libgc/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 /* work around a dlopen issue (bug #75390), undefs to avoid warnings with redefinitions */
  13 #undef PACKAGE_BUGREPORT
  14 #undef PACKAGE_NAME
  15 #undef PACKAGE_STRING
  16 #undef PACKAGE_TARNAME
  17 #undef PACKAGE_VERSION
  18 #include "mono/utils/mono-compiler.h"
  19
  20 #ifdef MONO_DEBUGGER_SUPPORTED
  21 #include "include/libgc-mono-debugger.h"
  22 #endif
  23
  24 #if DEBUG_THREADS
  25
  26 #ifndef NSIG
  27 # if defined(MAXSIG)
  28 #  define NSIG (MAXSIG+1)
  29 # elif defined(_NSIG)
  30 #  define NSIG _NSIG
  31 # elif defined(__SIGRTMAX)
  32 #  define NSIG (__SIGRTMAX+1)
  33 # else
  34   --> please fix it
  35 # endif
  36 #endif
  37
  38 void GC_print_sig_mask()
  39 {
  40     sigset_t blocked;
  41     int i;
  42
  43     if (pthread_sigmask(SIG_BLOCK, NULL, &blocked) != 0)
  44         ABORT("pthread_sigmask");
  45     GC_printf0("Blocked: ");
  46     for (i = 1; i < NSIG; i++) {
  47         if (sigismember(&blocked, i)) { GC_printf1("%ld ",(long) i); }
  48     }
  49     GC_printf0("\n");
  50 }
  51
  52 #endif
  53
  54 /* Remove the signals that we want to allow in thread stopping  */
  55 /* handler from a set.                                          */
  56 void GC_remove_allowed_signals(sigset_t *set)
  57 {
  58 #   ifdef NO_SIGNALS
  59       if (sigdelset(set, SIGINT) != 0
  60           || sigdelset(set, SIGQUIT) != 0
  61           || sigdelset(set, SIGABRT) != 0
  62           || sigdelset(set, SIGTERM) != 0) {
  63         ABORT("sigdelset() failed");
  64       }
  65 #   endif
  66
  67 #   ifdef MPROTECT_VDB
  68       /* Handlers write to the thread structure, which is in the heap,  */
  69       /* and hence can trigger a protection fault.                      */
  70       if (sigdelset(set, SIGSEGV) != 0
  71 #         ifdef SIGBUS
  72             || sigdelset(set, SIGBUS) != 0
  73 #         endif
  74           ) {
  75         ABORT("sigdelset() failed");
  76       }
  77 #   endif
  78 }
  79
  80 static sigset_t suspend_handler_mask;
  81
  82 word GC_stop_count;     /* Incremented at the beginning of GC_stop_world. */
  83
  84 #ifdef GC_OSF1_THREADS
  85   GC_bool GC_retry_signals = TRUE;
  86 #else
  87   GC_bool GC_retry_signals = FALSE;
  88 #endif
  89
  90 /*
  91  * We use signals to stop threads during GC.
  92  *
  93  * Suspended threads wait in signal handler for SIG_THR_RESTART.
  94  * That's more portable than semaphores or condition variables.
  95  * (We do use sem_post from a signal handler, but that should be portable.)
  96  *
  97  * The thread suspension signal SIG_SUSPEND is now defined in gc_priv.h.
  98  * Note that we can't just stop a thread; we need it to save its stack
  99  * pointer(s) and acknowledge.
