boehm-gc/pthread_stop_world.c

   1 #include "private/pthread_support.h"
   2
   3 #if defined(GC_PTHREADS) && !defined(GC_SOLARIS_THREADS) \
   4      && !defined(GC_IRIX_THREADS) && !defined(GC_WIN32_THREADS) \
   5      && !defined(GC_DARWIN_THREADS) && !defined(GC_AIX_THREADS)
   6
   7 #include <signal.h>
   8 #include <semaphore.h>
   9 #include <errno.h>
  10 #include <unistd.h>
  11
  12 #if DEBUG_THREADS
  13
  14 #ifndef NSIG
  15 # if defined(MAXSIG)
  16 #  define NSIG (MAXSIG+1)
  17 # elif defined(_NSIG)
  18 #  define NSIG _NSIG
  19 # elif defined(__SIGRTMAX)
  20 #  define NSIG (__SIGRTMAX+1)
  21 # else
  22   --> please fix it
  23 # endif
  24 #endif
  25
  26 void GC_print_sig_mask()
  27 {
  28     sigset_t blocked;
  29     int i;
  30
  31     if (pthread_sigmask(SIG_BLOCK, NULL, &blocked) != 0)
  32         ABORT("pthread_sigmask");
  33     GC_printf0("Blocked: ");
  34     for (i = 1; i < NSIG; i++) {
  35         if (sigismember(&blocked, i)) { GC_printf1("%ld ",(long) i); }
  36     }
  37     GC_printf0("\n");
  38 }
  39
  40 #endif
  41
  42 /* Remove the signals that we want to allow in thread stopping  */
  43 /* handler from a set.                                          */
  44 void GC_remove_allowed_signals(sigset_t *set)
  45 {
  46 #   ifdef NO_SIGNALS
  47       if (sigdelset(set, SIGINT) != 0
  48           || sigdelset(set, SIGQUIT) != 0
  49           || sigdelset(set, SIGABRT) != 0
  50           || sigdelset(set, SIGTERM) != 0) {
  51         ABORT("sigdelset() failed");
  52       }
  53 #   endif
  54
  55 #   ifdef MPROTECT_VDB
  56       /* Handlers write to the thread structure, which is in the heap,  */
  57       /* and hence can trigger a protection fault.                      */
  58       if (sigdelset(set, SIGSEGV) != 0
  59 #         ifdef SIGBUS
  60             || sigdelset(set, SIGBUS) != 0
  61 #         endif
  62           ) {
  63         ABORT("sigdelset() failed");
  64       }
  65 #   endif
  66 }
  67
  68 static sigset_t suspend_handler_mask;
  69
  70 word GC_stop_count;     /* Incremented at the beginning of GC_stop_world. */
  71
  72 #ifdef GC_OSF1_THREADS
  73   GC_bool GC_retry_signals = TRUE;
  74 #else
  75   GC_bool GC_retry_signals = FALSE;
  76 #endif
  77
  78 /*
  79  * We use signals to stop threads during GC.
  80  *
  81  * Suspended threads wait in signal handler for SIG_THR_RESTART.
  82  * That's more portable than semaphores or condition variables.
  83  * (We do use sem_post from a signal handler, but that should be portable.)
  84  *
  85  * The thread suspension signal SIG_SUSPEND is now defined in gc_priv.h.
  86  * Note that we can't just stop a thread; we need it to save its stack
  87  * pointer(s) and acknowledge.
