1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450
|
/*
* Copyright (C) 2011 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "thread_list.h"
#include <dirent.h>
#include <sys/types.h>
#include <unistd.h>
#include <sstream>
#include <vector>
#include "android-base/stringprintf.h"
#include "backtrace/BacktraceMap.h"
#include "nativehelper/scoped_local_ref.h"
#include "nativehelper/scoped_utf_chars.h"
#include "base/aborting.h"
#include "base/histogram-inl.h"
#include "base/mutex-inl.h"
#include "base/systrace.h"
#include "base/time_utils.h"
#include "base/timing_logger.h"
#include "debugger.h"
#include "gc/collector/concurrent_copying.h"
#include "gc/gc_pause_listener.h"
#include "gc/heap.h"
#include "gc/reference_processor.h"
#include "gc_root.h"
#include "jni/jni_internal.h"
#include "lock_word.h"
#include "monitor.h"
#include "native_stack_dump.h"
#include "scoped_thread_state_change-inl.h"
#include "thread.h"
#include "trace.h"
#include "well_known_classes.h"
#if ART_USE_FUTEXES
#include "linux/futex.h"
#include "sys/syscall.h"
#ifndef SYS_futex
#define SYS_futex __NR_futex
#endif
#endif // ART_USE_FUTEXES
namespace art {
using android::base::StringPrintf;
static constexpr uint64_t kLongThreadSuspendThreshold = MsToNs(5);
// Use 0 since we want to yield to prevent blocking for an unpredictable amount of time.
static constexpr useconds_t kThreadSuspendInitialSleepUs = 0;
static constexpr useconds_t kThreadSuspendMaxYieldUs = 3000;
static constexpr useconds_t kThreadSuspendMaxSleepUs = 5000;
// Whether we should try to dump the native stack of unattached threads. See commit ed8b723 for
// some history.
static constexpr bool kDumpUnattachedThreadNativeStackForSigQuit = true;
ThreadList::ThreadList(uint64_t thread_suspend_timeout_ns)
: suspend_all_count_(0),
unregistering_count_(0),
suspend_all_historam_("suspend all histogram", 16, 64),
long_suspend_(false),
shut_down_(false),
thread_suspend_timeout_ns_(thread_suspend_timeout_ns),
empty_checkpoint_barrier_(new Barrier(0)) {
CHECK(Monitor::IsValidLockWord(LockWord::FromThinLockId(kMaxThreadId, 1, 0U)));
}
ThreadList::~ThreadList() {
CHECK(shut_down_);
}
void ThreadList::ShutDown() {
ScopedTrace trace(__PRETTY_FUNCTION__);
// Detach the current thread if necessary. If we failed to start, there might not be any threads.
// We need to detach the current thread here in case there's another thread waiting to join with
// us.
bool contains = false;
Thread* self = Thread::Current();
{
MutexLock mu(self, *Locks::thread_list_lock_);
contains = Contains(self);
}
if (contains) {
Runtime::Current()->DetachCurrentThread();
}
WaitForOtherNonDaemonThreadsToExit();
// Disable GC and wait for GC to complete in case there are still daemon threads doing
// allocations.
gc::Heap* const heap = Runtime::Current()->GetHeap();
heap->DisableGCForShutdown();
// In case a GC is in progress, wait for it to finish.
heap->WaitForGcToComplete(gc::kGcCauseBackground, Thread::Current());
// TODO: there's an unaddressed race here where a thread may attach during shutdown, see
// Thread::Init.
SuspendAllDaemonThreadsForShutdown();
shut_down_ = true;
}
bool ThreadList::Contains(Thread* thread) {
return find(list_.begin(), list_.end(), thread) != list_.end();
}
bool ThreadList::Contains(pid_t tid) {
for (const auto& thread : list_) {
if (thread->GetTid() == tid) {
return true;
}
}
return false;
}
pid_t ThreadList::GetLockOwner() {
return Locks::thread_list_lock_->GetExclusiveOwnerTid();
}
void ThreadList::DumpNativeStacks(std::ostream& os) {
MutexLock mu(Thread::Current(), *Locks::thread_list_lock_);
std::unique_ptr<BacktraceMap> map(BacktraceMap::Create(getpid()));
for (const auto& thread : list_) {
os << "DUMPING THREAD " << thread->GetTid() << "\n";
DumpNativeStack(os, thread->GetTid(), map.get(), "\t");
os << "\n";
}
}
void ThreadList::DumpForSigQuit(std::ostream& os) {
{
ScopedObjectAccess soa(Thread::Current());
// Only print if we have samples.
if (suspend_all_historam_.SampleSize() > 0) {
Histogram<uint64_t>::CumulativeData data;
suspend_all_historam_.CreateHistogram(&data);
suspend_all_historam_.PrintConfidenceIntervals(os, 0.99, data); // Dump time to suspend.
}
}
bool dump_native_stack = Runtime::Current()->GetDumpNativeStackOnSigQuit();
Dump(os, dump_native_stack);
DumpUnattachedThreads(os, dump_native_stack && kDumpUnattachedThreadNativeStackForSigQuit);
}
static void DumpUnattachedThread(std::ostream& os, pid_t tid, bool dump_native_stack)
NO_THREAD_SAFETY_ANALYSIS {
// TODO: No thread safety analysis as DumpState with a null thread won't access fields, should
// refactor DumpState to avoid skipping analysis.
Thread::DumpState(os, nullptr, tid);
if (dump_native_stack) {
DumpNativeStack(os, tid, nullptr, " native: ");
}
os << std::endl;
}
void ThreadList::DumpUnattachedThreads(std::ostream& os, bool dump_native_stack) {
DIR* d = opendir("/proc/self/task");
if (!d) {
return;
}
Thread* self = Thread::Current();
dirent* e;
while ((e = readdir(d)) != nullptr) {
char* end;
pid_t tid = strtol(e->d_name, &end, 10);
if (!*end) {
bool contains;
{
MutexLock mu(self, *Locks::thread_list_lock_);
contains = Contains(tid);
}
if (!contains) {
DumpUnattachedThread(os, tid, dump_native_stack);
}
}
}
closedir(d);
}
// Dump checkpoint timeout in milliseconds. Larger amount on the target, since the device could be
// overloaded with ANR dumps.
static constexpr uint32_t kDumpWaitTimeout = kIsTargetBuild ? 100000 : 20000;
// A closure used by Thread::Dump.
class DumpCheckpoint final : public Closure {
public:
DumpCheckpoint(std::ostream* os, bool dump_native_stack)
: os_(os),
// Avoid verifying count in case a thread doesn't end up passing through the barrier.
