The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/subr_sleepqueue.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2004 John Baldwin <jhb@FreeBSD.org>
    3  * All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice, this list of conditions and the following disclaimer.
   10  * 2. Redistributions in binary form must reproduce the above copyright
   11  *    notice, this list of conditions and the following disclaimer in the
   12  *    documentation and/or other materials provided with the distribution.
   13  * 3. Neither the name of the author nor the names of any co-contributors
   14  *    may be used to endorse or promote products derived from this software
   15  *    without specific prior written permission.
   16  *
   17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   27  * SUCH DAMAGE.
   28  */
   29 
   30 /*
   31  * Implementation of sleep queues used to hold queue of threads blocked on
   32  * a wait channel.  Sleep queues different from turnstiles in that wait
   33  * channels are not owned by anyone, so there is no priority propagation.
   34  * Sleep queues can also provide a timeout and can also be interrupted by
   35  * signals.  That said, there are several similarities between the turnstile
   36  * and sleep queue implementations.  (Note: turnstiles were implemented
   37  * first.)  For example, both use a hash table of the same size where each
   38  * bucket is referred to as a "chain" that contains both a spin lock and
   39  * a linked list of queues.  An individual queue is located by using a hash
   40  * to pick a chain, locking the chain, and then walking the chain searching
   41  * for the queue.  This means that a wait channel object does not need to
   42  * embed it's queue head just as locks do not embed their turnstile queue
   43  * head.  Threads also carry around a sleep queue that they lend to the
   44  * wait channel when blocking.  Just as in turnstiles, the queue includes
   45  * a free list of the sleep queues of other threads blocked on the same
   46  * wait channel in the case of multiple waiters.
   47  *
   48  * Some additional functionality provided by sleep queues include the
   49  * ability to set a timeout.  The timeout is managed using a per-thread
   50  * callout that resumes a thread if it is asleep.  A thread may also
   51  * catch signals while it is asleep (aka an interruptible sleep).  The
   52  * signal code uses sleepq_abort() to interrupt a sleeping thread.  Finally,
   53  * sleep queues also provide some extra assertions.  One is not allowed to
   54  * mix the sleep/wakeup and cv APIs for a given wait channel.  Also, one
   55  * must consistently use the same lock to synchronize with a wait channel,
   56  * though this check is currently only a warning for sleep/wakeup due to
   57  * pre-existing abuse of that API.  The same lock must also be held when
   58  * awakening threads, though that is currently only enforced for condition
   59  * variables.
   60  */
   61 
   62 #include <sys/cdefs.h>
   63 __FBSDID("$FreeBSD$");
   64 
   65 #include "opt_sleepqueue_profiling.h"
   66 #include "opt_ddb.h"
   67 
   68 #include <sys/param.h>
   69 #include <sys/systm.h>
   70 #include <sys/lock.h>
   71 #include <sys/kernel.h>
   72 #include <sys/ktr.h>
   73 #include <sys/mutex.h>
   74 #include <sys/proc.h>
   75 #include <sys/sched.h>
   76 #include <sys/signalvar.h>
   77 #include <sys/sleepqueue.h>
   78 #include <sys/sysctl.h>
   79 
   80 #include <vm/uma.h>
   81 
   82 #ifdef DDB
   83 #include <ddb/ddb.h>
   84 #endif
   85 
   86 /*
   87  * Constants for the hash table of sleep queue chains.  These constants are
   88  * the same ones that 4BSD (and possibly earlier versions of BSD) used.
   89  * Basically, we ignore the lower 8 bits of the address since most wait
   90  * channel pointers are aligned and only look at the next 7 bits for the
   91  * hash.  SC_TABLESIZE must be a power of two for SC_MASK to work properly.
   92  */
   93 #define SC_TABLESIZE    128                     /* Must be power of 2. */
   94 #define SC_MASK         (SC_TABLESIZE - 1)
   95 #define SC_SHIFT        8
   96 #define SC_HASH(wc)     (((uintptr_t)(wc) >> SC_SHIFT) & SC_MASK)
   97 #define SC_LOOKUP(wc)   &sleepq_chains[SC_HASH(wc)]
   98 #define NR_SLEEPQS      2
   99 /*
  100  * There two different lists of sleep queues.  Both lists are connected
  101  * via the sq_hash entries.  The first list is the sleep queue chain list
  102  * that a sleep queue is on when it is attached to a wait channel.  The
  103  * second list is the free list hung off of a sleep queue that is attached
  104  * to a wait channel.
  105  *
  106  * Each sleep queue also contains the wait channel it is attached to, the
  107  * list of threads blocked on that wait channel, flags specific to the
  108  * wait channel, and the lock used to synchronize with a wait channel.
  109  * The flags are used to catch mismatches between the various consumers
  110  * of the sleep queue API (e.g. sleep/wakeup and condition variables).
  111  * The lock pointer is only used when invariants are enabled for various
  112  * debugging checks.
