nptl/sem_waitcommon.c

   1 /* sem_waitcommon -- wait on a semaphore, shared code.
   2    Copyright (C) 2003-2015 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4    Contributed by Paul Mackerras <paulus@au.ibm.com>, 2003.
   5
   6    The GNU C Library is free software; you can redistribute it and/or
   7    modify it under the terms of the GNU Lesser General Public
   8    License as published by the Free Software Foundation; either
   9    version 2.1 of the License, or (at your option) any later version.
  10
  11    The GNU C Library is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14    Lesser General Public License for more details.
  15
  16    You should have received a copy of the GNU Lesser General Public
  17    License along with the GNU C Library; if not, see
  18    <http://www.gnu.org/licenses/>.  */
  19
  20 #include <kernel-features.h>
  21 #include <errno.h>
  22 #include <sysdep.h>
  23 #include <futex-internal.h>
  24 #include <internaltypes.h>
  25 #include <semaphore.h>
  26 #include <sys/time.h>
  27
  28 #include <pthreadP.h>
  29 #include <shlib-compat.h>
  30 #include <atomic.h>
  31
  32
  33 /* The semaphore provides two main operations: sem_post adds a token to the
  34    semaphore; sem_wait grabs a token from the semaphore, potentially waiting
  35    until there is a token available.  A sem_wait needs to synchronize with
  36    the sem_post that provided the token, so that whatever lead to the sem_post
  37    happens before the code after sem_wait.
  38
  39    Conceptually, available tokens can simply be counted; let's call that the
  40    value of the semaphore.  However, we also want to know whether there might
  41    be a sem_wait that is blocked on the value because it was zero (using a
  42    futex with the value being the futex variable); if there is no blocked
  43    sem_wait, sem_post does not need to execute a futex_wake call.  Therefore,
  44    we also need to count the number of potentially blocked sem_wait calls
  45    (which we call nwaiters).
  46
  47    What makes this tricky is that POSIX requires that a semaphore can be
  48    destroyed as soon as the last remaining sem_wait has returned, and no
  49    other sem_wait or sem_post calls are executing concurrently.  However, the
  50    sem_post call whose token was consumed by the last sem_wait is considered
  51    to have finished once it provided the token to the sem_wait.
  52    Thus, sem_post must not access the semaphore struct anymore after it has
  53    made a token available; IOW, it needs to be able to atomically provide
  54    a token and check whether any blocked sem_wait calls might exist.
  55
  56    This is straightforward to do if the architecture provides 64b atomics
  57    because we can just put both the value and nwaiters into one variable that
  58    we access atomically: This is the data field, the value is in the
  59    least-significant 32 bits, and nwaiters in the other bits.  When sem_post
  60    makes a value available, it can atomically check nwaiters.
  61
  62    If we have only 32b atomics available, we cannot put both nwaiters and
  63    value into one 32b value because then we might have too few bits for both
  64    of those counters.  Therefore, we need to use two distinct fields.
  65
  66    To allow sem_post to atomically make a token available and check for
  67    blocked sem_wait calls, we use one bit in value to indicate whether
  68    nwaiters is nonzero.  That allows sem_post to use basically the same
  69    algorithm as with 64b atomics, but requires sem_wait to update the bit; it
  70    can't do this atomically with another access to nwaiters, but it can compute
  71    a conservative value for the bit because it's benign if the bit is set
  72    even if nwaiters is zero (all we get is an unnecessary futex wake call by
  73    sem_post).
  74    Specifically, sem_wait will unset the bit speculatively if it believes that
  75    there is no other concurrently executing sem_wait.  If it misspeculated,
  76    it will have to clean up by waking any other sem_wait call (i.e., what
  77    sem_post would do otherwise).  This does not conflict with the destruction
  78    requirement because the semaphore must not be destructed while any sem_wait
  79    is still executing.  */
  80
  81 #if !__HAVE_64B_ATOMICS
  82 static void
  83 __sem_wait_32_finish (struct new_sem *sem);
  84 #endif
  85
  86 static void
  87 __sem_wait_cleanup (void *arg)
  88 {
  89   struct new_sem *sem = (struct new_sem *) arg;
  90
  91 #if __HAVE_64B_ATOMICS
  92   /* Stop being registered as a waiter.  See below for MO.  */
  93   atomic_fetch_add_relaxed (&sem->data, -((uint64_t) 1 << SEM_NWAITERS_SHIFT));
  94 #else
  95   __sem_wait_32_finish (sem);
  96 #endif
  97 }
  98
  99 /* Wait until at least one token is available, possibly with a timeout.
