mono/metadata/sgen-stw.c

   1 /*
   2  * sgen-stw.c: Stop the world functionality
   3  *
   4  * Author:
   5  *      Paolo Molaro (lupus@ximian.com)
   6  *  Rodrigo Kumpera (kumpera@gmail.com)
   7  *
   8  * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
   9  * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
  10  * Copyright 2011 Xamarin, Inc.
  11  * Copyright (C) 2012 Xamarin Inc
  12  *
  13  * This library is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU Library General Public
  15  * License 2.0 as published by the Free Software Foundation;
  16  *
  17  * This library is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  20  * Library General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU Library General Public
  23  * License 2.0 along with this library; if not, write to the Free
  24  * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  25  */
  26
  27 #include "config.h"
  28 #ifdef HAVE_SGEN_GC
  29
  30 #include "sgen/sgen-gc.h"
  31 #include "sgen/sgen-protocol.h"
  32 #include "sgen/sgen-memory-governor.h"
  33 #include "sgen/sgen-thread-pool.h"
  34 #include "metadata/profiler-private.h"
  35 #include "sgen/sgen-client.h"
  36 #include "metadata/sgen-bridge-internal.h"
  37 #include "metadata/gc-internal.h"
  38
  39 #define TV_DECLARE SGEN_TV_DECLARE
  40 #define TV_GETTIME SGEN_TV_GETTIME
  41 #define TV_ELAPSED SGEN_TV_ELAPSED
  42
  43 static void sgen_unified_suspend_restart_world (void);
  44 static void sgen_unified_suspend_stop_world (void);
  45
  46 unsigned int sgen_global_stop_count = 0;
  47
  48 inline static void*
  49 align_pointer (void *ptr)
  50 {
  51         mword p = (mword)ptr;
  52         p += sizeof (gpointer) - 1;
  53         p &= ~ (sizeof (gpointer) - 1);
  54         return (void*)p;
  55 }
  56
  57 #ifdef USE_MONO_CTX
  58 static MonoContext cur_thread_ctx;
  59 #else
  60 static mword cur_thread_regs [ARCH_NUM_REGS];
  61 #endif
  62
  63 static void
  64 update_current_thread_stack (void *start)
  65 {
  66         int stack_guard = 0;
  67 #if !defined(USE_MONO_CTX)
  68         void *reg_ptr = cur_thread_regs;
  69 #endif
  70         SgenThreadInfo *info = mono_thread_info_current ();
  71
  72         info->client_info.stack_start = align_pointer (&stack_guard);
  73         g_assert (info->client_info.stack_start >= info->client_info.stack_start_limit && info->client_info.stack_start < info->client_info.stack_end);
  74 #ifdef USE_MONO_CTX
  75         MONO_CONTEXT_GET_CURRENT (cur_thread_ctx);
  76         memcpy (&info->client_info.ctx, &cur_thread_ctx, sizeof (MonoContext));
  77         if (mono_gc_get_gc_callbacks ()->thread_suspend_func)
  78                 mono_gc_get_gc_callbacks ()->thread_suspend_func (info->client_info.runtime_data, NULL, &info->client_info.ctx);
  79 #else
  80         ARCH_STORE_REGS (reg_ptr);
  81         memcpy (&info->client_info.regs, reg_ptr, sizeof (info->client_info.regs));
  82         if (mono_gc_get_gc_callbacks ()->thread_suspend_func)
  83                 mono_gc_get_gc_callbacks ()->thread_suspend_func (info->client_info.runtime_data, NULL, NULL);
  84 #endif
  85 }
  86
  87 static gboolean
  88 is_ip_in_managed_allocator (MonoDomain *domain, gpointer ip)
  89 {
  90         MonoJitInfo *ji;
  91
  92         if (!mono_thread_internal_current ())
  93                 /* Happens during thread attach */
  94                 return FALSE;
  95
  96         if (!ip || !domain)
  97                 return FALSE;
  98         if (!sgen_has_critical_method ())
  99                 return FALSE;
 100
 101         /*
 102          * mono_jit_info_table_find is not async safe since it calls into the AOT runtime to load information for
 103          * missing methods (#13951). To work around this, we disable the AOT fallback. For this to work, the JIT needs
 104          * to register the jit info for all GC critical methods after they are JITted/loaded.
