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_WIN32_THREADS) && !defined(GC_DARWIN_THREADS)
   5
   6 #include <signal.h>
   7 #include <semaphore.h>
   8 #include <errno.h>
   9 #include <unistd.h>
  10 #include <sys/time.h>
  11 #ifndef HPUX
  12 # include <sys/select.h>
  13   /* Doesn't exist on HP/UX 11.11. */
  14 #endif
  15
  16 void suspend_self();
  17
  18 #if DEBUG_THREADS
  19
  20 #ifndef NSIG
  21 # if defined(MAXSIG)
  22 #  define NSIG (MAXSIG+1)
  23 # elif defined(_NSIG)
  24 #  define NSIG _NSIG
  25 # elif defined(__SIGRTMAX)
  26 #  define NSIG (__SIGRTMAX+1)
  27 # else
  28   --> please fix it
  29 # endif
  30 #endif
  31
  32 void GC_print_sig_mask()
  33 {
  34     sigset_t blocked;
  35     int i;
  36
  37     if (pthread_sigmask(SIG_BLOCK, NULL, &blocked) != 0)
  38         ABORT("pthread_sigmask");
  39     GC_printf0("Blocked: ");
  40     for (i = 1; i < NSIG; i++) {
  41         if (sigismember(&blocked, i)) { GC_printf1("%ld ",(long) i); }
  42     }
  43     GC_printf0("\n");
  44 }
  45
  46 #endif
  47
  48 /* Remove the signals that we want to allow in thread stopping  */
  49 /* handler from a set.                                          */
  50 void GC_remove_allowed_signals(sigset_t *set)
  51 {
  52 #   ifdef NO_SIGNALS
  53       if (sigdelset(set, SIGINT) != 0
  54           || sigdelset(set, SIGQUIT) != 0
  55           || sigdelset(set, SIGABRT) != 0
  56           || sigdelset(set, SIGTERM) != 0) {
  57         ABORT("sigdelset() failed");
  58       }
  59 #   endif
  60
  61 #   ifdef MPROTECT_VDB
  62       /* Handlers write to the thread structure, which is in the heap,  */
  63       /* and hence can trigger a protection fault.                      */
  64       if (sigdelset(set, SIGSEGV) != 0
  65 #         ifdef SIGBUS
  66             || sigdelset(set, SIGBUS) != 0
  67 #         endif
  68           ) {
  69         ABORT("sigdelset() failed");
  70       }
  71 #   endif
  72 }
  73
  74 static sigset_t suspend_handler_mask;
  75
  76 volatile sig_atomic_t GC_stop_count;
  77                         /* Incremented at the beginning of GC_stop_world. */
  78
  79 volatile sig_atomic_t GC_world_is_stopped = FALSE;
  80                         /* FALSE ==> it is safe for threads to restart, i.e. */
  81                         /* they will see another suspend signal before they  */
  82                         /* are expected to stop (unless they have voluntarily */
  83                         /* stopped).                                         */
  84
  85 void GC_brief_async_signal_safe_sleep()
  86 {
  87     struct timeval tv;
  88     tv.tv_sec = 0;
  89     tv.tv_usec = 1000 * TIME_LIMIT / 2;
  90     select(0, 0, 0, 0, &tv);
  91 }
  92
  93 #ifdef GC_OSF1_THREADS
  94   GC_bool GC_retry_signals = TRUE;
  95 #else
  96   GC_bool GC_retry_signals = FALSE;
  97 #endif
  98
  99 /*
 100  * We use signals to stop threads during GC.
 101  *
 102  * Suspended threads wait in signal handler for SIG_THR_RESTART.
 103  * That's more portable than semaphores or condition variables.
 104  * (We do use sem_post from a signal handler, but that should be portable.)
 105  *
 106  * The thread suspension signal SIG_SUSPEND is now defined in gc_priv.h.
 107  * Note that we can't just stop a thread; we need it to save its stack
 108  * pointer(s) and acknowledge.