 100    This is in a separate function in order to make sure gcc
 101    puts the call site into an exception region, and thus the
 102    cleanups get properly run.  TODO still necessary?  Other futex_wait
 103    users don't seem to need it.  */
 104 static int
 105 __attribute__ ((noinline))
 106 do_futex_wait (struct new_sem *sem, const struct timespec *abstime)
 107 {
 108   int err;
 109
 110 #if __HAVE_64B_ATOMICS
 111   err = futex_abstimed_wait_cancelable (
 112       (unsigned int *) &sem->data + SEM_VALUE_OFFSET, 0, abstime,
 113       sem->private);
 114 #else
 115   err = futex_abstimed_wait_cancelable (&sem->value, SEM_NWAITERS_MASK,
 116                                         abstime, sem->private);
 117 #endif
 118
 119   return err;
 120 }
 121
 122 /* Fast path: Try to grab a token without blocking.  */
 123 static int
 124 __new_sem_wait_fast (struct new_sem *sem, int definitive_result)
 125 {
 126   /* We need acquire MO if we actually grab a token, so that this
 127      synchronizes with all token providers (i.e., the RMW operation we read
 128      from or all those before it in modification order; also see sem_post).
 129      We do not need to guarantee any ordering if we observed that there is
 130      no token (POSIX leaves it unspecified whether functions that fail
 131      synchronize memory); thus, relaxed MO is sufficient for the initial load
 132      and the failure path of the CAS.  If the weak CAS fails and we need a
 133      definitive result, retry.  */
 134 #if __HAVE_64B_ATOMICS
 135   uint64_t d = atomic_load_relaxed (&sem->data);
 136   do
 137     {
 138       if ((d & SEM_VALUE_MASK) == 0)
 139         break;
 140       if (atomic_compare_exchange_weak_acquire (&sem->data, &d, d - 1))
 141         return 0;
 142     }
 143   while (definitive_result);
 144   return -1;
 145 #else
 146   unsigned int v = atomic_load_relaxed (&sem->value);
 147   do
 148     {
 149       if ((v >> SEM_VALUE_SHIFT) == 0)
 150         break;
 151       if (atomic_compare_exchange_weak_acquire (&sem->value,
 152           &v, v - (1 << SEM_VALUE_SHIFT)))
 153         return 0;
 154     }
 155   while (definitive_result);
 156   return -1;
 157 #endif
 158 }
 159
 160 /* Slow path that blocks.  */
 161 static int
 162 __attribute__ ((noinline))
 163 __new_sem_wait_slow (struct new_sem *sem, const struct timespec *abstime)
 164 {
 165   int err = 0;
 166
 167 #if __HAVE_64B_ATOMICS
 168   /* Add a waiter.  Relaxed MO is sufficient because we can rely on the
 169      ordering provided by the RMW operations we use.  */
 170   uint64_t d = atomic_fetch_add_relaxed (&sem->data,
 171       (uint64_t) 1 << SEM_NWAITERS_SHIFT);
 172
 173   pthread_cleanup_push (__sem_wait_cleanup, sem);
 174
 175   /* Wait for a token to be available.  Retry until we can grab one.  */
 176   for (;;)
 177     {
 178       /* If there is no token available, sleep until there is.  */
 179       if ((d & SEM_VALUE_MASK) == 0)
 180         {
 181           err = do_futex_wait (sem, abstime);
 182           /* A futex return value of 0 or EAGAIN is due to a real or spurious
 183              wake-up, or due to a change in the number of tokens.  We retry in
 184              these cases.
