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_inner(ptr_t sig_arg);
 105
 106 #if defined(IA64) || defined(HP_PA)
 107 extern void GC_with_callee_saves_pushed();
 108
 109 void GC_suspend_handler(int sig)
 110 {
 111   GC_with_callee_saves_pushed(GC_suspend_handler_inner, (ptr_t)(word)sig);
 112 }
 113
 114 #else
 115 /* We believe that in all other cases the full context is already       */
 116 /* in the signal handler frame.                                         */
 117 void GC_suspend_handler(int sig)
 118 {
 119   GC_suspend_handler_inner((ptr_t)(word)sig);
 120 }
 121 #endif
 122
 123 void GC_suspend_handler_inner(ptr_t sig_arg)
 124 {
 125     int sig = (int)(word)sig_arg;
 126     int dummy;
 127     pthread_t my_thread = pthread_self();
 128     GC_thread me;
 129 #   ifdef PARALLEL_MARK
 130         word my_mark_no = GC_mark_no;
 131         /* Marker can't proceed until we acknowledge.  Thus this is     */
 132         /* guaranteed to be the mark_no correspending to our            */
 133         /* suspension, i.e. the marker can't have incremented it yet.   */
 134 #   endif
 135     word my_stop_count = GC_stop_count;
 136
 137     if (sig != SIG_SUSPEND) ABORT("Bad signal in suspend_handler");
 138
 139 #if DEBUG_THREADS
 140     GC_printf1("Suspending 0x%lx\n", my_thread);
 141 #endif
 142
 143     me = GC_lookup_thread(my_thread);
 144     /* The lookup here is safe, since I'm doing this on behalf  */
 145     /* of a thread which holds the allocation lock in order     */
 146     /* to stop the world.  Thus concurrent modification of the  */
 147     /* data structure is impossible.                            */
 148     if (me -> stop_info.last_stop_count == my_stop_count) {
 149         /* Duplicate signal.  OK if we are retrying.    */
 150         if (!GC_retry_signals) {
 151             WARN("Duplicate suspend signal in thread %lx\n",
 152                  pthread_self());
 153         }
 154         return;
 155     }
 156 #   ifdef SPARC
 157         me -> stop_info.stack_ptr = (ptr_t)GC_save_regs_in_stack();
 158 #   else
 159         me -> stop_info.stack_ptr = (ptr_t)(&dummy);
 160 #   endif
 161 #   ifdef IA64
 162         me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack();
 163 #   endif
 164
 165     /* Tell the thread that wants to stop the world that this   */
 166     /* thread has been stopped.  Note that sem_post() is        */
 167     /* the only async-signal-safe primitive in LinuxThreads.    */
 168     sem_post(&GC_suspend_ack_sem);
 169     me -> stop_info.last_stop_count = my_stop_count;
 170
 171     /* Wait until that thread tells us to restart by sending    */
 172     /* this thread a SIG_THR_RESTART signal.                    */
 173     /* SIG_THR_RESTART should be masked at this point.  Thus there      */
 174     /* is no race.                                              */
 175     do {
 176             me->stop_info.signal = 0;
 177             sigsuspend(&suspend_handler_mask);        /* Wait for signal */
 178     } while (me->stop_info.signal != SIG_THR_RESTART);
 179     /* If the RESTART signal gets lost, we can still lose.  That should be  */
 180     /* less likely than losing the SUSPEND signal, since we don't do much   */
 181     /* between the sem_post and sigsuspend.                                 */
 182     /* We'd need more handshaking to work around that, since we don't want  */
 183     /* to accidentally leave a RESTART signal pending, thus causing us to   */
 184     /* continue prematurely in a future round.                              */
 185
 186 #if DEBUG_THREADS
 187     GC_printf1("Continuing 0x%lx\n", my_thread);
 188 #endif
 189 }
 190
 191 void GC_restart_handler(int sig)
 192 {
 193     pthread_t my_thread = pthread_self();
 194     GC_thread me;
 195
 196     if (sig != SIG_THR_RESTART) ABORT("Bad signal in suspend_handler");
 197
 198     /* Let the GC_suspend_handler() know that we got a SIG_THR_RESTART. */
 199     /* The lookup here is safe, since I'm doing this on behalf  */
 200     /* of a thread which holds the allocation lock in order     */
 201     /* to stop the world.  Thus concurrent modification of the  */
 202     /* data structure is impossible.                            */
 203     me = GC_lookup_thread(my_thread);
 204     me->stop_info.signal = SIG_THR_RESTART;
 205
 206     /*
 207     ** Note: even if we didn't do anything useful here,
 208     ** it would still be necessary to have a signal handler,
 209     ** rather than ignoring the signals, otherwise
 210     ** the signals will not be delivered at all, and
 211     ** will thus not interrupt the sigsuspend() above.