  113  *
  114  * Locking key:
  115  *  c - sleep queue chain lock
  116  */
  117 struct sleepqueue {
  118         TAILQ_HEAD(, thread) sq_blocked[NR_SLEEPQS];    /* (c) Blocked threads. */
  119         LIST_ENTRY(sleepqueue) sq_hash;         /* (c) Chain and free list. */
  120         LIST_HEAD(, sleepqueue) sq_free;        /* (c) Free queues. */
  121         void    *sq_wchan;                      /* (c) Wait channel. */
  122 #ifdef INVARIANTS
  123         int     sq_type;                        /* (c) Queue type. */
  124         struct lock_object *sq_lock;            /* (c) Associated lock. */
  125 #endif
  126 };
  127 
  128 struct sleepqueue_chain {
  129         LIST_HEAD(, sleepqueue) sc_queues;      /* List of sleep queues. */
  130         struct mtx sc_lock;                     /* Spin lock for this chain. */
  131 #ifdef SLEEPQUEUE_PROFILING
  132         u_int   sc_depth;                       /* Length of sc_queues. */
  133         u_int   sc_max_depth;                   /* Max length of sc_queues. */
  134 #endif
  135 };
  136 
  137 #ifdef SLEEPQUEUE_PROFILING
  138 u_int sleepq_max_depth;
  139 SYSCTL_NODE(_debug, OID_AUTO, sleepq, CTLFLAG_RD, 0, "sleepq profiling");
  140 SYSCTL_NODE(_debug_sleepq, OID_AUTO, chains, CTLFLAG_RD, 0,
  141     "sleepq chain stats");
  142 SYSCTL_UINT(_debug_sleepq, OID_AUTO, max_depth, CTLFLAG_RD, &sleepq_max_depth,
  143     0, "maxmimum depth achieved of a single chain");
  144 #endif
  145 static struct sleepqueue_chain sleepq_chains[SC_TABLESIZE];
  146 static uma_zone_t sleepq_zone;
  147 
  148 /*
  149  * Prototypes for non-exported routines.
  150  */
  151 static int      sleepq_catch_signals(void *wchan);
  152 static int      sleepq_check_signals(void);
  153 static int      sleepq_check_timeout(void);
  154 #ifdef INVARIANTS
  155 static void     sleepq_dtor(void *mem, int size, void *arg);
  156 #endif
  157 static int      sleepq_init(void *mem, int size, int flags);
  158 static void     sleepq_resume_thread(struct sleepqueue *sq, struct thread *td,
  159                     int pri);
  160 static void     sleepq_switch(void *wchan);
  161 static void     sleepq_timeout(void *arg);
  162 
  163 /*
  164  * Early initialization of sleep queues that is called from the sleepinit()
  165  * SYSINIT.
  166  */
  167 void
  168 init_sleepqueues(void)
  169 {
  170 #ifdef SLEEPQUEUE_PROFILING
  171         struct sysctl_oid *chain_oid;
  172         char chain_name[10];
  173 #endif
  174         int i;
  175 
  176         for (i = 0; i < SC_TABLESIZE; i++) {
  177                 LIST_INIT(&sleepq_chains[i].sc_queues);
  178                 mtx_init(&sleepq_chains[i].sc_lock, "sleepq chain", NULL,
  179                     MTX_SPIN);
  180 #ifdef SLEEPQUEUE_PROFILING
  181                 snprintf(chain_name, sizeof(chain_name), "%d", i);
  182                 chain_oid = SYSCTL_ADD_NODE(NULL, 
  183                     SYSCTL_STATIC_CHILDREN(_debug_sleepq_chains), OID_AUTO,
  184                     chain_name, CTLFLAG_RD, NULL, "sleepq chain stats");
  185                 SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
  186                     "depth", CTLFLAG_RD, &sleepq_chains[i].sc_depth, 0, NULL);
  187                 SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
  188                     "max_depth", CTLFLAG_RD, &sleepq_chains[i].sc_max_depth, 0,
  189                     NULL);
  190 #endif
  191         }
  192         sleepq_zone = uma_zcreate("SLEEPQUEUE", sizeof(struct sleepqueue),
  193 #ifdef INVARIANTS
  194             NULL, sleepq_dtor, sleepq_init, NULL, UMA_ALIGN_CACHE, 0);
  195 #else
  196             NULL, NULL, sleepq_init, NULL, UMA_ALIGN_CACHE, 0);
  197 #endif
  198         
  199         thread0.td_sleepqueue = sleepq_alloc();
  200 }
  201 
  202 /*
  203  * Get a sleep queue for a new thread.
  204  */
  205 struct sleepqueue *
  206 sleepq_alloc(void)
  207 {
  208 
  209         return (uma_zalloc(sleepq_zone, M_WAITOK));
  210 }
  211 
  212 /*
  213  * Free a sleep queue when a thread is destroyed.