// This avoids a SIGABRT that would otherwise happen in the destructor.
barrier_(0, /*verify_count_on_shutdown=*/false),
backtrace_map_(dump_native_stack ? BacktraceMap::Create(getpid()) : nullptr),
dump_native_stack_(dump_native_stack) {
if (backtrace_map_ != nullptr) {
backtrace_map_->SetSuffixesToIgnore(std::vector<std::string> { "oat", "odex" });
}
}
void Run(Thread* thread) override {
// Note thread and self may not be equal if thread was already suspended at the point of the
// request.
Thread* self = Thread::Current();
CHECK(self != nullptr);
std::ostringstream local_os;
{
ScopedObjectAccess soa(self);
thread->Dump(local_os, dump_native_stack_, backtrace_map_.get());
}
{
// Use the logging lock to ensure serialization when writing to the common ostream.
MutexLock mu(self, *Locks::logging_lock_);
*os_ << local_os.str() << std::endl;
}
barrier_.Pass(self);
}
void WaitForThreadsToRunThroughCheckpoint(size_t threads_running_checkpoint) {
Thread* self = Thread::Current();
ScopedThreadStateChange tsc(self, kWaitingForCheckPointsToRun);
bool timed_out = barrier_.Increment(self, threads_running_checkpoint, kDumpWaitTimeout);
if (timed_out) {
// Avoid a recursive abort.
LOG((kIsDebugBuild && (gAborting == 0)) ? ::android::base::FATAL : ::android::base::ERROR)
<< "Unexpected time out during dump checkpoint.";
}
}
private:
// The common stream that will accumulate all the dumps.
std::ostream* const os_;
// The barrier to be passed through and for the requestor to wait upon.
Barrier barrier_;
// A backtrace map, so that all threads use a shared info and don't reacquire/parse separately.
std::unique_ptr<BacktraceMap> backtrace_map_;
// Whether we should dump the native stack.
const bool dump_native_stack_;
};
void ThreadList::Dump(std::ostream& os, bool dump_native_stack) {
Thread* self = Thread::Current();
{
MutexLock mu(self, *Locks::thread_list_lock_);
os << "DALVIK THREADS (" << list_.size() << "):\n";
}
if (self != nullptr) {
DumpCheckpoint checkpoint(&os, dump_native_stack);
size_t threads_running_checkpoint;
{
// Use SOA to prevent deadlocks if multiple threads are calling Dump() at the same time.
ScopedObjectAccess soa(self);
threads_running_checkpoint = RunCheckpoint(&checkpoint);
}
if (threads_running_checkpoint != 0) {
checkpoint.WaitForThreadsToRunThroughCheckpoint(threads_running_checkpoint);
}
} else {
DumpUnattachedThreads(os, dump_native_stack);
}
}
void ThreadList::AssertThreadsAreSuspended(Thread* self, Thread* ignore1, Thread* ignore2) {
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
for (const auto& thread : list_) {
if (thread != ignore1 && thread != ignore2) {
CHECK(thread->IsSuspended())
<< "\nUnsuspended thread: <<" << *thread << "\n"
<< "self: <<" << *Thread::Current();
}
}
}
#if HAVE_TIMED_RWLOCK
// Attempt to rectify locks so that we dump thread list with required locks before exiting.
NO_RETURN static void UnsafeLogFatalForThreadSuspendAllTimeout() {
// Increment gAborting before doing the thread list dump since we don't want any failures from
// AssertThreadSuspensionIsAllowable in cases where thread suspension is not allowed.
// See b/69044468.
++gAborting;
Runtime* runtime = Runtime::Current();
std::ostringstream ss;
ss << "Thread suspend timeout\n";
Locks::mutator_lock_->Dump(ss);
ss << "\n";
runtime->GetThreadList()->Dump(ss);
--gAborting;
LOG(FATAL) << ss.str();
exit(0);
}
#endif
// Unlike suspending all threads where we can wait to acquire the mutator_lock_, suspending an
// individual thread requires polling. delay_us is the requested sleep wait. If delay_us is 0 then
// we use sched_yield instead of calling usleep.
// Although there is the possibility, here and elsewhere, that usleep could return -1 and
// errno = EINTR, there should be no problem if interrupted, so we do not check.
static void ThreadSuspendSleep(useconds_t delay_us) {
if (delay_us == 0) {
sched_yield();
} else {
usleep(delay_us);
}
}
size_t ThreadList::RunCheckpoint(Closure* checkpoint_function, Closure* callback) {
Thread* self = Thread::Current();
Locks::mutator_lock_->AssertNotExclusiveHeld(self);
Locks::thread_list_lock_->AssertNotHeld(self);
Locks::thread_suspend_count_lock_->AssertNotHeld(self);
std::vector<Thread*> suspended_count_modified_threads;
size_t count = 0;
{
// Call a checkpoint function for each thread, threads which are suspend get their checkpoint
// manually called.
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
count = list_.size();
for (const auto& thread : list_) {
if (thread != self) {
bool requested_suspend = false;
while (true) {
if (thread->RequestCheckpoint(checkpoint_function)) {
// This thread will run its checkpoint some time in the near future.
if (requested_suspend) {
// The suspend request is now unnecessary.
bool updated =
thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
requested_suspend = false;
}
break;
} else {
// The thread is probably suspended, try to make sure that it stays suspended.
if (thread->GetState() == kRunnable) {
// Spurious fail, try again.
continue;
}
if (!requested_suspend) {
bool updated =
thread->ModifySuspendCount(self, +1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
requested_suspend = true;
if (thread->IsSuspended()) {
break;
}
// The thread raced us to become Runnable. Try to RequestCheckpoint() again.
} else {
// The thread previously raced our suspend request to become Runnable but
// since it is suspended again, it must honor that suspend request now.
DCHECK(thread->IsSuspended());
break;
}
}
}
if (requested_suspend) {
suspended_count_modified_threads.push_back(thread);
}
}
}
// Run the callback to be called inside this critical section.
if (callback != nullptr) {
callback->Run(self);
}
}
// Run the checkpoint on ourself while we wait for threads to suspend.
checkpoint_function->Run(self);
// Run the checkpoint on the suspended threads.
for (const auto& thread : suspended_count_modified_threads) {
// We know for sure that the thread is suspended at this point.
DCHECK(thread->IsSuspended());
checkpoint_function->Run(thread);
{
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
}
}
{
// Imitate ResumeAll, threads may be waiting on Thread::resume_cond_ since we raised their
// suspend count. Now the suspend_count_ is lowered so we must do the broadcast.
