blob: afb11d33e9ffe6fedd8fb243cfb9e917b72eb96c [file] [log] [blame]
Elliott Hughes8daa0922011-09-11 13:46:25 -07001/*
2 * Copyright (C) 2011 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17#include "thread_list.h"
18
Christopher Ferris6cff48f2014-01-26 21:36:13 -080019#include <backtrace/BacktraceMap.h>
Elliott Hughesabbe07d2012-06-05 17:42:23 -070020#include <dirent.h>
Ian Rogersd9c4fc92013-10-01 19:45:43 -070021#include <ScopedLocalRef.h>
22#include <ScopedUtfChars.h>
Elliott Hughesabbe07d2012-06-05 17:42:23 -070023#include <sys/types.h>
Elliott Hughes038a8062011-09-18 14:12:41 -070024#include <unistd.h>
25
Ian Rogersc7dd2952014-10-21 23:31:19 -070026#include <sstream>
27
Mathieu Chartier70a596d2014-12-17 14:56:47 -080028#include "base/histogram-inl.h"
Hiroshi Yamauchi967a0ad2013-09-10 16:24:21 -070029#include "base/mutex-inl.h"
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -080030#include "base/systrace.h"
Vladimir Marko80afd022015-05-19 18:08:00 +010031#include "base/time_utils.h"
Sameer Abu Asala8439542013-02-14 16:06:42 -080032#include "base/timing_logger.h"
Elliott Hughes475fc232011-10-25 15:00:35 -070033#include "debugger.h"
Hiroshi Yamauchi0b713572015-06-16 18:29:23 -070034#include "gc/collector/concurrent_copying.h"
Ian Rogersd9c4fc92013-10-01 19:45:43 -070035#include "jni_internal.h"
36#include "lock_word.h"
37#include "monitor.h"
38#include "scoped_thread_state_change.h"
Ian Rogers2dd0e2c2013-01-24 12:42:14 -080039#include "thread.h"
Jeff Haoe094b872014-10-14 13:12:01 -070040#include "trace.h"
Ian Rogersd9c4fc92013-10-01 19:45:43 -070041#include "well_known_classes.h"
Elliott Hughes475fc232011-10-25 15:00:35 -070042
Yu Lieac44242015-06-29 10:50:03 +080043#if ART_USE_FUTEXES
44#include "linux/futex.h"
45#include "sys/syscall.h"
46#ifndef SYS_futex
47#define SYS_futex __NR_futex
48#endif
49#endif // ART_USE_FUTEXES
50
Elliott Hughes8daa0922011-09-11 13:46:25 -070051namespace art {
52
Mathieu Chartier251755c2014-07-15 18:10:25 -070053static constexpr uint64_t kLongThreadSuspendThreshold = MsToNs(5);
Mathieu Chartier99143862015-02-03 14:26:46 -080054static constexpr uint64_t kThreadSuspendTimeoutMs = 30 * 1000; // 30s.
55// Use 0 since we want to yield to prevent blocking for an unpredictable amount of time.
56static constexpr useconds_t kThreadSuspendInitialSleepUs = 0;
57static constexpr useconds_t kThreadSuspendMaxYieldUs = 3000;
58static constexpr useconds_t kThreadSuspendMaxSleepUs = 5000;
Mathieu Chartier251755c2014-07-15 18:10:25 -070059
Andreas Gampe8d1594d2016-03-01 14:38:37 -080060// Whether we should try to dump the native stack of unattached threads. See commit ed8b723 for
61// some history.
62static constexpr bool kDumpUnattachedThreadNativeStack = true;
63
Elliott Hughes4dd9b4d2011-12-12 18:29:24 -080064ThreadList::ThreadList()
Mathieu Chartierb56200b2015-10-29 10:41:51 -070065 : suspend_all_count_(0),
66 debug_suspend_all_count_(0),
67 unregistering_count_(0),
68 suspend_all_historam_("suspend all histogram", 16, 64),
69 long_suspend_(false) {
Hiroshi Yamauchie15ea082015-02-09 17:11:42 -080070 CHECK(Monitor::IsValidLockWord(LockWord::FromThinLockId(kMaxThreadId, 1, 0U)));
Elliott Hughes8daa0922011-09-11 13:46:25 -070071}
72
73ThreadList::~ThreadList() {
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -080074 ScopedTrace trace(__PRETTY_FUNCTION__);
Elliott Hughese52e49b2012-04-02 16:05:44 -070075 // Detach the current thread if necessary. If we failed to start, there might not be any threads.
Elliott Hughes6a144332012-04-03 13:07:11 -070076 // We need to detach the current thread here in case there's another thread waiting to join with
77 // us.
Mathieu Chartierfec72f42014-10-09 12:57:58 -070078 bool contains = false;
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -080079 Thread* self = Thread::Current();
Mathieu Chartierfec72f42014-10-09 12:57:58 -070080 {
Mathieu Chartierfec72f42014-10-09 12:57:58 -070081 MutexLock mu(self, *Locks::thread_list_lock_);
82 contains = Contains(self);
83 }
84 if (contains) {
Elliott Hughes8daa0922011-09-11 13:46:25 -070085 Runtime::Current()->DetachCurrentThread();
86 }
Elliott Hughes6a144332012-04-03 13:07:11 -070087 WaitForOtherNonDaemonThreadsToExit();
Mathieu Chartier51168372015-08-12 16:40:32 -070088 // Disable GC and wait for GC to complete in case there are still daemon threads doing
89 // allocations.
90 gc::Heap* const heap = Runtime::Current()->GetHeap();
91 heap->DisableGCForShutdown();
92 // In case a GC is in progress, wait for it to finish.
93 heap->WaitForGcToComplete(gc::kGcCauseBackground, Thread::Current());
Ian Rogers00f7d0e2012-07-19 15:28:27 -070094 // TODO: there's an unaddressed race here where a thread may attach during shutdown, see
95 // Thread::Init.
Mathieu Chartier4d87df62016-01-07 15:14:19 -080096 SuspendAllDaemonThreadsForShutdown();
Elliott Hughes8daa0922011-09-11 13:46:25 -070097}
98
99bool ThreadList::Contains(Thread* thread) {
100 return find(list_.begin(), list_.end(), thread) != list_.end();
101}
102
Elliott Hughesabbe07d2012-06-05 17:42:23 -0700103bool ThreadList::Contains(pid_t tid) {
Mathieu Chartier02e25112013-08-14 16:14:24 -0700104 for (const auto& thread : list_) {
Ian Rogersdd7624d2014-03-14 17:43:00 -0700105 if (thread->GetTid() == tid) {
Elliott Hughesabbe07d2012-06-05 17:42:23 -0700106 return true;
107 }
108 }
109 return false;
110}
111
Brian Carlstrom24a3c2e2011-10-17 18:07:52 -0700112pid_t ThreadList::GetLockOwner() {
Ian Rogersb726dcb2012-09-05 08:57:23 -0700113 return Locks::thread_list_lock_->GetExclusiveOwnerTid();
Elliott Hughesaccd83d2011-10-17 14:25:58 -0700114}
115
Mathieu Chartier590fee92013-09-13 13:46:47 -0700116void ThreadList::DumpNativeStacks(std::ostream& os) {
117 MutexLock mu(Thread::Current(), *Locks::thread_list_lock_);
Christopher Ferris6cff48f2014-01-26 21:36:13 -0800118 std::unique_ptr<BacktraceMap> map(BacktraceMap::Create(getpid()));
Mathieu Chartier590fee92013-09-13 13:46:47 -0700119 for (const auto& thread : list_) {
Ian Rogersdd7624d2014-03-14 17:43:00 -0700120 os << "DUMPING THREAD " << thread->GetTid() << "\n";
Christopher Ferris6cff48f2014-01-26 21:36:13 -0800121 DumpNativeStack(os, thread->GetTid(), map.get(), "\t");
Mathieu Chartier590fee92013-09-13 13:46:47 -0700122 os << "\n";
123 }
124}
125
Elliott Hughesc967f782012-04-16 10:23:15 -0700126void ThreadList::DumpForSigQuit(std::ostream& os) {
Mathieu Chartier70a596d2014-12-17 14:56:47 -0800127 {
128 ScopedObjectAccess soa(Thread::Current());
Mathieu Chartier23f6e692014-12-18 18:24:39 -0800129 // Only print if we have samples.