 109  */
 110
 111 #ifndef SIG_THR_RESTART
 112 #  if defined(GC_HPUX_THREADS) || defined(GC_OSF1_THREADS)
 113 #    ifdef _SIGRTMIN
 114 #      define SIG_THR_RESTART _SIGRTMIN + 5
 115 #    else
 116 #      define SIG_THR_RESTART SIGRTMIN + 5
 117 #    endif
 118 #  else
 119 #   define SIG_THR_RESTART SIGXCPU
 120 #  endif
 121 #endif
 122
 123 sem_t GC_suspend_ack_sem;
 124
 125 void GC_suspend_handler_inner(ptr_t sig_arg);
 126
 127 #if defined(IA64) || defined(HP_PA) || defined(M68K)
 128 extern void GC_with_callee_saves_pushed();
 129
 130 void GC_suspend_handler(int sig)
 131 {
 132   GC_thread me = GC_lookup_thread (pthread_self());
 133   if (me -> flags & SUSPENDED)
 134     suspend_self();
 135   else {
 136     int old_errno = errno;
 137     GC_with_callee_saves_pushed(GC_suspend_handler_inner, (ptr_t)(word)sig);
 138     errno = old_errno;
 139   }
 140 }
 141
 142 #else
 143 /* We believe that in all other cases the full context is already       */
 144 /* in the signal handler frame.                                         */
 145 void GC_suspend_handler(int sig)
 146 {
 147   GC_thread me = GC_lookup_thread(pthread_self());
 148   if (me -> flags & SUSPENDED)
 149     suspend_self();
 150   else {
 151     int old_errno = errno;
 152     GC_suspend_handler_inner((ptr_t)(word)sig);
 153     errno = old_errno;
 154   }
 155 }
 156 #endif
 157
 158 void GC_suspend_handler_inner(ptr_t sig_arg)
 159 {
 160     int sig = (int)(word)sig_arg;
 161     int dummy;
 162     pthread_t my_thread = pthread_self();
 163     GC_thread me;
 164 #   ifdef PARALLEL_MARK
 165         word my_mark_no = GC_mark_no;
 166         /* Marker can't proceed until we acknowledge.  Thus this is     */
 167         /* guaranteed to be the mark_no correspending to our            */
 168         /* suspension, i.e. the marker can't have incremented it yet.   */
 169 #   endif
 170     word my_stop_count = GC_stop_count;
 171
 172     if (sig != SIG_SUSPEND) ABORT("Bad signal in suspend_handler");
 173
 174 #if DEBUG_THREADS
 175     GC_printf1("Suspending 0x%lx\n", my_thread);
 176 #endif
 177
 178     me = GC_lookup_thread(my_thread);
 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     if (me -> stop_info.last_stop_count == my_stop_count) {
 184         /* Duplicate signal.  OK if we are retrying.    */
 185         if (!GC_retry_signals) {
 186             WARN("Duplicate suspend signal in thread %lx\n",
 187                  pthread_self());
 188         }
 189         return;
 190     }
 191 #   ifdef SPARC
 192         me -> stop_info.stack_ptr = (ptr_t)GC_save_regs_in_stack();
 193 #   else
 194         me -> stop_info.stack_ptr = (ptr_t)(&dummy);
 195 #   endif
 196 #   ifdef IA64
 197         me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack();
 198 #   endif
 199
 200     /* Tell the thread that wants to stop the world that this   */
 201     /* thread has been stopped.  Note that sem_post() is        */
 202     /* the only async-signal-safe primitive in LinuxThreads.    */
 203     sem_post(&GC_suspend_ack_sem);
 204     me -> stop_info.last_stop_count = my_stop_count;
 205
 206     /* Wait until that thread tells us to restart by sending    */
 207     /* this thread a SIG_THR_RESTART signal.                    */
 208     /* SIG_THR_RESTART should be masked at this point.  Thus there      */
 209     /* is no race.                                              */
 210     /* We do not continue until we receive a SIG_THR_RESTART,   */
 211     /* but we do not take that as authoritative.  (We may be    */
 212     /* accidentally restarted by one of the user signals we     */
 213     /* don't block.)  After we receive the signal, we use a     */
 214     /* primitive and expensive mechanism to wait until it's     */
 215     /* really safe to proceed.  Under normal circumstances,     */
 216     /* this code should not be executed.                        */
 217     sigsuspend(&suspend_handler_mask);        /* Wait for signal */
 218     while (GC_world_is_stopped && GC_stop_count == my_stop_count) {
 219         GC_brief_async_signal_safe_sleep();
 220 #       if DEBUG_THREADS
 221           GC_err_printf0("Sleeping in signal handler");
 222 #       endif
 223     }
 224     /* If the RESTART signal gets lost, we can still lose.  That should be  */
 225     /* less likely than losing the SUSPEND signal, since we don't do much   */
 226     /* between the sem_post and sigsuspend.                                 */
 227     /* We'd need more handshaking to work around that.                      */
 228     /* Simply dropping the sigsuspend call should be safe, but is unlikely  */
 229     /* to be efficient.                                                     */
 230
 231 #if DEBUG_THREADS
 232     GC_printf1("Continuing 0x%lx\n", my_thread);
 233 #endif
 234 }
 235
 236 void GC_restart_handler(int sig)
 237 {
 238     pthread_t my_thread = pthread_self();
 239
 240     if (sig != SIG_THR_RESTART) ABORT("Bad signal in suspend_handler");
 241
 242     /*
 243     ** Note: even if we don't do anything useful here,
 244     ** it would still be necessary to have a signal handler,
 245     ** rather than ignoring the signals, otherwise
 246     ** the signals will not be delivered at all, and
 247     ** will thus not interrupt the sigsuspend() above.