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Thread::resume_cond_->Broadcast(self);
}
return count;
}
void ThreadList::RunEmptyCheckpoint() {
Thread* self = Thread::Current();
Locks::mutator_lock_->AssertNotExclusiveHeld(self);
Locks::thread_list_lock_->AssertNotHeld(self);
Locks::thread_suspend_count_lock_->AssertNotHeld(self);
std::vector<uint32_t> runnable_thread_ids;
size_t count = 0;
Barrier* barrier = empty_checkpoint_barrier_.get();
barrier->Init(self, 0);
{
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
for (Thread* thread : list_) {
if (thread != self) {
while (true) {
if (thread->RequestEmptyCheckpoint()) {
// This thread will run an empty checkpoint (decrement the empty checkpoint barrier)
// some time in the near future.
++count;
if (kIsDebugBuild) {
runnable_thread_ids.push_back(thread->GetThreadId());
}
break;
}
if (thread->GetState() != kRunnable) {
// It's seen suspended, we are done because it must not be in the middle of a mutator
// heap access.
break;
}
}
}
}
}
// Wake up the threads blocking for weak ref access so that they will respond to the empty
// checkpoint request. Otherwise we will hang as they are blocking in the kRunnable state.
Runtime::Current()->GetHeap()->GetReferenceProcessor()->BroadcastForSlowPath(self);
Runtime::Current()->BroadcastForNewSystemWeaks(/*broadcast_for_checkpoint=*/true);
{
ScopedThreadStateChange tsc(self, kWaitingForCheckPointsToRun);
uint64_t total_wait_time = 0;
bool first_iter = true;
while (true) {
// Wake up the runnable threads blocked on the mutexes that another thread, which is blocked
// on a weak ref access, holds (indirectly blocking for weak ref access through another thread
// and a mutex.) This needs to be done periodically because the thread may be preempted
// between the CheckEmptyCheckpointFromMutex call and the subsequent futex wait in
// Mutex::ExclusiveLock, etc. when the wakeup via WakeupToRespondToEmptyCheckpoint
// arrives. This could cause a *very rare* deadlock, if not repeated. Most of the cases are
// handled in the first iteration.
for (BaseMutex* mutex : Locks::expected_mutexes_on_weak_ref_access_) {
mutex->WakeupToRespondToEmptyCheckpoint();
}
static constexpr uint64_t kEmptyCheckpointPeriodicTimeoutMs = 100; // 100ms
static constexpr uint64_t kEmptyCheckpointTotalTimeoutMs = 600 * 1000; // 10 minutes.
size_t barrier_count = first_iter ? count : 0;
first_iter = false; // Don't add to the barrier count from the second iteration on.
bool timed_out = barrier->Increment(self, barrier_count, kEmptyCheckpointPeriodicTimeoutMs);
if (!timed_out) {
break; // Success
}
// This is a very rare case.
total_wait_time += kEmptyCheckpointPeriodicTimeoutMs;
if (kIsDebugBuild && total_wait_time > kEmptyCheckpointTotalTimeoutMs) {
std::ostringstream ss;
ss << "Empty checkpoint timeout\n";
ss << "Barrier count " << barrier->GetCount(self) << "\n";
ss << "Runnable thread IDs";
for (uint32_t tid : runnable_thread_ids) {
ss << " " << tid;
}
ss << "\n";
Locks::mutator_lock_->Dump(ss);
ss << "\n";
LOG(FATAL_WITHOUT_ABORT) << ss.str();
// Some threads in 'runnable_thread_ids' are probably stuck. Try to dump their stacks.
// Avoid using ThreadList::Dump() initially because it is likely to get stuck as well.
{
ScopedObjectAccess soa(self);
MutexLock mu1(self, *Locks::thread_list_lock_);
for (Thread* thread : GetList()) {
uint32_t tid = thread->GetThreadId();
bool is_in_runnable_thread_ids =
std::find(runnable_thread_ids.begin(), runnable_thread_ids.end(), tid) !=
runnable_thread_ids.end();
if (is_in_runnable_thread_ids &&
thread->ReadFlag(kEmptyCheckpointRequest)) {
// Found a runnable thread that hasn't responded to the empty checkpoint request.
// Assume it's stuck and safe to dump its stack.
thread->Dump(LOG_STREAM(FATAL_WITHOUT_ABORT),
/*dump_native_stack=*/ true,
/*backtrace_map=*/ nullptr,
/*force_dump_stack=*/ true);
}
}
}
LOG(FATAL_WITHOUT_ABORT)
<< "Dumped runnable threads that haven't responded to empty checkpoint.";
// Now use ThreadList::Dump() to dump more threads, noting it may get stuck.
Dump(LOG_STREAM(FATAL_WITHOUT_ABORT));
LOG(FATAL) << "Dumped all threads.";
}
}
}
}
// A checkpoint/suspend-all hybrid to switch thread roots from
// from-space to to-space refs. Used to synchronize threads at a point
// to mark the initiation of marking while maintaining the to-space
// invariant.
size_t ThreadList::FlipThreadRoots(Closure* thread_flip_visitor,
Closure* flip_callback,
gc::collector::GarbageCollector* collector,
gc::GcPauseListener* pause_listener) {
TimingLogger::ScopedTiming split("ThreadListFlip", collector->GetTimings());
Thread* self = Thread::Current();
Locks::mutator_lock_->AssertNotHeld(self);
Locks::thread_list_lock_->AssertNotHeld(self);
Locks::thread_suspend_count_lock_->AssertNotHeld(self);
CHECK_NE(self->GetState(), kRunnable);
collector->GetHeap()->ThreadFlipBegin(self); // Sync with JNI critical calls.
// ThreadFlipBegin happens before we suspend all the threads, so it does not count towards the
// pause.
const uint64_t suspend_start_time = NanoTime();
SuspendAllInternal(self, self, nullptr);
if (pause_listener != nullptr) {
pause_listener->StartPause();
}
// Run the flip callback for the collector.
