>
  43 #include <sys/sdt.h>
  44 #include <sys/callo.h>
  45 
  46 /*
  47  * CV_MAX_WAITERS is the maximum number of waiters we track; once
  48  * the number becomes higher than that, we look at the sleepq to
  49  * see whether there are *really* any waiters.
  50  */
  51 #define CV_MAX_WAITERS          1024            /* must be power of 2 */
  52 #define CV_WAITERS_MASK         (CV_MAX_WAITERS - 1)
  53 
  54 /*
  55  * Threads don't "own" condition variables.
  56  */
  57 /* ARGSUSED */
  58 static kthread_t *
  59 cv_owner(void *cvp)
  60 {
  61         return (NULL);
  62 }
  63 
  64 /*
  65  * Unsleep a thread that's blocked on a condition variable.
  66  */
  67 static void
  68 cv_unsleep(kthread_t *t)
  69 {
  70         condvar_impl_t *cvp = (condvar_impl_t *)t->t_wchan;
  71         sleepq_head_t *sqh = SQHASH(cvp);
  72 
  73         ASSERT(THREAD_LOCK_HELD(t));
  74 
  75         if (cvp == NULL)
  76                 panic("cv_unsleep: thread %p not on sleepq %p",
  77                     (void *)t, (void *)sqh);
  78         DTRACE_SCHED1(wakeup, kthread_t *, t);
  79         sleepq_unsleep(t);
  80         if (cvp->cv_waiters != CV_MAX_WAITERS)
  81                 cvp->cv_waiters--;
  82         disp_lock_exit_high(&sqh->sq_lock);
  83         CL_SETRUN(t);
  84 }
  85 
  86 /*
  87  * Change the priority of a thread that's blocked on a condition variable.
  88  */
  89 static void
  90 cv_change_pri(kthread_t *t, pri_t pri, pri_t *t_prip)
  91 {
  92         condvar_impl_t *cvp = (condvar_impl_t *)t->t_wchan;
  93         sleepq_t *sqp = t->t_sleepq;
  94 
  95         ASSERT(THREAD_LOCK_HELD(t));
  96         ASSERT(&SQHASH(cvp)->sq_queue == sqp);
  97 
  98         if (cvp == NULL)
  99                 panic("cv_change_pri: %p not on sleep queue", (void *)t);
 100         sleepq_dequeue(t);
 101         *t_prip = pri;
 102         sleepq_insert(sqp, t);
 103 }
 104 
 105 /*
 106  * The sobj_ops vector exports a set of functions needed when a thread
 107  * is asleep on a synchronization object of this type.
 108  */
 109 static sobj_ops_t cv_sobj_ops = {
 110         SOBJ_CV, cv_owner, cv_unsleep, cv_change_pri
 111 };
 112 
 113 /* ARGSUSED */
 114 void
 115 cv_init(kcondvar_t *cvp, char *name, kcv_type_t type, void *arg)
 116 {
 117         ((condvar_impl_t *)cvp)->cv_waiters = 0;
 118 }
 119 
 120 /*
 121  * cv_destroy is not currently needed, but is part of the DDI.
 122  * This is in case cv_init ever needs to allocate something for a cv.
 123  */
 124 /* ARGSUSED */
 125 void
 126 cv_destroy(kcondvar_t *cvp)
 127 {
 128         ASSERT((((condvar_impl_t *)cvp)->cv_waiters & CV_WAITERS_MASK) == 0);
 129 }
 130 
 131 /*
 132  * The cv_block() function blocks a thread on a condition variable
 133  * by putting it in a hashed sleep queue associated with the
 134  * synchronization object.
 135  *
 136  * Threads are taken off the hashed sleep queues via calls to
 137  * cv_signal(), cv_broadcast(), or cv_unsleep().
