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 "metadata/sgen-gc.h"
  31 #include "metadata/sgen-protocol.h"
  32 #include "metadata/sgen-memory-governor.h"
  33 #include "metadata/profiler-private.h"
  34 #include "utils/mono-time.h"
  35 #include "utils/dtrace.h"
  36 #include "utils/mono-counters.h"
  37
  38 #define TV_DECLARE SGEN_TV_DECLARE
  39 #define TV_GETTIME SGEN_TV_GETTIME
  40 #define TV_ELAPSED SGEN_TV_ELAPSED
  41
  42 inline static void*
  43 align_pointer (void *ptr)
  44 {
  45         mword p = (mword)ptr;
  46         p += sizeof (gpointer) - 1;
  47         p &= ~ (sizeof (gpointer) - 1);
  48         return (void*)p;
  49 }
  50
  51 #ifdef USE_MONO_CTX
  52 static MonoContext cur_thread_ctx;
  53 #else
  54 static mword cur_thread_regs [ARCH_NUM_REGS];
  55 #endif
  56
  57 static void
  58 update_current_thread_stack (void *start)
  59 {
  60         int stack_guard = 0;
  61 #if !defined(USE_MONO_CTX)
  62         void *reg_ptr = cur_thread_regs;
  63 #endif
  64         SgenThreadInfo *info = mono_thread_info_current ();
  65
  66         info->stack_start = align_pointer (&stack_guard);
  67         g_assert (info->stack_start >= info->stack_start_limit && info->stack_start < info->stack_end);
  68 #ifdef USE_MONO_CTX
  69         MONO_CONTEXT_GET_CURRENT (cur_thread_ctx);
  70         memcpy (&info->ctx, &cur_thread_ctx, sizeof (MonoContext));
  71         if (mono_gc_get_gc_callbacks ()->thread_suspend_func)
  72                 mono_gc_get_gc_callbacks ()->thread_suspend_func (info->runtime_data, NULL, &info->ctx);
  73 #else
  74         ARCH_STORE_REGS (reg_ptr);
  75         memcpy (&info->regs, reg_ptr, sizeof (info->regs));
  76         if (mono_gc_get_gc_callbacks ()->thread_suspend_func)
  77                 mono_gc_get_gc_callbacks ()->thread_suspend_func (info->runtime_data, NULL, NULL);
  78 #endif
  79 }
  80
  81 static gboolean
  82 is_ip_in_managed_allocator (MonoDomain *domain, gpointer ip)
  83 {
  84         MonoJitInfo *ji;
  85
  86         if (!mono_thread_internal_current ())
  87                 /* Happens during thread attach */
  88                 return FALSE;
  89
  90         if (!ip || !domain)
  91                 return FALSE;
  92         if (!sgen_has_critical_method ())
  93                 return FALSE;
  94
  95         /*
  96          * mono_jit_info_table_find is not async safe since it calls into the AOT runtime to load information for
  97          * missing methods (#13951). To work around this, we disable the AOT fallback. For this to work, the JIT needs
  98          * to register the jit info for all GC critical methods after they are JITted/loaded.