 185              If we timed out, forward this to the caller.
 186              EINTR is returned if we are interrupted by a signal; we
 187              forward this to the caller.  (See futex_wait and related
 188              documentation.  Before Linux 2.6.22, EINTR was also returned on
 189              spurious wake-ups; we only support more recent Linux versions,
 190              so do not need to consider this here.)  */
 191           if (err == ETIMEDOUT || err == EINTR)
 192             {
 193               __set_errno (err);
 194               err = -1;
 195               /* Stop being registered as a waiter.  */
 196               atomic_fetch_add_relaxed (&sem->data,
 197                   -((uint64_t) 1 << SEM_NWAITERS_SHIFT));
 198               break;
 199             }
 200           /* Relaxed MO is sufficient; see below.  */
 201           d = atomic_load_relaxed (&sem->data);
 202         }
 203       else
 204         {
 205           /* Try to grab both a token and stop being a waiter.  We need
 206              acquire MO so this synchronizes with all token providers (i.e.,
 207              the RMW operation we read from or all those before it in
 208              modification order; also see sem_post).  On the failure path,
 209              relaxed MO is sufficient because we only eventually need the
 210              up-to-date value; the futex_wait or the CAS perform the real
 211              work.  */
 212           if (atomic_compare_exchange_weak_acquire (&sem->data,
 213               &d, d - 1 - ((uint64_t) 1 << SEM_NWAITERS_SHIFT)))
 214             {
 215               err = 0;
 216               break;
 217             }
 218         }
 219     }
 220
 221   pthread_cleanup_pop (0);
 222 #else
 223   /* The main difference to the 64b-atomics implementation is that we need to
 224      access value and nwaiters in separate steps, and that the nwaiters bit
 225      in the value can temporarily not be set even if nwaiters is nonzero.
 226      We work around incorrectly unsetting the nwaiters bit by letting sem_wait
 227      set the bit again and waking the number of waiters that could grab a
 228      token.  There are two additional properties we need to ensure:
 229      (1) We make sure that whenever unsetting the bit, we see the increment of
 230      nwaiters by the other thread that set the bit.  IOW, we will notice if
 231      we make a mistake.
 232      (2) When setting the nwaiters bit, we make sure that we see the unsetting
 233      of the bit by another waiter that happened before us.  This avoids having
 234      to blindly set the bit whenever we need to block on it.  We set/unset
 235      the bit while having incremented nwaiters (i.e., are a registered
 236      waiter), and the problematic case only happens when one waiter indeed
 237      followed another (i.e., nwaiters was never larger than 1); thus, this
 238      works similarly as with a critical section using nwaiters (see the MOs
 239      and related comments below).
 240
 241      An alternative approach would be to unset the bit after decrementing
 242      nwaiters; however, that would result in needing Dekker-like
 243      synchronization and thus full memory barriers.  We also would not be able
 244      to prevent misspeculation, so this alternative scheme does not seem
 245      beneficial.  */
 246   unsigned int v;
 247
 248   /* Add a waiter.  We need acquire MO so this synchronizes with the release
 249      MO we use when decrementing nwaiters below; it ensures that if another
 250      waiter unset the bit before us, we see that and set it again.  Also see
 251      property (2) above.  */
 252   atomic_fetch_add_acquire (&sem->nwaiters, 1);
 253
 254   pthread_cleanup_push (__sem_wait_cleanup, sem);
 255
 256   /* Wait for a token to be available.  Retry until we can grab one.  */
 257   /* We do not need any ordering wrt. to this load's reads-from, so relaxed
 258      MO is sufficient.  The acquire MO above ensures that in the problematic
 259      case, we do see the unsetting of the bit by another waiter.  */
 260   v = atomic_load_relaxed (&sem->value);
 261   do
 262     {
 263       do
 264         {
 265           /* We are about to block, so make sure that the nwaiters bit is
 266              set.  We need release MO on the CAS to ensure that when another
