1 /*
   2  * Copyright (c) 1997, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "classfile/classLoaderDataGraph.inline.hpp"
  27 #include "classfile/stringTable.hpp"
  28 #include "classfile/symbolTable.hpp"
  29 #include "classfile/systemDictionary.hpp"
  30 #include "code/codeCache.hpp"
  31 #include "code/icBuffer.hpp"
  32 #include "code/nmethod.hpp"
  33 #include "code/pcDesc.hpp"
  34 #include "code/scopeDesc.hpp"
  35 #include "gc/shared/collectedHeap.hpp"
  36 #include "gc/shared/gcLocker.hpp"
  37 #include "gc/shared/strongRootsScope.hpp"
  38 #include "gc/shared/workgroup.hpp"
  39 #include "interpreter/interpreter.hpp"
  40 #include "jfr/jfrEvents.hpp"
  41 #include "logging/log.hpp"
  42 #include "logging/logStream.hpp"
  43 #include "memory/resourceArea.hpp"
  44 #include "memory/universe.hpp"
  45 #include "oops/oop.inline.hpp"
  46 #include "oops/symbol.hpp"
  47 #include "runtime/atomic.hpp"
  48 #include "runtime/compilationPolicy.hpp"
  49 #include "runtime/deoptimization.hpp"
  50 #include "runtime/frame.inline.hpp"
  51 #include "runtime/handles.inline.hpp"
  52 #include "runtime/interfaceSupport.inline.hpp"
  53 #include "runtime/mutexLocker.hpp"
  54 #include "runtime/orderAccess.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/safepoint.hpp"
  57 #include "runtime/safepointMechanism.inline.hpp"
  58 #include "runtime/signature.hpp"
  59 #include "runtime/stubCodeGenerator.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "runtime/sweeper.hpp"
  62 #include "runtime/synchronizer.hpp"
  63 #include "runtime/thread.inline.hpp"
  64 #include "runtime/threadSMR.hpp"
  65 #include "runtime/timerTrace.hpp"
  66 #include "services/runtimeService.hpp"
  67 #include "utilities/events.hpp"
  68 #include "utilities/macros.hpp"
  69 #ifdef COMPILER1
  70 #include "c1/c1_globals.hpp"
  71 #endif
  72 
  73 static void post_safepoint_begin_event(EventSafepointBegin& event,
  74                                        uint64_t safepoint_id,
  75                                        int thread_count,
  76                                        int critical_thread_count) {
  77   if (event.should_commit()) {
  78     event.set_safepointId(safepoint_id);
  79     event.set_totalThreadCount(thread_count);
  80     event.set_jniCriticalThreadCount(critical_thread_count);
  81     event.commit();
  82   }
  83 }
  84 
  85 static void post_safepoint_cleanup_event(EventSafepointCleanup& event, uint64_t safepoint_id) {
  86   if (event.should_commit()) {
  87     event.set_safepointId(safepoint_id);
  88     event.commit();
  89   }
  90 }
  91 
  92 static void post_safepoint_synchronize_event(EventSafepointStateSynchronization& event,
  93                                              uint64_t safepoint_id,
  94                                              int initial_number_of_threads,
  95                                              int threads_waiting_to_block,
  96                                              uint64_t iterations) {
  97   if (event.should_commit()) {
  98     event.set_safepointId(safepoint_id);
  99     event.set_initialThreadCount(initial_number_of_threads);
 100     event.set_runningThreadCount(threads_waiting_to_block);
 101     event.set_iterations(iterations);
 102     event.commit();
 103   }
 104 }
 105 
 106 static void post_safepoint_cleanup_task_event(EventSafepointCleanupTask& event,
 107                                               uint64_t safepoint_id,
 108                                               const char* name) {
 109   if (event.should_commit()) {
 110     event.set_safepointId(safepoint_id);
 111     event.set_name(name);
 112     event.commit();
 113   }
 114 }
 115 
 116 static void post_safepoint_end_event(EventSafepointEnd& event, uint64_t safepoint_id) {
 117   if (event.should_commit()) {
 118     event.set_safepointId(safepoint_id);
 119     event.commit();
 120   }
 121 }
 122 
 123 // --------------------------------------------------------------------------------------------------
 124 // Implementation of Safepoint begin/end
 125 
 126 SafepointSynchronize::SynchronizeState volatile SafepointSynchronize::_state = SafepointSynchronize::_not_synchronized;
 127 int SafepointSynchronize::_waiting_to_block = 0;
 128 volatile uint64_t SafepointSynchronize::_safepoint_counter = 0;
 129 const uint64_t SafepointSynchronize::InactiveSafepointCounter = 0;
 130 int SafepointSynchronize::_current_jni_active_count = 0;
 131 
 132 WaitBarrier* SafepointSynchronize::_wait_barrier;
 133 
 134 static volatile bool PageArmed = false;        // safepoint polling page is RO|RW vs PROT_NONE
 135 static bool timeout_error_printed = false;
 136 
 137 // Statistic related
 138 static jlong _safepoint_begin_time = 0;
 139 static volatile int _nof_threads_hit_polling_page = 0;
 140 
 141 void SafepointSynchronize::init(Thread* vmthread) {
 142   // WaitBarrier should never be destroyed since we will have
 143   // threads waiting on it while exiting.
 144   _wait_barrier = new WaitBarrier(vmthread);
 145   SafepointTracing::init();
 146 }
 147 
 148 void SafepointSynchronize::increment_jni_active_count() {
 149   assert(Thread::current()->is_VM_thread(), "Only VM thread may increment");
 150   ++_current_jni_active_count;
 151 }
 152 
 153 void SafepointSynchronize::decrement_waiting_to_block() {
 154   assert(_waiting_to_block > 0, "sanity check");
 155   assert(Thread::current()->is_VM_thread(), "Only VM thread may decrement");
 156   --_waiting_to_block;
 157 }
 158 
 159 static bool thread_not_running(ThreadSafepointState *cur_state) {
 160   if (!cur_state->is_running()) {
 161     return true;
 162   }
 163   cur_state->examine_state_of_thread(SafepointSynchronize::safepoint_counter());
 164   if (!cur_state->is_running()) {
 165     return true;
 166   }
 167   LogTarget(Trace, safepoint) lt;
 168   if (lt.is_enabled()) {
 169     ResourceMark rm;
 170     LogStream ls(lt);
 171     cur_state->print_on(&ls);
 172   }
 173   return false;
 174 }
 175 
 176 #ifdef ASSERT
 177 static void assert_list_is_valid(const ThreadSafepointState* tss_head, int still_running) {
 178   int a = 0;
 179   const ThreadSafepointState *tmp_tss = tss_head;
 180   while (tmp_tss != NULL) {
 181     ++a;
 182     assert(tmp_tss->is_running(), "Illegal initial state");
 183     tmp_tss = tmp_tss->get_next();
 184   }
 185   assert(a == still_running, "Must be the same");
 186 }
 187 #endif // ASSERT
 188 
 189 static void back_off(int iteration) {
 190   // iteration will be 1 the first time we enter this spin back-off.
 191   // naked_short_nanosleep takes tenths of micros which means that
 192   // number of nanoseconds is irrelevant if it's below that. We do
 193   // 20 1 ns sleeps with a total cost of ~1 ms, then we do 1 ms sleeps.