 100  */
 101
 102 #ifndef SIG_THR_RESTART
 103 #  if defined(GC_HPUX_THREADS) || defined(GC_OSF1_THREADS)
 104 #    ifdef _SIGRTMIN
 105 #      define SIG_THR_RESTART _SIGRTMIN + 5
 106 #    else
 107 #      define SIG_THR_RESTART SIGRTMIN + 5
 108 #    endif
 109 #  else
 110 #   define SIG_THR_RESTART SIGXCPU
 111 #  endif
 112 #endif
 113
 114 sem_t GC_suspend_ack_sem;
 115
 116 static void _GC_suspend_handler(int sig)
 117 {
 118     int dummy;
 119     pthread_t my_thread = pthread_self();
 120     GC_thread me;
 121 #   ifdef PARALLEL_MARK
 122         word my_mark_no = GC_mark_no;
 123         /* Marker can't proceed until we acknowledge.  Thus this is     */
 124         /* guaranteed to be the mark_no correspending to our            */
 125         /* suspension, i.e. the marker can't have incremented it yet.   */
 126 #   endif
 127     word my_stop_count = GC_stop_count;
 128
 129     if (sig != SIG_SUSPEND) ABORT("Bad signal in suspend_handler");
 130
 131 #if DEBUG_THREADS
 132     GC_printf1("Suspending 0x%lx\n", my_thread);
 133 #endif
 134
 135     me = GC_lookup_thread(my_thread);
 136     /* The lookup here is safe, since I'm doing this on behalf  */
 137     /* of a thread which holds the allocation lock in order     */
 138     /* to stop the world.  Thus concurrent modification of the  */
 139     /* data structure is impossible.                            */
 140     if (me -> stop_info.last_stop_count == my_stop_count) {
 141         /* Duplicate signal.  OK if we are retrying.    */
 142         if (!GC_retry_signals) {
 143             WARN("Duplicate suspend signal in thread %lx\n",
 144                  pthread_self());
 145         }
 146         return;
 147     }
 148 #   ifdef SPARC
 149         me -> stop_info.stack_ptr = (ptr_t)GC_save_regs_in_stack();
 150 #   else
 151         me -> stop_info.stack_ptr = (ptr_t)(&dummy);
 152 #   endif
 153 #   ifdef IA64
 154         me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack();
 155 #   endif
 156
 157     /* Tell the thread that wants to stop the world that this   */
 158     /* thread has been stopped.  Note that sem_post() is        */
 159     /* the only async-signal-safe primitive in LinuxThreads.    */
 160     sem_post(&GC_suspend_ack_sem);
 161     me -> stop_info.last_stop_count = my_stop_count;
 162
 163     /* Wait until that thread tells us to restart by sending    */
 164     /* this thread a SIG_THR_RESTART signal.                    */
 165     /* SIG_THR_RESTART should be masked at this point.  Thus there      */
 166     /* is no race.                                              */
 167     do {
 168             me->stop_info.signal = 0;
 169             sigsuspend(&suspend_handler_mask);        /* Wait for signal */
 170     } while (me->stop_info.signal != SIG_THR_RESTART);
 171     /* If the RESTART signal gets lost, we can still lose.  That should be  */
 172     /* less likely than losing the SUSPEND signal, since we don't do much   */
 173     /* between the sem_post and sigsuspend.                                 */
 174     /* We'd need more handshaking to work around that, since we don't want  */
 175     /* to accidentally leave a RESTART signal pending, thus causing us to   */
 176     /* continue prematurely in a future round.                              */
 177
 178     /* Tell the thread that wants to start the world that this  */
 179     /* thread has been started.  Note that sem_post() is        */
 180     /* the only async-signal-safe primitive in LinuxThreads.    */
 181     sem_post(&GC_suspend_ack_sem);
 182
 183
 184 #if DEBUG_THREADS
 185     GC_printf1("Continuing 0x%lx\n", my_thread);
 186 #endif
 187 }
 188
 189 void GC_suspend_handler(int sig)
 190 {
 191         int old_errno = errno;
 192         _GC_suspend_handler(sig);
 193         errno = old_errno;
 194 }
 195
 196 static void _GC_restart_handler(int sig)
 197 {
 198     pthread_t my_thread = pthread_self();
 199     GC_thread me;
 200
 201     if (sig != SIG_THR_RESTART) ABORT("Bad signal in suspend_handler");
 202
 203     /* Let the GC_suspend_handler() know that we got a SIG_THR_RESTART. */
 204     /* The lookup here is safe, since I'm doing this on behalf  */
 205     /* of a thread which holds the allocation lock in order     */
 206     /* to stop the world.  Thus concurrent modification of the  */
 207     /* data structure is impossible.                            */
 208     me = GC_lookup_thread(my_thread);
 209     me->stop_info.signal = SIG_THR_RESTART;
 210
 211     /*
 212     ** Note: even if we didn't do anything useful here,
 213     ** it would still be necessary to have a signal handler,
 214     ** rather than ignoring the signals, otherwise
 215     ** the signals will not be delivered at all, and
 216     ** will thus not interrupt the sigsuspend() above.