 138  */
 139 static void
 140 cv_block(condvar_impl_t *cvp)
 141 {
 142         kthread_t *t = curthread;
 143         klwp_t *lwp = ttolwp(t);
 144         sleepq_head_t *sqh;
 145 
 146         ASSERT(THREAD_LOCK_HELD(t));
 147         ASSERT(t != CPU->cpu_idle_thread);
 148         ASSERT(CPU_ON_INTR(CPU) == 0);
 149         ASSERT(t->t_wchan0 == NULL && t->t_wchan == NULL);
 150         ASSERT(t->t_state == TS_ONPROC);
 151 
 152         t->t_schedflag &= ~TS_SIGNALLED;
 153         CL_SLEEP(t);                    /* assign kernel priority */
 154         t->t_wchan = (caddr_t)cvp;
 155         t->t_sobj_ops = &cv_sobj_ops;
 156         DTRACE_SCHED(sleep);
 157 
 158         /*
 159          * The check for t_intr is to avoid doing the
 160          * account for an interrupt thread on the still-pinned
 161          * lwp's statistics.
 162          */
 163         if (lwp != NULL && t->t_intr == NULL) {
 164                 lwp->lwp_ru.nvcsw++;
 165                 (void) new_mstate(t, LMS_SLEEP);
 166         }
 167 
 168         sqh = SQHASH(cvp);
 169         disp_lock_enter_high(&sqh->sq_lock);
 170         if (cvp->cv_waiters < CV_MAX_WAITERS)
 171                 cvp->cv_waiters++;
 172         ASSERT(cvp->cv_waiters <= CV_MAX_WAITERS);
 173         THREAD_SLEEP(t, &sqh->sq_lock);
 174         sleepq_insert(&sqh->sq_queue, t);
 175         /*
 176          * THREAD_SLEEP() moves curthread->t_lockp to point to the
 177          * lock sqh->sq_lock. This lock is later released by the caller
 178          * when it calls thread_unlock() on curthread.
 179          */
 180 }
 181 
 182 #define cv_block_sig(t, cvp)    \
 183         { (t)->t_flag |= T_WAKEABLE; cv_block(cvp); }
 184 
 185 /*
 186  * Block on the indicated condition variable and release the
 187  * associated kmutex while blocked.
 188  */
 189 void
 190 cv_wait(kcondvar_t *cvp, kmutex_t *mp)
 191 {
 192         if (panicstr)
 193                 return;
 194         ASSERT(!quiesce_active);
 195 
 196         thread_lock(curthread);                 /* lock the thread */
 197         cv_block((condvar_impl_t *)cvp);
 198         thread_unlock_nopreempt(curthread);     /* unlock the waiters field */
 199         mutex_exit(mp);
 200         swtch();
 201         mutex_enter(mp);
 202 }
 203 
 204 static void
 205 cv_wakeup(void *arg)
 206 {
 207         kthread_t *t = arg;
 208 
 209         /*
 210          * This mutex is acquired and released in order to make sure that
 211          * the wakeup does not happen before the block itself happens.
 212          */
 213         mutex_enter(&t->t_wait_mutex);
 214         mutex_exit(&t->t_wait_mutex);
 215         setrun(t);
 216 }
 217 
 218 /*
 219  * Same as cv_wait except the thread will unblock at 'tim'
 220  * (an absolute time) if it hasn't already unblocked.
 221  *
 222  * Returns the amount of time left from the original 'tim' value
 223  * when it was unblocked.
 224  */
 225 clock_t
 226 cv_timedwait(kcondvar_t *cvp, kmutex_t *mp, clock_t tim)
 227 {
 228         hrtime_t hrtim;
 229         clock_t now = ddi_get_lbolt();
 230 
 231         if (tim <= now)
 232                 return (-1);
 233 
 234         hrtim = TICK_TO_NSEC(tim - now);
 235         return (cv_timedwait_hires(cvp, mp, hrtim, nsec_per_tick, 0));
 236 }
 237 
 238 /*
 239  * Same as cv_timedwait() except that the third argument is a relative
 240  * timeout value, as opposed to an absolute one. There is also a fourth
 241  * argument that specifies how accurately the timeout must be implemented.