 194   jlong sleep_ns = 1;
 195   if (iteration > 20) {
 196     sleep_ns = NANOUNITS / MILLIUNITS;  // 1 ms
 197   }
 198   os::naked_short_nanosleep(sleep_ns);
 199 }
 200 
 201 int SafepointSynchronize::synchronize_threads(jlong safepoint_limit_time, int nof_threads, int* initial_running)
 202 {
 203   JavaThreadIteratorWithHandle jtiwh;
 204 
 205 #ifdef ASSERT
 206   for (; JavaThread *cur = jtiwh.next(); ) {
 207     assert(cur->safepoint_state()->is_running(), "Illegal initial state");
 208   }
 209   jtiwh.rewind();
 210 #endif // ASSERT
 211 
 212   // Iterate through all threads until it has been determined how to stop them all at a safepoint.
 213   int still_running = nof_threads;
 214   ThreadSafepointState *tss_head = NULL;
 215   ThreadSafepointState **p_prev = &tss_head;
 216   for (; JavaThread *cur = jtiwh.next(); ) {
 217     ThreadSafepointState *cur_tss = cur->safepoint_state();
 218     assert(cur_tss->get_next() == NULL, "Must be NULL");
 219     if (thread_not_running(cur_tss)) {
 220       --still_running;
 221     } else {
 222       *p_prev = cur_tss;
 223       p_prev = cur_tss->next_ptr();
 224     }
 225   }
 226   *p_prev = NULL;
 227 
 228   DEBUG_ONLY(assert_list_is_valid(tss_head, still_running);)
 229 
 230   *initial_running = still_running;
 231 
 232   int iterations = 1; // The first iteration is above.
 233 
 234   while (still_running > 0) {
 235     // Check if this has taken too long:
 236     if (SafepointTimeout && safepoint_limit_time < os::javaTimeNanos()) {
 237       print_safepoint_timeout();
 238     }
 239     if (int(iterations) == -1) { // overflow - something is wrong.
 240       // We can only overflow here when we are using global
 241       // polling pages. We keep this guarantee in its original
 242       // form so that searches of the bug database for this
 243       // failure mode find the right bugs.
 244       guarantee (!PageArmed, "invariant");
 245     }
 246 
 247     p_prev = &tss_head;
 248     ThreadSafepointState *cur_tss = tss_head;
 249     while (cur_tss != NULL) {
 250       assert(cur_tss->is_running(), "Illegal initial state");
 251       if (thread_not_running(cur_tss)) {
 252         --still_running;
 253         *p_prev = NULL;
 254         ThreadSafepointState *tmp = cur_tss;
 255         cur_tss = cur_tss->get_next();
 256         tmp->set_next(NULL);
 257       } else {
 258         *p_prev = cur_tss;
 259         p_prev = cur_tss->next_ptr();
 260         cur_tss = cur_tss->get_next();
 261       }
 262     }
 263 
 264     DEBUG_ONLY(assert_list_is_valid(tss_head, still_running);)
 265 
 266     if (still_running > 0) {
 267       back_off(iterations);
 268     }
 269 
 270     iterations++;
 271   }
 272 
 273   assert(tss_head == NULL, "Must be empty");
 274 
 275   return iterations;
 276 }
 277 
 278 void SafepointSynchronize::arm_safepoint() {
 279   // Begin the process of bringing the system to a safepoint.
 280   // Java threads can be in several different states and are
 281   // stopped by different mechanisms:
 282   //
 283   //  1. Running interpreted
 284   //     When executing branching/returning byte codes interpreter
 285   //     checks if the poll is armed, if so blocks in SS::block().
 286   //     When using global polling the interpreter dispatch table
 287   //     is changed to force it to check for a safepoint condition
 288   //     between bytecodes.
 289   //  2. Running in native code
 290   //     When returning from the native code, a Java thread must check
 291   //     the safepoint _state to see if we must block.  If the
 292   //     VM thread sees a Java thread in native, it does
 293   //     not wait for this thread to block.  The order of the memory
 294   //     writes and reads of both the safepoint state and the Java
 295   //     threads state is critical.  In order to guarantee that the
 296   //     memory writes are serialized with respect to each other,
 297   //     the VM thread issues a memory barrier instruction.
 298   //  3. Running compiled Code
 299   //     Compiled code reads the local polling page that
 300   //     is set to fault if we are trying to get to a safepoint.
 301   //  4. Blocked
 302   //     A thread which is blocked will not be allowed to return from the
 303   //     block condition until the safepoint operation is complete.
 304   //  5. In VM or Transitioning between states
 305   //     If a Java thread is currently running in the VM or transitioning
 306   //     between states, the safepointing code will poll the thread state
 307   //     until the thread blocks itself when it attempts transitions to a
 308   //     new state or locking a safepoint checked monitor.
 309 
 310   // We must never miss a thread with correct safepoint id, so we must make sure we arm
 311   // the wait barrier for the next safepoint id/counter.
 312   // Arming must be done after resetting _current_jni_active_count, _waiting_to_block.
 313   _wait_barrier->arm(static_cast<int>(_safepoint_counter + 1));
 314 
 315   assert((_safepoint_counter & 0x1) == 0, "must be even");
 316   // The store to _safepoint_counter must happen after any stores in arming.
 317   OrderAccess::release_store(&_safepoint_counter, _safepoint_counter + 1);
 318 
 319   // We are synchronizing
 320   OrderAccess::storestore(); // Ordered with _safepoint_counter
 321   _state = _synchronizing;
 322 
 323   if (SafepointMechanism::uses_thread_local_poll()) {
 324     // Arming the per thread poll while having _state != _not_synchronized means safepointing
 325     log_trace(safepoint)("Setting thread local yield flag for threads");
 326     OrderAccess::storestore(); // storestore, global state -> local state
 327     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *cur = jtiwh.next(); ) {
 328       // Make sure the threads start polling, it is time to yield.
 329       SafepointMechanism::arm_local_poll(cur);
 330     }
 331   }
 332   OrderAccess::fence(); // storestore|storeload, global state -> local state
 333 
 334   if (SafepointMechanism::uses_global_page_poll()) {
 335     // Make interpreter safepoint aware
 336     Interpreter::notice_safepoints();
 337 
 338     // Make polling safepoint aware
 339     guarantee (!PageArmed, "invariant") ;
 340     PageArmed = true;
 341     os::make_polling_page_unreadable();
 342   }
 343 }
 344 
 345 // Roll all threads forward to a safepoint and suspend them all
 346 void SafepointSynchronize::begin() {
 347   assert(Thread::current()->is_VM_thread(), "Only VM thread may execute a safepoint");
 348 
 349   EventSafepointBegin begin_event;
 350   SafepointTracing::begin(VMThread::vm_op_type());
 351 
 352   Universe::heap()->safepoint_synchronize_begin();
 353 
 354   // By getting the Threads_lock, we assure that no threads are about to start or
 355   // exit. It is released again in SafepointSynchronize::end().