130 if (suspend_all_historam_.SampleSize() > 0) {
131 Histogram<uint64_t>::CumulativeData data;
132 suspend_all_historam_.CreateHistogram(&data);
133 suspend_all_historam_.PrintConfidenceIntervals(os, 0.99, data); // Dump time to suspend.
134 }
Mathieu Chartier70a596d2014-12-17 14:56:47 -0800135 }
Nicolas Geoffraya73280d2016-02-15 13:05:16 +0000136 Dump(os, Runtime::Current()->GetDumpNativeStackOnSigQuit());
Elliott Hughesabbe07d2012-06-05 17:42:23 -0700137 DumpUnattachedThreads(os);
138}
139
Ian Rogerscfaa4552012-11-26 21:00:08 -0800140static void DumpUnattachedThread(std::ostream& os, pid_t tid) NO_THREAD_SAFETY_ANALYSIS {
Mathieu Chartier2cebb242015-04-21 16:50:40 -0700141 // TODO: No thread safety analysis as DumpState with a null thread won't access fields, should
Ian Rogerscfaa4552012-11-26 21:00:08 -0800142 // refactor DumpState to avoid skipping analysis.
Mathieu Chartier2cebb242015-04-21 16:50:40 -0700143 Thread::DumpState(os, nullptr, tid);
Elliott Hughesabbe07d2012-06-05 17:42:23 -0700144 DumpKernelStack(os, tid, " kernel: ", false);
Andreas Gampe8d1594d2016-03-01 14:38:37 -0800145 if (kDumpUnattachedThreadNativeStack) {
Christopher Ferris6cff48f2014-01-26 21:36:13 -0800146 DumpNativeStack(os, tid, nullptr, " native: ");
Brian Carlstromed8b7232012-06-27 17:54:47 -0700147 }
Elliott Hughesabbe07d2012-06-05 17:42:23 -0700148 os << "\n";
149}
150
151void ThreadList::DumpUnattachedThreads(std::ostream& os) {
152 DIR* d = opendir("/proc/self/task");
153 if (!d) {
154 return;
155 }
156
Ian Rogers50b35e22012-10-04 10:09:15 -0700157 Thread* self = Thread::Current();
Elliott Hughes4696b5b2012-10-30 10:35:10 -0700158 dirent* e;
Mathieu Chartier2cebb242015-04-21 16:50:40 -0700159 while ((e = readdir(d)) != nullptr) {
Elliott Hughesabbe07d2012-06-05 17:42:23 -0700160 char* end;
Elliott Hughes4696b5b2012-10-30 10:35:10 -0700161 pid_t tid = strtol(e->d_name, &end, 10);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700162 if (!*end) {
163 bool contains;
164 {
Ian Rogers50b35e22012-10-04 10:09:15 -0700165 MutexLock mu(self, *Locks::thread_list_lock_);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700166 contains = Contains(tid);
167 }
168 if (!contains) {
169 DumpUnattachedThread(os, tid);
170 }
Elliott Hughesabbe07d2012-06-05 17:42:23 -0700171 }
172 }
173 closedir(d);
Elliott Hughesff738062012-02-03 15:00:42 -0800174}
175
Mathieu Chartier47c19592016-03-07 11:59:01 -0800176// Dump checkpoint timeout in milliseconds. Larger amount on the target, since the device could be
177// overloaded with ANR dumps.
178static constexpr uint32_t kDumpWaitTimeout = kIsTargetBuild ? 100000 : 20000;
Andreas Gampe4a3d19b2015-01-09 17:54:51 -0800179
Ian Rogers7b078e82014-09-10 14:44:24 -0700180// A closure used by Thread::Dump.
181class DumpCheckpoint FINAL : public Closure {
182 public:
Nicolas Geoffraya73280d2016-02-15 13:05:16 +0000183 DumpCheckpoint(std::ostream* os, bool dump_native_stack)
184 : os_(os),
185 barrier_(0),
186 backtrace_map_(dump_native_stack ? BacktraceMap::Create(getpid()) : nullptr),
187 dump_native_stack_(dump_native_stack) {}
Ian Rogers7b078e82014-09-10 14:44:24 -0700188
189 void Run(Thread* thread) OVERRIDE {
190 // Note thread and self may not be equal if thread was already suspended at the point of the
191 // request.
192 Thread* self = Thread::Current();
193 std::ostringstream local_os;
194 {
195 ScopedObjectAccess soa(self);
Nicolas Geoffraya73280d2016-02-15 13:05:16 +0000196 thread->Dump(local_os, dump_native_stack_, backtrace_map_.get());
Ian Rogers7b078e82014-09-10 14:44:24 -0700197 }
198 local_os << "\n";
199 {
200 // Use the logging lock to ensure serialization when writing to the common ostream.
201 MutexLock mu(self, *Locks::logging_lock_);
202 *os_ << local_os.str();
203 }
Mathieu Chartier10d25082015-10-28 18:36:09 -0700204 barrier_.Pass(self);
Elliott Hughes8daa0922011-09-11 13:46:25 -0700205 }
Ian Rogers7b078e82014-09-10 14:44:24 -0700206
207 void WaitForThreadsToRunThroughCheckpoint(size_t threads_running_checkpoint) {
208 Thread* self = Thread::Current();
209 ScopedThreadStateChange tsc(self, kWaitingForCheckPointsToRun);
Andreas Gampe1e4b0ca2015-01-14 09:06:32 -0800210 bool timed_out = barrier_.Increment(self, threads_running_checkpoint, kDumpWaitTimeout);
Ian Rogers2156ff12014-09-13 19:20:54 -0700211 if (timed_out) {
Nicolas Geoffraydb978712014-12-09 13:33:38 +0000212 // Avoid a recursive abort.
213 LOG((kIsDebugBuild && (gAborting == 0)) ? FATAL : ERROR)
214 << "Unexpected time out during dump checkpoint.";
Ian Rogers2156ff12014-09-13 19:20:54 -0700215 }
Ian Rogers7b078e82014-09-10 14:44:24 -0700216 }
217
218 private:
219 // The common stream that will accumulate all the dumps.