  99          */
 100         ji = mono_jit_info_table_find_internal (domain, ip, FALSE);
 101         if (!ji)
 102                 return FALSE;
 103
 104         return sgen_is_critical_method (mono_jit_info_get_method (ji));
 105 }
 106
 107 static int
 108 restart_threads_until_none_in_managed_allocator (void)
 109 {
 110         SgenThreadInfo *info;
 111         int num_threads_died = 0;
 112         int sleep_duration = -1;
 113
 114         for (;;) {
 115                 int restart_count = 0, restarted_count = 0;
 116                 /* restart all threads that stopped in the
 117                    allocator */
 118                 FOREACH_THREAD_SAFE (info) {
 119                         gboolean result;
 120                         if (info->skip || info->gc_disabled)
 121                                 continue;
 122                         if (mono_thread_info_run_state (info) == STATE_RUNNING && (!info->stack_start || info->in_critical_region || info->info.inside_critical_region ||
 123                                         is_ip_in_managed_allocator (info->stopped_domain, info->stopped_ip))) {
 124                                 binary_protocol_thread_restart ((gpointer)mono_thread_info_get_tid (info));
 125                                 SGEN_LOG (3, "thread %p resumed.", (void*) (size_t) info->info.native_handle);
 126                                 result = sgen_resume_thread (info);
 127                                 if (result) {
 128                                         ++restart_count;
 129                                 } else {
 130                                         info->skip = 1;
 131                                 }
 132                         } else {
 133                                 /* we set the stopped_ip to
 134                                    NULL for threads which
 135                                    we're not restarting so
 136                                    that we can easily identify
 137                                    the others */
 138                                 info->stopped_ip = NULL;
 139                                 info->stopped_domain = NULL;
 140                         }
 141                 } END_FOREACH_THREAD_SAFE
 142                 /* if no threads were restarted, we're done */
 143                 if (restart_count == 0)
 144                         break;
 145
 146                 /* wait for the threads to signal their restart */
 147                 sgen_wait_for_suspend_ack (restart_count);
 148
 149                 if (sleep_duration < 0) {
 150                         mono_thread_info_yield ();
 151                         sleep_duration = 0;
 152                 } else {
 153                         g_usleep (sleep_duration);
 154                         sleep_duration += 10;
 155                 }
 156
 157                 /* stop them again */
 158                 FOREACH_THREAD (info) {
 159                         gboolean result;
 160                         if (info->skip || info->stopped_ip == NULL)
 161                                 continue;
 162                         result = sgen_suspend_thread (info);
 163
 164                         if (result) {
 165                                 ++restarted_count;
 166                         } else {
 167                                 info->skip = 1;
 168                         }
 169                 } END_FOREACH_THREAD
 170                 /* some threads might have died */
 171                 num_threads_died += restart_count - restarted_count;
 172                 /* wait for the threads to signal their suspension
 173                    again */
 174                 sgen_wait_for_suspend_ack (restarted_count);
 175         }
 176
 177         return num_threads_died;
 178 }
 179
 180 static void
 181 acquire_gc_locks (void)
 182 {
 183         LOCK_INTERRUPTION;
 184         mono_thread_info_suspend_lock ();
 185 }
 186
 187 static void
 188 release_gc_locks (void)
 189 {
 190         mono_thread_info_suspend_unlock ();
 191         UNLOCK_INTERRUPTION;
 192 }
 193
 194 static void
 195 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
 196 {
 197         sgen_get_major_collector ()->count_cards (major_total, major_marked);
 198         sgen_los_count_cards (los_total, los_marked);
 199 }
 200
 201 static TV_DECLARE (stop_world_time);
 202 static unsigned long max_pause_usec = 0;
 203
 204 static guint64 time_stop_world;
 205 static guint64 time_restart_world;
 206
 207 /* LOCKING: assumes the GC lock is held */
 208 int
 209 sgen_stop_world (int generation)
 210 {
 211         TV_DECLARE (end_handshake);
 212         int count, dead;
 213
 214         mono_profiler_gc_event (MONO_GC_EVENT_PRE_STOP_WORLD, generation);
 215         MONO_GC_WORLD_STOP_BEGIN ();
 216         binary_protocol_world_stopping (sgen_timestamp ());
 217         acquire_gc_locks ();
 218
 219         /* We start to scan after locks are taking, this ensures we won't be interrupted. */
 220         sgen_process_togglerefs ();
 221
 222         update_current_thread_stack (&count);