 356   Threads_lock->lock();
 357 
 358   assert( _state == _not_synchronized, "trying to safepoint synchronize with wrong state");
 359 
 360   int nof_threads = Threads::number_of_threads();
 361 
 362   _nof_threads_hit_polling_page = 0;
 363 
 364   log_debug(safepoint)("Safepoint synchronization initiated using %s wait barrier. (%d threads)", _wait_barrier->description(), nof_threads);
 365 
 366   // Reset the count of active JNI critical threads
 367   _current_jni_active_count = 0;
 368 
 369   // Set number of threads to wait for
 370   _waiting_to_block = nof_threads;
 371 
 372   jlong safepoint_limit_time = 0;
 373   if (SafepointTimeout) {
 374     // Set the limit time, so that it can be compared to see if this has taken
 375     // too long to complete.
 376     safepoint_limit_time = SafepointTracing::start_of_safepoint() + (jlong)SafepointTimeoutDelay * (NANOUNITS / MILLIUNITS);
 377     timeout_error_printed = false;
 378   }
 379 
 380   EventSafepointStateSynchronization sync_event;
 381   int initial_running = 0;
 382 
 383   // Arms the safepoint, _current_jni_active_count and _waiting_to_block must be set before.
 384   arm_safepoint();
 385 
 386   // Will spin until all threads are safe.
 387   int iterations = synchronize_threads(safepoint_limit_time, nof_threads, &initial_running);
 388   assert(_waiting_to_block == 0, "No thread should be running");
 389 
 390 #ifndef PRODUCT
 391   if (safepoint_limit_time != 0) {
 392     jlong current_time = os::javaTimeNanos();
 393     if (safepoint_limit_time < current_time) {
 394       log_warning(safepoint)("# SafepointSynchronize: Finished after "
 395                     INT64_FORMAT_W(6) " ms",
 396                     (int64_t)(current_time - SafepointTracing::start_of_safepoint()) / (NANOUNITS / MILLIUNITS));
 397     }
 398   }
 399 #endif
 400 
 401   assert(Threads_lock->owned_by_self(), "must hold Threads_lock");
 402 
 403   // Record state
 404   _state = _synchronized;
 405 
 406   OrderAccess::fence();
 407 
 408 #ifdef ASSERT
 409   // Make sure all the threads were visited.
 410   for (JavaThreadIteratorWithHandle jtiwh; JavaThread *cur = jtiwh.next(); ) {
 411     assert(cur->was_visited_for_critical_count(_safepoint_counter), "missed a thread");
 412   }
 413 #endif // ASSERT
 414 
 415   // Update the count of active JNI critical regions
 416   GCLocker::set_jni_lock_count(_current_jni_active_count);
 417 
 418   post_safepoint_synchronize_event(sync_event,
 419                                    _safepoint_counter,
 420                                    initial_running,
 421                                    _waiting_to_block, iterations);
 422 
 423   SafepointTracing::synchronized(nof_threads, initial_running, _nof_threads_hit_polling_page);
 424 
 425   // We do the safepoint cleanup first since a GC related safepoint
 426   // needs cleanup to be completed before running the GC op.
 427   EventSafepointCleanup cleanup_event;
 428   do_cleanup_tasks();
 429   post_safepoint_cleanup_event(cleanup_event, _safepoint_counter);
 430 
 431   post_safepoint_begin_event(begin_event, _safepoint_counter, nof_threads, _current_jni_active_count);
 432   SafepointTracing::cleanup();
 433 }
 434 
 435 void SafepointSynchronize::disarm_safepoint() {
 436   uint64_t safepoint_id = _safepoint_counter;
 437   {
 438     JavaThreadIteratorWithHandle jtiwh;
 439 #ifdef ASSERT
 440     // A pending_exception cannot be installed during a safepoint.  The threads
 441     // may install an async exception after they come back from a safepoint into
 442     // pending_exception after they unblock.  But that should happen later.
 443     for (; JavaThread *cur = jtiwh.next(); ) {
 444       assert (!(cur->has_pending_exception() &&
 445                 cur->safepoint_state()->is_at_poll_safepoint()),
 446               "safepoint installed a pending exception");
 447     }
 448 #endif // ASSERT
 449 
 450     if (SafepointMechanism::uses_global_page_poll()) {
 451       guarantee (PageArmed, "invariant");
 452       // Make polling safepoint aware
 453       os::make_polling_page_readable();
 454       PageArmed = false;
 455       // Remove safepoint check from interpreter
 456       Interpreter::ignore_safepoints();
 457     }
 458 
 459     OrderAccess::fence(); // keep read and write of _state from floating up
 460     assert(_state == _synchronized, "must be synchronized before ending safepoint synchronization");
 461 
 462     // Change state first to _not_synchronized.
 463     // No threads should see _synchronized when running.
 464     _state = _not_synchronized;
 465 
 466     // Set the next dormant (even) safepoint id.
 467     assert((_safepoint_counter & 0x1) == 1, "must be odd");
 468     OrderAccess::release_store(&_safepoint_counter, _safepoint_counter + 1);
 469 
 470     OrderAccess::fence(); // Keep the local state from floating up.
 471 
 472     jtiwh.rewind();
 473     for (; JavaThread *current = jtiwh.next(); ) {
 474       // Clear the visited flag to ensure that the critical counts are collected properly.
 475       DEBUG_ONLY(current->reset_visited_for_critical_count(safepoint_id);)
 476       ThreadSafepointState* cur_state = current->safepoint_state();
 477       assert(!cur_state->is_running(), "Thread not suspended at safepoint");
 478       cur_state->restart(); // TSS _running
 479       assert(cur_state->is_running(), "safepoint state has not been reset");
 480 
 481       SafepointMechanism::disarm_if_needed(current, false /* NO release */);
 482     }
 483   } // ~JavaThreadIteratorWithHandle
 484 
 485   // Release threads lock, so threads can be created/destroyed again.
 486   Threads_lock->unlock();
 487 
 488   // Wake threads after local state is correctly set.
 489   _wait_barrier->disarm();
 490 }
 491 
 492 // Wake up all threads, so they are ready to resume execution after the safepoint
 493 // operation has been carried out
 494 void SafepointSynchronize::end() {
 495   assert(Threads_lock->owned_by_self(), "must hold Threads_lock");
 496   EventSafepointEnd event;
 497   uint64_t safepoint_id = _safepoint_counter;
 498   assert(Thread::current()->is_VM_thread(), "Only VM thread can execute a safepoint");
 499 
 500   disarm_safepoint();
 501 
 502   Universe::heap()->safepoint_synchronize_end();
 503 
 504   SafepointTracing::end();
 505 
 506   post_safepoint_end_event(event, safepoint_id);
 507 }
 508 
 509 bool SafepointSynchronize::is_cleanup_needed() {
 510   // Need a safepoint if there are many monitors to deflate.
 511   if (ObjectSynchronizer::is_cleanup_needed()) return true;
 512   // Need a safepoint if some inline cache buffers is non-empty
 513   if (!InlineCacheBuffer::is_empty()) return true;
 514   return false;
 515 }
 516 
 517 class ParallelSPCleanupThreadClosure : public ThreadClosure {
 518 private:
 519   CodeBlobClosure* _nmethod_cl;
 520   DeflateMonitorCounters* _counters;
 521 
 522 public:
 523   ParallelSPCleanupThreadClosure(DeflateMonitorCounters* counters) :
 524     _nmethod_cl(UseCodeAging ? NMethodSweeper::prepare_reset_hotness_counters() : NULL),
 525     _counters(counters) {}
 526 
 527   void do_thread(Thread* thread) {
 528     // deflate_thread_local_monitors() handles or requests deflation of
 529     // this thread's idle monitors. If !AsyncDeflateIdleMonitors or if
 530     // there is a special cleanup request, deflation is handled now.