  214  */
  215 void
  216 sleepq_free(struct sleepqueue *sq)
  217 {
  218 
  219         uma_zfree(sleepq_zone, sq);
  220 }
  221 
  222 /*
  223  * Lock the sleep queue chain associated with the specified wait channel.
  224  */
  225 void
  226 sleepq_lock(void *wchan)
  227 {
  228         struct sleepqueue_chain *sc;
  229 
  230         sc = SC_LOOKUP(wchan);
  231         mtx_lock_spin(&sc->sc_lock);
  232 }
  233 
  234 /*
  235  * Look up the sleep queue associated with a given wait channel in the hash
  236  * table locking the associated sleep queue chain.  If no queue is found in
  237  * the table, NULL is returned.
  238  */
  239 struct sleepqueue *
  240 sleepq_lookup(void *wchan)
  241 {
  242         struct sleepqueue_chain *sc;
  243         struct sleepqueue *sq;
  244 
  245         KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
  246         sc = SC_LOOKUP(wchan);
  247         mtx_assert(&sc->sc_lock, MA_OWNED);
  248         LIST_FOREACH(sq, &sc->sc_queues, sq_hash)
  249                 if (sq->sq_wchan == wchan)
  250                         return (sq);
  251         return (NULL);
  252 }
  253 
  254 /*
  255  * Unlock the sleep queue chain associated with a given wait channel.
  256  */
  257 void
  258 sleepq_release(void *wchan)
  259 {
  260         struct sleepqueue_chain *sc;
  261 
  262         sc = SC_LOOKUP(wchan);
  263         mtx_unlock_spin(&sc->sc_lock);
  264 }
  265 
  266 /*
  267  * Places the current thread on the sleep queue for the specified wait
  268  * channel.  If INVARIANTS is enabled, then it associates the passed in
  269  * lock with the sleepq to make sure it is held when that sleep queue is
  270  * woken up.
  271  */
  272 void
  273 sleepq_add(void *wchan, struct lock_object *lock, const char *wmesg, int flags,
  274     int queue)
  275 {
  276         struct sleepqueue_chain *sc;
  277         struct sleepqueue *sq;
  278         struct thread *td;
  279 
  280         td = curthread;
  281         sc = SC_LOOKUP(wchan);
  282         mtx_assert(&sc->sc_lock, MA_OWNED);
  283         MPASS(td->td_sleepqueue != NULL);
  284         MPASS(wchan != NULL);
  285         MPASS((queue >= 0) && (queue < NR_SLEEPQS));
  286 
  287         /* If this thread is not allowed to sleep, die a horrible death. */
  288         KASSERT(!(td->td_pflags & TDP_NOSLEEPING),
  289             ("Trying sleep, but thread marked as sleeping prohibited"));
  290 
  291         /* Look up the sleep queue associated with the wait channel 'wchan'. */
  292         sq = sleepq_lookup(wchan);
  293 
  294         /*
  295          * If the wait channel does not already have a sleep queue, use
  296          * this thread's sleep queue.  Otherwise, insert the current thread
  297          * into the sleep queue already in use by this wait channel.
  298          */
  299         if (sq == NULL) {
  300 #ifdef INVARIANTS
  301                 int i;
  302 
  303                 sq = td->td_sleepqueue;
  304                 for (i = 0; i < NR_SLEEPQS; i++)
  305                         KASSERT(TAILQ_EMPTY(&sq->sq_blocked[i]),
  306                                 ("thread's sleep queue %d is not empty", i));
  307                 KASSERT(LIST_EMPTY(&sq->sq_free),
  308                     ("thread's sleep queue has a non-empty free list"));
  309                 KASSERT(sq->sq_wchan == NULL, ("stale sq_wchan pointer"));
  310                 sq->sq_lock = lock;
  311                 sq->sq_type = flags & SLEEPQ_TYPE;
  312 #endif
  313 #ifdef SLEEPQUEUE_PROFILING
  314                 sc->sc_depth++;
  315                 if (sc->sc_depth > sc->sc_max_depth) {
  316                         sc->sc_max_depth = sc->sc_depth;
  317                         if (sc->sc_max_depth > sleepq_max_depth)
  318                                 sleepq_max_depth = sc->sc_max_depth;
  319                 }
  320 #endif
  321                 sq = td->td_sleepqueue;
  322                 LIST_INSERT_HEAD(&sc->sc_queues, sq, sq_hash);
  323                 sq->sq_wchan = wchan;
  324         } else {
  325                 MPASS(wchan == sq->sq_wchan);
  326                 MPASS(lock == sq->sq_lock);
  327                 MPASS((flags & SLEEPQ_TYPE) == sq->sq_type);
  328                 LIST_INSERT_HEAD(&sq->sq_free, td->td_sleepqueue, sq_hash);
  329         }
  330         TAILQ_INSERT_TAIL(&sq->sq_blocked[queue], td, td_slpq);
  331         td->td_sleepqueue = NULL;
  332         mtx_lock_spin(&sched_lock);
  333         td->td_sqqueue = queue;
  334         td->td_wchan = wchan;
  335         td->td_wmesg = wmesg;
  336         if (flags & SLEEPQ_INTERRUPTIBLE) {
  337                 td->td_flags |= TDF_SINTR;
  338                 td->td_flags &= ~TDF_SLEEPABORT;
  339         }
  340         mtx_unlock_spin(&sched_lock);
  341 }
  342 
  343 /*
  344  * Sets a timeout that will remove the current thread from the specified
  345  * sleep queue after timo ticks if the thread has not already been awakened.