  88  */
  89
  90 #ifndef SIG_THR_RESTART
  91 #  if defined(GC_HPUX_THREADS) || defined(GC_OSF1_THREADS)
  92 #    ifdef _SIGRTMIN
  93 #      define SIG_THR_RESTART _SIGRTMIN + 5
  94 #    else
  95 #      define SIG_THR_RESTART SIGRTMIN + 5
  96 #    endif
  97 #  else
  98 #   define SIG_THR_RESTART SIGXCPU
  99 #  endif
 100 #endif
 101
 102 sem_t GC_suspend_ack_sem;
 103
 104 void GC_suspend_handler(int sig)
 105 {
 106     int dummy;
 107     pthread_t my_thread = pthread_self();
 108     GC_thread me;
 109 #   ifdef PARALLEL_MARK
 110         word my_mark_no = GC_mark_no;
 111         /* Marker can't proceed until we acknowledge.  Thus this is     */
 112         /* guaranteed to be the mark_no correspending to our            */
 113         /* suspension, i.e. the marker can't have incremented it yet.   */
 114 #   endif
 115     word my_stop_count = GC_stop_count;
 116
 117     if (sig != SIG_SUSPEND) ABORT("Bad signal in suspend_handler");
 118
 119 #if DEBUG_THREADS
 120     GC_printf1("Suspending 0x%lx\n", my_thread);
 121 #endif
 122
 123     me = GC_lookup_thread(my_thread);
 124     /* The lookup here is safe, since I'm doing this on behalf  */
 125     /* of a thread which holds the allocation lock in order     */
 126     /* to stop the world.  Thus concurrent modification of the  */
 127     /* data structure is impossible.                            */
 128     if (me -> stop_info.last_stop_count == my_stop_count) {
 129         /* Duplicate signal.  OK if we are retrying.    */
 130         if (!GC_retry_signals) {
 131             WARN("Duplicate suspend signal in thread %lx\n",
 132                  pthread_self());
 133         }
 134         return;
 135     }
 136 #   ifdef SPARC
 137         me -> stop_info.stack_ptr = (ptr_t)GC_save_regs_in_stack();
 138 #   else
 139         me -> stop_info.stack_ptr = (ptr_t)(&dummy);
 140 #   endif
 141 #   ifdef IA64
 142         me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack();
 143 #   endif
 144
 145     /* Tell the thread that wants to stop the world that this   */
 146     /* thread has been stopped.  Note that sem_post() is        */
 147     /* the only async-signal-safe primitive in LinuxThreads.    */
 148     sem_post(&GC_suspend_ack_sem);
 149     me -> stop_info.last_stop_count = my_stop_count;
 150
 151     /* Wait until that thread tells us to restart by sending    */
 152     /* this thread a SIG_THR_RESTART signal.                    */
 153     /* SIG_THR_RESTART should be masked at this point.  Thus there      */
 154     /* is no race.                                              */
 155     do {
 156             me->stop_info.signal = 0;
 157             sigsuspend(&suspend_handler_mask);        /* Wait for signal */
 158     } while (me->stop_info.signal != SIG_THR_RESTART);
 159     /* If the RESTART signal gets lost, we can still lose.  That should be  */
 160     /* less likely than losing the SUSPEND signal, since we don't do much   */
 161     /* between the sem_post and sigsuspend.                                 */
 162     /* We'd need more handshaking to work around that, since we don't want  */
 163     /* to accidentally leave a RESTART signal pending, thus causing us to   */
 164     /* continue prematurely in a future round.                              */
 165
 166 #if DEBUG_THREADS
 167     GC_printf1("Continuing 0x%lx\n", my_thread);
 168 #endif
 169 }
 170
 171 void GC_restart_handler(int sig)
 172 {
 173     pthread_t my_thread = pthread_self();
 174     GC_thread me;
 175
 176     if (sig != SIG_THR_RESTART) ABORT("Bad signal in suspend_handler");
 177
 178     /* Let the GC_suspend_handler() know that we got a SIG_THR_RESTART. */
 179     /* The lookup here is safe, since I'm doing this on behalf  */
 180     /* of a thread which holds the allocation lock in order     */
 181     /* to stop the world.  Thus concurrent modification of the  */
 182     /* data structure is impossible.                            */
 183     me = GC_lookup_thread(my_thread);
 184     me->stop_info.signal = SIG_THR_RESTART;
 185
 186     /*
 187     ** Note: even if we didn't do anything useful here,
 188     ** it would still be necessary to have a signal handler,
 189     ** rather than ignoring the signals, otherwise
 190     ** the signals will not be delivered at all, and
 191     ** will thus not interrupt the sigsuspend() above.
 192     */
 193
 194 #if DEBUG_THREADS
 195     GC_printf1("In GC_restart_handler for 0x%lx\n", pthread_self());
 196 #endif
 197 }
 198
 199 # ifdef IA64
 200 #   define IF_IA64(x) x
 201 # else
 202 #   define IF_IA64(x)
 203 # endif
 204 /* We hold allocation lock.  Should do exactly the right thing if the   */
 205 /* world is stopped.  Should not fail if it isn't.                      */
 206 void GC_push_all_stacks()
 207 {
 208     GC_bool found_me = FALSE;
 209     int i;
 210     GC_thread p;
 211     ptr_t lo, hi;