 531     // Otherwise, async deflation is requested via a flag.
 532     ObjectSynchronizer::deflate_thread_local_monitors(thread, _counters);
 533     if (_nmethod_cl != NULL && thread->is_Java_thread() &&
 534         ! thread->is_Code_cache_sweeper_thread()) {
 535       JavaThread* jt = (JavaThread*) thread;
 536       jt->nmethods_do(_nmethod_cl);
 537     }
 538   }
 539 };
 540 
 541 class ParallelSPCleanupTask : public AbstractGangTask {
 542 private:
 543   SubTasksDone _subtasks;
 544   ParallelSPCleanupThreadClosure _cleanup_threads_cl;
 545   uint _num_workers;
 546   DeflateMonitorCounters* _counters;
 547 public:
 548   ParallelSPCleanupTask(uint num_workers, DeflateMonitorCounters* counters) :
 549     AbstractGangTask("Parallel Safepoint Cleanup"),
 550     _subtasks(SubTasksDone(SafepointSynchronize::SAFEPOINT_CLEANUP_NUM_TASKS)),
 551     _cleanup_threads_cl(ParallelSPCleanupThreadClosure(counters)),
 552     _num_workers(num_workers),
 553     _counters(counters) {}
 554 
 555   void work(uint worker_id) {
 556     uint64_t safepoint_id = SafepointSynchronize::safepoint_counter();
 557     // All threads deflate monitors and mark nmethods (if necessary).
 558     Threads::possibly_parallel_threads_do(true, &_cleanup_threads_cl);
 559 
 560     if (_subtasks.try_claim_task(SafepointSynchronize::SAFEPOINT_CLEANUP_DEFLATE_MONITORS)) {
 561       const char* name = "deflating global idle monitors";
 562       EventSafepointCleanupTask event;
 563       TraceTime timer(name, TRACETIME_LOG(Info, safepoint, cleanup));
 564       // AsyncDeflateIdleMonitors only uses DeflateMonitorCounters
 565       // when a special cleanup has been requested.
 566       // Note: This logging output will include global idle monitor
 567       // elapsed times, but not global idle monitor deflation count.
 568       ObjectSynchronizer::do_safepoint_work(_counters);
 569 
 570       post_safepoint_cleanup_task_event(event, safepoint_id, name);
 571     }
 572 
 573     if (_subtasks.try_claim_task(SafepointSynchronize::SAFEPOINT_CLEANUP_UPDATE_INLINE_CACHES)) {
 574       const char* name = "updating inline caches";
 575       EventSafepointCleanupTask event;
 576       TraceTime timer(name, TRACETIME_LOG(Info, safepoint, cleanup));
 577       InlineCacheBuffer::update_inline_caches();
 578 
 579       post_safepoint_cleanup_task_event(event, safepoint_id, name);
 580     }
 581 
 582     if (_subtasks.try_claim_task(SafepointSynchronize::SAFEPOINT_CLEANUP_COMPILATION_POLICY)) {
 583       const char* name = "compilation policy safepoint handler";
 584       EventSafepointCleanupTask event;
 585       TraceTime timer(name, TRACETIME_LOG(Info, safepoint, cleanup));
 586       CompilationPolicy::policy()->do_safepoint_work();
 587 
 588       post_safepoint_cleanup_task_event(event, safepoint_id, name);
 589     }
 590 
 591     if (_subtasks.try_claim_task(SafepointSynchronize::SAFEPOINT_CLEANUP_SYMBOL_TABLE_REHASH)) {
 592       if (SymbolTable::needs_rehashing()) {
 593         const char* name = "rehashing symbol table";
 594         EventSafepointCleanupTask event;
 595         TraceTime timer(name, TRACETIME_LOG(Info, safepoint, cleanup));
 596         SymbolTable::rehash_table();
 597 
 598         post_safepoint_cleanup_task_event(event, safepoint_id, name);
 599       }
 600     }
 601 
 602     if (_subtasks.try_claim_task(SafepointSynchronize::SAFEPOINT_CLEANUP_STRING_TABLE_REHASH)) {
 603       if (StringTable::needs_rehashing()) {
 604         const char* name = "rehashing string table";
 605         EventSafepointCleanupTask event;
 606         TraceTime timer(name, TRACETIME_LOG(Info, safepoint, cleanup));
 607         StringTable::rehash_table();
 608 
 609         post_safepoint_cleanup_task_event(event, safepoint_id, name);
 610       }
 611     }
 612 
 613     if (_subtasks.try_claim_task(SafepointSynchronize::SAFEPOINT_CLEANUP_CLD_PURGE)) {
 614       // CMS delays purging the CLDG until the beginning of the next safepoint and to
 615       // make sure concurrent sweep is done
 616       const char* name = "purging class loader data graph";
 617       EventSafepointCleanupTask event;
 618       TraceTime timer(name, TRACETIME_LOG(Info, safepoint, cleanup));
 619       ClassLoaderDataGraph::purge_if_needed();
 620 
 621       post_safepoint_cleanup_task_event(event, safepoint_id, name);
 622     }
 623 
 624     if (_subtasks.try_claim_task(SafepointSynchronize::SAFEPOINT_CLEANUP_SYSTEM_DICTIONARY_RESIZE)) {
 625       const char* name = "resizing system dictionaries";
 626       EventSafepointCleanupTask event;
 627       TraceTime timer(name, TRACETIME_LOG(Info, safepoint, cleanup));
 628       ClassLoaderDataGraph::resize_if_needed();
 629 
 630       post_safepoint_cleanup_task_event(event, safepoint_id, name);
 631     }
 632 
 633     _subtasks.all_tasks_completed(_num_workers);
 634   }
 635 };
 636 
 637 // Various cleaning tasks that should be done periodically at safepoints.
 638 void SafepointSynchronize::do_cleanup_tasks() {
 639 
 640   TraceTime timer("safepoint cleanup tasks", TRACETIME_LOG(Info, safepoint, cleanup));
 641 
 642   // Prepare for monitor deflation.
 643   DeflateMonitorCounters deflate_counters;
 644   ObjectSynchronizer::prepare_deflate_idle_monitors(&deflate_counters);
 645 
 646   CollectedHeap* heap = Universe::heap();
 647   assert(heap != NULL, "heap not initialized yet?");
 648   WorkGang* cleanup_workers = heap->get_safepoint_workers();
 649   if (cleanup_workers != NULL) {
 650     // Parallel cleanup using GC provided thread pool.
 651     uint num_cleanup_workers = cleanup_workers->active_workers();
 652     ParallelSPCleanupTask cleanup(num_cleanup_workers, &deflate_counters);
 653     StrongRootsScope srs(num_cleanup_workers);
 654     cleanup_workers->run_task(&cleanup);
 655   } else {
 656     // Serial cleanup using VMThread.