 267              waiter unsets the nwaiters bit, it will also observe that we
 268              incremented nwaiters in the meantime (also see the unsetting of
 269              the bit below).  Relaxed MO on CAS failure is sufficient (see
 270              above).  */
 271           do
 272             {
 273               if ((v & SEM_NWAITERS_MASK) != 0)
 274                 break;
 275             }
 276           while (!atomic_compare_exchange_weak_release (&sem->value,
 277               &v, v | SEM_NWAITERS_MASK));
 278           /* If there is no token, wait.  */
 279           if ((v >> SEM_VALUE_SHIFT) == 0)
 280             {
 281               /* See __HAVE_64B_ATOMICS variant.  */
 282               err = do_futex_wait(sem, abstime);
 283               if (err == ETIMEDOUT || err == EINTR)
 284                 {
 285                   __set_errno (err);
 286                   err = -1;
 287                   goto error;
 288                 }
 289               err = 0;
 290               /* We blocked, so there might be a token now.  Relaxed MO is
 291                  sufficient (see above).  */
 292               v = atomic_load_relaxed (&sem->value);
 293             }
 294         }
 295       /* If there is no token, we must not try to grab one.  */
 296       while ((v >> SEM_VALUE_SHIFT) == 0);
 297     }
 298   /* Try to grab a token.  We need acquire MO so this synchronizes with
 299      all token providers (i.e., the RMW operation we read from or all those
 300      before it in modification order; also see sem_post).  */
 301   while (!atomic_compare_exchange_weak_acquire (&sem->value,
 302       &v, v - (1 << SEM_VALUE_SHIFT)));
 303
 304 error:
 305   pthread_cleanup_pop (0);
 306
 307   __sem_wait_32_finish (sem);
 308 #endif
 309
 310   return err;
 311 }
 312
 313 /* Stop being a registered waiter (non-64b-atomics code only).  */
 314 #if !__HAVE_64B_ATOMICS
 315 static void
 316 __sem_wait_32_finish (struct new_sem *sem)
 317 {
 318   /* The nwaiters bit is still set, try to unset it now if this seems
 319      necessary.  We do this before decrementing nwaiters so that the unsetting
 320      is visible to other waiters entering after us.  Relaxed MO is sufficient
 321      because we are just speculating here; a stronger MO would not prevent
 322      misspeculation.  */
 323   unsigned int wguess = atomic_load_relaxed (&sem->nwaiters);
 324   if (wguess == 1)
 325     /* We might be the last waiter, so unset.  This needs acquire MO so that
 326        it syncronizes with the release MO when setting the bit above; if we
 327        overwrite someone else that set the bit, we'll read in the following
 328        decrement of nwaiters at least from that release sequence, so we'll
 329        see if the other waiter is still active or if another writer entered
 330        in the meantime (i.e., using the check below).  */
 331     atomic_fetch_and_acquire (&sem->value, ~SEM_NWAITERS_MASK);
 332
 333   /* Now stop being a waiter, and see whether our guess was correct.
 334      This needs release MO so that it synchronizes with the acquire MO when
 335      a waiter increments nwaiters; this makes sure that newer writers see that
 336      we reset the waiters_present bit.  */
 337   unsigned int wfinal = atomic_fetch_add_release (&sem->nwaiters, -1);
 338   if (wfinal > 1 && wguess == 1)
 339     {
 340       /* We guessed wrong, and so need to clean up after the mistake and
 341          unblock any waiters that could have not been woken.  There is no
 342          additional ordering that we need to set up, so relaxed MO is
 343          sufficient.  */
 344       unsigned int v = atomic_fetch_or_relaxed (&sem->value,
 345                                                 SEM_NWAITERS_MASK);
 346       /* If there are available tokens, then wake as many waiters.  If there
 347          aren't any, then there is no need to wake anyone because there is
 348          none to grab for another waiter.  If tokens become available
 349          subsequently, then the respective sem_post calls will do the wake-up
 350          due to us having set the nwaiters bit again.  */
 351       v >>= SEM_VALUE_SHIFT;
 352       if (v > 0)
 353         futex_wake (&sem->value, v, sem->private);
 354     }
 355 }
 356 #endif