  346  */
  347 void
  348 sleepq_set_timeout(void *wchan, int timo)
  349 {
  350         struct sleepqueue_chain *sc;
  351         struct thread *td;
  352 
  353         td = curthread;
  354         sc = SC_LOOKUP(wchan);
  355         mtx_assert(&sc->sc_lock, MA_OWNED);
  356         MPASS(TD_ON_SLEEPQ(td));
  357         MPASS(td->td_sleepqueue == NULL);
  358         MPASS(wchan != NULL);
  359         callout_reset(&td->td_slpcallout, timo, sleepq_timeout, td);
  360 }
  361 
  362 /*
  363  * Marks the pending sleep of the current thread as interruptible and
  364  * makes an initial check for pending signals before putting a thread
  365  * to sleep. Return with sleep queue and scheduler lock held.
  366  */
  367 static int
  368 sleepq_catch_signals(void *wchan)
  369 {
  370         struct sleepqueue_chain *sc;
  371         struct sleepqueue *sq;
  372         struct thread *td;
  373         struct proc *p;
  374         struct sigacts *ps;
  375         int sig, ret;
  376 
  377         td = curthread;
  378         p = curproc;
  379         sc = SC_LOOKUP(wchan);
  380         mtx_assert(&sc->sc_lock, MA_OWNED);
  381         MPASS(wchan != NULL);
  382         CTR3(KTR_PROC, "sleepq catching signals: thread %p (pid %ld, %s)",
  383                 (void *)td, (long)p->p_pid, p->p_comm);
  384 
  385         mtx_unlock_spin(&sc->sc_lock);
  386 
  387         /* See if there are any pending signals for this thread. */
  388         PROC_LOCK(p);
  389         ps = p->p_sigacts;
  390         mtx_lock(&ps->ps_mtx);
  391         sig = cursig(td);
  392         if (sig == 0) {
  393                 mtx_unlock(&ps->ps_mtx);
  394                 ret = thread_suspend_check(1);
  395                 MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
  396         } else {
  397                 if (SIGISMEMBER(ps->ps_sigintr, sig))
  398                         ret = EINTR;
  399                 else
  400                         ret = ERESTART;
  401                 mtx_unlock(&ps->ps_mtx);
  402         }
  403 
  404         if (ret == 0) {
  405                 mtx_lock_spin(&sc->sc_lock);
  406                 /*
  407                  * Lock sched_lock before unlocking proc lock,
  408                  * without this, we could lose a race.
  409                  */
  410                 mtx_lock_spin(&sched_lock);
  411                 PROC_UNLOCK(p);
  412                 if (!(td->td_flags & TDF_INTERRUPT))
  413                         return (0);
  414                 /* KSE threads tried unblocking us. */
  415                 ret = td->td_intrval;
  416                 mtx_unlock_spin(&sched_lock);
  417                 MPASS(ret == EINTR || ret == ERESTART);
  418         } else {
  419                 PROC_UNLOCK(p);
  420                 mtx_lock_spin(&sc->sc_lock);
  421         }
  422         /*
  423          * There were pending signals and this thread is still
  424          * on the sleep queue, remove it from the sleep queue.
  425          */
  426         sq = sleepq_lookup(wchan);
  427         mtx_lock_spin(&sched_lock);
  428         if (TD_ON_SLEEPQ(td))
  429                 sleepq_resume_thread(sq, td, -1);
  430         return (ret);
  431 }
  432 
  433 /*
  434  * Switches to another thread if we are still asleep on a sleep queue and
  435  * drop the lock on the sleep queue chain.  Returns with sched_lock held.
  436  */
  437 static void
  438 sleepq_switch(void *wchan)
  439 {
  440         struct sleepqueue_chain *sc;
  441         struct thread *td;
  442 
  443         td = curthread;
  444         sc = SC_LOOKUP(wchan);
  445         mtx_assert(&sc->sc_lock, MA_OWNED);
  446         mtx_assert(&sched_lock, MA_OWNED);
  447 
  448         /* 
  449          * If we have a sleep queue, then we've already been woken up, so
  450          * just return.
  451          */
  452         if (td->td_sleepqueue != NULL) {
  453                 MPASS(!TD_ON_SLEEPQ(td));
  454                 mtx_unlock_spin(&sc->sc_lock);
  455                 return;
  456         }
  457 
  458         /*
  459          * Otherwise, actually go to sleep.