 657     ParallelSPCleanupTask cleanup(1, &deflate_counters);
 658     StrongRootsScope srs(1);
 659     cleanup.work(0);
 660   }
 661 
 662   // Needs to be done single threaded by the VMThread.  This walks
 663   // the thread stacks looking for references to metadata before
 664   // deciding to remove it from the metaspaces.
 665   if (ClassLoaderDataGraph::should_clean_metaspaces_and_reset()) {
 666     const char* name = "cleanup live ClassLoaderData metaspaces";
 667     TraceTime timer(name, TRACETIME_LOG(Info, safepoint, cleanup));
 668     ClassLoaderDataGraph::walk_metadata_and_clean_metaspaces();
 669   }
 670 
 671   // Finish monitor deflation.
 672   ObjectSynchronizer::finish_deflate_idle_monitors(&deflate_counters);
 673 
 674   assert(InlineCacheBuffer::is_empty(), "should have cleaned up ICBuffer");
 675 }
 676 
 677 // Methods for determining if a JavaThread is safepoint safe.
 678 
 679 // False means unsafe with undetermined state.
 680 // True means a determined state, but it may be an unsafe state.
 681 // If called from a non-safepoint context safepoint_count MUST be InactiveSafepointCounter.
 682 bool SafepointSynchronize::try_stable_load_state(JavaThreadState *state, JavaThread *thread, uint64_t safepoint_count) {
 683   assert((safepoint_count != InactiveSafepointCounter &&
 684           Thread::current() == (Thread*)VMThread::vm_thread() &&
 685           SafepointSynchronize::_state != _not_synchronized)
 686          || safepoint_count == InactiveSafepointCounter, "Invalid check");
 687 
 688   // To handle the thread_blocked state on the backedge of the WaitBarrier from
 689   // previous safepoint and reading the reset value (0/InactiveSafepointCounter) we
 690   // re-read state after we read thread safepoint id. The JavaThread changes its
 691   // thread state from thread_blocked before resetting safepoint id to 0.
 692   // This guarantees the second read will be from an updated thread state. It can
 693   // either be different state making this an unsafe state or it can see blocked
 694   // again. When we see blocked twice with a 0 safepoint id, either:
 695   // - It is normally blocked, e.g. on Mutex, TBIVM.
 696   // - It was in SS:block(), looped around to SS:block() and is blocked on the WaitBarrier.
 697   // - It was in SS:block() but now on a Mutex.
 698   // All of these cases are safe.
 699 
 700   *state = thread->thread_state();
 701   OrderAccess::loadload();
 702   uint64_t sid = thread->safepoint_state()->get_safepoint_id();  // Load acquire
 703   if (sid != InactiveSafepointCounter && sid != safepoint_count) {
 704     // In an old safepoint, state not relevant.
 705     return false;
 706   }
 707   return *state == thread->thread_state();
 708 }
 709 
 710 static bool safepoint_safe_with(JavaThread *thread, JavaThreadState state) {
 711   switch(state) {
 712   case _thread_in_native:
 713     // native threads are safe if they have no java stack or have walkable stack
 714     return !thread->has_last_Java_frame() || thread->frame_anchor()->walkable();
 715 
 716   case _thread_blocked:
 717     // On wait_barrier or blocked.
 718     // Blocked threads should already have walkable stack.
 719     assert(!thread->has_last_Java_frame() || thread->frame_anchor()->walkable(), "blocked and not walkable");
 720     return true;
 721 
 722   default:
 723     return false;
 724   }
 725 }
 726 
 727 bool SafepointSynchronize::handshake_safe(JavaThread *thread) {
 728   // This function must be called with the Threads_lock held so an externally
 729   // suspended thread cannot be resumed thus it is safe.
 730   assert(Threads_lock->owned_by_self() && Thread::current()->is_VM_thread(),
 731          "Must hold Threads_lock and be VMThread");
 732   if (thread->is_ext_suspended() || thread->is_terminated()) {
 733     return true;
 734   }
 735   JavaThreadState stable_state;
 736   if (try_stable_load_state(&stable_state, thread, InactiveSafepointCounter)) {
 737     return safepoint_safe_with(thread, stable_state);
 738   }
 739   return false;
 740 }
 741 
 742 // See if the thread is running inside a lazy critical native and
 743 // update the thread critical count if so.  Also set a suspend flag to
 744 // cause the native wrapper to return into the JVM to do the unlock
 745 // once the native finishes.
 746 static void check_for_lazy_critical_native(JavaThread *thread, JavaThreadState state) {
 747   if (state == _thread_in_native &&
 748       thread->has_last_Java_frame() &&
 749       thread->frame_anchor()->walkable()) {
 750     // This thread might be in a critical native nmethod so look at
 751     // the top of the stack and increment the critical count if it
 752     // is.
 753     frame wrapper_frame = thread->last_frame();
 754     CodeBlob* stub_cb = wrapper_frame.cb();
 755     if (stub_cb != NULL &&
 756         stub_cb->is_nmethod() &&
 757         stub_cb->as_nmethod_or_null()->is_lazy_critical_native()) {
 758       // A thread could potentially be in a critical native across
 759       // more than one safepoint, so only update the critical state on
 760       // the first one.  When it returns it will perform the unlock.
 761       if (!thread->do_critical_native_unlock()) {
 762 #ifdef ASSERT
 763         if (!thread->in_critical()) {
 764           GCLocker::increment_debug_jni_lock_count();
 765         }
 766 #endif
 767         thread->enter_critical();
 768         // Make sure the native wrapper calls back on return to
 769         // perform the needed critical unlock.
 770         thread->set_critical_native_unlock();
 771       }
 772     }
 773   }
 774 }
 775 
 776 // -------------------------------------------------------------------------------------------------------
 777 // Implementation of Safepoint blocking point
 778 
 779 void SafepointSynchronize::block(JavaThread *thread) {
 780   assert(thread != NULL, "thread must be set");
 781   assert(thread->is_Java_thread(), "not a Java thread");
 782 
 783   // Threads shouldn't block if they are in the middle of printing, but...
 784   ttyLocker::break_tty_lock_for_safepoint(os::current_thread_id());
 785 
 786   // Only bail from the block() call if the thread is gone from the
 787   // thread list; starting to exit should still block.
 788   if (thread->is_terminated()) {
 789      // block current thread if we come here from native code when VM is gone
 790      thread->block_if_vm_exited();
 791 
 792      // otherwise do nothing
 793      return;
 794   }
 795 
 796   JavaThreadState state = thread->thread_state();
 797   thread->frame_anchor()->make_walkable(thread);
 798 
 799   uint64_t safepoint_id = SafepointSynchronize::safepoint_counter();
 800   // Check that we have a valid thread_state at this point
 801   switch(state) {
 802     case _thread_in_vm_trans:
 803     case _thread_in_Java:        // From compiled code
 804     case _thread_in_native_trans:
 805     case _thread_blocked_trans:
 806     case _thread_new_trans:
 807 
 808       // We have no idea where the VMThread is, it might even be at next safepoint.
 809       // So we can miss this poll, but stop at next.
 810 
 811       // Load dependent store, it must not pass loading of safepoint_id.
 812       thread->safepoint_state()->set_safepoint_id(safepoint_id); // Release store
 813 
 814       // This part we can skip if we notice we miss or are in a future safepoint.