 217     */
 218
 219 #if DEBUG_THREADS
 220     GC_printf1("In GC_restart_handler for 0x%lx\n", pthread_self());
 221 #endif
 222 }
 223
 224 # ifdef IA64
 225 #   define IF_IA64(x) x
 226 # else
 227 #   define IF_IA64(x)
 228 # endif
 229 /* We hold allocation lock.  Should do exactly the right thing if the   */
 230 /* world is stopped.  Should not fail if it isn't.                      */
 231 static void pthread_push_all_stacks()
 232 {
 233     GC_bool found_me = FALSE;
 234     int i;
 235     GC_thread p;
 236     ptr_t lo, hi;
 237     /* On IA64, we also need to scan the register backing store. */
 238     IF_IA64(ptr_t bs_lo; ptr_t bs_hi;)
 239     pthread_t me = pthread_self();
 240
 241     if (!GC_thr_initialized) GC_thr_init();
 242     #if DEBUG_THREADS
 243         GC_printf1("Pushing stacks from thread 0x%lx\n", (unsigned long) me);
 244     #endif
 245     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 246       for (p = GC_threads[i]; p != 0; p = p -> next) {
 247         if (p -> flags & FINISHED) continue;
 248         if (pthread_equal(p -> id, me)) {
 249 #           ifdef SPARC
 250                 lo = (ptr_t)GC_save_regs_in_stack();
 251 #           else
 252                 lo = GC_approx_sp();
 253 #           endif
 254             found_me = TRUE;
 255             IF_IA64(bs_hi = (ptr_t)GC_save_regs_in_stack();)
 256         } else {
 257             lo = p -> stop_info.stack_ptr;
 258             IF_IA64(bs_hi = p -> backing_store_ptr;)
 259         }
 260         if ((p -> flags & MAIN_THREAD) == 0) {
 261             hi = p -> stack_end;
 262             IF_IA64(bs_lo = p -> backing_store_end);
 263         } else {
 264             /* The original stack. */
 265             hi = GC_stackbottom;
 266             IF_IA64(bs_lo = BACKING_STORE_BASE;)
 267         }
 268         #if DEBUG_THREADS
 269             GC_printf3("Stack for thread 0x%lx = [%lx,%lx)\n",
 270                 (unsigned long) p -> id,
 271                 (unsigned long) lo, (unsigned long) hi);
 272         #endif
 273         if (0 == lo) ABORT("GC_push_all_stacks: sp not set!\n");
 274 #       ifdef STACK_GROWS_UP
 275           /* We got them backwards! */
 276           GC_push_all_stack(hi, lo);
 277 #       else
 278           GC_push_all_stack(lo, hi);
 279 #       endif
 280 #       ifdef IA64
 281 #         if DEBUG_THREADS
 282             GC_printf3("Reg stack for thread 0x%lx = [%lx,%lx)\n",
 283                 (unsigned long) p -> id,
 284                 (unsigned long) bs_lo, (unsigned long) bs_hi);
 285 #         endif
 286           if (pthread_equal(p -> id, me)) {
 287             GC_push_all_eager(bs_lo, bs_hi);
 288           } else {
 289             GC_push_all_stack(bs_lo, bs_hi);
 290           }
 291 #       endif
 292       }
 293     }
 294     if (!found_me && !GC_in_thread_creation)
 295       ABORT("Collecting from unknown thread.");
 296 }
 297
 298 void GC_restart_handler(int sig)
 299 {
 300         int old_errno = errno;
 301         _GC_restart_handler (sig);
 302         errno = old_errno;
 303 }
 304
 305 /* We hold allocation lock.  Should do exactly the right thing if the   */
 306 /* world is stopped.  Should not fail if it isn't.                      */
 307 void GC_push_all_stacks()
 308 {
 309     pthread_push_all_stacks();
 310 }
 311
 312 /* There seems to be a very rare thread stopping problem.  To help us  */
 313 /* debug that, we save the ids of the stopping thread. */
 314 pthread_t GC_stopping_thread;
 315 int GC_stopping_pid;
 316
 317 #ifdef PLATFORM_ANDROID
 318 static
 319 int android_thread_kill(pid_t tid, int sig)
 320 {
 321     int  ret;
 322     int  old_errno = errno;
 323
 324     ret = tkill(tid, sig);
 325     if (ret < 0) {
 326         ret = errno;
 327         errno = old_errno;
 328     }
 329
 330     return ret;
 331 }
 332 #endif
 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 #ifndef PLATFORM_ANDROID
 359         result = pthread_kill(p -> id, SIG_SUSPEND);
 360 #else
 361         result = android_thread_kill(p -> kernel_id, SIG_SUSPEND);
 362 #endif
 363             switch(result) {
 364                 case ESRCH:
 365                     /* Not really there anymore.  Possible? */
 366                     n_live_threads--;
 367                     break;
 368                 case 0:
 369                     break;
 370                 default:
 371                     ABORT("pthread_kill failed");
 372             }
 373         }
 374       }
 375     }
 376     return n_live_threads;
 377 }
 378
 379 /* Caller holds allocation lock.        */
 380 static void pthread_stop_world()
 381 {
 382     int i;
 383     int n_live_threads;
 384     int code;
 385
 386     #if DEBUG_THREADS
 387     GC_printf1("Stopping the world from 0x%lx\n", pthread_self());
 388     #endif
 389
 390     n_live_threads = GC_suspend_all();
 391
 392       if (GC_retry_signals) {
 393           unsigned long wait_usecs = 0;  /* Total wait since retry.     */
 394 #         define WAIT_UNIT 3000
 395 #         define RETRY_INTERVAL 100000
 396           for (;;) {
 397               int ack_count;
 398
 399               sem_getvalue(&GC_suspend_ack_sem, &ack_count);
 400               if (ack_count == n_live_threads) break;
 401               if (wait_usecs > RETRY_INTERVAL) {
 402                   int newly_sent = GC_suspend_all();
 403
 404 #                 ifdef CONDPRINT
 405                     if (GC_print_stats) {
 406                       GC_printf1("Resent %ld signals after timeout\n",
 407                                  newly_sent);
 408                     }
 409 #                 endif
 410                   sem_getvalue(&GC_suspend_ack_sem, &ack_count);
 411                   if (newly_sent < n_live_threads - ack_count) {
 412                       WARN("Lost some threads during GC_stop_world?!\n",0);
 413                       n_live_threads = ack_count + newly_sent;
 414                   }
 415                   wait_usecs = 0;
 416               }
 417               usleep(WAIT_UNIT);
 418               wait_usecs += WAIT_UNIT;
 419           }
 420       }
 421     for (i = 0; i < n_live_threads; i++) {
 422           while (0 != (code = sem_wait(&GC_suspend_ack_sem))) {
 423               if (errno != EINTR) {
 424                  GC_err_printf1("Sem_wait returned %ld\n", (unsigned long)code);
 425                  ABORT("sem_wait for handler failed");
 426               }
 427           }
 428     }
 429     #if DEBUG_THREADS
 430       GC_printf1("World stopped from 0x%lx\n", pthread_self());
 431     #endif
 432     GC_stopping_thread = 0;  /* debugging only */
 433 }
 434
 435 /* Caller holds allocation lock.        */
 436 void GC_stop_world()
 437 {
 438     if (GC_notify_event)
 439         GC_notify_event (GC_EVENT_PRE_STOP_WORLD);
 440     /* Make sure all free list construction has stopped before we start. */
 441     /* No new construction can start, since free list construction is   */
 442     /* required to acquire and release the GC lock before it starts,    */
 443     /* and we have the lock.                                            */
 444 #   ifdef PARALLEL_MARK
 445       GC_acquire_mark_lock();
 446       GC_ASSERT(GC_fl_builder_count == 0);
 447       /* We should have previously waited for it to become zero. */
 448 #   endif /* PARALLEL_MARK */
 449     ++GC_stop_count;
 450 #ifdef MONO_DEBUGGER_SUPPORTED
 451     if (gc_thread_vtable && gc_thread_vtable->stop_world)
 452         gc_thread_vtable->stop_world ();
 453     else
 454 #endif
 455         pthread_stop_world ();
 456 #   ifdef PARALLEL_MARK
 457       GC_release_mark_lock();
 458 #   endif
 459     if (GC_notify_event)
 460         GC_notify_event (GC_EVENT_POST_STOP_WORLD);
 461 }
 462
 463 /* Caller holds allocation lock, and has held it continuously since     */
 464 /* the world stopped.                                                   */
 465 static void pthread_start_world()
 466 {
 467     pthread_t my_thread = pthread_self();
 468     register int i;
 469     register GC_thread p;
 470     register int n_live_threads = 0;
 471     register int result;
 472     int code;
 473
 474 #   if DEBUG_THREADS
 475       GC_printf0("World starting\n");
 476 #   endif
 477     if (GC_notify_event)
 478         GC_notify_event (GC_EVENT_PRE_START_WORLD);
 479