  460          */
  461         mtx_unlock_spin(&sc->sc_lock);
  462         sched_sleep(td);
  463         TD_SET_SLEEPING(td);
  464         mi_switch(SW_VOL, NULL);
  465         KASSERT(TD_IS_RUNNING(td), ("running but not TDS_RUNNING"));
  466         CTR3(KTR_PROC, "sleepq resume: thread %p (pid %ld, %s)",
  467             (void *)td, (long)td->td_proc->p_pid, (void *)td->td_proc->p_comm);
  468 }
  469 
  470 /*
  471  * Check to see if we timed out.
  472  */
  473 static int
  474 sleepq_check_timeout(void)
  475 {
  476         struct thread *td;
  477 
  478         mtx_assert(&sched_lock, MA_OWNED);
  479         td = curthread;
  480 
  481         /*
  482          * If TDF_TIMEOUT is set, we timed out.
  483          */
  484         if (td->td_flags & TDF_TIMEOUT) {
  485                 td->td_flags &= ~TDF_TIMEOUT;
  486                 return (EWOULDBLOCK);
  487         }
  488 
  489         /*
  490          * If TDF_TIMOFAIL is set, the timeout ran after we had
  491          * already been woken up.
  492          */
  493         if (td->td_flags & TDF_TIMOFAIL)
  494                 td->td_flags &= ~TDF_TIMOFAIL;
  495 
  496         /*
  497          * If callout_stop() fails, then the timeout is running on
  498          * another CPU, so synchronize with it to avoid having it
  499          * accidentally wake up a subsequent sleep.
  500          */
  501         else if (callout_stop(&td->td_slpcallout) == 0) {
  502                 td->td_flags |= TDF_TIMEOUT;
  503                 TD_SET_SLEEPING(td);
  504                 mi_switch(SW_INVOL, NULL);
  505         }
  506         return (0);
  507 }
  508 
  509 /*
  510  * Check to see if we were awoken by a signal.
  511  */
  512 static int
  513 sleepq_check_signals(void)
  514 {
  515         struct thread *td;
  516 
  517         mtx_assert(&sched_lock, MA_OWNED);
  518         td = curthread;
  519 
  520         /* We are no longer in an interruptible sleep. */
  521         if (td->td_flags & TDF_SINTR)
  522                 td->td_flags &= ~TDF_SINTR;
  523 
  524         if (td->td_flags & TDF_SLEEPABORT) {
  525                 td->td_flags &= ~TDF_SLEEPABORT;
  526                 return (td->td_intrval);
  527         }
  528 
  529         if (td->td_flags & TDF_INTERRUPT)
  530                 return (td->td_intrval);
  531 
  532         return (0);
  533 }
  534 
  535 /*
  536  * Block the current thread until it is awakened from its sleep queue.
  537  */
  538 void
  539 sleepq_wait(void *wchan)
  540 {
  541 
  542         MPASS(!(curthread->td_flags & TDF_SINTR));
  543         mtx_lock_spin(&sched_lock);
  544         sleepq_switch(wchan);
  545         mtx_unlock_spin(&sched_lock);
  546 }
  547 
  548 /*
  549  * Block the current thread until it is awakened from its sleep queue
  550  * or it is interrupted by a signal.
  551  */
  552 int
  553 sleepq_wait_sig(void *wchan)
  554 {
  555         int rcatch;
  556         int rval;
  557 
  558         rcatch = sleepq_catch_signals(wchan);
  559         if (rcatch == 0)
  560                 sleepq_switch(wchan);
  561         else
  562                 sleepq_release(wchan);
  563         rval = sleepq_check_signals();
  564         mtx_unlock_spin(&sched_lock); 
  565         if (rcatch)
  566                 return (rcatch);
  567         return (rval);
  568 }
  569 
  570 /*
  571  * Block the current thread until it is awakened from its sleep queue
  572  * or it times out while waiting.
  573  */
  574 int
  575 sleepq_timedwait(void *wchan)
  576 {
  577         int rval;
  578 
  579         MPASS(!(curthread->td_flags & TDF_SINTR));
  580         mtx_lock_spin(&sched_lock);
  581         sleepq_switch(wchan);
  582         rval = sleepq_check_timeout();
  583         mtx_unlock_spin(&sched_lock);
  584         return (rval);
  585 }
  586 
  587 /*
  588  * Block the current thread until it is awakened from its sleep queue,
  589  * it is interrupted by a signal, or it times out waiting to be awakened.
  590  */
  591 int
  592 sleepq_timedwait_sig(void *wchan)
  593 {
  594         int rcatch, rvalt, rvals;
  595 
  596         rcatch = sleepq_catch_signals(wchan);
  597         if (rcatch == 0)
  598                 sleepq_switch(wchan);
  599         else
  600                 sleepq_release(wchan);
  601         rvalt = sleepq_check_timeout();
  602         rvals = sleepq_check_signals();
  603         mtx_unlock_spin(&sched_lock);
  604         if (rcatch)
  605                 return (rcatch);
  606         if (rvals)
  607                 return (rvals);
  608         return (rvalt);
  609 }
  610 
  611 /*
  612  * Removes a thread from a sleep queue and makes it
  613  * runnable.