 212     */
 213
 214 #if DEBUG_THREADS
 215     GC_printf1("In GC_restart_handler for 0x%lx\n", pthread_self());
 216 #endif
 217 }
 218
 219 # ifdef IA64
 220 #   define IF_IA64(x) x
 221 # else
 222 #   define IF_IA64(x)
 223 # endif
 224 /* We hold allocation lock.  Should do exactly the right thing if the   */
 225 /* world is stopped.  Should not fail if it isn't.                      */
 226 void GC_push_all_stacks()
 227 {
 228     GC_bool found_me = FALSE;
 229     int i;
 230     GC_thread p;
 231     ptr_t lo, hi;
 232     /* On IA64, we also need to scan the register backing store. */
 233     IF_IA64(ptr_t bs_lo; ptr_t bs_hi;)
 234     pthread_t me = pthread_self();
 235
 236     if (!GC_thr_initialized) GC_thr_init();
 237     #if DEBUG_THREADS
 238         GC_printf1("Pushing stacks from thread 0x%lx\n", (unsigned long) me);
 239     #endif
 240     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 241       for (p = GC_threads[i]; p != 0; p = p -> next) {
 242         if (p -> flags & FINISHED) continue;
 243         if (pthread_equal(p -> id, me)) {
 244 #           ifdef SPARC
 245                 lo = (ptr_t)GC_save_regs_in_stack();
 246 #           else
 247                 lo = GC_approx_sp();
 248 #           endif
 249             found_me = TRUE;
 250             IF_IA64(bs_hi = (ptr_t)GC_save_regs_in_stack();)
 251         } else {
 252             lo = p -> stop_info.stack_ptr;
 253             IF_IA64(bs_hi = p -> backing_store_ptr;)
 254         }
 255         if ((p -> flags & MAIN_THREAD) == 0) {
 256             hi = p -> stack_end;
 257             IF_IA64(bs_lo = p -> backing_store_end);
 258         } else {
 259             /* The original stack. */
 260             hi = GC_stackbottom;
 261             IF_IA64(bs_lo = BACKING_STORE_BASE;)
 262         }
 263         #if DEBUG_THREADS
 264             GC_printf3("Stack for thread 0x%lx = [%lx,%lx)\n",
 265                 (unsigned long) p -> id,
 266                 (unsigned long) lo, (unsigned long) hi);
 267         #endif
 268         if (0 == lo) ABORT("GC_push_all_stacks: sp not set!\n");
 269 #       ifdef STACK_GROWS_UP
 270           /* We got them backwards! */
 271           GC_push_all_stack(hi, lo);
 272 #       else
 273           GC_push_all_stack(lo, hi);
 274 #       endif
 275 #       ifdef IA64
 276 #         if DEBUG_THREADS
 277             GC_printf3("Reg stack for thread 0x%lx = [%lx,%lx)\n",
 278                 (unsigned long) p -> id,
 279                 (unsigned long) bs_lo, (unsigned long) bs_hi);
 280 #         endif
 281           if (pthread_equal(p -> id, me)) {
 282             GC_push_all_eager(bs_lo, bs_hi);
 283           } else {
 284             GC_push_all_stack(bs_lo, bs_hi);
 285           }
 286 #       endif
 287       }
 288     }
 289     if (!found_me && !GC_in_thread_creation)
 290       ABORT("Collecting from unknown thread.");
 291 }
 292
 293 /* There seems to be a very rare thread stopping problem.  To help us  */
 294 /* debug that, we save the ids of the stopping thread. */
 295 pthread_t GC_stopping_thread;
 296 int GC_stopping_pid;
 297
 298 /* We hold the allocation lock.  Suspend all threads that might */
 299 /* still be running.  Return the number of suspend signals that */
 300 /* were sent. */
 301 int GC_suspend_all()
 302 {
 303     int n_live_threads = 0;
 304     int i;
 305     GC_thread p;
 306     int result;
 307     pthread_t my_thread = pthread_self();
 308
 309     GC_stopping_thread = my_thread;    /* debugging only.      */