Locks::mutator_lock_->ExclusiveLock(self);
suspend_all_historam_.AdjustAndAddValue(NanoTime() - suspend_start_time);
flip_callback->Run(self);
Locks::mutator_lock_->ExclusiveUnlock(self);
collector->RegisterPause(NanoTime() - suspend_start_time);
if (pause_listener != nullptr) {
pause_listener->EndPause();
}
// Resume runnable threads.
size_t runnable_thread_count = 0;
std::vector<Thread*> other_threads;
{
TimingLogger::ScopedTiming split2("ResumeRunnableThreads", collector->GetTimings());
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
--suspend_all_count_;
for (const auto& thread : list_) {
// Set the flip function for all threads because Thread::DumpState/DumpJavaStack() (invoked by
// a checkpoint) may cause the flip function to be run for a runnable/suspended thread before
// a runnable thread runs it for itself or we run it for a suspended thread below.
thread->SetFlipFunction(thread_flip_visitor);
if (thread == self) {
continue;
}
// Resume early the threads that were runnable but are suspended just for this thread flip or
// about to transition from non-runnable (eg. kNative at the SOA entry in a JNI function) to
// runnable (both cases waiting inside Thread::TransitionFromSuspendedToRunnable), or waiting
// for the thread flip to end at the JNI critical section entry (kWaitingForGcThreadFlip),
ThreadState state = thread->GetState();
if ((state == kWaitingForGcThreadFlip || thread->IsTransitioningToRunnable()) &&
thread->GetSuspendCount() == 1) {
// The thread will resume right after the broadcast.
bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
++runnable_thread_count;
} else {
other_threads.push_back(thread);
}
}
Thread::resume_cond_->Broadcast(self);
}
collector->GetHeap()->ThreadFlipEnd(self);
// Run the closure on the other threads and let them resume.
{
TimingLogger::ScopedTiming split3("FlipOtherThreads", collector->GetTimings());
ReaderMutexLock mu(self, *Locks::mutator_lock_);
for (const auto& thread : other_threads) {
Closure* flip_func = thread->GetFlipFunction();
if (flip_func != nullptr) {
flip_func->Run(thread);
}
}
// Run it for self.
Closure* flip_func = self->GetFlipFunction();
if (flip_func != nullptr) {
flip_func->Run(self);
}
}
// Resume other threads.
{
TimingLogger::ScopedTiming split4("ResumeOtherThreads", collector->GetTimings());
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
for (const auto& thread : other_threads) {
bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
}
Thread::resume_cond_->Broadcast(self);
}
return runnable_thread_count + other_threads.size() + 1; // +1 for self.
}
void ThreadList::SuspendAll(const char* cause, bool long_suspend) {
Thread* self = Thread::Current();
if (self != nullptr) {
VLOG(threads) << *self << " SuspendAll for " << cause << " starting...";
} else {
VLOG(threads) << "Thread[null] SuspendAll for " << cause << " starting...";
}
{
ScopedTrace trace("Suspending mutator threads");
const uint64_t start_time = NanoTime();
SuspendAllInternal(self, self);
// All threads are known to have suspended (but a thread may still own the mutator lock)
// Make sure this thread grabs exclusive access to the mutator lock and its protected data.
#if HAVE_TIMED_RWLOCK
while (true) {
if (Locks::mutator_lock_->ExclusiveLockWithTimeout(self,
NsToMs(thread_suspend_timeout_ns_),
0)) {
break;
} else if (!long_suspend_) {
// Reading long_suspend without the mutator lock is slightly racy, in some rare cases, this
// could result in a thread suspend timeout.
// Timeout if we wait more than thread_suspend_timeout_ns_ nanoseconds.
UnsafeLogFatalForThreadSuspendAllTimeout();
}
}
#else
Locks::mutator_lock_->ExclusiveLock(self);
#endif
long_suspend_ = long_suspend;
const uint64_t end_time = NanoTime();
const uint64_t suspend_time = end_time - start_time;
suspend_all_historam_.AdjustAndAddValue(suspend_time);
if (suspend_time > kLongThreadSuspendThreshold) {
LOG(WARNING) << "Suspending all threads took: " << PrettyDuration(suspend_time);
}
if (kDebugLocking) {
// Debug check that all threads are suspended.
AssertThreadsAreSuspended(self, self);
}
}
ATraceBegin((std::string("Mutator threads suspended for ") + cause).c_str());
if (self != nullptr) {
VLOG(threads) << *self << " SuspendAll complete";
} else {
VLOG(threads) << "Thread[null] SuspendAll complete";
}
}
// Ensures all threads running Java suspend and that those not running Java don't start.
void ThreadList::SuspendAllInternal(Thread* self,
Thread* ignore1,
Thread* ignore2,
SuspendReason reason) {
Locks::mutator_lock_->AssertNotExclusiveHeld(self);
Locks::thread_list_lock_->AssertNotHeld(self);
Locks::thread_suspend_count_lock_->AssertNotHeld(self);
if (kDebugLocking && self != nullptr) {
CHECK_NE(self->GetState(), kRunnable);
}
// First request that all threads suspend, then wait for them to suspend before
// returning. This suspension scheme also relies on other behaviour:
// 1. Threads cannot be deleted while they are suspended or have a suspend-
// request flag set - (see Unregister() below).
// 2. When threads are created, they are created in a suspended state (actually
// kNative) and will never begin executing Java code without first checking
// the suspend-request flag.
// The atomic counter for number of threads that need to pass the barrier.
AtomicInteger pending_threads;
uint32_t num_ignored = 0;
if (ignore1 != nullptr) {
++num_ignored;
}
if (ignore2 != nullptr && ignore1 != ignore2) {
++num_ignored;
}
{
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
// Update global suspend all state for attaching threads.
++suspend_all_count_;
pending_threads.store(list_.size() - num_ignored, std::memory_order_relaxed);
// Increment everybody's suspend count (except those that should be ignored).
for (const auto& thread : list_) {
if (thread == ignore1 || thread == ignore2) {
continue;
}
VLOG(threads) << "requesting thread suspend: " << *thread;
bool updated = thread->ModifySuspendCount(self, +1, &pending_threads, reason);
DCHECK(updated);
// Must install the pending_threads counter first, then check thread->IsSuspend() and clear
// the counter. Otherwise there's a race with Thread::TransitionFromRunnableToSuspended()
// that can lead a thread to miss a call to PassActiveSuspendBarriers().
if (thread->IsSuspended()) {
// Only clear the counter for the current thread.
thread->ClearSuspendBarrier(&pending_threads);
pending_threads.fetch_sub(1, std::memory_order_seq_cst);
}
}
}
// Wait for the barrier to be passed by all runnable threads. This wait
// is done with a timeout so that we can detect problems.