 242  */
 243 clock_t
 244 cv_reltimedwait(kcondvar_t *cvp, kmutex_t *mp, clock_t delta, time_res_t res)
 245 {
 246         hrtime_t exp;
 247 
 248         ASSERT(TIME_RES_VALID(res));
 249 
 250         if (delta <= 0)
 251                 return (-1);
 252 
 253         if ((exp = TICK_TO_NSEC(delta)) < 0)
 254                 exp = CY_INFINITY;
 255 
 256         return (cv_timedwait_hires(cvp, mp, exp, time_res[res], 0));
 257 }
 258 
 259 clock_t
 260 cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t tim,
 261     hrtime_t res, int flag)
 262 {
 263         kthread_t *t = curthread;
 264         callout_id_t id;
 265         clock_t timeleft;
 266         hrtime_t limit;
 267         int signalled;
 268 
 269         if (panicstr)
 270                 return (-1);
 271         ASSERT(!quiesce_active);
 272 
 273         limit = (flag & CALLOUT_FLAG_ABSOLUTE) ? gethrtime() : 0;
 274         if (tim <= limit)
 275                 return (-1);
 276         mutex_enter(&t->t_wait_mutex);
 277         id = timeout_generic(CALLOUT_REALTIME, (void (*)(void *))cv_wakeup, t,
 278             tim, res, flag);
 279         thread_lock(t);         /* lock the thread */
 280         cv_block((condvar_impl_t *)cvp);
 281         thread_unlock_nopreempt(t);
 282         mutex_exit(&t->t_wait_mutex);
 283         mutex_exit(mp);
 284         swtch();
 285         signalled = (t->t_schedflag & TS_SIGNALLED);
 286         /*
 287          * Get the time left. untimeout() returns -1 if the timeout has
 288          * occured or the time remaining.  If the time remaining is zero,
 289          * the timeout has occured between when we were awoken and
 290          * we called untimeout.  We will treat this as if the timeout
 291          * has occured and set timeleft to -1.
 292          */
 293         timeleft = untimeout_default(id, 0);
 294         mutex_enter(mp);
 295         if (timeleft <= 0) {
 296                 timeleft = -1;
 297                 if (signalled)  /* avoid consuming the cv_signal() */
 298                         cv_signal(cvp);
 299         }
 300         return (timeleft);
 301 }
 302 
 303 int
 304 cv_wait_sig(kcondvar_t *cvp, kmutex_t *mp)
 305 {
 306         kthread_t *t = curthread;
 307         proc_t *p = ttoproc(t);
 308         klwp_t *lwp = ttolwp(t);
 309         int cancel_pending;
 310         int rval = 1;
 311         int signalled = 0;
 312 
 313         if (panicstr)
 314                 return (rval);
 315         ASSERT(!quiesce_active);
 316 
 317         /*
 318          * Threads in system processes don't process signals.  This is
 319          * true both for standard threads of system processes and for
 320          * interrupt threads which have borrowed their pinned thread's LWP.