 105          */
 106         ji = mono_jit_info_table_find_internal (domain, ip, FALSE, FALSE);
 107         if (!ji)
 108                 return FALSE;
 109
 110         return sgen_is_critical_method (mono_jit_info_get_method (ji));
 111 }
 112
 113 static int
 114 restart_threads_until_none_in_managed_allocator (void)
 115 {
 116         SgenThreadInfo *info;
 117         int num_threads_died = 0;
 118         int sleep_duration = -1;
 119
 120         for (;;) {
 121                 int restart_count = 0, restarted_count = 0;
 122                 /* restart all threads that stopped in the
 123                    allocator */
 124                 FOREACH_THREAD_SAFE (info) {
 125                         gboolean result;
 126                         if (info->client_info.skip || info->client_info.gc_disabled || info->client_info.suspend_done)
 127                                 continue;
 128                         if (mono_thread_info_is_live (info) &&
 129                                         (!info->client_info.stack_start || info->client_info.in_critical_region || info->client_info.info.inside_critical_region ||
 130                                         is_ip_in_managed_allocator (info->client_info.stopped_domain, info->client_info.stopped_ip))) {
 131                                 binary_protocol_thread_restart ((gpointer)mono_thread_info_get_tid (info));
 132                                 SGEN_LOG (3, "thread %p resumed.", (void*) (size_t) info->client_info.info.native_handle);
 133                                 result = sgen_resume_thread (info);
 134                                 if (result) {
 135                                         ++restart_count;
 136                                 } else {
 137                                         info->client_info.skip = 1;
 138                                 }
 139                         } else {
 140                                 /* we set the stopped_ip to
 141                                    NULL for threads which
 142                                    we're not restarting so
 143                                    that we can easily identify
 144                                    the others */
 145                                 info->client_info.stopped_ip = NULL;
 146                                 info->client_info.stopped_domain = NULL;
 147                                 info->client_info.suspend_done = TRUE;
 148                         }
 149                 } END_FOREACH_THREAD_SAFE
 150                 /* if no threads were restarted, we're done */
 151                 if (restart_count == 0)
 152                         break;
 153
 154                 /* wait for the threads to signal their restart */
 155                 sgen_wait_for_suspend_ack (restart_count);
 156
 157                 if (sleep_duration < 0) {
 158                         mono_thread_info_yield ();
 159                         sleep_duration = 0;
 160                 } else {
 161                         g_usleep (sleep_duration);
 162                         sleep_duration += 10;
 163                 }
 164
 165                 /* stop them again */
 166                 FOREACH_THREAD (info) {
 167                         gboolean result;
 168                         if (info->client_info.skip || info->client_info.stopped_ip == NULL)
 169                                 continue;
 170                         result = sgen_suspend_thread (info);
 171
 172                         if (result) {
 173                                 ++restarted_count;
 174                         } else {
 175                                 info->client_info.skip = 1;
 176                         }
 177                 } END_FOREACH_THREAD
 178                 /* some threads might have died */
 179                 num_threads_died += restart_count - restarted_count;
 180                 /* wait for the threads to signal their suspension
 181                    again */
 182                 sgen_wait_for_suspend_ack (restarted_count);
 183         }
 184
 185         return num_threads_died;
 186 }
 187
 188 static void
 189 acquire_gc_locks (void)
 190 {
 191         LOCK_INTERRUPTION;
 192         mono_thread_info_suspend_lock ();
 193 }
 194
 195 static void
 196 release_gc_locks (void)
 197 {
 198         mono_thread_info_suspend_unlock ();
 199         UNLOCK_INTERRUPTION;
 200 }
 201
 202 static TV_DECLARE (stop_world_time);
 203 static unsigned long max_pause_usec = 0;
 204
 205 static guint64 time_stop_world;
 206 static guint64 time_restart_world;
 207