#if ART_USE_FUTEXES
timespec wait_timeout;
InitTimeSpec(false, CLOCK_MONOTONIC, NsToMs(thread_suspend_timeout_ns_), 0, &wait_timeout);
#endif
const uint64_t start_time = NanoTime();
while (true) {
int32_t cur_val = pending_threads.load(std::memory_order_relaxed);
if (LIKELY(cur_val > 0)) {
#if ART_USE_FUTEXES
if (futex(pending_threads.Address(), FUTEX_WAIT_PRIVATE, cur_val, &wait_timeout, nullptr, 0)
!= 0) {
if ((errno == EAGAIN) || (errno == EINTR)) {
// EAGAIN and EINTR both indicate a spurious failure, try again from the beginning.
continue;
}
if (errno == ETIMEDOUT) {
const uint64_t wait_time = NanoTime() - start_time;
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
std::ostringstream oss;
for (const auto& thread : list_) {
if (thread == ignore1 || thread == ignore2) {
continue;
}
if (!thread->IsSuspended()) {
oss << std::endl << "Thread not suspended: " << *thread;
}
}
LOG(kIsDebugBuild ? ::android::base::FATAL : ::android::base::ERROR)
<< "Timed out waiting for threads to suspend, waited for "
<< PrettyDuration(wait_time)
<< oss.str();
} else {
PLOG(FATAL) << "futex wait failed for SuspendAllInternal()";
}
} // else re-check pending_threads in the next iteration (this may be a spurious wake-up).
#else
// Spin wait. This is likely to be slow, but on most architecture ART_USE_FUTEXES is set.
UNUSED(start_time);
#endif
} else {
CHECK_EQ(cur_val, 0);
break;
}
}
}
void ThreadList::ResumeAll() {
Thread* self = Thread::Current();
if (self != nullptr) {
VLOG(threads) << *self << " ResumeAll starting";
} else {
VLOG(threads) << "Thread[null] ResumeAll starting";
}
ATraceEnd();
ScopedTrace trace("Resuming mutator threads");
if (kDebugLocking) {
// Debug check that all threads are suspended.
AssertThreadsAreSuspended(self, self);
}
long_suspend_ = false;
Locks::mutator_lock_->ExclusiveUnlock(self);
{
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
// Update global suspend all state for attaching threads.
--suspend_all_count_;
// Decrement the suspend counts for all threads.
for (const auto& thread : list_) {
if (thread == self) {
continue;
}
bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
}
// Broadcast a notification to all suspended threads, some or all of
// which may choose to wake up. No need to wait for them.
if (self != nullptr) {
VLOG(threads) << *self << " ResumeAll waking others";
} else {
VLOG(threads) << "Thread[null] ResumeAll waking others";
}
Thread::resume_cond_->Broadcast(self);
}
if (self != nullptr) {
VLOG(threads) << *self << " ResumeAll complete";
} else {
VLOG(threads) << "Thread[null] ResumeAll complete";
}
}
bool ThreadList::Resume(Thread* thread, SuspendReason reason) {
// This assumes there was an ATraceBegin when we suspended the thread.
ATraceEnd();
Thread* self = Thread::Current();
DCHECK_NE(thread, self);
VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") starting..." << reason;
{
// To check Contains.
MutexLock mu(self, *Locks::thread_list_lock_);
// To check IsSuspended.
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
if (UNLIKELY(!thread->IsSuspended())) {
LOG(ERROR) << "Resume(" << reinterpret_cast<void*>(thread)
<< ") thread not suspended";
return false;
}
if (!Contains(thread)) {
// We only expect threads within the thread-list to have been suspended otherwise we can't
// stop such threads from delete-ing themselves.
LOG(ERROR) << "Resume(" << reinterpret_cast<void*>(thread)
<< ") thread not within thread list";
return false;
}
if (UNLIKELY(!thread->ModifySuspendCount(self, -1, nullptr, reason))) {
LOG(ERROR) << "Resume(" << reinterpret_cast<void*>(thread)
<< ") could not modify suspend count.";
return false;
}
}
{
VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") waking others";
MutexLock mu(self, *Locks::thread_suspend_count_lock_);
Thread::resume_cond_->Broadcast(self);
}
VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") complete";
return true;
}
static void ThreadSuspendByPeerWarning(Thread* self,
LogSeverity severity,
const char* message,
jobject peer) {
JNIEnvExt* env = self->GetJniEnv();
ScopedLocalRef<jstring>
scoped_name_string(env, static_cast<jstring>(env->GetObjectField(
peer, WellKnownClasses::java_lang_Thread_name)));
ScopedUtfChars scoped_name_chars(env, scoped_name_string.get());
if (scoped_name_chars.c_str() == nullptr) {
LOG(severity) << message << ": " << peer;
env->ExceptionClear();
} else {
LOG(severity) << message << ": " << peer << ":" << scoped_name_chars.c_str();
}
}
Thread* ThreadList::SuspendThreadByPeer(jobject peer,
bool request_suspension,
SuspendReason reason,
bool* timed_out) {
const uint64_t start_time = NanoTime();
useconds_t sleep_us = kThreadSuspendInitialSleepUs;
*timed_out = false;
Thread* const self = Thread::Current();
Thread* suspended_thread = nullptr;
VLOG(threads) << "SuspendThreadByPeer starting";
while (true) {
Thread* thread;
{
// Note: this will transition to runnable and potentially suspend. We ensure only one thread
// is requesting another suspend, to avoid deadlock, by requiring this function be called
// holding Locks::thread_list_suspend_thread_lock_. Its important this thread suspend rather
// than request thread suspension, to avoid potential cycles in threads requesting each other
// suspend.
ScopedObjectAccess soa(self);
MutexLock thread_list_mu(self, *Locks::thread_list_lock_);
thread = Thread::FromManagedThread(soa, peer);
if (thread == nullptr) {
if (suspended_thread != nullptr) {
MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_);
// If we incremented the suspend count but the thread reset its peer, we need to
// re-decrement it since it is shutting down and may deadlock the runtime in
// ThreadList::WaitForOtherNonDaemonThreadsToExit.
bool updated = suspended_thread->ModifySuspendCount(soa.Self(),
-1,
nullptr,
reason);
DCHECK(updated);
}
ThreadSuspendByPeerWarning(self,
::android::base::WARNING,
"No such thread for suspend",
peer);
return nullptr;
}
if (!Contains(thread)) {
CHECK(suspended_thread == nullptr);
VLOG(threads) << "SuspendThreadByPeer failed for unattached thread: "
<< reinterpret_cast<void*>(thread);
return nullptr;
}
VLOG(threads) << "SuspendThreadByPeer found thread: " << *thread;
{
MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_);
if (request_suspension) {
if (self->GetSuspendCount() > 0) {
// We hold the suspend count lock but another thread is trying to suspend us. Its not
// safe to try to suspend another thread in case we get a cycle. Start the loop again
// which will allow this thread to be suspended.
continue;
}
CHECK(suspended_thread == nullptr);
suspended_thread = thread;
bool updated = suspended_thread->ModifySuspendCount(self, +1, nullptr, reason);
DCHECK(updated);
request_suspension = false;
} else {
// If the caller isn't requesting suspension, a suspension should have already occurred.