 248     */
 249
 250 #if DEBUG_THREADS
 251     GC_printf1("In GC_restart_handler for 0x%lx\n", pthread_self());
 252 #endif
 253 }
 254
 255 # ifdef IA64
 256 #   define IF_IA64(x) x
 257 # else
 258 #   define IF_IA64(x)
 259 # endif
 260 /* We hold allocation lock.  Should do exactly the right thing if the   */
 261 /* world is stopped.  Should not fail if it isn't.                      */
 262 void GC_push_all_stacks()
 263 {
 264     GC_bool found_me = FALSE;
 265     int i;
 266     GC_thread p;
 267     ptr_t lo, hi;
 268     /* On IA64, we also need to scan the register backing store. */
 269     IF_IA64(ptr_t bs_lo; ptr_t bs_hi;)
 270     pthread_t me = pthread_self();
 271
 272     if (!GC_thr_initialized) GC_thr_init();
 273     #if DEBUG_THREADS
 274         GC_printf1("Pushing stacks from thread 0x%lx\n", (unsigned long) me);
 275     #endif
 276     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 277       for (p = GC_threads[i]; p != 0; p = p -> next) {
 278         if (p -> flags & FINISHED) continue;
 279         if (pthread_equal(p -> id, me)) {
 280 #           ifdef SPARC
 281                 lo = (ptr_t)GC_save_regs_in_stack();
 282 #           else
 283                 lo = GC_approx_sp();
 284 #           endif
 285             found_me = TRUE;
 286             IF_IA64(bs_hi = (ptr_t)GC_save_regs_in_stack();)
 287         } else {
 288             lo = p -> stop_info.stack_ptr;
 289             IF_IA64(bs_hi = p -> backing_store_ptr;)
 290         }
 291         if ((p -> flags & MAIN_THREAD) == 0) {
 292             hi = p -> stack_end;
 293             IF_IA64(bs_lo = p -> backing_store_end);
 294         } else {
 295             /* The original stack. */
 296             hi = GC_stackbottom;
 297             IF_IA64(bs_lo = BACKING_STORE_BASE;)
 298         }
 299         #if DEBUG_THREADS
 300             GC_printf3("Stack for thread 0x%lx = [%lx,%lx)\n",
 301                 (unsigned long) p -> id,
 302                 (unsigned long) lo, (unsigned long) hi);
 303         #endif
 304         if (0 == lo) ABORT("GC_push_all_stacks: sp not set!\n");
 305 #       ifdef STACK_GROWS_UP
 306           /* We got them backwards! */
 307           GC_push_all_stack(hi, lo);
 308 #       else
 309           GC_push_all_stack(lo, hi);
 310 #       endif
 311 #       ifdef IA64
 312 #         if DEBUG_THREADS
 313             GC_printf3("Reg stack for thread 0x%lx = [%lx,%lx)\n",
 314                 (unsigned long) p -> id,
 315                 (unsigned long) bs_lo, (unsigned long) bs_hi);
 316 #         endif
 317           if (pthread_equal(p -> id, me)) {
 318             GC_push_all_eager(bs_lo, bs_hi);
 319           } else {
 320             GC_push_all_stack(bs_lo, bs_hi);
 321           }
 322 #       endif
 323       }
 324     }
 325     if (!found_me && !GC_in_thread_creation)
 326       ABORT("Collecting from unknown thread.");
 327 }
 328
 329 /* There seems to be a very rare thread stopping problem.  To help us  */
 330 /* debug that, we save the ids of the stopping thread. */
 331 pthread_t GC_stopping_thread;
 332 int GC_stopping_pid;
 333
 334 /* We hold the allocation lock.  Suspend all threads that might */
 335 /* still be running.  Return the number of suspend signals that */
 336 /* were sent. */
 337 int GC_suspend_all()
 338 {
 339     int n_live_threads = 0;
 340     int i;
 341     GC_thread p;
 342     int result;
 343     pthread_t my_thread = pthread_self();
 344
 345     GC_stopping_thread = my_thread;    /* debugging only.      */
 346     GC_stopping_pid = getpid();                /* debugging only.      */
 347     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 348       for (p = GC_threads[i]; p != 0; p = p -> next) {
 349         if (p -> id != my_thread) {
 350             if (p -> flags & FINISHED) continue;
 351             if (p -> stop_info.last_stop_count == GC_stop_count) continue;