220 std::ostream* const os_;
221 // The barrier to be passed through and for the requestor to wait upon.
222 Barrier barrier_;
Christopher Ferris6cff48f2014-01-26 21:36:13 -0800223 // A backtrace map, so that all threads use a shared info and don't reacquire/parse separately.
224 std::unique_ptr<BacktraceMap> backtrace_map_;
Nicolas Geoffraya73280d2016-02-15 13:05:16 +0000225 // Whether we should dump the native stack.
226 const bool dump_native_stack_;
Ian Rogers7b078e82014-09-10 14:44:24 -0700227};
228
Nicolas Geoffraya73280d2016-02-15 13:05:16 +0000229void ThreadList::Dump(std::ostream& os, bool dump_native_stack) {
Ian Rogers7b078e82014-09-10 14:44:24 -0700230 {
231 MutexLock mu(Thread::Current(), *Locks::thread_list_lock_);
232 os << "DALVIK THREADS (" << list_.size() << "):\n";
233 }
Nicolas Geoffraya73280d2016-02-15 13:05:16 +0000234 DumpCheckpoint checkpoint(&os, dump_native_stack);
Ian Rogers7b078e82014-09-10 14:44:24 -0700235 size_t threads_running_checkpoint = RunCheckpoint(&checkpoint);
Lei Lidd9943d2015-02-02 14:24:44 +0800236 if (threads_running_checkpoint != 0) {
237 checkpoint.WaitForThreadsToRunThroughCheckpoint(threads_running_checkpoint);
238 }
Elliott Hughes8daa0922011-09-11 13:46:25 -0700239}
240
Ian Rogers50b35e22012-10-04 10:09:15 -0700241void ThreadList::AssertThreadsAreSuspended(Thread* self, Thread* ignore1, Thread* ignore2) {
242 MutexLock mu(self, *Locks::thread_list_lock_);
243 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Mathieu Chartier02e25112013-08-14 16:14:24 -0700244 for (const auto& thread : list_) {
jeffhao725a9572012-11-13 18:20:12 -0800245 if (thread != ignore1 && thread != ignore2) {
Ian Rogers01ae5802012-09-28 16:14:01 -0700246 CHECK(thread->IsSuspended())
247 << "\nUnsuspended thread: <<" << *thread << "\n"
248 << "self: <<" << *Thread::Current();
249 }
Elliott Hughes8d768a92011-09-14 16:35:25 -0700250 }
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700251}
252
Ian Rogers66aee5c2012-08-15 17:17:47 -0700253#if HAVE_TIMED_RWLOCK
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700254// Attempt to rectify locks so that we dump thread list with required locks before exiting.
Andreas Gampe794ad762015-02-23 08:12:24 -0800255NO_RETURN static void UnsafeLogFatalForThreadSuspendAllTimeout() {
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700256 Runtime* runtime = Runtime::Current();
257 std::ostringstream ss;
258 ss << "Thread suspend timeout\n";
Mathieu Chartier5869a2c2014-10-08 14:26:23 -0700259 Locks::mutator_lock_->Dump(ss);
260 ss << "\n";
Ian Rogers7b078e82014-09-10 14:44:24 -0700261 runtime->GetThreadList()->Dump(ss);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700262 LOG(FATAL) << ss.str();
Ian Rogers719d1a32014-03-06 12:13:39 -0800263 exit(0);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700264}
Ian Rogers66aee5c2012-08-15 17:17:47 -0700265#endif
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700266
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800267// Unlike suspending all threads where we can wait to acquire the mutator_lock_, suspending an
Mathieu Chartier99143862015-02-03 14:26:46 -0800268// individual thread requires polling. delay_us is the requested sleep wait. If delay_us is 0 then
269// we use sched_yield instead of calling usleep.
270static void ThreadSuspendSleep(useconds_t delay_us) {
271 if (delay_us == 0) {
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800272 sched_yield();
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800273 } else {
Mathieu Chartier99143862015-02-03 14:26:46 -0800274 usleep(delay_us);
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800275 }
276}
277
Mathieu Chartier0e4627e2012-10-23 16:13:36 -0700278size_t ThreadList::RunCheckpoint(Closure* checkpoint_function) {
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700279 Thread* self = Thread::Current();
Mathieu Chartier6dda8982014-03-06 11:11:48 -0800280 Locks::mutator_lock_->AssertNotExclusiveHeld(self);
281 Locks::thread_list_lock_->AssertNotHeld(self);
282 Locks::thread_suspend_count_lock_->AssertNotHeld(self);
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700283
284 std::vector<Thread*> suspended_count_modified_threads;
285 size_t count = 0;
286 {
287 // Call a checkpoint function for each thread, threads which are suspend get their checkpoint
288 // manually called.
289 MutexLock mu(self, *Locks::thread_list_lock_);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700290 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Mathieu Chartier10d25082015-10-28 18:36:09 -0700291 count = list_.size();
Mathieu Chartier02e25112013-08-14 16:14:24 -0700292 for (const auto& thread : list_) {
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700293 if (thread != self) {
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700294 while (true) {
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700295 if (thread->RequestCheckpoint(checkpoint_function)) {
Dave Allison0aded082013-11-07 13:15:11 -0800296 // This thread will run its checkpoint some time in the near future.
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700297 break;
298 } else {
299 // We are probably suspended, try to make sure that we stay suspended.
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700300 // The thread switched back to runnable.
301 if (thread->GetState() == kRunnable) {
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700302 // Spurious fail, try again.
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700303 continue;
304 }
Yu Lieac44242015-06-29 10:50:03 +0800305 thread->ModifySuspendCount(self, +1, nullptr, false);
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700306 suspended_count_modified_threads.push_back(thread);
307 break;
308 }
309 }
310 }
311 }
312 }
313
314 // Run the checkpoint on ourself while we wait for threads to suspend.
315 checkpoint_function->Run(self);
316
317 // Run the checkpoint on the suspended threads.
Mathieu Chartier02e25112013-08-14 16:14:24 -0700318 for (const auto& thread : suspended_count_modified_threads) {
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700319 if (!thread->IsSuspended()) {
Mathieu Chartier99143862015-02-03 14:26:46 -0800320 if (ATRACE_ENABLED()) {
321 std::ostringstream oss;
322 thread->ShortDump(oss);
323 ATRACE_BEGIN((std::string("Waiting for suspension of thread ") + oss.str()).c_str());
324 }
325 // Busy wait until the thread is suspended.
326 const uint64_t start_time = NanoTime();
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700327 do {
Mathieu Chartier99143862015-02-03 14:26:46 -0800328 ThreadSuspendSleep(kThreadSuspendInitialSleepUs);
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700329 } while (!thread->IsSuspended());
Mathieu Chartier99143862015-02-03 14:26:46 -0800330 const uint64_t total_delay = NanoTime() - start_time;
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800331 // Shouldn't need to wait for longer than 1000 microseconds.