 208 /* LOCKING: assumes the GC lock is held */
 209 void
 210 sgen_client_stop_world (int generation)
 211 {
 212         TV_DECLARE (end_handshake);
 213
 214         /* notify the profiler of the leftovers */
 215         /* FIXME this is the wrong spot at we can STW for non collection reasons. */
 216         if (G_UNLIKELY (mono_profiler_events & MONO_PROFILE_GC_MOVES))
 217                 mono_sgen_gc_event_moves ();
 218
 219         acquire_gc_locks ();
 220
 221         /* We start to scan after locks are taking, this ensures we won't be interrupted. */
 222         sgen_process_togglerefs ();
 223
 224         update_current_thread_stack (&generation);
 225
 226         sgen_global_stop_count++;
 227         SGEN_LOG (3, "stopping world n %d from %p %p", sgen_global_stop_count, mono_thread_info_current (), (gpointer)mono_native_thread_id_get ());
 228         TV_GETTIME (stop_world_time);
 229
 230         if (mono_thread_info_unified_management_enabled ()) {
 231                 sgen_unified_suspend_stop_world ();
 232         } else {
 233                 int count, dead;
 234                 count = sgen_thread_handshake (TRUE);
 235                 dead = restart_threads_until_none_in_managed_allocator ();
 236                 if (count < dead)
 237                         g_error ("More threads have died (%d) that been initialy suspended %d", dead, count);
 238         }
 239
 240         SGEN_LOG (3, "world stopped");
 241
 242         TV_GETTIME (end_handshake);
 243         time_stop_world += TV_ELAPSED (stop_world_time, end_handshake);
 244
 245         sgen_memgov_collection_start (generation);
 246         if (sgen_need_bridge_processing ())
 247                 sgen_bridge_reset_data ();
 248 }
 249
 250 /* LOCKING: assumes the GC lock is held */
 251 void
 252 sgen_client_restart_world (int generation, GGTimingInfo *timing)
 253 {
 254         SgenThreadInfo *info;
 255         TV_DECLARE (end_sw);
 256         TV_DECLARE (start_handshake);
 257         TV_DECLARE (end_bridge);
 258         unsigned long usec, bridge_usec;
 259
 260         /* notify the profiler of the leftovers */
 261         /* FIXME this is the wrong spot at we can STW for non collection reasons. */
 262         if (G_UNLIKELY (mono_profiler_events & MONO_PROFILE_GC_MOVES))
 263                 mono_sgen_gc_event_moves ();
 264
 265         FOREACH_THREAD (info) {
 266                 info->client_info.stack_start = NULL;
 267 #ifdef USE_MONO_CTX
 268                 memset (&info->client_info.ctx, 0, sizeof (MonoContext));
 269 #else
 270                 memset (&info->client_info.regs, 0, sizeof (info->client_info.regs));
 271 #endif
 272         } END_FOREACH_THREAD
 273
 274         TV_GETTIME (start_handshake);
 275
 276         if (mono_thread_info_unified_management_enabled ())
 277                 sgen_unified_suspend_restart_world ();
 278         else
 279                 sgen_thread_handshake (FALSE);
 280
 281         TV_GETTIME (end_sw);
 282         time_restart_world += TV_ELAPSED (start_handshake, end_sw);
 283         usec = TV_ELAPSED (stop_world_time, end_sw);
 284         max_pause_usec = MAX (usec, max_pause_usec);
 285
 286         SGEN_LOG (2, "restarted (pause time: %d usec, max: %d)", (int)usec, (int)max_pause_usec);
 287
 288         /*
 289          * We must release the thread info suspend lock after doing
 290          * the thread handshake.  Otherwise, if the GC stops the world
 291          * and a thread is in the process of starting up, but has not
 292          * yet registered (it's not in the thread_list), it is
 293          * possible that the thread does register while the world is
 294          * stopped.  When restarting the GC will then try to restart
 295          * said thread, but since it never got the suspend signal, it
 296          * cannot answer the restart signal, so a deadlock results.
 297          */
 298         release_gc_locks ();
 299
 300         TV_GETTIME (end_bridge);
 301         bridge_usec = TV_ELAPSED (end_sw, end_bridge);
 302
 303         if (timing) {
 304                 timing [0].stw_time = usec;
 305                 timing [0].bridge_time = bridge_usec;
 306         }
 307 }
 308
 309 void
 310 mono_sgen_init_stw (void)
 311 {
 312         mono_counters_register ("World stop", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_stop_world);
 313         mono_counters_register ("World restart", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_restart_world);
 314 }
 315
 316 /* Unified suspend code */
 317
 318 static gboolean
 319 sgen_is_thread_in_current_stw (SgenThreadInfo *info)