 321          */
 322         if (lwp == NULL || (p->p_flag & SSYS)) {
 323                 cv_wait(cvp, mp);
 324                 return (rval);
 325         }
 326         ASSERT(t->t_intr == NULL);
 327 
 328         cancel_pending = schedctl_cancel_pending();
 329         lwp->lwp_asleep = 1;
 330         lwp->lwp_sysabort = 0;
 331         thread_lock(t);
 332         cv_block_sig(t, (condvar_impl_t *)cvp);
 333         thread_unlock_nopreempt(t);
 334         mutex_exit(mp);
 335         if (ISSIG(t, JUSTLOOKING) || MUSTRETURN(p, t) || cancel_pending)
 336                 setrun(t);
 337         /* ASSERT(no locks are held) */
 338         swtch();
 339         signalled = (t->t_schedflag & TS_SIGNALLED);
 340         t->t_flag &= ~T_WAKEABLE;
 341         mutex_enter(mp);
 342         if (ISSIG_PENDING(t, lwp, p)) {
 343                 mutex_exit(mp);
 344                 if (issig(FORREAL))
 345                         rval = 0;
 346                 mutex_enter(mp);
 347         }
 348         if (lwp->lwp_sysabort || MUSTRETURN(p, t))
 349                 rval = 0;
 350         if (rval != 0 && cancel_pending) {
 351                 schedctl_cancel_eintr();
 352                 rval = 0;
 353         }
 354         lwp->lwp_asleep = 0;
 355         lwp->lwp_sysabort = 0;
 356         if (rval == 0 && signalled)     /* avoid consuming the cv_signal() */
 357                 cv_signal(cvp);
 358         return (rval);
 359 }
 360 
 361 static clock_t
 362 cv_timedwait_sig_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t tim,
 363     hrtime_t res, int flag)
 364 {
 365         kthread_t *t = curthread;
 366         proc_t *p = ttoproc(t);
 367         klwp_t *lwp = ttolwp(t);
 368         int cancel_pending = 0;
 369         callout_id_t id;
 370         clock_t rval = 1;
 371         hrtime_t limit;
 372         int signalled = 0;
 373 
 374         if (panicstr)
 375                 return (rval);
 376         ASSERT(!quiesce_active);
 377 
 378         /*
 379          * Threads in system processes don't process signals.  This is
 380          * true both for standard threads of system processes and for
 381          * interrupt threads which have borrowed their pinned thread's LWP.
 382          */
 383         if (lwp == NULL || (p->p_flag & SSYS))
 384                 return (cv_timedwait_hires(cvp, mp, tim, res, flag));
 385         ASSERT(t->t_intr == NULL);
 386 
 387         /*
 388          * If tim is less than or equal to current hrtime, then the timeout
 389          * has already occured.  So just check to see if there is a signal
 390          * pending.  If so return 0 indicating that there is a signal pending.
 391          * Else return -1 indicating that the timeout occured. No need to
 392          * wait on anything.
 393          */
 394         limit = (flag & CALLOUT_FLAG_ABSOLUTE) ? gethrtime() : 0;
 395         if (tim <= limit) {
 396                 lwp->lwp_asleep = 1;
 397                 lwp->lwp_sysabort = 0;
 398                 rval = -1;
 399                 goto out;
 400         }
 401 
 402         /*
 403          * Set the timeout and wait.
 404          */
 405         cancel_pending = schedctl_cancel_pending();
 406         mutex_enter(&t->t_wait_mutex);
 407         id = timeout_generic(CALLOUT_REALTIME, (void (*)(void *))cv_wakeup, t,
 408             tim, res, flag);
 409         lwp->lwp_asleep = 1;
 410         lwp->lwp_sysabort = 0;
 411         thread_lock(t);
 412         cv_block_sig(t, (condvar_impl_t *)cvp);
 413         thread_unlock_nopreempt(t);
 414         mutex_exit(&t->t_wait_mutex);
 415         mutex_exit(mp);
 416         if (ISSIG(t, JUSTLOOKING) || MUSTRETURN(p, t) || cancel_pending)
 417                 setrun(t);
 418         /* ASSERT(no locks are held) */
 419         swtch();
 420         signalled = (t->t_schedflag & TS_SIGNALLED);
 421         t->t_flag &= ~T_WAKEABLE;
 422 
 423         /*
 424          * Untimeout the thread.  untimeout() returns -1 if the timeout has
 425          * occured or the time remaining.  If the time remaining is zero,
 426          * the timeout has occured between when we were awoken and
 427          * we called untimeout.  We will treat this as if the timeout
 428          * has occured and set rval to -1.
 429          */
 430         rval = untimeout_default(id, 0);
 431         mutex_enter(mp);
 432         if (rval <= 0)
 433                 rval = -1;
 434 
 435         /*
 436          * Check to see if a signal is pending.  If so, regardless of whether
 437          * or not we were awoken due to the signal, the signal is now pending
 438          * and a return of 0 has the highest priority.