  614  */
  615 static void
  616 sleepq_resume_thread(struct sleepqueue *sq, struct thread *td, int pri)
  617 {
  618         struct sleepqueue_chain *sc;
  619 
  620         MPASS(td != NULL);
  621         MPASS(sq->sq_wchan != NULL);
  622         MPASS(td->td_wchan == sq->sq_wchan);
  623         MPASS(td->td_sqqueue < NR_SLEEPQS && td->td_sqqueue >= 0);
  624         sc = SC_LOOKUP(sq->sq_wchan);
  625         mtx_assert(&sc->sc_lock, MA_OWNED);
  626         mtx_assert(&sched_lock, MA_OWNED);
  627 
  628         /* Remove the thread from the queue. */
  629         TAILQ_REMOVE(&sq->sq_blocked[(int)td->td_sqqueue], td, td_slpq);
  630 
  631         /*
  632          * Get a sleep queue for this thread.  If this is the last waiter,
  633          * use the queue itself and take it out of the chain, otherwise,
  634          * remove a queue from the free list.
  635          */
  636         if (LIST_EMPTY(&sq->sq_free)) {
  637                 td->td_sleepqueue = sq;
  638 #ifdef INVARIANTS
  639                 sq->sq_wchan = NULL;
  640 #endif
  641 #ifdef SLEEPQUEUE_PROFILING
  642                 sc->sc_depth--;
  643 #endif
  644         } else
  645                 td->td_sleepqueue = LIST_FIRST(&sq->sq_free);
  646         LIST_REMOVE(td->td_sleepqueue, sq_hash);
  647 
  648         td->td_wmesg = NULL;
  649         td->td_wchan = NULL;
  650         td->td_flags &= ~TDF_SINTR;
  651 
  652         /*
  653          * Note that thread td might not be sleeping if it is running
  654          * sleepq_catch_signals() on another CPU or is blocked on
  655          * its proc lock to check signals.  It doesn't hurt to clear
  656          * the sleeping flag if it isn't set though, so we just always
  657          * do it.  However, we can't assert that it is set.
  658          */
  659         CTR3(KTR_PROC, "sleepq_wakeup: thread %p (pid %ld, %s)",
  660             (void *)td, (long)td->td_proc->p_pid, td->td_proc->p_comm);
  661         TD_CLR_SLEEPING(td);
  662 
  663         /* Adjust priority if requested. */
  664         MPASS(pri == -1 || (pri >= PRI_MIN && pri <= PRI_MAX));
  665         if (pri != -1 && td->td_priority > pri)
  666                 sched_prio(td, pri);
  667         setrunnable(td);
  668 }
  669 
  670 #ifdef INVARIANTS
  671 /*
  672  * UMA zone item deallocator.
  673  */
  674 static void
  675 sleepq_dtor(void *mem, int size, void *arg)
  676 {
  677         struct sleepqueue *sq;
  678         int i;
  679 
  680         sq = mem;
  681         for (i = 0; i < NR_SLEEPQS; i++)
  682                 MPASS(TAILQ_EMPTY(&sq->sq_blocked[i]));
  683 }
  684 #endif
  685 
  686 /*
  687  * UMA zone item initializer.
  688  */
  689 static int
  690 sleepq_init(void *mem, int size, int flags)
  691 {
  692         struct sleepqueue *sq;
  693         int i;
  694 
  695         bzero(mem, size);
  696         sq = mem;
  697         for (i = 0; i < NR_SLEEPQS; i++)
  698                 TAILQ_INIT(&sq->sq_blocked[i]);
  699         LIST_INIT(&sq->sq_free);
  700         return (0);
  701 }
  702 
  703 /*
  704  * Find the highest priority thread sleeping on a wait channel and resume it.
  705  */
  706 void
  707 sleepq_signal(void *wchan, int flags, int pri, int queue)
  708 {
  709         struct sleepqueue *sq;
  710         struct thread *td, *besttd;
  711 
  712         CTR2(KTR_PROC, "sleepq_signal(%p, %d)", wchan, flags);
  713         KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
  714         MPASS((queue >= 0) && (queue < NR_SLEEPQS));
  715         sq = sleepq_lookup(wchan);
  716         if (sq == NULL) {
  717                 sleepq_release(wchan);
  718                 return;
  719         }
  720         KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE),
  721             ("%s: mismatch between sleep/wakeup and cv_*", __func__));
  722 
  723         /*
  724          * Find the highest priority thread on the queue.  If there is a
  725          * tie, use the thread that first appears in the queue as it has
  726          * been sleeping the longest since threads are always added to
  727          * the tail of sleep queues.