 223
 224         sgen_global_stop_count++;
 225         SGEN_LOG (3, "stopping world n %d from %p %p", sgen_global_stop_count, mono_thread_info_current (), (gpointer)mono_native_thread_id_get ());
 226         TV_GETTIME (stop_world_time);
 227         count = sgen_thread_handshake (TRUE);
 228         dead = restart_threads_until_none_in_managed_allocator ();
 229         if (count < dead)
 230                 g_error ("More threads have died (%d) that been initialy suspended %d", dead, count);
 231         count -= dead;
 232
 233         SGEN_LOG (3, "world stopped %d thread(s)", count);
 234         mono_profiler_gc_event (MONO_GC_EVENT_POST_STOP_WORLD, generation);
 235         MONO_GC_WORLD_STOP_END ();
 236         if (binary_protocol_is_enabled ()) {
 237                 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
 238                 if (binary_protocol_is_heavy_enabled ())
 239                         count_cards (&major_total, &major_marked, &los_total, &los_marked);
 240                 binary_protocol_world_stopped (sgen_timestamp (), major_total, major_marked, los_total, los_marked);
 241         }
 242
 243         TV_GETTIME (end_handshake);
 244         time_stop_world += TV_ELAPSED (stop_world_time, end_handshake);
 245
 246         sgen_memgov_collection_start (generation);
 247         if (sgen_need_bridge_processing ())
 248                 sgen_bridge_reset_data ();
 249
 250         return count;
 251 }
 252
 253 /* LOCKING: assumes the GC lock is held */
 254 int
 255 sgen_restart_world (int generation, GGTimingInfo *timing)
 256 {
 257         int count;
 258         SgenThreadInfo *info;
 259         TV_DECLARE (end_sw);
 260         TV_DECLARE (start_handshake);
 261         TV_DECLARE (end_bridge);
 262         unsigned long usec, bridge_usec;
 263
 264         if (binary_protocol_is_enabled ()) {
 265                 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
 266                 if (binary_protocol_is_heavy_enabled ())
 267                         count_cards (&major_total, &major_marked, &los_total, &los_marked);
 268                 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
 269         }
 270
 271         /* notify the profiler of the leftovers */
 272         /* FIXME this is the wrong spot at we can STW for non collection reasons. */
 273         if (G_UNLIKELY (mono_profiler_events & MONO_PROFILE_GC_MOVES))
 274                 sgen_gc_event_moves ();
 275         mono_profiler_gc_event (MONO_GC_EVENT_PRE_START_WORLD, generation);
 276         MONO_GC_WORLD_RESTART_BEGIN (generation);
 277         FOREACH_THREAD (info) {
 278                 info->stack_start = NULL;
 279 #ifdef USE_MONO_CTX
 280                 memset (&info->ctx, 0, sizeof (MonoContext));
 281 #else
 282                 memset (&info->regs, 0, sizeof (info->regs));
 283 #endif
 284         } END_FOREACH_THREAD
 285
 286         TV_GETTIME (start_handshake);
 287         count = sgen_thread_handshake (FALSE);
 288         TV_GETTIME (end_sw);
 289         time_restart_world += TV_ELAPSED (start_handshake, end_sw);
 290         usec = TV_ELAPSED (stop_world_time, end_sw);
 291         max_pause_usec = MAX (usec, max_pause_usec);
 292         SGEN_LOG (2, "restarted %d thread(s) (pause time: %d usec, max: %d)", count, (int)usec, (int)max_pause_usec);
 293         mono_profiler_gc_event (MONO_GC_EVENT_POST_START_WORLD, generation);
 294         MONO_GC_WORLD_RESTART_END (generation);
 295         binary_protocol_world_restarted (generation, sgen_timestamp ());
 296
 297         /*
 298          * We must release the thread info suspend lock after doing
 299          * the thread handshake.  Otherwise, if the GC stops the world
 300          * and a thread is in the process of starting up, but has not
 301          * yet registered (it's not in the thread_list), it is
 302          * possible that the thread does register while the world is
 303          * stopped.  When restarting the GC will then try to restart
 304          * said thread, but since it never got the suspend signal, it
 305          * cannot answer the restart signal, so a deadlock results.
 306          */
 307         release_gc_locks ();
 308
 309         sgen_try_free_some_memory = TRUE;
 310
 311         if (sgen_need_bridge_processing ())
 312                 sgen_bridge_processing_finish (generation);
 313
 314         TV_GETTIME (end_bridge);
 315         bridge_usec = TV_ELAPSED (end_sw, end_bridge);
 316
 317         if (timing) {
 318                 timing [0].stw_time = usec;
 319                 timing [0].bridge_time = bridge_usec;
 320         }
 321
 322         sgen_memgov_collection_end (generation, timing, timing ? 2 : 0);
 323
 324         return count;
 325 }
 326
 327 void
 328 sgen_init_stw (void)
 329 {
 330         mono_counters_register ("World stop", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_stop_world);
 331         mono_counters_register ("World restart", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_restart_world);
 332 }
 333
 334 #endif