 439          */
 440 out:
 441         if (ISSIG_PENDING(t, lwp, p)) {
 442                 mutex_exit(mp);
 443                 if (issig(FORREAL))
 444                         rval = 0;
 445                 mutex_enter(mp);
 446         }
 447         if (lwp->lwp_sysabort || MUSTRETURN(p, t))
 448                 rval = 0;
 449         if (rval != 0 && cancel_pending) {
 450                 schedctl_cancel_eintr();
 451                 rval = 0;
 452         }
 453         lwp->lwp_asleep = 0;
 454         lwp->lwp_sysabort = 0;
 455         if (rval <= 0 && signalled)  /* avoid consuming the cv_signal() */
 456                 cv_signal(cvp);
 457         return (rval);
 458 }
 459 
 460 /*
 461  * Returns:
 462  *      Function result in order of precedence:
 463  *               0 if a signal was received
 464  *              -1 if timeout occured
 465  *              >0 if awakened via cv_signal() or cv_broadcast().
 466  *                 (returns time remaining)
 467  *
 468  * cv_timedwait_sig() is now part of the DDI.
 469  *
 470  * This function is now just a wrapper for cv_timedwait_sig_hires().
 471  */
 472 clock_t
 473 cv_timedwait_sig(kcondvar_t *cvp, kmutex_t *mp, clock_t tim)
 474 {
 475         hrtime_t hrtim;
 476 
 477         hrtim = TICK_TO_NSEC(tim - ddi_get_lbolt());
 478         return (cv_timedwait_sig_hires(cvp, mp, hrtim, nsec_per_tick, 0));
 479 }
 480 
 481 /*
 482  * Wait until the specified time.
 483  * If tim == -1, waits without timeout using cv_wait_sig_swap().
 484  */
 485 int
 486 cv_timedwait_sig_hrtime(kcondvar_t *cvp, kmutex_t *mp, hrtime_t tim)
 487 {
 488         if (tim == -1) {
 489                 return (cv_wait_sig_swap(cvp, mp));
 490         } else {
 491                 return (cv_timedwait_sig_hires(cvp, mp, tim, 1,
 492                     CALLOUT_FLAG_ABSOLUTE | CALLOUT_FLAG_ROUNDUP));
 493         }
 494 }
 495 
 496 /*
 497  * Same as cv_timedwait_sig() except that the third argument is a relative
 498  * timeout value, as opposed to an absolute one. There is also a fourth
 499  * argument that specifies how accurately the timeout must be implemented.
 500  */
 501 clock_t
 502 cv_reltimedwait_sig(kcondvar_t *cvp, kmutex_t *mp, clock_t delta,
 503     time_res_t res)
 504 {
 505         hrtime_t exp = 0;
 506 
 507         ASSERT(TIME_RES_VALID(res));
 508 
 509         if (delta > 0) {
 510                 if ((exp = TICK_TO_NSEC(delta)) < 0)
 511                         exp = CY_INFINITY;
 512         }
 513 
 514         return (cv_timedwait_sig_hires(cvp, mp, exp, time_res[res], 0));
 515 }
 516 
 517 /*
 518  * Like cv_wait_sig_swap but allows the caller to indicate (with a
 519  * non-NULL sigret) that they will take care of signalling the cv
 520  * after wakeup, if necessary.  This is a vile hack that should only
 521  * be used when no other option is available; almost all callers
 522  * should just use cv_wait_sig_swap (which takes care of the cv_signal
 523  * stuff automatically) instead.
 524  */
 525 int
 526 cv_wait_sig_swap_core(kcondvar_t *cvp, kmutex_t *mp, int *sigret)
 527 {
 528         kthread_t *t = curthread;
 529         proc_t *p = ttoproc(t);
 530         klwp_t *lwp = ttolwp(t);
 531         int cancel_pending;
 532         int rval = 1;
 533         int signalled = 0;
 534 
 535         if (panicstr)
 536                 return (rval);
 537 
 538         /*
 539          * Threads in system processes don't process signals.  This is
 540          * true both for standard threads of system processes and for
 541          * interrupt threads which have borrowed their pinned thread's LWP.