Mathieu Chartier99143862015-02-03 14:26:46 -0800332 constexpr uint64_t kLongWaitThreshold = MsToNs(1);
333 ATRACE_END();
334 if (UNLIKELY(total_delay > kLongWaitThreshold)) {
335 LOG(WARNING) << "Long wait of " << PrettyDuration(total_delay) << " for "
336 << *thread << " suspension!";
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700337 }
338 }
339 // We know for sure that the thread is suspended at this point.
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700340 checkpoint_function->Run(thread);
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700341 {
342 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Yu Lieac44242015-06-29 10:50:03 +0800343 thread->ModifySuspendCount(self, -1, nullptr, false);
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700344 }
345 }
346
Mathieu Chartier664bebf2012-11-12 16:54:11 -0800347 {
348 // Imitate ResumeAll, threads may be waiting on Thread::resume_cond_ since we raised their
349 // suspend count. Now the suspend_count_ is lowered so we must do the broadcast.
350 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
351 Thread::resume_cond_->Broadcast(self);
352 }
353
Lei Lidd9943d2015-02-02 14:24:44 +0800354 return count;
Mathieu Chartier858f1c52012-10-17 17:45:55 -0700355}
356
Dave Allison39c3bfb2014-01-28 18:33:52 -0800357// Request that a checkpoint function be run on all active (non-suspended)
358// threads. Returns the number of successful requests.
359size_t ThreadList::RunCheckpointOnRunnableThreads(Closure* checkpoint_function) {
360 Thread* self = Thread::Current();
Ian Rogers7b078e82014-09-10 14:44:24 -0700361 Locks::mutator_lock_->AssertNotExclusiveHeld(self);
362 Locks::thread_list_lock_->AssertNotHeld(self);
363 Locks::thread_suspend_count_lock_->AssertNotHeld(self);
364 CHECK_NE(self->GetState(), kRunnable);
Dave Allison39c3bfb2014-01-28 18:33:52 -0800365
366 size_t count = 0;
367 {
368 // Call a checkpoint function for each non-suspended thread.
369 MutexLock mu(self, *Locks::thread_list_lock_);
370 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
371 for (const auto& thread : list_) {
372 if (thread != self) {
373 if (thread->RequestCheckpoint(checkpoint_function)) {
374 // This thread will run its checkpoint some time in the near future.
375 count++;
376 }
377 }
378 }
379 }
380
381 // Return the number of threads that will run the checkpoint function.
382 return count;
383}
384
Hiroshi Yamauchi2cd334a2015-01-09 14:03:35 -0800385// A checkpoint/suspend-all hybrid to switch thread roots from
386// from-space to to-space refs. Used to synchronize threads at a point
387// to mark the initiation of marking while maintaining the to-space
388// invariant.
Mathieu Chartierb56200b2015-10-29 10:41:51 -0700389size_t ThreadList::FlipThreadRoots(Closure* thread_flip_visitor,
390 Closure* flip_callback,
Hiroshi Yamauchi2cd334a2015-01-09 14:03:35 -0800391 gc::collector::GarbageCollector* collector) {
392 TimingLogger::ScopedTiming split("ThreadListFlip", collector->GetTimings());
393 const uint64_t start_time = NanoTime();
394 Thread* self = Thread::Current();
395 Locks::mutator_lock_->AssertNotHeld(self);
396 Locks::thread_list_lock_->AssertNotHeld(self);
397 Locks::thread_suspend_count_lock_->AssertNotHeld(self);
398 CHECK_NE(self->GetState(), kRunnable);
399
Mathieu Chartierb19ccb12015-07-15 10:24:16 -0700400 SuspendAllInternal(self, self, nullptr);
Hiroshi Yamauchi2cd334a2015-01-09 14:03:35 -0800401
402 // Run the flip callback for the collector.
403 Locks::mutator_lock_->ExclusiveLock(self);
404 flip_callback->Run(self);
405 Locks::mutator_lock_->ExclusiveUnlock(self);
406 collector->RegisterPause(NanoTime() - start_time);
407
408 // Resume runnable threads.
Mathieu Chartierb19ccb12015-07-15 10:24:16 -0700409 std::vector<Thread*> runnable_threads;
410 std::vector<Thread*> other_threads;
Hiroshi Yamauchi2cd334a2015-01-09 14:03:35 -0800411 {
412 MutexLock mu(self, *Locks::thread_list_lock_);
413 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
414 --suspend_all_count_;
415 for (const auto& thread : list_) {
416 if (thread == self) {
417 continue;
418 }
419 // Set the flip function for both runnable and suspended threads
420 // because Thread::DumpState/DumpJavaStack() (invoked by a
421 // checkpoint) may cause the flip function to be run for a
422 // runnable/suspended thread before a runnable threads runs it
423 // for itself or we run it for a suspended thread below.
424 thread->SetFlipFunction(thread_flip_visitor);
425 if (thread->IsSuspendedAtSuspendCheck()) {
426 // The thread will resume right after the broadcast.
Yu Lieac44242015-06-29 10:50:03 +0800427 thread->ModifySuspendCount(self, -1, nullptr, false);
Hiroshi Yamauchi2cd334a2015-01-09 14:03:35 -0800428 runnable_threads.push_back(thread);
429 } else {
430 other_threads.push_back(thread);
431 }
432 }
433 Thread::resume_cond_->Broadcast(self);
434 }
435
436 // Run the closure on the other threads and let them resume.
437 {
438 ReaderMutexLock mu(self, *Locks::mutator_lock_);
439 for (const auto& thread : other_threads) {
440 Closure* flip_func = thread->GetFlipFunction();
441 if (flip_func != nullptr) {
442 flip_func->Run(thread);
443 }
444 }
445 // Run it for self.
446 thread_flip_visitor->Run(self);
447 }
448
449 // Resume other threads.
450 {
451 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
452 for (const auto& thread : other_threads) {
Yu Lieac44242015-06-29 10:50:03 +0800453 thread->ModifySuspendCount(self, -1, nullptr, false);
Hiroshi Yamauchi2cd334a2015-01-09 14:03:35 -0800454 }
455 Thread::resume_cond_->Broadcast(self);
456 }
457
458 return runnable_threads.size() + other_threads.size() + 1; // +1 for self.
459}
460
Mathieu Chartierbf44d422015-06-02 11:42:18 -0700461void ThreadList::SuspendAll(const char* cause, bool long_suspend) {
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700462 Thread* self = Thread::Current();
463
Jeff Haoc5d824a2014-07-28 18:35:38 -0700464 if (self != nullptr) {
Mathieu Chartierbf9fc582015-03-13 17:21:25 -0700465 VLOG(threads) << *self << " SuspendAll for " << cause << " starting...";
Jeff Haoc5d824a2014-07-28 18:35:38 -0700466 } else {
Mathieu Chartierbf9fc582015-03-13 17:21:25 -0700467 VLOG(threads) << "Thread[null] SuspendAll for " << cause << " starting...";
Jeff Haoc5d824a2014-07-28 18:35:38 -0700468 }
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -0800469 {
470 ScopedTrace trace("Suspending mutator threads");
471 const uint64_t start_time = NanoTime();
Mathieu Chartier6f365cc2014-04-23 12:42:27 -0700472
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -0800473 SuspendAllInternal(self, self);
474 // All threads are known to have suspended (but a thread may still own the mutator lock)
475 // Make sure this thread grabs exclusive access to the mutator lock and its protected data.