 212     /* On IA64, we also need to scan the register backing store. */
 213     IF_IA64(ptr_t bs_lo; ptr_t bs_hi;)
 214     pthread_t me = pthread_self();
 215
 216     if (!GC_thr_initialized) GC_thr_init();
 217     #if DEBUG_THREADS
 218         GC_printf1("Pushing stacks from thread 0x%lx\n", (unsigned long) me);
 219     #endif
 220     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 221       for (p = GC_threads[i]; p != 0; p = p -> next) {
 222         if (p -> flags & FINISHED) continue;
 223         if (pthread_equal(p -> id, me)) {
 224 #           ifdef SPARC
 225                 lo = (ptr_t)GC_save_regs_in_stack();
 226 #           else
 227                 lo = GC_approx_sp();
 228 #           endif
 229             found_me = TRUE;
 230             IF_IA64(bs_hi = (ptr_t)GC_save_regs_in_stack();)
 231         } else {
 232             lo = p -> stop_info.stack_ptr;
 233             IF_IA64(bs_hi = p -> backing_store_ptr;)
 234         }
 235         if ((p -> flags & MAIN_THREAD) == 0) {
 236             hi = p -> stack_end;
 237             IF_IA64(bs_lo = p -> backing_store_end);
 238         } else {
 239             /* The original stack. */
 240             hi = GC_stackbottom;
 241             IF_IA64(bs_lo = BACKING_STORE_BASE;)
 242         }
 243         #if DEBUG_THREADS
 244             GC_printf3("Stack for thread 0x%lx = [%lx,%lx)\n",
 245                 (unsigned long) p -> id,
 246                 (unsigned long) lo, (unsigned long) hi);
 247         #endif
 248         if (0 == lo) ABORT("GC_push_all_stacks: sp not set!\n");
 249 #       ifdef STACK_GROWS_UP
 250           /* We got them backwards! */
 251           GC_push_all_stack(hi, lo);
 252 #       else
 253           GC_push_all_stack(lo, hi);
 254 #       endif
 255 #       ifdef IA64
 256 #         if DEBUG_THREADS
 257             GC_printf3("Reg stack for thread 0x%lx = [%lx,%lx)\n",
 258                 (unsigned long) p -> id,
 259                 (unsigned long) bs_lo, (unsigned long) bs_hi);
 260 #         endif
 261           if (pthread_equal(p -> id, me)) {
 262             GC_push_all_eager(bs_lo, bs_hi);
 263           } else {
 264             GC_push_all_stack(bs_lo, bs_hi);
 265           }
 266 #       endif
 267       }
 268     }
 269     if (!found_me && !GC_in_thread_creation)
 270       ABORT("Collecting from unknown thread.");
 271 }
 272
 273 /* There seems to be a very rare thread stopping problem.  To help us  */
 274 /* debug that, we save the ids of the stopping thread. */
 275 pthread_t GC_stopping_thread;
 276 int GC_stopping_pid;
 277
 278 /* We hold the allocation lock.  Suspend all threads that might */
 279 /* still be running.  Return the number of suspend signals that */
 280 /* were sent. */
 281 int GC_suspend_all()
 282 {
 283     int n_live_threads = 0;
 284     int i;
 285     GC_thread p;
 286     int result;
 287     pthread_t my_thread = pthread_self();
 288
 289     GC_stopping_thread = my_thread;    /* debugging only.      */
 290     GC_stopping_pid = getpid();                /* debugging only.      */
 291     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 292       for (p = GC_threads[i]; p != 0; p = p -> next) {
 293         if (p -> id != my_thread) {
 294             if (p -> flags & FINISHED) continue;
 295             if (p -> stop_info.last_stop_count == GC_stop_count) continue;
 296             if (p -> thread_blocked) /* Will wait */ continue;
 297             n_live_threads++;
 298             #if DEBUG_THREADS
 299               GC_printf1("Sending suspend signal to 0x%lx\n", p -> id);
 300             #endif
 301
 302         result = pthread_kill(p -> id, SIG_SUSPEND);
 303             switch(result) {
 304                 case ESRCH:
 305                     /* Not really there anymore.  Possible? */
 306                     n_live_threads--;
 307                     break;
 308                 case 0:
 309                     break;
 310                 default:
 311                     ABORT("pthread_kill failed");
 312             }
 313         }
 314       }
 315     }
 316     return n_live_threads;
 317 }
 318
 319 /* Caller holds allocation lock.        */
 320 void GC_stop_world()
 321 {
 322     int i;
 323     int n_live_threads;
 324     int code;
 325
 326     #if DEBUG_THREADS
 327     GC_printf1("Stopping the world from 0x%lx\n", pthread_self());
 328     #endif
 329
 330     /* Make sure all free list construction has stopped before we start. */
 331     /* No new construction can start, since free list construction is   */
 332     /* required to acquire and release the GC lock before it starts,    */
 333     /* and we have the lock.                                            */
 334 #   ifdef PARALLEL_MARK
 335       GC_acquire_mark_lock();
 336       GC_ASSERT(GC_fl_builder_count == 0);
 337       /* We should have previously waited for it to become zero. */
 338 #   endif /* PARALLEL_MARK */
 339     ++GC_stop_count;