 815       OrderAccess::storestore();
 816       // Load in wait barrier should not float up
 817       thread->set_thread_state_fence(_thread_blocked);
 818 
 819       _wait_barrier->wait(static_cast<int>(safepoint_id));
 820       assert(_state != _synchronized, "Can't be");
 821 
 822       // If barrier is disarmed stop store from floating above loads in barrier.
 823       OrderAccess::loadstore();
 824       thread->set_thread_state(state);
 825 
 826       // Then we reset the safepoint id to inactive.
 827       thread->safepoint_state()->reset_safepoint_id(); // Release store
 828 
 829       OrderAccess::fence();
 830 
 831       break;
 832 
 833     default:
 834      fatal("Illegal threadstate encountered: %d", state);
 835   }
 836   guarantee(thread->safepoint_state()->get_safepoint_id() == InactiveSafepointCounter,
 837             "The safepoint id should be set only in block path");
 838 
 839   // Check for pending. async. exceptions or suspends - except if the
 840   // thread was blocked inside the VM. has_special_runtime_exit_condition()
 841   // is called last since it grabs a lock and we only want to do that when
 842   // we must.
 843   //
 844   // Note: we never deliver an async exception at a polling point as the
 845   // compiler may not have an exception handler for it. The polling
 846   // code will notice the async and deoptimize and the exception will
 847   // be delivered. (Polling at a return point is ok though). Sure is
 848   // a lot of bother for a deprecated feature...
 849   //
 850   // We don't deliver an async exception if the thread state is
 851   // _thread_in_native_trans so JNI functions won't be called with
 852   // a surprising pending exception. If the thread state is going back to java,
 853   // async exception is checked in check_special_condition_for_native_trans().
 854 
 855   if (state != _thread_blocked_trans &&
 856       state != _thread_in_vm_trans &&
 857       thread->has_special_runtime_exit_condition()) {
 858     thread->handle_special_runtime_exit_condition(
 859       !thread->is_at_poll_safepoint() && (state != _thread_in_native_trans));
 860   }
 861 
 862   // cross_modify_fence is done by SafepointMechanism::block_if_requested_slow
 863   // which is the only caller here.
 864 }
 865 
 866 // ------------------------------------------------------------------------------------------------------
 867 // Exception handlers
 868 
 869 
 870 void SafepointSynchronize::handle_polling_page_exception(JavaThread *thread) {
 871   assert(thread->is_Java_thread(), "polling reference encountered by VM thread");
 872   assert(thread->thread_state() == _thread_in_Java, "should come from Java code");
 873   if (!ThreadLocalHandshakes) {
 874     assert(SafepointSynchronize::is_synchronizing(), "polling encountered outside safepoint synchronization");
 875   }
 876 
 877   if (log_is_enabled(Info, safepoint, stats)) {
 878     Atomic::inc(&_nof_threads_hit_polling_page);
 879   }
 880 
 881   ThreadSafepointState* state = thread->safepoint_state();
 882 
 883   state->handle_polling_page_exception();
 884 }
 885 
 886 
 887 void SafepointSynchronize::print_safepoint_timeout() {
 888   if (!timeout_error_printed) {
 889     timeout_error_printed = true;
 890     // Print out the thread info which didn't reach the safepoint for debugging
 891     // purposes (useful when there are lots of threads in the debugger).
 892     LogTarget(Warning, safepoint) lt;
 893     if (lt.is_enabled()) {
 894       ResourceMark rm;
 895       LogStream ls(lt);
 896 
 897       ls.cr();
 898       ls.print_cr("# SafepointSynchronize::begin: Timeout detected:");
 899       ls.print_cr("# SafepointSynchronize::begin: Timed out while spinning to reach a safepoint.");
 900       ls.print_cr("# SafepointSynchronize::begin: Threads which did not reach the safepoint:");
 901       for (JavaThreadIteratorWithHandle jtiwh; JavaThread *cur_thread = jtiwh.next(); ) {
 902         if (cur_thread->safepoint_state()->is_running()) {
 903           ls.print("# ");
 904           cur_thread->print_on(&ls);
 905           ls.cr();
 906         }
 907       }
 908       ls.print_cr("# SafepointSynchronize::begin: (End of list)");
 909     }
 910   }
 911 
 912   // To debug the long safepoint, specify both AbortVMOnSafepointTimeout &
 913   // ShowMessageBoxOnError.
 914   if (AbortVMOnSafepointTimeout) {
 915     // Send the blocking thread a signal to terminate and write an error file.
 916     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *cur_thread = jtiwh.next(); ) {
 917       if (cur_thread->safepoint_state()->is_running()) {
 918         if (!os::signal_thread(cur_thread, SIGILL, "blocking a safepoint")) {
 919           break; // Could not send signal. Report fatal error.
 920         }
 921         // Give cur_thread a chance to report the error and terminate the VM.
 922         os::sleep(Thread::current(), 3000, false);
 923       }
 924     }
 925     fatal("Safepoint sync time longer than " INTX_FORMAT "ms detected when executing %s.",
 926           SafepointTimeoutDelay, VMThread::vm_operation()->name());
 927   }
 928 }
 929 
 930 // -------------------------------------------------------------------------------------------------------
 931 // Implementation of ThreadSafepointState
 932 
 933 ThreadSafepointState::ThreadSafepointState(JavaThread *thread)
 934   : _at_poll_safepoint(false), _thread(thread), _safepoint_safe(false),
 935     _safepoint_id(SafepointSynchronize::InactiveSafepointCounter),
 936     _orig_thread_state(_thread_uninitialized), _next(NULL) {
 937 }
 938 
 939 void ThreadSafepointState::create(JavaThread *thread) {
 940   ThreadSafepointState *state = new ThreadSafepointState(thread);
 941   thread->set_safepoint_state(state);
 942 }
 943 
 944 void ThreadSafepointState::destroy(JavaThread *thread) {
 945   if (thread->safepoint_state()) {
 946     delete(thread->safepoint_state());
 947     thread->set_safepoint_state(NULL);
 948   }
 949 }
 950 
 951 uint64_t ThreadSafepointState::get_safepoint_id() const {
 952   return OrderAccess::load_acquire(&_safepoint_id);
 953 }
 954 
 955 void ThreadSafepointState::reset_safepoint_id() {
 956   OrderAccess::release_store(&_safepoint_id, SafepointSynchronize::InactiveSafepointCounter);
 957 }
 958 
 959 void ThreadSafepointState::set_safepoint_id(uint64_t safepoint_id) {
 960   OrderAccess::release_store(&_safepoint_id, safepoint_id);
 961 }
 962 
 963 void ThreadSafepointState::examine_state_of_thread(uint64_t safepoint_count) {
 964   assert(is_running(), "better be running or just have hit safepoint poll");
 965 
 966   JavaThreadState stable_state;
 967   if (!SafepointSynchronize::try_stable_load_state(&stable_state, _thread, safepoint_count)) {
 968     // We could not get stable state of the JavaThread.
 969     // Consider it running and just return.
 970     return;
 971   }
 972 
 973   // Save the state at the start of safepoint processing.
 974   _orig_thread_state = stable_state;
 975 
 976   // Check for a thread that is suspended. Note that thread resume tries
 977   // to grab the Threads_lock which we own here, so a thread cannot be
 978   // resumed during safepoint synchronization.