 320 {
 321         /*
 322         A thread explicitly asked to be skiped because it holds no managed state.
 323         This is used by TP and finalizer threads.
 324         FIXME Use an atomic variable for this to avoid everyone taking the GC LOCK.
 325         */
 326         if (info->client_info.gc_disabled) {
 327                 return FALSE;
 328         }
 329
 330         /*
 331         We have detected that this thread is failing/dying, ignore it.
 332         FIXME: can't we merge this with thread_is_dying?
 333         */
 334         if (info->client_info.skip) {
 335                 return FALSE;
 336         }
 337
 338         /*
 339         Suspending the current thread will deadlock us, bad idea.
 340         */
 341         if (info == mono_thread_info_current ()) {
 342                 return FALSE;
 343         }
 344
 345         /*
 346         We can't suspend the workers that will do all the heavy lifting.
 347         FIXME Use some state bit in SgenThreadInfo for this.
 348         */
 349         if (sgen_thread_pool_is_thread_pool_thread (mono_thread_info_get_tid (info))) {
 350                 return FALSE;
 351         }
 352
 353         /*
 354         The thread has signaled that it started to detach, ignore it.
 355         FIXME: can't we merge this with skip
 356         */
 357         if (!mono_thread_info_is_live (info)) {
 358                 return FALSE;
 359         }
 360
 361         return TRUE;
 362 }
 363
 364 static void
 365 update_sgen_info (SgenThreadInfo *info)
 366 {
 367         char *stack_start;
 368
 369         /* Once we remove the old suspend code, we should move sgen to directly access the state in MonoThread */
 370         info->client_info.stopped_domain = mono_thread_info_tls_get (info, TLS_KEY_DOMAIN);
 371         info->client_info.stopped_ip = (gpointer) MONO_CONTEXT_GET_IP (&mono_thread_info_get_suspend_state (info)->ctx);
 372         stack_start = (char*)MONO_CONTEXT_GET_SP (&mono_thread_info_get_suspend_state (info)->ctx) - REDZONE_SIZE;
 373
 374         /* altstack signal handler, sgen can't handle them, mono-threads should have handled this. */
 375         if (stack_start < (char*)info->client_info.stack_start_limit || stack_start >= (char*)info->client_info.stack_end)
 376                 g_error ("BAD STACK");
 377
 378         info->client_info.stack_start = stack_start;
 379 #ifdef USE_MONO_CTX
 380         info->client_info.ctx = mono_thread_info_get_suspend_state (info)->ctx;
 381 #else
 382         g_assert_not_reached ();
 383 #endif
 384 }
 385
 386 static void
 387 sgen_unified_suspend_stop_world (void)
 388 {
 389         int restart_counter;
 390         SgenThreadInfo *info;
 391         int sleep_duration = -1;
 392
 393         mono_threads_begin_global_suspend ();
 394         THREADS_STW_DEBUG ("[GC-STW-BEGIN] *** BEGIN SUSPEND *** \n");
 395
 396         FOREACH_THREAD_SAFE (info) {
 397                 info->client_info.skip = FALSE;
 398                 info->client_info.suspend_done = FALSE;
 399                 if (sgen_is_thread_in_current_stw (info)) {
 400                         info->client_info.skip = !mono_thread_info_begin_suspend (info, FALSE);
 401                         THREADS_STW_DEBUG ("[GC-STW-BEGIN-SUSPEND] SUSPEND thread %p skip %d\n", mono_thread_info_get_tid (info), info->client_info.skip);
 402                 } else {
 403                         THREADS_STW_DEBUG ("[GC-STW-BEGIN-SUSPEND] IGNORE thread %p skip %d\n", mono_thread_info_get_tid (info), info->skip);
 404                 }
 405         } END_FOREACH_THREAD_SAFE
 406
 407         mono_thread_info_current ()->client_info.suspend_done = TRUE;
 408         mono_threads_wait_pending_operations ();
 409
 410         for (;;) {
 411                 restart_counter = 0;
 412                 FOREACH_THREAD_SAFE (info) {
 413                         if (info->client_info.suspend_done || !sgen_is_thread_in_current_stw (info)) {
 414                                 THREADS_STW_DEBUG ("[GC-STW-RESTART] IGNORE thread %p not been processed done %d current %d\n", mono_thread_info_get_tid (info), info->client_info.suspend_done, !sgen_is_thread_in_current_stw (info));
 415                                 continue;
 416                         }
 417
 418                         /*
 419                         All threads that reach here are pristine suspended. This means the following:
 420
 421                         - We haven't accepted the previous suspend as good.