CHECK_GT(thread->GetSuspendCount(), 0);
}
// IsSuspended on the current thread will fail as the current thread is changed into
// Runnable above. As the suspend count is now raised if this is the current thread
// it will self suspend on transition to Runnable, making it hard to work with. It's simpler
// to just explicitly handle the current thread in the callers to this code.
CHECK_NE(thread, self) << "Attempt to suspend the current thread for the debugger";
// If thread is suspended (perhaps it was already not Runnable but didn't have a suspend
// count, or else we've waited and it has self suspended) or is the current thread, we're
// done.
if (thread->IsSuspended()) {
VLOG(threads) << "SuspendThreadByPeer thread suspended: " << *thread;
if (ATraceEnabled()) {
std::string name;
thread->GetThreadName(name);
ATraceBegin(StringPrintf("SuspendThreadByPeer suspended %s for peer=%p", name.c_str(),
peer).c_str());
}
return thread;
}
const uint64_t total_delay = NanoTime() - start_time;
if (total_delay >= thread_suspend_timeout_ns_) {
ThreadSuspendByPeerWarning(self,
::android::base::FATAL,
"Thread suspension timed out",
peer);
if (suspended_thread != nullptr) {
CHECK_EQ(suspended_thread, thread);
bool updated = suspended_thread->ModifySuspendCount(soa.Self(),
-1,
nullptr,
reason);
DCHECK(updated);
}
*timed_out = true;
return nullptr;
} else if (sleep_us == 0 &&
total_delay > static_cast<uint64_t>(kThreadSuspendMaxYieldUs) * 1000) {
// We have spun for kThreadSuspendMaxYieldUs time, switch to sleeps to prevent
// excessive CPU usage.
sleep_us = kThreadSuspendMaxYieldUs / 2;
}
}
// Release locks and come out of runnable state.
}
VLOG(threads) << "SuspendThreadByPeer waiting to allow thread chance to suspend";
ThreadSuspendSleep(sleep_us);
// This may stay at 0 if sleep_us == 0, but this is WAI since we want to avoid using usleep at
// all if possible. This shouldn't be an issue since time to suspend should always be small.
sleep_us = std::min(sleep_us * 2, kThreadSuspendMaxSleepUs);
}
}
static void ThreadSuspendByThreadIdWarning(LogSeverity severity,
const char* message,
uint32_t thread_id) {
LOG(severity) << StringPrintf("%s: %d", message, thread_id);
}
Thread* ThreadList::SuspendThreadByThreadId(uint32_t thread_id,
SuspendReason reason,
bool* timed_out) {
const uint64_t start_time = NanoTime();
useconds_t sleep_us = kThreadSuspendInitialSleepUs;
*timed_out = false;
Thread* suspended_thread = nullptr;
Thread* const self = Thread::Current();
CHECK_NE(thread_id, kInvalidThreadId);
VLOG(threads) << "SuspendThreadByThreadId starting";
while (true) {
{
// Note: this will transition to runnable and potentially suspend. We ensure only one thread
// is requesting another suspend, to avoid deadlock, by requiring this function be called
// holding Locks::thread_list_suspend_thread_lock_. Its important this thread suspend rather
// than request thread suspension, to avoid potential cycles in threads requesting each other
// suspend.
ScopedObjectAccess soa(self);
MutexLock thread_list_mu(self, *Locks::thread_list_lock_);
Thread* thread = nullptr;
for (const auto& it : list_) {
if (it->GetThreadId() == thread_id) {
thread = it;
break;
}
}
if (thread == nullptr) {
CHECK(suspended_thread == nullptr) << "Suspended thread " << suspended_thread
<< " no longer in thread list";
// There's a race in inflating a lock and the owner giving up ownership and then dying.
ThreadSuspendByThreadIdWarning(::android::base::WARNING,
"No such thread id for suspend",
thread_id);
return nullptr;
}
VLOG(threads) << "SuspendThreadByThreadId found thread: " << *thread;
DCHECK(Contains(thread));
{
MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_);
if (suspended_thread == nullptr) {
if (self->GetSuspendCount() > 0) {
// We hold the suspend count lock but another thread is trying to suspend us. Its not
// safe to try to suspend another thread in case we get a cycle. Start the loop again
// which will allow this thread to be suspended.
continue;
}
bool updated = thread->ModifySuspendCount(self, +1, nullptr, reason);
DCHECK(updated);
suspended_thread = thread;
} else {
CHECK_EQ(suspended_thread, thread);
// If the caller isn't requesting suspension, a suspension should have already occurred.
CHECK_GT(thread->GetSuspendCount(), 0);
}
// IsSuspended on the current thread will fail as the current thread is changed into
// Runnable above. As the suspend count is now raised if this is the current thread
// it will self suspend on transition to Runnable, making it hard to work with. It's simpler
// to just explicitly handle the current thread in the callers to this code.
CHECK_NE(thread, self) << "Attempt to suspend the current thread for the debugger";
// If thread is suspended (perhaps it was already not Runnable but didn't have a suspend
// count, or else we've waited and it has self suspended) or is the current thread, we're
// done.
if (thread->IsSuspended()) {
if (ATraceEnabled()) {
std::string name;
thread->GetThreadName(name);
ATraceBegin(StringPrintf("SuspendThreadByThreadId suspended %s id=%d",
name.c_str(), thread_id).c_str());
}
VLOG(threads) << "SuspendThreadByThreadId thread suspended: " << *thread;
return thread;
}
const uint64_t total_delay = NanoTime() - start_time;
if (total_delay >= thread_suspend_timeout_ns_) {
ThreadSuspendByThreadIdWarning(::android::base::WARNING,
"Thread suspension timed out",
thread_id);
if (suspended_thread != nullptr) {
bool updated = thread->ModifySuspendCount(soa.Self(), -1, nullptr, reason);
DCHECK(updated);
}
*timed_out = true;
return nullptr;
} else if (sleep_us == 0 &&
total_delay > static_cast<uint64_t>(kThreadSuspendMaxYieldUs) * 1000) {
// We have spun for kThreadSuspendMaxYieldUs time, switch to sleeps to prevent
// excessive CPU usage.
sleep_us = kThreadSuspendMaxYieldUs / 2;
}
}
// Release locks and come out of runnable state.