 542          */
 543         if (lwp == NULL || (p->p_flag & SSYS)) {
 544                 cv_wait(cvp, mp);
 545                 return (rval);
 546         }
 547         ASSERT(t->t_intr == NULL);
 548 
 549         cancel_pending = schedctl_cancel_pending();
 550         lwp->lwp_asleep = 1;
 551         lwp->lwp_sysabort = 0;
 552         thread_lock(t);
 553         t->t_kpri_req = 0;   /* don't need kernel priority */
 554         cv_block_sig(t, (condvar_impl_t *)cvp);
 555         thread_unlock_nopreempt(t);
 556         mutex_exit(mp);
 557         if (ISSIG(t, JUSTLOOKING) || MUSTRETURN(p, t) || cancel_pending)
 558                 setrun(t);
 559         /* ASSERT(no locks are held) */
 560         swtch();
 561         signalled = (t->t_schedflag & TS_SIGNALLED);
 562         t->t_flag &= ~T_WAKEABLE;
 563         mutex_enter(mp);
 564         if (ISSIG_PENDING(t, lwp, p)) {
 565                 mutex_exit(mp);
 566                 if (issig(FORREAL))
 567                         rval = 0;
 568                 mutex_enter(mp);
 569         }
 570         if (lwp->lwp_sysabort || MUSTRETURN(p, t))
 571                 rval = 0;
 572         if (rval != 0 && cancel_pending) {
 573                 schedctl_cancel_eintr();
 574                 rval = 0;
 575         }
 576         lwp->lwp_asleep = 0;
 577         lwp->lwp_sysabort = 0;
 578         if (rval == 0) {
 579                 if (sigret != NULL)
 580                         *sigret = signalled;    /* just tell the caller */
 581                 else if (signalled)
 582                         cv_signal(cvp); /* avoid consuming the cv_signal() */
 583         }
 584         return (rval);
 585 }
 586 
 587 /*
 588  * Same as cv_wait_sig but the thread can be swapped out while waiting.
 589  * This should only be used when we know we aren't holding any locks.
 590  */
 591 int
 592 cv_wait_sig_swap(kcondvar_t *cvp, kmutex_t *mp)
 593 {
 594         return (cv_wait_sig_swap_core(cvp, mp, NULL));
 595 }
 596 
 597 void
 598 cv_signal(kcondvar_t *cvp)
 599 {
 600         condvar_impl_t *cp = (condvar_impl_t *)cvp;
 601 
 602         /* make sure the cv_waiters field looks sane */
 603         ASSERT(cp->cv_waiters <= CV_MAX_WAITERS);
 604         if (cp->cv_waiters > 0) {
 605                 sleepq_head_t *sqh = SQHASH(cp);
 606                 disp_lock_enter(&sqh->sq_lock);
 607                 ASSERT(CPU_ON_INTR(CPU) == 0);
 608                 if (cp->cv_waiters & CV_WAITERS_MASK) {
 609                         kthread_t *t;
 610                         cp->cv_waiters--;
 611                         t = sleepq_wakeone_chan(&sqh->sq_queue, cp);
 612                         /*
 613                          * If cv_waiters is non-zero (and less than
 614                          * CV_MAX_WAITERS) there should be a thread
 615                          * in the queue.