 480     for (i = 0; i < THREAD_TABLE_SZ; i++) {
 481       for (p = GC_threads[i]; p != 0; p = p -> next) {
 482         if (p -> id != my_thread) {
 483             if (p -> flags & FINISHED) continue;
 484             if (p -> thread_blocked) continue;
 485             n_live_threads++;
 486             #if DEBUG_THREADS
 487               GC_printf1("Sending restart signal to 0x%lx\n", p -> id);
 488             #endif
 489
 490 #ifndef PLATFORM_ANDROID
 491         result = pthread_kill(p -> id, SIG_THR_RESTART);
 492 #else
 493         result = android_thread_kill(p -> kernel_id, SIG_THR_RESTART);
 494 #endif
 495             switch(result) {
 496                 case ESRCH:
 497                     /* Not really there anymore.  Possible? */
 498                     n_live_threads--;
 499                     break;
 500                 case 0:
 501                     break;
 502                 default:
 503                     ABORT("pthread_kill failed");
 504             }
 505         }
 506       }
 507     }
 508
 509     #if DEBUG_THREADS
 510     GC_printf0 ("All threads signaled");
 511     #endif
 512
 513     for (i = 0; i < n_live_threads; i++) {
 514         while (0 != (code = sem_wait(&GC_suspend_ack_sem))) {
 515             if (errno != EINTR) {
 516                 GC_err_printf1("Sem_wait returned %ld\n", (unsigned long)code);
 517                 ABORT("sem_wait for handler failed");
 518             }
 519         }
 520     }
 521
 522     if (GC_notify_event)
 523         GC_notify_event (GC_EVENT_POST_START_WORLD);
 524     #if DEBUG_THREADS
 525       GC_printf0("World started\n");
 526     #endif
 527 }
 528
 529 void GC_start_world()
 530 {
 531 #ifdef MONO_DEBUGGER_SUPPORTED
 532     if (gc_thread_vtable && gc_thread_vtable->start_world)
 533         gc_thread_vtable->start_world();
 534     else
 535 #endif
 536         pthread_start_world ();
 537 }
 538
 539 static void pthread_stop_init() {
 540     struct sigaction act;
 541
 542     if (sem_init(&GC_suspend_ack_sem, 0, 0) != 0)
 543         ABORT("sem_init failed");
 544
 545     act.sa_flags = SA_RESTART;
 546     if (sigfillset(&act.sa_mask) != 0) {
 547         ABORT("sigfillset() failed");
 548     }
 549     GC_remove_allowed_signals(&act.sa_mask);
 550     /* SIG_THR_RESTART is set in the resulting mask.            */
 551     /* It is unmasked by the handler when necessary.            */
 552     act.sa_handler = GC_suspend_handler;
 553     if (sigaction(SIG_SUSPEND, &act, NULL) != 0) {
 554         ABORT("Cannot set SIG_SUSPEND handler");
 555     }
 556
 557     act.sa_handler = GC_restart_handler;
 558     if (sigaction(SIG_THR_RESTART, &act, NULL) != 0) {
 559         ABORT("Cannot set SIG_THR_RESTART handler");
 560     }
 561
 562     /* Inititialize suspend_handler_mask. It excludes SIG_THR_RESTART. */
 563       if (sigfillset(&suspend_handler_mask) != 0) ABORT("sigfillset() failed");
 564       GC_remove_allowed_signals(&suspend_handler_mask);
 565       if (sigdelset(&suspend_handler_mask, SIG_THR_RESTART) != 0)
 566           ABORT("sigdelset() failed");
 567
 568     /* Check for GC_RETRY_SIGNALS.      */
 569       if (0 != GETENV("GC_RETRY_SIGNALS")) {
 570           GC_retry_signals = TRUE;
 571       }
 572       if (0 != GETENV("GC_NO_RETRY_SIGNALS")) {
 573           GC_retry_signals = FALSE;
 574       }
 575 #     ifdef CONDPRINT
 576           if (GC_print_stats && GC_retry_signals) {
 577               GC_printf0("Will retry suspend signal if necessary.\n");
 578           }
 579 #     endif
 580 }
 581
 582 /* We hold the allocation lock. */
 583 void GC_stop_init()
 584 {
 585 #ifdef MONO_DEBUGGER_SUPPORTED
 586     if (gc_thread_vtable && gc_thread_vtable->initialize)
 587         gc_thread_vtable->initialize ();
 588     else
 589 #endif
 590         pthread_stop_init ();
 591 }
 592
 593 #ifdef MONO_DEBUGGER_SUPPORTED
 594
 595 GCThreadFunctions *gc_thread_vtable = NULL;
 596
 597 void *
 598 GC_mono_debugger_get_stack_ptr (void)
 599 {
 600         GC_thread me;
 601
 602         me = GC_lookup_thread (pthread_self ());
 603         return &me->stop_info.stack_ptr;
 604 }
 605
 606 #endif
 607
 608 #endif