 422                         - We haven't gave up on it for this STW (it's either bad or asked not to)
 423                         */
 424                         if (!mono_threads_core_check_suspend_result (info)) {
 425                                 THREADS_STW_DEBUG ("[GC-STW-RESTART] SKIP thread %p failed to finish to suspend\n", mono_thread_info_get_tid (info));
 426                                 info->client_info.skip = TRUE;
 427                         } else if (mono_thread_info_in_critical_location (info)) {
 428                                 gboolean res;
 429                                 g_assert (mono_thread_info_suspend_count (info) == 1);
 430                                 res = mono_thread_info_begin_resume (info);
 431                                 THREADS_STW_DEBUG ("[GC-STW-RESTART] RESTART thread %p skip %d\n", mono_thread_info_get_tid (info), res);
 432                                 if (res)
 433                                         ++restart_counter;
 434                                 else
 435                                         info->client_info.skip = TRUE;
 436                         } else {
 437                                 THREADS_STW_DEBUG ("[GC-STW-RESTART] DONE thread %p deemed fully suspended\n", mono_thread_info_get_tid (info));
 438                                 g_assert (!info->client_info.in_critical_region);
 439                                 info->client_info.suspend_done = TRUE;
 440                         }
 441                 } END_FOREACH_THREAD_SAFE
 442
 443                 if (restart_counter == 0)
 444                         break;
 445                 mono_threads_wait_pending_operations ();
 446
 447                 if (sleep_duration < 0) {
 448 #ifdef HOST_WIN32
 449                         SwitchToThread ();
 450 #else
 451                         sched_yield ();
 452 #endif
 453                         sleep_duration = 0;
 454                 } else {
 455                         g_usleep (sleep_duration);
 456                         sleep_duration += 10;
 457                 }
 458
 459                 FOREACH_THREAD_SAFE (info) {
 460                         if (sgen_is_thread_in_current_stw (info) && mono_thread_info_is_running (info)) {
 461                                 gboolean res = mono_thread_info_begin_suspend (info, FALSE);
 462                                 THREADS_STW_DEBUG ("[GC-STW-RESTART] SUSPEND thread %p skip %d\n", mono_thread_info_get_tid (info), res);
 463                                 if (!res)
 464                                         info->client_info.skip = TRUE;
 465                         }
 466                 } END_FOREACH_THREAD_SAFE
 467
 468                 mono_threads_wait_pending_operations ();
 469         }
 470
 471         FOREACH_THREAD_SAFE (info) {
 472                 if (sgen_is_thread_in_current_stw (info)) {
 473                         THREADS_STW_DEBUG ("[GC-STW-SUSPEND-END] thread %p is suspended\n", mono_thread_info_get_tid (info));
 474                         g_assert (info->client_info.suspend_done);
 475                         update_sgen_info (info);
 476                 } else {
 477                         g_assert (!info->client_info.suspend_done || info == mono_thread_info_current ());
 478                 }
 479         } END_FOREACH_THREAD_SAFE
 480 }
 481
 482 static void
 483 sgen_unified_suspend_restart_world (void)
 484 {
 485         SgenThreadInfo *info;
 486
 487         THREADS_STW_DEBUG ("[GC-STW-END] *** BEGIN RESUME ***\n");
 488         FOREACH_THREAD_SAFE (info) {
 489                 if (sgen_is_thread_in_current_stw (info)) {
 490                         g_assert (mono_thread_info_begin_resume (info));
 491                         THREADS_STW_DEBUG ("[GC-STW-RESUME-WORLD] RESUME thread %p\n", mono_thread_info_get_tid (info));
 492                 } else {
 493                         THREADS_STW_DEBUG ("[GC-STW-RESUME-WORLD] IGNORE thread %p\n", mono_thread_info_get_tid (info));
 494                 }
 495         } END_FOREACH_THREAD_SAFE
 496
 497         mono_threads_wait_pending_operations ();
 498         mono_threads_end_global_suspend ();
 499 }
 500 #endif