 616                          */
 617                         ASSERT(t != NULL);
 618                 } else if (sleepq_wakeone_chan(&sqh->sq_queue, cp) == NULL) {
 619                         cp->cv_waiters = 0;
 620                 }
 621                 disp_lock_exit(&sqh->sq_lock);
 622         }
 623 }
 624 
 625 void
 626 cv_broadcast(kcondvar_t *cvp)
 627 {
 628         condvar_impl_t *cp = (condvar_impl_t *)cvp;
 629 
 630         /* make sure the cv_waiters field looks sane */
 631         ASSERT(cp->cv_waiters <= CV_MAX_WAITERS);
 632         if (cp->cv_waiters > 0) {
 633                 sleepq_head_t *sqh = SQHASH(cp);
 634                 disp_lock_enter(&sqh->sq_lock);
 635                 ASSERT(CPU_ON_INTR(CPU) == 0);
 636                 sleepq_wakeall_chan(&sqh->sq_queue, cp);
 637                 cp->cv_waiters = 0;
 638                 disp_lock_exit(&sqh->sq_lock);
 639         }
 640 }
 641 
 642 /*
 643  * Same as cv_wait(), but wakes up (after wakeup_time milliseconds) to check
 644  * for requests to stop, like cv_wait_sig() but without dealing with signals.
 645  * This is a horrible kludge.  It is evil.  It is vile.  It is swill.
 646  * If your code has to call this function then your code is the same.
 647  */
 648 void
 649 cv_wait_stop(kcondvar_t *cvp, kmutex_t *mp, int wakeup_time)
 650 {
 651         kthread_t *t = curthread;
 652         klwp_t *lwp = ttolwp(t);
 653         proc_t *p = ttoproc(t);
 654         callout_id_t id;
 655         clock_t tim;
 656 
 657         if (panicstr)
 658                 return;
 659 
 660         /*
 661          * Threads in system processes don't process signals.  This is
 662          * true both for standard threads of system processes and for
 663          * interrupt threads which have borrowed their pinned thread's LWP.
 664          */
 665         if (lwp == NULL || (p->p_flag & SSYS)) {
 666                 cv_wait(cvp, mp);
 667                 return;
 668         }
 669         ASSERT(t->t_intr == NULL);
 670 
 671         /*
 672          * Wakeup in wakeup_time milliseconds, i.e., human time.
 673          */
 674         tim = ddi_get_lbolt() + MSEC_TO_TICK(wakeup_time);
 675         mutex_enter(&t->t_wait_mutex);
 676         id = realtime_timeout_default((void (*)(void *))cv_wakeup, t,
 677             tim - ddi_get_lbolt());
 678         thread_lock(t);                 /* lock the thread */
 679         cv_block((condvar_impl_t *)cvp);
 680         thread_unlock_nopreempt(t);
 681         mutex_exit(&t->t_wait_mutex);
 682         mutex_exit(mp);
 683         /* ASSERT(no locks are held); */
 684         swtch();
 685         (void) untimeout_default(id, 0);
 686 
 687         /*
 688          * Check for reasons to stop, if lwp_nostop is not true.
 689          * See issig_forreal() for explanations of the various stops.
 690          */
 691         mutex_enter(&p->p_lock);
 692         while (lwp->lwp_nostop == 0 && !(p->p_flag & SEXITLWPS)) {
 693                 /*
 694                  * Hold the lwp here for watchpoint manipulation.
 695                  */
 696                 if (t->t_proc_flag & TP_PAUSE) {
 697                         stop(PR_SUSPENDED, SUSPEND_PAUSE);
 698                         continue;
 699                 }
 700                 /*
 701                  * System checkpoint.
 702                  */
 703                 if (t->t_proc_flag & TP_CHKPT) {
 704                         stop(PR_CHECKPOINT, 0);
 705                         continue;
 706                 }
 707                 /*
 708                  * Honor fork1(), watchpoint activity (remapping a page),
 709                  * and lwp_suspend() requests.
 710                  */
 711                 if ((p->p_flag & (SHOLDFORK1|SHOLDWATCH)) ||
 712                     (t->t_proc_flag & TP_HOLDLWP)) {
 713                         stop(PR_SUSPENDED, SUSPEND_NORMAL);
 714                         continue;
 715                 }
 716                 /*
 717                  * Honor /proc requested stop.