 979 
 980   // We check to see if this thread is suspended without locking to
 981   // avoid deadlocking with a third thread that is waiting for this
 982   // thread to be suspended. The third thread can notice the safepoint
 983   // that we're trying to start at the beginning of its SR_lock->wait()
 984   // call. If that happens, then the third thread will block on the
 985   // safepoint while still holding the underlying SR_lock. We won't be
 986   // able to get the SR_lock and we'll deadlock.
 987   //
 988   // We don't need to grab the SR_lock here for two reasons:
 989   // 1) The suspend flags are both volatile and are set with an
 990   //    Atomic::cmpxchg() call so we should see the suspended
 991   //    state right away.
 992   // 2) We're being called from the safepoint polling loop; if
 993   //    we don't see the suspended state on this iteration, then
 994   //    we'll come around again.
 995   //
 996   bool is_suspended = _thread->is_ext_suspended();
 997   if (is_suspended) {
 998     account_safe_thread();
 999     return;
1000   }
1001 
1002   if (safepoint_safe_with(_thread, stable_state)) {
1003     check_for_lazy_critical_native(_thread, stable_state);
1004     account_safe_thread();
1005     return;
1006   }
1007 
1008   // All other thread states will continue to run until they
1009   // transition and self-block in state _blocked
1010   // Safepoint polling in compiled code causes the Java threads to do the same.
1011   // Note: new threads may require a malloc so they must be allowed to finish
1012 
1013   assert(is_running(), "examine_state_of_thread on non-running thread");
1014   return;
1015 }
1016 
1017 void ThreadSafepointState::account_safe_thread() {
1018   SafepointSynchronize::decrement_waiting_to_block();
1019   if (_thread->in_critical()) {
1020     // Notice that this thread is in a critical section
1021     SafepointSynchronize::increment_jni_active_count();
1022   }
1023   DEBUG_ONLY(_thread->set_visited_for_critical_count(SafepointSynchronize::safepoint_counter());)
1024   assert(!_safepoint_safe, "Must be unsafe before safe");
1025   _safepoint_safe = true;
1026 }
1027 
1028 void ThreadSafepointState::restart() {
1029   assert(_safepoint_safe, "Must be safe before unsafe");
1030   _safepoint_safe = false;
1031 }
1032 
1033 void ThreadSafepointState::print_on(outputStream *st) const {
1034   const char *s = _safepoint_safe ? "_at_safepoint" : "_running";
1035 
1036   st->print_cr("Thread: " INTPTR_FORMAT
1037               "  [0x%2x] State: %s _at_poll_safepoint %d",
1038                p2i(_thread), _thread->osthread()->thread_id(), s, _at_poll_safepoint);
1039 
1040   _thread->print_thread_state_on(st);
1041 }
1042 
1043 // ---------------------------------------------------------------------------------------------------------------------
1044 
1045 // Block the thread at poll or poll return for safepoint/handshake.
1046 void ThreadSafepointState::handle_polling_page_exception() {
1047 
1048   // If we're using a global poll, then the thread should not be
1049   // marked as safepoint safe yet.
1050   assert(!SafepointMechanism::uses_global_page_poll() || !_safepoint_safe,
1051          "polling page exception on thread safepoint safe");
1052 
1053   // Step 1: Find the nmethod from the return address
1054   address real_return_addr = thread()->saved_exception_pc();
1055 
1056   CodeBlob *cb = CodeCache::find_blob(real_return_addr);
1057   assert(cb != NULL && cb->is_compiled(), "return address should be in nmethod");
1058   CompiledMethod* nm = (CompiledMethod*)cb;
1059 
1060   // Find frame of caller
1061   frame stub_fr = thread()->last_frame();
1062   CodeBlob* stub_cb = stub_fr.cb();
1063   assert(stub_cb->is_safepoint_stub(), "must be a safepoint stub");
1064   RegisterMap map(thread(), true);
1065   frame caller_fr = stub_fr.sender(&map);
1066 
1067   // Should only be poll_return or poll
1068   assert( nm->is_at_poll_or_poll_return(real_return_addr), "should not be at call" );
1069 
1070   // This is a poll immediately before a return. The exception handling code
1071   // has already had the effect of causing the return to occur, so the execution
1072   // will continue immediately after the call. In addition, the oopmap at the
1073   // return point does not mark the return value as an oop (if it is), so
1074   // it needs a handle here to be updated.
1075   if( nm->is_at_poll_return(real_return_addr) ) {
1076     // See if return type is an oop.
1077     bool return_oop = nm->method()->is_returning_oop();
1078     Handle return_value;
1079     if (return_oop) {
1080       // The oop result has been saved on the stack together with all
1081       // the other registers. In order to preserve it over GCs we need
1082       // to keep it in a handle.
1083       oop result = caller_fr.saved_oop_result(&map);
1084       assert(oopDesc::is_oop_or_null(result), "must be oop");
1085       return_value = Handle(thread(), result);
1086       assert(Universe::heap()->is_in_or_null(result), "must be heap pointer");
1087     }
1088 
1089     // Block the thread
1090     SafepointMechanism::block_if_requested(thread());
1091 
1092     // restore oop result, if any
1093     if (return_oop) {
1094       caller_fr.set_saved_oop_result(&map, return_value());
1095     }
1096   }
1097 
1098   // This is a safepoint poll. Verify the return address and block.
1099   else {
1100     set_at_poll_safepoint(true);
1101 
1102     // verify the blob built the "return address" correctly
1103     assert(real_return_addr == caller_fr.pc(), "must match");
1104 
1105     // Block the thread
1106     SafepointMechanism::block_if_requested(thread());
1107     set_at_poll_safepoint(false);
1108 
1109     // If we have a pending async exception deoptimize the frame
1110     // as otherwise we may never deliver it.
1111     if (thread()->has_async_condition()) {
1112       ThreadInVMfromJavaNoAsyncException __tiv(thread());
1113       Deoptimization::deoptimize_frame(thread(), caller_fr.id());
1114     }
1115 
1116     // If an exception has been installed we must check for a pending deoptimization
1117     // Deoptimize frame if exception has been thrown.
1118 
1119     if (thread()->has_pending_exception() ) {
1120       RegisterMap map(thread(), true);
1121       frame caller_fr = stub_fr.sender(&map);
1122       if (caller_fr.is_deoptimized_frame()) {
1123         // The exception patch will destroy registers that are still
1124         // live and will be needed during deoptimization. Defer the
1125         // Async exception should have deferred the exception until the
1126         // next safepoint which will be detected when we get into
1127         // the interpreter so if we have an exception now things
1128         // are messed up.