  728          */
  729         besttd = NULL;
  730         TAILQ_FOREACH(td, &sq->sq_blocked[queue], td_slpq) {
  731                 if (besttd == NULL || td->td_priority < besttd->td_priority)
  732                         besttd = td;
  733         }
  734         MPASS(besttd != NULL);
  735         mtx_lock_spin(&sched_lock);
  736         sleepq_resume_thread(sq, besttd, pri);
  737         mtx_unlock_spin(&sched_lock);
  738         sleepq_release(wchan);
  739 }
  740 
  741 /*
  742  * Resume all threads sleeping on a specified wait channel.
  743  */
  744 void
  745 sleepq_broadcast(void *wchan, int flags, int pri, int queue)
  746 {
  747         struct sleepqueue *sq;
  748 
  749         CTR2(KTR_PROC, "sleepq_broadcast(%p, %d)", wchan, flags);
  750         KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
  751         MPASS((queue >= 0) && (queue < NR_SLEEPQS));
  752         sq = sleepq_lookup(wchan);
  753         if (sq == NULL) {
  754                 sleepq_release(wchan);
  755                 return;
  756         }
  757         KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE),
  758             ("%s: mismatch between sleep/wakeup and cv_*", __func__));
  759 
  760         /* Resume all blocked threads on the sleep queue. */
  761         mtx_lock_spin(&sched_lock);
  762         while (!TAILQ_EMPTY(&sq->sq_blocked[queue]))
  763                 sleepq_resume_thread(sq, TAILQ_FIRST(&sq->sq_blocked[queue]),
  764                     pri);
  765         mtx_unlock_spin(&sched_lock);
  766         sleepq_release(wchan);
  767 }
  768 
  769 /*
  770  * Time sleeping threads out.  When the timeout expires, the thread is
  771  * removed from the sleep queue and made runnable if it is still asleep.
  772  */
  773 static void
  774 sleepq_timeout(void *arg)
  775 {
  776         struct sleepqueue *sq;
  777         struct thread *td;
  778         void *wchan;
  779 
  780         td = arg;
  781         CTR3(KTR_PROC, "sleepq_timeout: thread %p (pid %ld, %s)",
  782             (void *)td, (long)td->td_proc->p_pid, (void *)td->td_proc->p_comm);
  783 
  784         /*
  785          * First, see if the thread is asleep and get the wait channel if
  786          * it is.
  787          */
  788         mtx_lock_spin(&sched_lock);
  789         if (TD_ON_SLEEPQ(td)) {
  790                 wchan = td->td_wchan;
  791                 mtx_unlock_spin(&sched_lock);
  792                 sleepq_lock(wchan);
  793                 sq = sleepq_lookup(wchan);
  794                 mtx_lock_spin(&sched_lock);
  795         } else {
  796                 wchan = NULL;
  797                 sq = NULL;
  798         }
  799 
  800         /*
  801          * At this point, if the thread is still on the sleep queue,
  802          * we have that sleep queue locked as it cannot migrate sleep
  803          * queues while we dropped sched_lock.  If it had resumed and
  804          * was on another CPU while the lock was dropped, it would have
  805          * seen that TDF_TIMEOUT and TDF_TIMOFAIL are clear and the
  806          * call to callout_stop() to stop this routine would have failed
  807          * meaning that it would have already set TDF_TIMEOUT to
  808          * synchronize with this function.
  809          */
  810         if (TD_ON_SLEEPQ(td)) {
  811                 MPASS(td->td_wchan == wchan);
  812                 MPASS(sq != NULL);
  813                 td->td_flags |= TDF_TIMEOUT;
  814                 sleepq_resume_thread(sq, td, -1);
  815                 mtx_unlock_spin(&sched_lock);
  816                 sleepq_release(wchan);
  817                 return;
  818         } else if (wchan != NULL)
  819                 sleepq_release(wchan);
  820 
  821         /*
  822          * Now check for the edge cases.  First, if TDF_TIMEOUT is set,
  823          * then the other thread has already yielded to us, so clear
  824          * the flag and resume it.  If TDF_TIMEOUT is not set, then the
  825          * we know that the other thread is not on a sleep queue, but it
  826          * hasn't resumed execution yet.  In that case, set TDF_TIMOFAIL
  827          * to let it know that the timeout has already run and doesn't
  828          * need to be canceled.
  829          */
  830         if (td->td_flags & TDF_TIMEOUT) {
  831                 MPASS(TD_IS_SLEEPING(td));
  832                 td->td_flags &= ~TDF_TIMEOUT;
  833                 TD_CLR_SLEEPING(td);
  834                 setrunnable(td);
  835         } else
  836                 td->td_flags |= TDF_TIMOFAIL;
  837         mtx_unlock_spin(&sched_lock);
  838 }
  839 
  840 /*
  841  * Resumes a specific thread from the sleep queue associated with a specific
  842  * wait channel if it is on that queue.