 310     GC_stopping_pid = getpid();                /* debugging only.      */
 311     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 312       for (p = GC_threads[i]; p != 0; p = p -> next) {
 313         if (p -> id != my_thread) {
 314             if (p -> flags & FINISHED) continue;
 315             if (p -> stop_info.last_stop_count == GC_stop_count) continue;
 316             if (p -> thread_blocked) /* Will wait */ continue;
 317             n_live_threads++;
 318             #if DEBUG_THREADS
 319               GC_printf1("Sending suspend signal to 0x%lx\n", p -> id);
 320             #endif
 321
 322         result = pthread_kill(p -> id, SIG_SUSPEND);
 323             switch(result) {
 324                 case ESRCH:
 325                     /* Not really there anymore.  Possible? */
 326                     n_live_threads--;
 327                     break;
 328                 case 0:
 329                     break;
 330                 default:
 331                     ABORT("pthread_kill failed");
 332             }
 333         }
 334       }
 335     }
 336     return n_live_threads;
 337 }
 338
 339 /* Caller holds allocation lock.        */
 340 void GC_stop_world()
 341 {
 342     int i;
 343     int n_live_threads;
 344     int code;
 345
 346     #if DEBUG_THREADS
 347     GC_printf1("Stopping the world from 0x%lx\n", pthread_self());
 348     #endif
 349
 350     /* Make sure all free list construction has stopped before we start. */
 351     /* No new construction can start, since free list construction is   */
 352     /* required to acquire and release the GC lock before it starts,    */
 353     /* and we have the lock.                                            */
 354 #   ifdef PARALLEL_MARK
 355       GC_acquire_mark_lock();
 356       GC_ASSERT(GC_fl_builder_count == 0);
 357       /* We should have previously waited for it to become zero. */
 358 #   endif /* PARALLEL_MARK */
 359     ++GC_stop_count;
 360     n_live_threads = GC_suspend_all();
 361
 362       if (GC_retry_signals) {
 363           unsigned long wait_usecs = 0;  /* Total wait since retry.     */
 364 #         define WAIT_UNIT 3000
 365 #         define RETRY_INTERVAL 100000
 366           for (;;) {
 367               int ack_count;
 368
 369               sem_getvalue(&GC_suspend_ack_sem, &ack_count);
 370               if (ack_count == n_live_threads) break;
 371               if (wait_usecs > RETRY_INTERVAL) {
 372                   int newly_sent = GC_suspend_all();
 373
 374 #                 ifdef CONDPRINT
 375                     if (GC_print_stats) {
 376                       GC_printf1("Resent %ld signals after timeout\n",
 377                                  newly_sent);
 378                     }
 379 #                 endif
 380                   sem_getvalue(&GC_suspend_ack_sem, &ack_count);
 381                   if (newly_sent < n_live_threads - ack_count) {
 382                       WARN("Lost some threads during GC_stop_world?!\n",0);
 383                       n_live_threads = ack_count + newly_sent;
 384                   }
 385                   wait_usecs = 0;
 386               }
 387               usleep(WAIT_UNIT);
 388               wait_usecs += WAIT_UNIT;
 389           }
 390       }
 391     for (i = 0; i < n_live_threads; i++) {
 392           while (0 != (code = sem_wait(&GC_suspend_ack_sem))) {
 393             if (errno != EINTR) {
 394               GC_err_printf1("Sem_wait returned %ld\n", (unsigned long)code);