Ian Rogers66aee5c2012-08-15 17:17:47 -0700476#if HAVE_TIMED_RWLOCK
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -0800477 while (true) {
478 if (Locks::mutator_lock_->ExclusiveLockWithTimeout(self, kThreadSuspendTimeoutMs, 0)) {
479 break;
480 } else if (!long_suspend_) {
481 // Reading long_suspend without the mutator lock is slightly racy, in some rare cases, this
482 // could result in a thread suspend timeout.
483 // Timeout if we wait more than kThreadSuspendTimeoutMs seconds.
484 UnsafeLogFatalForThreadSuspendAllTimeout();
485 }
Mathieu Chartierbf44d422015-06-02 11:42:18 -0700486 }
Ian Rogers66aee5c2012-08-15 17:17:47 -0700487#else
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -0800488 Locks::mutator_lock_->ExclusiveLock(self);
Ian Rogers66aee5c2012-08-15 17:17:47 -0700489#endif
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700490
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -0800491 long_suspend_ = long_suspend;
Mathieu Chartierbf44d422015-06-02 11:42:18 -0700492
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -0800493 const uint64_t end_time = NanoTime();
494 const uint64_t suspend_time = end_time - start_time;
495 suspend_all_historam_.AdjustAndAddValue(suspend_time);
496 if (suspend_time > kLongThreadSuspendThreshold) {
497 LOG(WARNING) << "Suspending all threads took: " << PrettyDuration(suspend_time);
498 }
499
500 if (kDebugLocking) {
501 // Debug check that all threads are suspended.
502 AssertThreadsAreSuspended(self, self);
503 }
Mathieu Chartier251755c2014-07-15 18:10:25 -0700504 }
Mathieu Chartierbf9fc582015-03-13 17:21:25 -0700505 ATRACE_BEGIN((std::string("Mutator threads suspended for ") + cause).c_str());
Mathieu Chartier6f365cc2014-04-23 12:42:27 -0700506
Jeff Haoc5d824a2014-07-28 18:35:38 -0700507 if (self != nullptr) {
508 VLOG(threads) << *self << " SuspendAll complete";
509 } else {
510 VLOG(threads) << "Thread[null] SuspendAll complete";
511 }
Elliott Hughes8d768a92011-09-14 16:35:25 -0700512}
513
Yu Lieac44242015-06-29 10:50:03 +0800514// Ensures all threads running Java suspend and that those not running Java don't start.
515// Debugger thread might be set to kRunnable for a short period of time after the
516// SuspendAllInternal. This is safe because it will be set back to suspended state before
517// the SuspendAll returns.
Mathieu Chartierb56200b2015-10-29 10:41:51 -0700518void ThreadList::SuspendAllInternal(Thread* self,
519 Thread* ignore1,
520 Thread* ignore2,
Yu Lieac44242015-06-29 10:50:03 +0800521 bool debug_suspend) {
522 Locks::mutator_lock_->AssertNotExclusiveHeld(self);
523 Locks::thread_list_lock_->AssertNotHeld(self);
524 Locks::thread_suspend_count_lock_->AssertNotHeld(self);
525 if (kDebugLocking && self != nullptr) {
526 CHECK_NE(self->GetState(), kRunnable);
527 }
528
529 // First request that all threads suspend, then wait for them to suspend before
530 // returning. This suspension scheme also relies on other behaviour:
531 // 1. Threads cannot be deleted while they are suspended or have a suspend-
532 // request flag set - (see Unregister() below).
533 // 2. When threads are created, they are created in a suspended state (actually
534 // kNative) and will never begin executing Java code without first checking
535 // the suspend-request flag.
536
537 // The atomic counter for number of threads that need to pass the barrier.
538 AtomicInteger pending_threads;
539 uint32_t num_ignored = 0;
540 if (ignore1 != nullptr) {
541 ++num_ignored;
542 }
543 if (ignore2 != nullptr && ignore1 != ignore2) {
544 ++num_ignored;
545 }
546 {
547 MutexLock mu(self, *Locks::thread_list_lock_);
548 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
549 // Update global suspend all state for attaching threads.
550 ++suspend_all_count_;
551 if (debug_suspend)
552 ++debug_suspend_all_count_;
553 pending_threads.StoreRelaxed(list_.size() - num_ignored);
554 // Increment everybody's suspend count (except those that should be ignored).
555 for (const auto& thread : list_) {
556 if (thread == ignore1 || thread == ignore2) {
557 continue;
558 }
559 VLOG(threads) << "requesting thread suspend: " << *thread;
560 while (true) {
561 if (LIKELY(thread->ModifySuspendCount(self, +1, &pending_threads, debug_suspend))) {
562 break;
563 } else {
564 // Failure means the list of active_suspend_barriers is full, we should release the
565 // thread_suspend_count_lock_ (to avoid deadlock) and wait till the target thread has
566 // executed Thread::PassActiveSuspendBarriers(). Note that we could not simply wait for
567 // the thread to change to a suspended state, because it might need to run checkpoint
568 // function before the state change, which also needs thread_suspend_count_lock_.
569
570 // This is very unlikely to happen since more than kMaxSuspendBarriers threads need to
571 // execute SuspendAllInternal() simultaneously, and target thread stays in kRunnable
572 // in the mean time.
573 Locks::thread_suspend_count_lock_->ExclusiveUnlock(self);
574 NanoSleep(100000);
575 Locks::thread_suspend_count_lock_->ExclusiveLock(self);
576 }
577 }
578
579 // Must install the pending_threads counter first, then check thread->IsSuspend() and clear
580 // the counter. Otherwise there's a race with Thread::TransitionFromRunnableToSuspended()
581 // that can lead a thread to miss a call to PassActiveSuspendBarriers().
582 if (thread->IsSuspended()) {
583 // Only clear the counter for the current thread.
584 thread->ClearSuspendBarrier(&pending_threads);
585 pending_threads.FetchAndSubSequentiallyConsistent(1);
586 }
587 }
588 }
589
590 // Wait for the barrier to be passed by all runnable threads. This wait
591 // is done with a timeout so that we can detect problems.
Mathieu Chartier19af1172015-07-14 10:05:45 -0700592#if ART_USE_FUTEXES
Yu Lieac44242015-06-29 10:50:03 +0800593 timespec wait_timeout;
594 InitTimeSpec(true, CLOCK_MONOTONIC, 10000, 0, &wait_timeout);
Mathieu Chartier19af1172015-07-14 10:05:45 -0700595#endif
Yu Lieac44242015-06-29 10:50:03 +0800596 while (true) {
597 int32_t cur_val = pending_threads.LoadRelaxed();
598 if (LIKELY(cur_val > 0)) {
599#if ART_USE_FUTEXES
600 if (futex(pending_threads.Address(), FUTEX_WAIT, cur_val, &wait_timeout, nullptr, 0) != 0) {
601 // EAGAIN and EINTR both indicate a spurious failure, try again from the beginning.