 352             if (p -> thread_blocked) /* Will wait */ continue;
 353             n_live_threads++;
 354             #if DEBUG_THREADS
 355               GC_printf1("Sending suspend signal to 0x%lx\n", p -> id);
 356             #endif
 357
 358         result = pthread_kill(p -> id, SIG_SUSPEND);
 359             switch(result) {
 360                 case ESRCH:
 361                     /* Not really there anymore.  Possible? */
 362                     n_live_threads--;
 363                     break;
 364                 case 0:
 365                     break;
 366                 default:
 367                     ABORT("pthread_kill failed");
 368             }
 369         }
 370       }
 371     }
 372     return n_live_threads;
 373 }
 374
 375 /* Caller holds allocation lock.        */
 376 void GC_stop_world()
 377 {
 378     int i;
 379     int n_live_threads;
 380     int code;
 381
 382     #if DEBUG_THREADS
 383     GC_printf1("Stopping the world from 0x%lx\n", pthread_self());
 384     #endif
 385
 386     /* Make sure all free list construction has stopped before we start. */
 387     /* No new construction can start, since free list construction is   */
 388     /* required to acquire and release the GC lock before it starts,    */
 389     /* and we have the lock.                                            */
 390 #   ifdef PARALLEL_MARK
 391       GC_acquire_mark_lock();
 392       GC_ASSERT(GC_fl_builder_count == 0);
 393       /* We should have previously waited for it to become zero. */
 394 #   endif /* PARALLEL_MARK */
 395     ++GC_stop_count;
 396     GC_world_is_stopped = TRUE;
 397     n_live_threads = GC_suspend_all();
 398
 399       if (GC_retry_signals) {
 400           unsigned long wait_usecs = 0;  /* Total wait since retry.     */
 401 #         define WAIT_UNIT 3000
 402 #         define RETRY_INTERVAL 100000
 403           for (;;) {
 404               int ack_count;
 405
 406               sem_getvalue(&GC_suspend_ack_sem, &ack_count);
 407               if (ack_count == n_live_threads) break;
 408               if (wait_usecs > RETRY_INTERVAL) {
 409                   int newly_sent = GC_suspend_all();
 410
 411 #                 ifdef CONDPRINT
 412                     if (GC_print_stats) {
 413                       GC_printf1("Resent %ld signals after timeout\n",
 414                                  newly_sent);
 415                     }
 416 #                 endif
 417                   sem_getvalue(&GC_suspend_ack_sem, &ack_count);
 418                   if (newly_sent < n_live_threads - ack_count) {
 419                       WARN("Lost some threads during GC_stop_world?!\n",0);
 420                       n_live_threads = ack_count + newly_sent;
 421                   }
 422                   wait_usecs = 0;
 423               }
 424               usleep(WAIT_UNIT);
 425               wait_usecs += WAIT_UNIT;
 426           }
 427       }
 428     for (i = 0; i < n_live_threads; i++) {
 429           while (0 != (code = sem_wait(&GC_suspend_ack_sem))) {
 430               if (errno != EINTR) {
 431                  GC_err_printf1("Sem_wait returned %ld\n", (unsigned long)code);
 432                  ABORT("sem_wait for handler failed");
 433               }
 434           }
 435     }
 436 #   ifdef PARALLEL_MARK
 437       GC_release_mark_lock();
 438 #   endif
 439     #if DEBUG_THREADS
 440       GC_printf1("World stopped from 0x%lx\n", pthread_self());
 441     #endif
 442     GC_stopping_thread = 0;  /* debugging only */
 443 }
 444
 445 void suspend_self() {
 446   GC_thread me = GC_lookup_thread(pthread_self());
 447   if (me == NULL)
 448     ABORT("attempting to suspend unknown thread");
 449
 450   me -> flags |= SUSPENDED;
 451   GC_start_blocking();
 452   while (me -> flags & SUSPENDED)
 453     GC_brief_async_signal_safe_sleep();
 454   GC_end_blocking();
 455 }
 456