 395               ABORT("sem_wait for handler failed");
 396             }
 397           }
 398     }
 399 #   ifdef PARALLEL_MARK
 400       GC_release_mark_lock();
 401 #   endif
 402     #if DEBUG_THREADS
 403       GC_printf1("World stopped from 0x%lx\n", pthread_self());
 404     #endif
 405     GC_stopping_thread = 0;  /* debugging only */
 406 }
 407
 408 /* Caller holds allocation lock, and has held it continuously since     */
 409 /* the world stopped.                                                   */
 410 void GC_start_world()
 411 {
 412     pthread_t my_thread = pthread_self();
 413     register int i;
 414     register GC_thread p;
 415     register int n_live_threads = 0;
 416     register int result;
 417
 418 #   if DEBUG_THREADS
 419       GC_printf0("World starting\n");
 420 #   endif
 421
 422     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 423       for (p = GC_threads[i]; p != 0; p = p -> next) {
 424         if (p -> id != my_thread) {
 425             if (p -> flags & FINISHED) continue;
 426             if (p -> thread_blocked) continue;
 427             n_live_threads++;
 428             #if DEBUG_THREADS
 429               GC_printf1("Sending restart signal to 0x%lx\n", p -> id);
 430             #endif
 431
 432         result = pthread_kill(p -> id, SIG_THR_RESTART);
 433             switch(result) {
 434                 case ESRCH:
 435                     /* Not really there anymore.  Possible? */
 436                     n_live_threads--;
 437                     break;
 438                 case 0:
 439                     break;
 440                 default:
 441                     ABORT("pthread_kill failed");
 442             }
 443         }
 444       }
 445     }
 446     #if DEBUG_THREADS
 447       GC_printf0("World started\n");
 448     #endif
 449 }
 450
 451 void GC_stop_init() {
 452     struct sigaction act;
 453
 454     if (sem_init(&GC_suspend_ack_sem, 0, 0) != 0)
 455         ABORT("sem_init failed");
 456
 457     act.sa_flags = SA_RESTART;
 458     if (sigfillset(&act.sa_mask) != 0) {
 459         ABORT("sigfillset() failed");
 460     }
 461     GC_remove_allowed_signals(&act.sa_mask);
 462     /* SIG_THR_RESTART is set in the resulting mask.            */
 463     /* It is unmasked by the handler when necessary.            */
 464     act.sa_handler = GC_suspend_handler;
 465     if (sigaction(SIG_SUSPEND, &act, NULL) != 0) {
 466         ABORT("Cannot set SIG_SUSPEND handler");
 467     }
 468
 469     act.sa_handler = GC_restart_handler;
 470     if (sigaction(SIG_THR_RESTART, &act, NULL) != 0) {
 471         ABORT("Cannot set SIG_THR_RESTART handler");
 472     }
 473
 474     /* Inititialize suspend_handler_mask. It excludes SIG_THR_RESTART. */
 475       if (sigfillset(&suspend_handler_mask) != 0) ABORT("sigfillset() failed");
 476       GC_remove_allowed_signals(&suspend_handler_mask);
 477       if (sigdelset(&suspend_handler_mask, SIG_THR_RESTART) != 0)
 478           ABORT("sigdelset() failed");
 479
 480     /* Check for GC_RETRY_SIGNALS.      */
 481       if (0 != GETENV("GC_RETRY_SIGNALS")) {
 482           GC_retry_signals = TRUE;
 483       }
 484       if (0 != GETENV("GC_NO_RETRY_SIGNALS")) {
 485           GC_retry_signals = FALSE;
 486       }
 487 #     ifdef CONDPRINT
 488           if (GC_print_stats && GC_retry_signals) {
 489               GC_printf0("Will retry suspend signal if necessary.\n");
 490           }
 491 #     endif
 492 }
 493
 494 #endif