  843  */
  844 void
  845 sleepq_remove(struct thread *td, void *wchan)
  846 {
  847         struct sleepqueue *sq;
  848 
  849         /*
  850          * Look up the sleep queue for this wait channel, then re-check
  851          * that the thread is asleep on that channel, if it is not, then
  852          * bail.
  853          */
  854         MPASS(wchan != NULL);
  855         sleepq_lock(wchan);
  856         sq = sleepq_lookup(wchan);
  857         mtx_lock_spin(&sched_lock);
  858         if (!TD_ON_SLEEPQ(td) || td->td_wchan != wchan) {
  859                 mtx_unlock_spin(&sched_lock);
  860                 sleepq_release(wchan);
  861                 return;
  862         }
  863         MPASS(sq != NULL);
  864 
  865         /* Thread is asleep on sleep queue sq, so wake it up. */
  866         sleepq_resume_thread(sq, td, -1);
  867         sleepq_release(wchan);
  868         mtx_unlock_spin(&sched_lock);
  869 }
  870 
  871 /*
  872  * Abort a thread as if an interrupt had occurred.  Only abort
  873  * interruptible waits (unfortunately it isn't safe to abort others).
  874  *
  875  * XXX: What in the world does the comment below mean?
  876  * Also, whatever the signal code does...
  877  */
  878 void
  879 sleepq_abort(struct thread *td, int intrval)
  880 {
  881         void *wchan;
  882 
  883         mtx_assert(&sched_lock, MA_OWNED);
  884         MPASS(TD_ON_SLEEPQ(td));
  885         MPASS(td->td_flags & TDF_SINTR);
  886         MPASS(intrval == EINTR || intrval == ERESTART);
  887 
  888         /*
  889          * If the TDF_TIMEOUT flag is set, just leave. A
  890          * timeout is scheduled anyhow.
  891          */
  892         if (td->td_flags & TDF_TIMEOUT)
  893                 return;
  894 
  895         CTR3(KTR_PROC, "sleepq_abort: thread %p (pid %ld, %s)",
  896             (void *)td, (long)td->td_proc->p_pid, (void *)td->td_proc->p_comm);
  897         wchan = td->td_wchan;
  898         if (wchan != NULL) {
  899                 td->td_intrval = intrval;
  900                 td->td_flags |= TDF_SLEEPABORT;
  901         }
  902         mtx_unlock_spin(&sched_lock);
  903         sleepq_remove(td, wchan);
  904         mtx_lock_spin(&sched_lock);
  905 }
  906 
  907 #ifdef DDB
  908 DB_SHOW_COMMAND(sleepq, db_show_sleepqueue)
  909 {
  910         struct sleepqueue_chain *sc;
  911         struct sleepqueue *sq;
  912 #ifdef INVARIANTS
  913         struct lock_object *lock;
  914 #endif
  915         struct thread *td;
  916         void *wchan;
  917         int i;
  918 
  919         if (!have_addr)
  920                 return;
  921 
  922         /*
  923          * First, see if there is an active sleep queue for the wait channel
  924          * indicated by the address.
  925          */
  926         wchan = (void *)addr;
  927         sc = SC_LOOKUP(wchan);
  928         LIST_FOREACH(sq, &sc->sc_queues, sq_hash)
  929                 if (sq->sq_wchan == wchan)
  930                         goto found;
  931 
  932         /*
  933          * Second, see if there is an active sleep queue at the address
  934          * indicated.
  935          */
  936         for (i = 0; i < SC_TABLESIZE; i++)
  937                 LIST_FOREACH(sq, &sleepq_chains[i].sc_queues, sq_hash) {
  938                         if (sq == (struct sleepqueue *)addr)
  939                                 goto found;
  940                 }
  941 
  942         db_printf("Unable to locate a sleep queue via %p\n", (void *)addr);
  943         return;
  944 found:
  945         db_printf("Wait channel: %p\n", sq->sq_wchan);
  946 #ifdef INVARIANTS
  947         db_printf("Queue type: %d\n", sq->sq_type);
  948         if (sq->sq_lock) {
  949                 lock = sq->sq_lock;
  950                 db_printf("Associated Interlock: %p - (%s) %s\n", lock,
  951                     LOCK_CLASS(lock)->lc_name, lock->lo_name);
  952         }
  953 #endif
  954         db_printf("Blocked threads:\n");
  955         for (i = 0; i < NR_SLEEPQS; i++) {
  956                 db_printf("\nQueue[%d]:\n", i);
  957                 if (TAILQ_EMPTY(&sq->sq_blocked[i]))
  958                         db_printf("\tempty\n");
  959                 else
  960                         TAILQ_FOREACH(td, &sq->sq_blocked[0],
  961                                       td_slpq) {
  962                                 db_printf("\t%p (tid %d, pid %d, \"%s\")\n", td,
  963                                           td->td_tid, td->td_proc->p_pid,
  964                                           td->td_proc->p_comm);
  965                         }
  966         }
  967 }
  968 #endif

Cache object: 1bac85700eeed0f56879a422d8d82697


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.