 340     n_live_threads = GC_suspend_all();
 341
 342       if (GC_retry_signals) {
 343           unsigned long wait_usecs = 0;  /* Total wait since retry.     */
 344 #         define WAIT_UNIT 3000
 345 #         define RETRY_INTERVAL 100000
 346           for (;;) {
 347               int ack_count;
 348
 349               sem_getvalue(&GC_suspend_ack_sem, &ack_count);
 350               if (ack_count == n_live_threads) break;
 351               if (wait_usecs > RETRY_INTERVAL) {
 352                   int newly_sent = GC_suspend_all();
 353
 354 #                 ifdef CONDPRINT
 355                     if (GC_print_stats) {
 356                       GC_printf1("Resent %ld signals after timeout\n",
 357                                  newly_sent);
 358                     }
 359 #                 endif
 360                   sem_getvalue(&GC_suspend_ack_sem, &ack_count);
 361                   if (newly_sent < n_live_threads - ack_count) {
 362                       WARN("Lost some threads during GC_stop_world?!\n",0);
 363                       n_live_threads = ack_count + newly_sent;
 364                   }
 365                   wait_usecs = 0;
 366               }
 367               usleep(WAIT_UNIT);
 368               wait_usecs += WAIT_UNIT;
 369           }
 370       }
 371     for (i = 0; i < n_live_threads; i++) {
 372           if (0 != (code = sem_wait(&GC_suspend_ack_sem))) {
 373               GC_err_printf1("Sem_wait returned %ld\n", (unsigned long)code);
 374               ABORT("sem_wait for handler failed");
 375           }
 376     }
 377 #   ifdef PARALLEL_MARK
 378       GC_release_mark_lock();
 379 #   endif
 380     #if DEBUG_THREADS
 381       GC_printf1("World stopped from 0x%lx\n", pthread_self());
 382     #endif
 383     GC_stopping_thread = 0;  /* debugging only */
 384 }
 385
 386 /* Caller holds allocation lock, and has held it continuously since     */
 387 /* the world stopped.                                                   */
 388 void GC_start_world()
 389 {
 390     pthread_t my_thread = pthread_self();
 391     register int i;
 392     register GC_thread p;
 393     register int n_live_threads = 0;
 394     register int result;
 395
 396 #   if DEBUG_THREADS
 397       GC_printf0("World starting\n");
 398 #   endif
 399
 400     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 401       for (p = GC_threads[i]; p != 0; p = p -> next) {
 402         if (p -> id != my_thread) {
 403             if (p -> flags & FINISHED) continue;
 404             if (p -> thread_blocked) continue;
 405             n_live_threads++;
 406             #if DEBUG_THREADS
 407               GC_printf1("Sending restart signal to 0x%lx\n", p -> id);
 408             #endif
 409
 410         result = pthread_kill(p -> id, SIG_THR_RESTART);
 411             switch(result) {
 412                 case ESRCH:
 413                     /* Not really there anymore.  Possible? */
 414                     n_live_threads--;
 415                     break;
 416                 case 0:
 417                     break;
 418                 default:
 419                     ABORT("pthread_kill failed");
 420             }
 421         }
 422       }
 423     }
 424     #if DEBUG_THREADS
 425       GC_printf0("World started\n");
 426     #endif
 427 }
 428
 429 void GC_stop_init() {
 430     struct sigaction act;
 431
 432     if (sem_init(&GC_suspend_ack_sem, 0, 0) != 0)
 433         ABORT("sem_init failed");
 434
 435     act.sa_flags = SA_RESTART;
 436     if (sigfillset(&act.sa_mask) != 0) {
 437         ABORT("sigfillset() failed");
 438     }
 439     GC_remove_allowed_signals(&act.sa_mask);
 440     /* SIG_THR_RESTART is set in the resulting mask.            */
 441     /* It is unmasked by the handler when necessary.            */
 442     act.sa_handler = GC_suspend_handler;
 443     if (sigaction(SIG_SUSPEND, &act, NULL) != 0) {
 444         ABORT("Cannot set SIG_SUSPEND handler");
 445     }
 446
 447     act.sa_handler = GC_restart_handler;
 448     if (sigaction(SIG_THR_RESTART, &act, NULL) != 0) {
 449         ABORT("Cannot set SIG_THR_RESTART handler");
 450     }
 451
 452     /* Inititialize suspend_handler_mask. It excludes SIG_THR_RESTART. */
 453       if (sigfillset(&suspend_handler_mask) != 0) ABORT("sigfillset() failed");
 454       GC_remove_allowed_signals(&suspend_handler_mask);
 455       if (sigdelset(&suspend_handler_mask, SIG_THR_RESTART) != 0)
 456           ABORT("sigdelset() failed");
 457
 458     /* Check for GC_RETRY_SIGNALS.      */
 459       if (0 != GETENV("GC_RETRY_SIGNALS")) {
 460           GC_retry_signals = TRUE;
 461       }
 462       if (0 != GETENV("GC_NO_RETRY_SIGNALS")) {
 463           GC_retry_signals = FALSE;
 464       }
 465 #     ifdef CONDPRINT
 466           if (GC_print_stats && GC_retry_signals) {
 467               GC_printf0("Will retry suspend signal if necessary.\n");
 468           }
 469 #     endif
 470 }
 471
 472 #endif