}
VLOG(threads) << "SuspendThreadByThreadId waiting to allow thread chance to suspend";
ThreadSuspendSleep(sleep_us);
sleep_us = std::min(sleep_us * 2, kThreadSuspendMaxSleepUs);
}
}
Thread* ThreadList::FindThreadByThreadId(uint32_t thread_id) {
for (const auto& thread : list_) {
if (thread->GetThreadId() == thread_id) {
return thread;
}
}
return nullptr;
}
void ThreadList::WaitForOtherNonDaemonThreadsToExit(bool check_no_birth) {
ScopedTrace trace(__PRETTY_FUNCTION__);
Thread* self = Thread::Current();
Locks::mutator_lock_->AssertNotHeld(self);
while (true) {
Locks::runtime_shutdown_lock_->Lock(self);
if (check_no_birth) {
// No more threads can be born after we start to shutdown.
CHECK(Runtime::Current()->IsShuttingDownLocked());
CHECK_EQ(Runtime::Current()->NumberOfThreadsBeingBorn(), 0U);
} else {
if (Runtime::Current()->NumberOfThreadsBeingBorn() != 0U) {
// Awkward. Shutdown_cond_ is private, but the only live thread may not be registered yet.
// Fortunately, this is used mostly for testing, and not performance-critical.
Locks::runtime_shutdown_lock_->Unlock(self);
usleep(1000);
continue;
}
}
MutexLock mu(self, *Locks::thread_list_lock_);
Locks::runtime_shutdown_lock_->Unlock(self);
// Also wait for any threads that are unregistering to finish. This is required so that no
// threads access the thread list after it is deleted. TODO: This may not work for user daemon
// threads since they could unregister at the wrong time.
bool done = unregistering_count_ == 0;
if (done) {
for (const auto& thread : list_) {
if (thread != self && !thread->IsDaemon()) {
done = false;
break;
}
}
}
if (done) {
break;
}
// Wait for another thread to exit before re-checking.
Locks::thread_exit_cond_->Wait(self);
}
}
void ThreadList::SuspendAllDaemonThreadsForShutdown() {
ScopedTrace trace(__PRETTY_FUNCTION__);
Thread* self = Thread::Current();
size_t daemons_left = 0;
{
// Tell all the daemons it's time to suspend.
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
for (const auto& thread : list_) {
// This is only run after all non-daemon threads have exited, so the remainder should all be
// daemons.
CHECK(thread->IsDaemon()) << *thread;
if (thread != self) {
bool updated = thread->ModifySuspendCount(self, +1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
++daemons_left;
}
// We are shutting down the runtime, set the JNI functions of all the JNIEnvs to be
// the sleep forever one.
thread->GetJniEnv()->SetFunctionsToRuntimeShutdownFunctions();
}
}
if (daemons_left == 0) {
// No threads left; safe to shut down.
return;
}
// There is not a clean way to shut down if we have daemons left. We have no mechanism for
// killing them and reclaiming thread stacks. We also have no mechanism for waiting until they
// have truly finished touching the memory we are about to deallocate. We do the best we can with
// timeouts.
//
// If we have any daemons left, wait until they are (a) suspended and (b) they are not stuck
// in a place where they are about to access runtime state and are not in a runnable state.
// We attempt to do the latter by just waiting long enough for things to
// quiesce. Examples: Monitor code or waking up from a condition variable.
//
// Give the threads a chance to suspend, complaining if they're slow. (a)
bool have_complained = false;
static constexpr size_t kTimeoutMicroseconds = 2000 * 1000;
static constexpr size_t kSleepMicroseconds = 1000;
bool all_suspended = false;
for (size_t i = 0; !all_suspended && i < kTimeoutMicroseconds / kSleepMicroseconds; ++i) {
bool found_running = false;
{
MutexLock mu(self, *Locks::thread_list_lock_);
for (const auto& thread : list_) {
if (thread != self && thread->GetState() == kRunnable) {
if (!have_complained) {
LOG(WARNING) << "daemon thread not yet suspended: " << *thread;
have_complained = true;
}
found_running = true;
}
}
}
if (found_running) {
// Sleep briefly before checking again. Max total sleep time is kTimeoutMicroseconds.
usleep(kSleepMicroseconds);
} else {
all_suspended = true;
}
}
if (!all_suspended) {
// We can get here if a daemon thread executed a fastnative native call, so that it
// remained in runnable state, and then made a JNI call after we called
// SetFunctionsToRuntimeShutdownFunctions(), causing it to permanently stay in a harmless
// but runnable state. See b/147804269 .
LOG(WARNING) << "timed out suspending all daemon threads";
}
// Assume all threads are either suspended or somehow wedged.
// Wait again for all the now "suspended" threads to actually quiesce. (b)
static constexpr size_t kDaemonSleepTime = 200 * 1000;
usleep(kDaemonSleepTime);
std::list<Thread*> list_copy;
{
MutexLock mu(self, *Locks::thread_list_lock_);
// Half-way through the wait, set the "runtime deleted" flag, causing any newly awoken
// threads to immediately go back to sleep without touching memory. This prevents us from
// touching deallocated memory, but it also prevents mutexes from getting released. Thus we
// only do this once we're reasonably sure that no system mutexes are still held.
for (const auto& thread : list_) {
DCHECK(thread == self || !all_suspended || thread->GetState() != kRunnable);
// In the !all_suspended case, the target is probably sleeping.
thread->GetJniEnv()->SetRuntimeDeleted();
// Possibly contended Mutex acquisitions are unsafe after this.
// Releasing thread_list_lock_ is OK, since it can't block.
}
}
// Finally wait for any threads woken before we set the "runtime deleted" flags to finish
// touching memory.
usleep(kDaemonSleepTime);
#if defined(__has_feature)
#if __has_feature(address_sanitizer) || __has_feature(hwaddress_sanitizer)
// Sleep a bit longer with -fsanitize=address, since everything is slower.
usleep(2 * kDaemonSleepTime);
#endif
#endif
// At this point no threads should be touching our data structures anymore.
}
void ThreadList::Register(Thread* self) {
DCHECK_EQ(self, Thread::Current());
CHECK(!shut_down_);
if (VLOG_IS_ON(threads)) {
std::ostringstream oss;
self->ShortDump(oss); // We don't hold the mutator_lock_ yet and so cannot call Dump.
LOG(INFO) << "ThreadList::Register() " << *self << "\n" << oss.str();
}
// Atomically add self to the thread list and make its thread_suspend_count_ reflect ongoing
// SuspendAll requests.