602 if ((errno != EAGAIN) && (errno != EINTR)) {
603 if (errno == ETIMEDOUT) {
604 LOG(kIsDebugBuild ? FATAL : ERROR) << "Unexpected time out during suspend all.";
605 } else {
606 PLOG(FATAL) << "futex wait failed for SuspendAllInternal()";
607 }
608 }
609 } else {
610 cur_val = pending_threads.LoadRelaxed();
611 CHECK_EQ(cur_val, 0);
612 break;
613 }
614#else
615 // Spin wait. This is likely to be slow, but on most architecture ART_USE_FUTEXES is set.
616#endif
617 } else {
618 CHECK_EQ(cur_val, 0);
619 break;
620 }
621 }
622}
623
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700624void ThreadList::ResumeAll() {
625 Thread* self = Thread::Current();
626
Jeff Haoc5d824a2014-07-28 18:35:38 -0700627 if (self != nullptr) {
628 VLOG(threads) << *self << " ResumeAll starting";
629 } else {
630 VLOG(threads) << "Thread[null] ResumeAll starting";
631 }
Ian Rogers01ae5802012-09-28 16:14:01 -0700632
Mathieu Chartier6f365cc2014-04-23 12:42:27 -0700633 ATRACE_END();
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -0800634
635 ScopedTrace trace("Resuming mutator threads");
Mathieu Chartier6f365cc2014-04-23 12:42:27 -0700636
Mathieu Chartier6dda8982014-03-06 11:11:48 -0800637 if (kDebugLocking) {
638 // Debug check that all threads are suspended.
639 AssertThreadsAreSuspended(self, self);
640 }
Ian Rogers01ae5802012-09-28 16:14:01 -0700641
Mathieu Chartierbf44d422015-06-02 11:42:18 -0700642 long_suspend_ = false;
643
Ian Rogers81d425b2012-09-27 16:03:43 -0700644 Locks::mutator_lock_->ExclusiveUnlock(self);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700645 {
Ian Rogers81d425b2012-09-27 16:03:43 -0700646 MutexLock mu(self, *Locks::thread_list_lock_);
647 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700648 // Update global suspend all state for attaching threads.
649 --suspend_all_count_;
650 // Decrement the suspend counts for all threads.
Mathieu Chartier02e25112013-08-14 16:14:24 -0700651 for (const auto& thread : list_) {
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700652 if (thread == self) {
653 continue;
654 }
Yu Lieac44242015-06-29 10:50:03 +0800655 thread->ModifySuspendCount(self, -1, nullptr, false);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700656 }
657
658 // Broadcast a notification to all suspended threads, some or all of
659 // which may choose to wake up. No need to wait for them.
Jeff Haoc5d824a2014-07-28 18:35:38 -0700660 if (self != nullptr) {
661 VLOG(threads) << *self << " ResumeAll waking others";
662 } else {
663 VLOG(threads) << "Thread[null] ResumeAll waking others";
664 }
Ian Rogersc604d732012-10-14 16:09:54 -0700665 Thread::resume_cond_->Broadcast(self);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700666 }
Jeff Haoc5d824a2014-07-28 18:35:38 -0700667
668 if (self != nullptr) {
669 VLOG(threads) << *self << " ResumeAll complete";
670 } else {
671 VLOG(threads) << "Thread[null] ResumeAll complete";
672 }
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700673}
674
675void ThreadList::Resume(Thread* thread, bool for_debugger) {
Mathieu Chartierf0dc8b52014-12-17 10:13:30 -0800676 // This assumes there was an ATRACE_BEGIN when we suspended the thread.
677 ATRACE_END();
678
Ian Rogers81d425b2012-09-27 16:03:43 -0700679 Thread* self = Thread::Current();
680 DCHECK_NE(thread, self);
Brian Carlstromba32de42014-08-27 23:43:46 -0700681 VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") starting..."
682 << (for_debugger ? " (debugger)" : "");
Elliott Hughes01158d72011-09-19 19:47:10 -0700683
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700684 {
685 // To check Contains.
Ian Rogers81d425b2012-09-27 16:03:43 -0700686 MutexLock mu(self, *Locks::thread_list_lock_);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700687 // To check IsSuspended.
Ian Rogers81d425b2012-09-27 16:03:43 -0700688 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
689 DCHECK(thread->IsSuspended());
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700690 if (!Contains(thread)) {
Brian Carlstromba32de42014-08-27 23:43:46 -0700691 // We only expect threads within the thread-list to have been suspended otherwise we can't
692 // stop such threads from delete-ing themselves.
693 LOG(ERROR) << "Resume(" << reinterpret_cast<void*>(thread)
694 << ") thread not within thread list";
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700695 return;
696 }
Yu Lieac44242015-06-29 10:50:03 +0800697 thread->ModifySuspendCount(self, -1, nullptr, for_debugger);
Elliott Hughes01158d72011-09-19 19:47:10 -0700698 }
699
700 {
Brian Carlstromba32de42014-08-27 23:43:46 -0700701 VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") waking others";
Ian Rogers81d425b2012-09-27 16:03:43 -0700702 MutexLock mu(self, *Locks::thread_suspend_count_lock_);
Ian Rogersc604d732012-10-14 16:09:54 -0700703 Thread::resume_cond_->Broadcast(self);
Elliott Hughes01158d72011-09-19 19:47:10 -0700704 }
705
Brian Carlstromba32de42014-08-27 23:43:46 -0700706 VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") complete";
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700707}
Elliott Hughes01158d72011-09-19 19:47:10 -0700708
Mathieu Chartierb56200b2015-10-29 10:41:51 -0700709static void ThreadSuspendByPeerWarning(Thread* self,
710 LogSeverity severity,
711 const char* message,
Ian Rogersc7dd2952014-10-21 23:31:19 -0700712 jobject peer) {
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700713 JNIEnvExt* env = self->GetJniEnv();
714 ScopedLocalRef<jstring>
Mathieu Chartierb56200b2015-10-29 10:41:51 -0700715 scoped_name_string(env, static_cast<jstring>(env->GetObjectField(
716 peer, WellKnownClasses::java_lang_Thread_name)));
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700717 ScopedUtfChars scoped_name_chars(env, scoped_name_string.get());
Mathieu Chartier2cebb242015-04-21 16:50:40 -0700718 if (scoped_name_chars.c_str() == nullptr) {
Ian Rogersc7dd2952014-10-21 23:31:19 -0700719 LOG(severity) << message << ": " << peer;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700720 env->ExceptionClear();
721 } else {
Ian Rogersc7dd2952014-10-21 23:31:19 -0700722 LOG(severity) << message << ": " << peer << ":" << scoped_name_chars.c_str();
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700723 }
724}
725
Mathieu Chartierb56200b2015-10-29 10:41:51 -0700726Thread* ThreadList::SuspendThreadByPeer(jobject peer,
727 bool request_suspension,
728 bool debug_suspension,
729 bool* timed_out) {
Mathieu Chartier3a958aa2015-02-04 12:52:34 -0800730 const uint64_t start_time = NanoTime();
Mathieu Chartier99143862015-02-03 14:26:46 -0800731 useconds_t sleep_us = kThreadSuspendInitialSleepUs;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700732 *timed_out = false;
Mathieu Chartier99143862015-02-03 14:26:46 -0800733 Thread* const self = Thread::Current();
Mathieu Chartier82a800d2014-12-15 15:59:49 -0800734 Thread* suspended_thread = nullptr;
Brian Carlstromba32de42014-08-27 23:43:46 -0700735 VLOG(threads) << "SuspendThreadByPeer starting";
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700736 while (true) {
737 Thread* thread;
738 {
Ian Rogersf3d874c2014-07-17 18:52:42 -0700739 // Note: this will transition to runnable and potentially suspend. We ensure only one thread
740 // is requesting another suspend, to avoid deadlock, by requiring this function be called
741 // holding Locks::thread_list_suspend_thread_lock_. Its important this thread suspend rather
742 // than request thread suspension, to avoid potential cycles in threads requesting each other
743 // suspend.