 457 void GC_suspend_thread(pthread_t thread) {
 458   if (thread == pthread_self())
 459     suspend_self();
 460   else {
 461     int result;
 462     GC_thread t = GC_lookup_thread(thread);
 463     if (t == NULL)
 464       ABORT("attempting to suspend unknown thread");
 465
 466     t -> flags |= SUSPENDED;
 467     result = pthread_kill (t -> id, SIG_SUSPEND);
 468     switch (result) {
 469     case ESRCH:
 470     case 0:
 471       break;
 472     default:
 473       ABORT("pthread_kill failed");
 474     }
 475   }
 476 }
 477
 478 void GC_resume_thread(pthread_t thread) {
 479   GC_thread t = GC_lookup_thread(thread);
 480   if (t == NULL)
 481     ABORT("attempting to resume unknown thread");
 482
 483   t -> flags &= ~SUSPENDED;
 484 }
 485
 486 /* Caller holds allocation lock, and has held it continuously since     */
 487 /* the world stopped.                                                   */
 488 void GC_start_world()
 489 {
 490     pthread_t my_thread = pthread_self();
 491     register int i;
 492     register GC_thread p;
 493     register int n_live_threads = 0;
 494     register int result;
 495
 496 #   if DEBUG_THREADS
 497       GC_printf0("World starting\n");
 498 #   endif
 499
 500     GC_world_is_stopped = FALSE;
 501     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 502       for (p = GC_threads[i]; p != 0; p = p -> next) {
 503         if (p -> id != my_thread) {
 504             if (p -> flags & FINISHED) continue;
 505             if (p -> thread_blocked) continue;
 506             n_live_threads++;
 507             #if DEBUG_THREADS
 508               GC_printf1("Sending restart signal to 0x%lx\n", p -> id);
 509             #endif
 510             result = pthread_kill(p -> id, SIG_THR_RESTART);
 511             switch(result) {
 512                 case ESRCH:
 513                     /* Not really there anymore.  Possible? */
 514                     n_live_threads--;
 515                     break;
 516                 case 0:
 517                     break;
 518                 default:
 519                     ABORT("pthread_kill failed");
 520             }
 521         }
 522       }
 523     }
 524     #if DEBUG_THREADS
 525       GC_printf0("World started\n");
 526     #endif
 527 }
 528
 529 void GC_stop_init() {
 530     struct sigaction act;
 531
 532     if (sem_init(&GC_suspend_ack_sem, 0, 0) != 0)
 533         ABORT("sem_init failed");
 534
 535     act.sa_flags = SA_RESTART;
 536     if (sigfillset(&act.sa_mask) != 0) {
 537         ABORT("sigfillset() failed");
 538     }
 539     GC_remove_allowed_signals(&act.sa_mask);
 540     /* SIG_THR_RESTART is set in the resulting mask.            */
 541     /* It is unmasked by the handler when necessary.            */
 542     act.sa_handler = GC_suspend_handler;
 543     if (sigaction(SIG_SUSPEND, &act, NULL) != 0) {
 544         ABORT("Cannot set SIG_SUSPEND handler");
 545     }
 546
 547     act.sa_handler = GC_restart_handler;
 548     if (sigaction(SIG_THR_RESTART, &act, NULL) != 0) {
 549         ABORT("Cannot set SIG_THR_RESTART handler");
 550     }
 551
 552     /* Inititialize suspend_handler_mask. It excludes SIG_THR_RESTART. */
 553       if (sigfillset(&suspend_handler_mask) != 0) ABORT("sigfillset() failed");
 554       GC_remove_allowed_signals(&suspend_handler_mask);
 555       if (sigdelset(&suspend_handler_mask, SIG_THR_RESTART) != 0)
 556           ABORT("sigdelset() failed");
 557
 558     /* Check for GC_RETRY_SIGNALS.      */
 559       if (0 != GETENV("GC_RETRY_SIGNALS")) {
 560           GC_retry_signals = TRUE;
 561       }
 562       if (0 != GETENV("GC_NO_RETRY_SIGNALS")) {
 563           GC_retry_signals = FALSE;
 564       }
 565 #     ifdef CONDPRINT
 566           if (GC_print_stats && GC_retry_signals) {
 567               GC_printf0("Will retry suspend signal if necessary.\n");
 568           }
 569 #     endif
 570 }
 571
 572 #endif