 718                  */
 719                 if (t->t_proc_flag & TP_PRSTOP) {
 720                         stop(PR_REQUESTED, 0);
 721                 }
 722                 /*
 723                  * If some lwp in the process has already stopped
 724                  * showing PR_JOBCONTROL, stop in sympathy with it.
 725                  */
 726                 if (p->p_stopsig && t != p->p_agenttp) {
 727                         stop(PR_JOBCONTROL, p->p_stopsig);
 728                         continue;
 729                 }
 730                 break;
 731         }
 732         mutex_exit(&p->p_lock);
 733         mutex_enter(mp);
 734 }
 735 
 736 /*
 737  * Like cv_timedwait_sig(), but takes an absolute hires future time
 738  * rather than a future time in clock ticks.  Will not return showing
 739  * that a timeout occurred until the future time is passed.
 740  * If 'when' is a NULL pointer, no timeout will occur.
 741  * Returns:
 742  *      Function result in order of precedence:
 743  *               0 if a signal was received
 744  *              -1 if timeout occured
 745  *              >0 if awakened via cv_signal() or cv_broadcast()
 746  *                 or by a spurious wakeup.
 747  *                 (might return time remaining)
 748  * As a special test, if someone abruptly resets the system time
 749  * (but not through adjtime(2); drifting of the clock is allowed and
 750  * expected [see timespectohz_adj()]), then we force a return of -1
 751  * so the caller can return a premature timeout to the calling process
 752  * so it can reevaluate the situation in light of the new system time.
 753  * (The system clock has been reset if timecheck != timechanged.)
 754  *
 755  * Generally, cv_timedwait_sig_hrtime() should be used instead of this
 756  * routine.  It waits based on hrtime rather than wall-clock time and therefore
 757  * does not need to deal with the time changing.
 758  */
 759 int
 760 cv_waituntil_sig(kcondvar_t *cvp, kmutex_t *mp,
 761         timestruc_t *when, int timecheck)
 762 {
 763         timestruc_t now;
 764         timestruc_t delta;
 765         hrtime_t interval;
 766         int rval;
 767 
 768         if (when == NULL)
 769                 return (cv_wait_sig_swap(cvp, mp));
 770 
 771         gethrestime(&now);
 772         delta = *when;
 773         timespecsub(&delta, &now);
 774         if (delta.tv_sec < 0 || (delta.tv_sec == 0 && delta.tv_nsec == 0)) {
 775                 /*
 776                  * We have already reached the absolute future time.
 777                  * Call cv_timedwait_sig() just to check for signals.
 778                  * We will return immediately with either 0 or -1.
 779                  */
 780                 rval = cv_timedwait_sig_hires(cvp, mp, 0, 1, 0);
 781         } else {
 782                 if (timecheck == timechanged) {
 783                         /*
 784                          * Make sure that the interval is atleast one tick.
 785                          * This is to prevent a user from flooding the system
 786                          * with very small, high resolution timers.
 787                          */
 788                         interval = ts2hrt(&delta);
 789                         if (interval < nsec_per_tick)
 790                                 interval = nsec_per_tick;
 791                         rval = cv_timedwait_sig_hires(cvp, mp, interval, 1,
 792                             CALLOUT_FLAG_HRESTIME);
 793                 } else {
 794                         /*
 795                          * Someone reset the system time;
 796                          * just force an immediate timeout.
 797                          */
 798                         rval = -1;
 799                 }
 800                 if (rval == -1 && timecheck == timechanged) {
 801                         /*
 802                          * Even though cv_timedwait_sig() returned showing a
 803                          * timeout, the future time may not have passed yet.
 804                          * If not, change rval to indicate a normal wakeup.
 805                          */
 806                         gethrestime(&now);
 807                         delta = *when;
 808                         timespecsub(&delta, &now);
 809                         if (delta.tv_sec > 0 || (delta.tv_sec == 0 &&
 810                             delta.tv_nsec > 0))
 811                                 rval = 1;
 812                 }
 813         }
 814         return (rval);
 815 }