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700744 ScopedObjectAccess soa(self);
Andreas Gampe277ccbd2014-11-03 21:36:10 -0800745 MutexLock thread_list_mu(self, *Locks::thread_list_lock_);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700746 thread = Thread::FromManagedThread(soa, peer);
Brian Carlstromba32de42014-08-27 23:43:46 -0700747 if (thread == nullptr) {
Mathieu Chartier82a800d2014-12-15 15:59:49 -0800748 if (suspended_thread != nullptr) {
749 MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_);
750 // If we incremented the suspend count but the thread reset its peer, we need to
751 // re-decrement it since it is shutting down and may deadlock the runtime in
752 // ThreadList::WaitForOtherNonDaemonThreadsToExit.
Yu Lieac44242015-06-29 10:50:03 +0800753 suspended_thread->ModifySuspendCount(soa.Self(), -1, nullptr, debug_suspension);
Mathieu Chartier82a800d2014-12-15 15:59:49 -0800754 }
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700755 ThreadSuspendByPeerWarning(self, WARNING, "No such thread for suspend", peer);
Brian Carlstromba32de42014-08-27 23:43:46 -0700756 return nullptr;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700757 }
Brian Carlstromba32de42014-08-27 23:43:46 -0700758 if (!Contains(thread)) {
Mathieu Chartier82a800d2014-12-15 15:59:49 -0800759 CHECK(suspended_thread == nullptr);
Brian Carlstromba32de42014-08-27 23:43:46 -0700760 VLOG(threads) << "SuspendThreadByPeer failed for unattached thread: "
761 << reinterpret_cast<void*>(thread);
762 return nullptr;
763 }
764 VLOG(threads) << "SuspendThreadByPeer found thread: " << *thread;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700765 {
Andreas Gampe277ccbd2014-11-03 21:36:10 -0800766 MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700767 if (request_suspension) {
Ian Rogers4ad5cd32014-11-11 23:08:07 -0800768 if (self->GetSuspendCount() > 0) {
769 // We hold the suspend count lock but another thread is trying to suspend us. Its not
770 // safe to try to suspend another thread in case we get a cycle. Start the loop again
771 // which will allow this thread to be suspended.
772 continue;
773 }
Mathieu Chartier82a800d2014-12-15 15:59:49 -0800774 CHECK(suspended_thread == nullptr);
775 suspended_thread = thread;
Yu Lieac44242015-06-29 10:50:03 +0800776 suspended_thread->ModifySuspendCount(self, +1, nullptr, debug_suspension);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700777 request_suspension = false;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700778 } else {
779 // If the caller isn't requesting suspension, a suspension should have already occurred.
780 CHECK_GT(thread->GetSuspendCount(), 0);
781 }
782 // IsSuspended on the current thread will fail as the current thread is changed into
783 // Runnable above. As the suspend count is now raised if this is the current thread
784 // it will self suspend on transition to Runnable, making it hard to work with. It's simpler
785 // to just explicitly handle the current thread in the callers to this code.
786 CHECK_NE(thread, self) << "Attempt to suspend the current thread for the debugger";
787 // If thread is suspended (perhaps it was already not Runnable but didn't have a suspend
788 // count, or else we've waited and it has self suspended) or is the current thread, we're
789 // done.
790 if (thread->IsSuspended()) {
Brian Carlstromba32de42014-08-27 23:43:46 -0700791 VLOG(threads) << "SuspendThreadByPeer thread suspended: " << *thread;
Mathieu Chartierf0dc8b52014-12-17 10:13:30 -0800792 if (ATRACE_ENABLED()) {
793 std::string name;
794 thread->GetThreadName(name);
795 ATRACE_BEGIN(StringPrintf("SuspendThreadByPeer suspended %s for peer=%p", name.c_str(),
796 peer).c_str());
797 }
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700798 return thread;
799 }
Mathieu Chartier99143862015-02-03 14:26:46 -0800800 const uint64_t total_delay = NanoTime() - start_time;
801 if (total_delay >= MsToNs(kThreadSuspendTimeoutMs)) {
Mathieu Chartier590fee92013-09-13 13:46:47 -0700802 ThreadSuspendByPeerWarning(self, FATAL, "Thread suspension timed out", peer);
Mathieu Chartier82a800d2014-12-15 15:59:49 -0800803 if (suspended_thread != nullptr) {
804 CHECK_EQ(suspended_thread, thread);
Yu Lieac44242015-06-29 10:50:03 +0800805 suspended_thread->ModifySuspendCount(soa.Self(), -1, nullptr, debug_suspension);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700806 }
807 *timed_out = true;
Brian Carlstromba32de42014-08-27 23:43:46 -0700808 return nullptr;
Mathieu Chartier99143862015-02-03 14:26:46 -0800809 } else if (sleep_us == 0 &&
810 total_delay > static_cast<uint64_t>(kThreadSuspendMaxYieldUs) * 1000) {
811 // We have spun for kThreadSuspendMaxYieldUs time, switch to sleeps to prevent
812 // excessive CPU usage.
813 sleep_us = kThreadSuspendMaxYieldUs / 2;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700814 }
815 }
816 // Release locks and come out of runnable state.
817 }
Mathieu Chartier99143862015-02-03 14:26:46 -0800818 VLOG(threads) << "SuspendThreadByPeer waiting to allow thread chance to suspend";
819 ThreadSuspendSleep(sleep_us);
820 // This may stay at 0 if sleep_us == 0, but this is WAI since we want to avoid using usleep at
821 // all if possible. This shouldn't be an issue since time to suspend should always be small.
822 sleep_us = std::min(sleep_us * 2, kThreadSuspendMaxSleepUs);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700823 }
824}
825
Mathieu Chartierb56200b2015-10-29 10:41:51 -0700826static void ThreadSuspendByThreadIdWarning(LogSeverity severity,
827 const char* message,
Ian Rogersc7dd2952014-10-21 23:31:19 -0700828 uint32_t thread_id) {
829 LOG(severity) << StringPrintf("%s: %d", message, thread_id);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700830}
831
Mathieu Chartierb56200b2015-10-29 10:41:51 -0700832Thread* ThreadList::SuspendThreadByThreadId(uint32_t thread_id,
833 bool debug_suspension,
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700834 bool* timed_out) {
Mathieu Chartier3a958aa2015-02-04 12:52:34 -0800835 const uint64_t start_time = NanoTime();
Mathieu Chartier99143862015-02-03 14:26:46 -0800836 useconds_t sleep_us = kThreadSuspendInitialSleepUs;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700837 *timed_out = false;
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800838 Thread* suspended_thread = nullptr;
Mathieu Chartier99143862015-02-03 14:26:46 -0800839 Thread* const self = Thread::Current();
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700840 CHECK_NE(thread_id, kInvalidThreadId);
Brian Carlstromba32de42014-08-27 23:43:46 -0700841 VLOG(threads) << "SuspendThreadByThreadId starting";
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700842 while (true) {
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700843 {
Ian Rogersf3d874c2014-07-17 18:52:42 -0700844 // Note: this will transition to runnable and potentially suspend. We ensure only one thread
845 // is requesting another suspend, to avoid deadlock, by requiring this function be called
846 // holding Locks::thread_list_suspend_thread_lock_. Its important this thread suspend rather
847 // than request thread suspension, to avoid potential cycles in threads requesting each other
848 // suspend.