1129 
1130         fatal("Exception installed and deoptimization is pending");
1131       }
1132     }
1133   }
1134 }
1135 
1136 
1137 // -------------------------------------------------------------------------------------------------------
1138 // Implementation of SafepointTracing
1139 
1140 jlong SafepointTracing::_last_safepoint_begin_time_ns = 0;
1141 jlong SafepointTracing::_last_safepoint_sync_time_ns = 0;
1142 jlong SafepointTracing::_last_safepoint_cleanup_time_ns = 0;
1143 jlong SafepointTracing::_last_safepoint_end_time_ns = 0;
1144 jlong SafepointTracing::_last_safepoint_end_time_epoch_ms = 0;
1145 jlong SafepointTracing::_last_app_time_ns = 0;
1146 int SafepointTracing::_nof_threads = 0;
1147 int SafepointTracing::_nof_running = 0;
1148 int SafepointTracing::_page_trap = 0;
1149 VM_Operation::VMOp_Type SafepointTracing::_current_type;
1150 jlong     SafepointTracing::_max_sync_time = 0;
1151 jlong     SafepointTracing::_max_vmop_time = 0;
1152 uint64_t  SafepointTracing::_op_count[VM_Operation::VMOp_Terminating] = {0};
1153 
1154 void SafepointTracing::init() {
1155   // Application start
1156   _last_safepoint_end_time_ns = os::javaTimeNanos();
1157   // amount of time since epoch
1158   _last_safepoint_end_time_epoch_ms = os::javaTimeMillis();
1159 }
1160 
1161 // Helper method to print the header.
1162 static void print_header(outputStream* st) {
1163   // The number of spaces is significant here, and should match the format
1164   // specifiers in print_statistics().
1165 
1166   st->print("VM Operation                 "
1167             "[ threads: total initial_running ]"
1168             "[ time:       sync    cleanup       vmop      total ]");
1169 
1170   st->print_cr(" page_trap_count");
1171 }
1172 
1173 // This prints a nice table.  To get the statistics to not shift due to the logging uptime
1174 // decorator, use the option as: -Xlog:safepoint+stats:[outputfile]:none
1175 void SafepointTracing::statistics_log() {
1176   LogTarget(Info, safepoint, stats) lt;
1177   assert (lt.is_enabled(), "should only be called when printing statistics is enabled");
1178   LogStream ls(lt);
1179 
1180   static int _cur_stat_index = 0;
1181 
1182   // Print header every 30 entries
1183   if ((_cur_stat_index % 30) == 0) {
1184     print_header(&ls);
1185     _cur_stat_index = 1;  // wrap
1186   } else {
1187     _cur_stat_index++;
1188   }
1189 
1190   ls.print("%-28s [       "
1191            INT32_FORMAT_W(8) "        " INT32_FORMAT_W(8) " "
1192            "]",
1193            VM_Operation::name(_current_type),
1194            _nof_threads,
1195            _nof_running);
1196   ls.print("[       "
1197            INT64_FORMAT_W(10) " " INT64_FORMAT_W(10) " "
1198            INT64_FORMAT_W(10) " " INT64_FORMAT_W(10) " ]",
1199            (int64_t)(_last_safepoint_sync_time_ns - _last_safepoint_begin_time_ns),
1200            (int64_t)(_last_safepoint_cleanup_time_ns - _last_safepoint_sync_time_ns),
1201            (int64_t)(_last_safepoint_end_time_ns - _last_safepoint_cleanup_time_ns),
1202            (int64_t)(_last_safepoint_end_time_ns - _last_safepoint_begin_time_ns));
1203 
1204   ls.print_cr(INT32_FORMAT_W(16), _page_trap);
1205 }
1206 
1207 // This method will be called when VM exits. This tries to summarize the sampling.
1208 // Current thread may already be deleted, so don't use ResourceMark.
1209 void SafepointTracing::statistics_exit_log() {
1210   if (!log_is_enabled(Info, safepoint, stats)) {
1211     return;
1212   }
1213   for (int index = 0; index < VM_Operation::VMOp_Terminating; index++) {
1214     if (_op_count[index] != 0) {
1215       log_info(safepoint, stats)("%-28s" UINT64_FORMAT_W(10), VM_Operation::name(index),
1216                _op_count[index]);
1217     }
1218   }
1219 
1220   log_info(safepoint, stats)("VM operations coalesced during safepoint " INT64_FORMAT,
1221                               VMThread::get_coalesced_count());
1222   log_info(safepoint, stats)("Maximum sync time  " INT64_FORMAT" ns",
1223                               (int64_t)(_max_sync_time));
1224   log_info(safepoint, stats)("Maximum vm operation time (except for Exit VM operation)  "
1225                               INT64_FORMAT " ns",
1226                               (int64_t)(_max_vmop_time));
1227 }
1228 
1229 void SafepointTracing::begin(VM_Operation::VMOp_Type type) {
1230   _op_count[type]++;
1231   _current_type = type;
1232 
1233   // update the time stamp to begin recording safepoint time
1234   _last_safepoint_begin_time_ns = os::javaTimeNanos();
1235   _last_safepoint_sync_time_ns = 0;
1236   _last_safepoint_cleanup_time_ns = 0;
1237 
1238   _last_app_time_ns = _last_safepoint_begin_time_ns - _last_safepoint_end_time_ns;
1239   _last_safepoint_end_time_ns = 0;
1240 
1241   RuntimeService::record_safepoint_begin(_last_app_time_ns);
1242 }
1243 
1244 void SafepointTracing::synchronized(int nof_threads, int nof_running, int traps) {
1245   _last_safepoint_sync_time_ns = os::javaTimeNanos();
1246   _nof_threads = nof_threads;
1247   _nof_running = nof_running;
1248   _page_trap   = traps;
1249   RuntimeService::record_safepoint_synchronized(_last_safepoint_sync_time_ns - _last_safepoint_begin_time_ns);
1250 }
1251 
1252 void SafepointTracing::cleanup() {
1253   _last_safepoint_cleanup_time_ns = os::javaTimeNanos();
1254 }
1255 
1256 void SafepointTracing::end() {
1257   _last_safepoint_end_time_ns = os::javaTimeNanos();
1258   // amount of time since epoch
1259   _last_safepoint_end_time_epoch_ms = os::javaTimeMillis();
1260 
1261   if (_max_sync_time < (_last_safepoint_sync_time_ns - _last_safepoint_begin_time_ns)) {
1262     _max_sync_time = _last_safepoint_sync_time_ns - _last_safepoint_begin_time_ns;
1263   }
1264   if (_max_vmop_time < (_last_safepoint_end_time_ns - _last_safepoint_sync_time_ns)) {
1265     _max_vmop_time = _last_safepoint_end_time_ns - _last_safepoint_sync_time_ns;
1266   }
1267   if (log_is_enabled(Info, safepoint, stats)) {
1268     statistics_log();
1269   }
1270 
1271   log_info(safepoint)(
1272      "Safepoint \"%s\", "
1273      "Time since last: " JLONG_FORMAT " ns, "
1274      "Reaching safepoint: " JLONG_FORMAT " ns, "
1275      "At safepoint: " JLONG_FORMAT " ns, "
1276      "Total: " JLONG_FORMAT " ns",
1277       VM_Operation::name(_current_type),
1278       _last_app_time_ns,
1279       _last_safepoint_cleanup_time_ns - _last_safepoint_begin_time_ns,
1280       _last_safepoint_end_time_ns     - _last_safepoint_cleanup_time_ns,
1281       _last_safepoint_end_time_ns     - _last_safepoint_begin_time_ns
1282      );
1283 
1284   RuntimeService::record_safepoint_end(_last_safepoint_end_time_ns - _last_safepoint_cleanup_time_ns);
1285 }