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
// Modify suspend count in increments of 1 to maintain invariants in ModifySuspendCount. While
// this isn't particularly efficient the suspend counts are most commonly 0 or 1.
for (int delta = suspend_all_count_; delta > 0; delta--) {
bool updated = self->ModifySuspendCount(self, +1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
}
CHECK(!Contains(self));
list_.push_back(self);
if (kUseReadBarrier) {
gc::collector::ConcurrentCopying* const cc =
Runtime::Current()->GetHeap()->ConcurrentCopyingCollector();
// Initialize according to the state of the CC collector.
self->SetIsGcMarkingAndUpdateEntrypoints(cc->IsMarking());
if (cc->IsUsingReadBarrierEntrypoints()) {
self->SetReadBarrierEntrypoints();
}
self->SetWeakRefAccessEnabled(cc->IsWeakRefAccessEnabled());
}
self->NotifyInTheadList();
}
void ThreadList::Unregister(Thread* self) {
DCHECK_EQ(self, Thread::Current());
CHECK_NE(self->GetState(), kRunnable);
Locks::mutator_lock_->AssertNotHeld(self);
VLOG(threads) << "ThreadList::Unregister() " << *self;
{
MutexLock mu(self, *Locks::thread_list_lock_);
++unregistering_count_;
}
// Any time-consuming destruction, plus anything that can call back into managed code or
// suspend and so on, must happen at this point, and not in ~Thread. The self->Destroy is what
// causes the threads to join. It is important to do this after incrementing unregistering_count_
// since we want the runtime to wait for the daemon threads to exit before deleting the thread
// list.
self->Destroy();
// If tracing, remember thread id and name before thread exits.
Trace::StoreExitingThreadInfo(self);
uint32_t thin_lock_id = self->GetThreadId();
while (true) {
// Remove and delete the Thread* while holding the thread_list_lock_ and
// thread_suspend_count_lock_ so that the unregistering thread cannot be suspended.
// Note: deliberately not using MutexLock that could hold a stale self pointer.
{
MutexLock mu(self, *Locks::thread_list_lock_);
if (!Contains(self)) {
std::string thread_name;
self->GetThreadName(thread_name);
std::ostringstream os;
DumpNativeStack(os, GetTid(), nullptr, " native: ", nullptr);
LOG(ERROR) << "Request to unregister unattached thread " << thread_name << "\n" << os.str();
break;
} else {
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
if (!self->IsSuspended()) {
list_.remove(self);
break;
}
}
}
// In the case where we are not suspended yet, sleep to leave other threads time to execute.
// This is important if there are realtime threads. b/111277984
usleep(1);
// We failed to remove the thread due to a suspend request, loop and try again.
}
delete self;
// Release the thread ID after the thread is finished and deleted to avoid cases where we can
// temporarily have multiple threads with the same thread id. When this occurs, it causes
// problems in FindThreadByThreadId / SuspendThreadByThreadId.
ReleaseThreadId(nullptr, thin_lock_id);
// Clear the TLS data, so that the underlying native thread is recognizably detached.
// (It may wish to reattach later.)
#ifdef __BIONIC__
__get_tls()[TLS_SLOT_ART_THREAD_SELF] = nullptr;
#else
CHECK_PTHREAD_CALL(pthread_setspecific, (Thread::pthread_key_self_, nullptr), "detach self");
Thread::self_tls_ = nullptr;
#endif
// Signal that a thread just detached.
MutexLock mu(nullptr, *Locks::thread_list_lock_);
--unregistering_count_;
Locks::thread_exit_cond_->Broadcast(nullptr);
}
void ThreadList::ForEach(void (*callback)(Thread*, void*), void* context) {
for (const auto& thread : list_) {
callback(thread, context);
}
}
void ThreadList::VisitRootsForSuspendedThreads(RootVisitor* visitor) {
Thread* const self = Thread::Current();
std::vector<Thread*> threads_to_visit;
// Tell threads to suspend and copy them into list.
{
MutexLock mu(self, *Locks::thread_list_lock_);
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
for (Thread* thread : list_) {
bool suspended = thread->ModifySuspendCount(self, +1, nullptr, SuspendReason::kInternal);
DCHECK(suspended);
if (thread == self || thread->IsSuspended()) {
threads_to_visit.push_back(thread);
} else {
bool resumed = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal);
DCHECK(resumed);
}
}
}
// Visit roots without holding thread_list_lock_ and thread_suspend_count_lock_ to prevent lock
// order violations.
for (Thread* thread : threads_to_visit) {
thread->VisitRoots(visitor, kVisitRootFlagAllRoots);
}
// Restore suspend counts.
{
MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
for (Thread* thread : threads_to_visit) {
bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal);
DCHECK(updated);
}
}
}
void ThreadList::VisitRoots(RootVisitor* visitor, VisitRootFlags flags) const {
MutexLock mu(Thread::Current(), *Locks::thread_list_lock_);
for (const auto& thread : list_) {
thread->VisitRoots(visitor, flags);
}
}
void ThreadList::SweepInterpreterCaches(IsMarkedVisitor* visitor) const {
MutexLock mu(Thread::Current(), *Locks::thread_list_lock_);
for (const auto& thread : list_) {
thread->SweepInterpreterCache(visitor);
}
}
void ThreadList::VisitReflectiveTargets(ReflectiveValueVisitor *visitor) const {
MutexLock mu(Thread::Current(), *Locks::thread_list_lock_);
for (const auto& thread : list_) {
thread->VisitReflectiveTargets(visitor);
}
}
uint32_t ThreadList::AllocThreadId(Thread* self) {
MutexLock mu(self, *Locks::allocated_thread_ids_lock_);
for (size_t i = 0; i < allocated_ids_.size(); ++i) {
if (!allocated_ids_[i]) {
allocated_ids_.set(i);
return i + 1; // Zero is reserved to mean "invalid".
}
}
LOG(FATAL) << "Out of internal thread ids";
UNREACHABLE();
}
void ThreadList::ReleaseThreadId(Thread* self, uint32_t id) {
MutexLock mu(self, *Locks::allocated_thread_ids_lock_);
--id; // Zero is reserved to mean "invalid".
DCHECK(allocated_ids_[id]) << id;
allocated_ids_.reset(id);
}
ScopedSuspendAll::ScopedSuspendAll(const char* cause, bool long_suspend) {
Runtime::Current()->GetThreadList()->SuspendAll(cause, long_suspend);
}
ScopedSuspendAll::~ScopedSuspendAll() {
Runtime::Current()->GetThreadList()->ResumeAll();
}
} // namespace art
|