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700849 ScopedObjectAccess soa(self);
Andreas Gampe277ccbd2014-11-03 21:36:10 -0800850 MutexLock thread_list_mu(self, *Locks::thread_list_lock_);
Ian Rogersf3d874c2014-07-17 18:52:42 -0700851 Thread* thread = nullptr;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700852 for (const auto& it : list_) {
853 if (it->GetThreadId() == thread_id) {
854 thread = it;
855 break;
856 }
857 }
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800858 if (thread == nullptr) {
859 CHECK(suspended_thread == nullptr) << "Suspended thread " << suspended_thread
860 << " no longer in thread list";
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700861 // There's a race in inflating a lock and the owner giving up ownership and then dying.
862 ThreadSuspendByThreadIdWarning(WARNING, "No such thread id for suspend", thread_id);
Brian Carlstromba32de42014-08-27 23:43:46 -0700863 return nullptr;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700864 }
Brian Carlstromba32de42014-08-27 23:43:46 -0700865 VLOG(threads) << "SuspendThreadByThreadId found thread: " << *thread;
866 DCHECK(Contains(thread));
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700867 {
Andreas Gampe277ccbd2014-11-03 21:36:10 -0800868 MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_);
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800869 if (suspended_thread == nullptr) {
Ian Rogers4ad5cd32014-11-11 23:08:07 -0800870 if (self->GetSuspendCount() > 0) {
871 // We hold the suspend count lock but another thread is trying to suspend us. Its not
872 // safe to try to suspend another thread in case we get a cycle. Start the loop again
873 // which will allow this thread to be suspended.
874 continue;
875 }
Yu Lieac44242015-06-29 10:50:03 +0800876 thread->ModifySuspendCount(self, +1, nullptr, debug_suspension);
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800877 suspended_thread = thread;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700878 } else {
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800879 CHECK_EQ(suspended_thread, thread);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700880 // If the caller isn't requesting suspension, a suspension should have already occurred.
881 CHECK_GT(thread->GetSuspendCount(), 0);
882 }
883 // IsSuspended on the current thread will fail as the current thread is changed into
884 // Runnable above. As the suspend count is now raised if this is the current thread
885 // it will self suspend on transition to Runnable, making it hard to work with. It's simpler
886 // to just explicitly handle the current thread in the callers to this code.
887 CHECK_NE(thread, self) << "Attempt to suspend the current thread for the debugger";
888 // If thread is suspended (perhaps it was already not Runnable but didn't have a suspend
889 // count, or else we've waited and it has self suspended) or is the current thread, we're
890 // done.
891 if (thread->IsSuspended()) {
Mathieu Chartierf0dc8b52014-12-17 10:13:30 -0800892 if (ATRACE_ENABLED()) {
893 std::string name;
894 thread->GetThreadName(name);
895 ATRACE_BEGIN(StringPrintf("SuspendThreadByThreadId suspended %s id=%d",
896 name.c_str(), thread_id).c_str());
897 }
Brian Carlstromba32de42014-08-27 23:43:46 -0700898 VLOG(threads) << "SuspendThreadByThreadId thread suspended: " << *thread;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700899 return thread;
900 }
Mathieu Chartier99143862015-02-03 14:26:46 -0800901 const uint64_t total_delay = NanoTime() - start_time;
902 if (total_delay >= MsToNs(kThreadSuspendTimeoutMs)) {
Mathieu Chartier590fee92013-09-13 13:46:47 -0700903 ThreadSuspendByThreadIdWarning(WARNING, "Thread suspension timed out", thread_id);
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -0800904 if (suspended_thread != nullptr) {
Yu Lieac44242015-06-29 10:50:03 +0800905 thread->ModifySuspendCount(soa.Self(), -1, nullptr, debug_suspension);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700906 }
907 *timed_out = true;
Brian Carlstromba32de42014-08-27 23:43:46 -0700908 return nullptr;
Mathieu Chartier99143862015-02-03 14:26:46 -0800909 } else if (sleep_us == 0 &&
910 total_delay > static_cast<uint64_t>(kThreadSuspendMaxYieldUs) * 1000) {
911 // We have spun for kThreadSuspendMaxYieldUs time, switch to sleeps to prevent
912 // excessive CPU usage.
913 sleep_us = kThreadSuspendMaxYieldUs / 2;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700914 }
915 }
916 // Release locks and come out of runnable state.
917 }
Mathieu Chartier99143862015-02-03 14:26:46 -0800918 VLOG(threads) << "SuspendThreadByThreadId waiting to allow thread chance to suspend";
919 ThreadSuspendSleep(sleep_us);
920 sleep_us = std::min(sleep_us * 2, kThreadSuspendMaxSleepUs);
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700921 }
922}
923
924Thread* ThreadList::FindThreadByThreadId(uint32_t thin_lock_id) {
925 Thread* self = Thread::Current();
926 MutexLock mu(self, *Locks::thread_list_lock_);
927 for (const auto& thread : list_) {
928 if (thread->GetThreadId() == thin_lock_id) {
929 CHECK(thread == self || thread->IsSuspended());
930 return thread;
931 }
932 }
Mathieu Chartier2cebb242015-04-21 16:50:40 -0700933 return nullptr;
Ian Rogersd9c4fc92013-10-01 19:45:43 -0700934}
935
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700936void ThreadList::SuspendAllForDebugger() {
937 Thread* self = Thread::Current();
938 Thread* debug_thread = Dbg::GetDebugThread();
939
940 VLOG(threads) << *self << " SuspendAllForDebugger starting...";
941
Yu Lieac44242015-06-29 10:50:03 +0800942 SuspendAllInternal(self, self, debug_thread, true);
Ian Rogers66aee5c2012-08-15 17:17:47 -0700943 // Block on the mutator lock until all Runnable threads release their share of access then
944 // immediately unlock again.
945#if HAVE_TIMED_RWLOCK
946 // Timeout if we wait more than 30 seconds.
Ian Rogersc604d732012-10-14 16:09:54 -0700947 if (!Locks::mutator_lock_->ExclusiveLockWithTimeout(self, 30 * 1000, 0)) {
Sebastien Hertzbae182c2013-12-17 10:42:03 +0100948 UnsafeLogFatalForThreadSuspendAllTimeout();
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700949 } else {
Ian Rogers81d425b2012-09-27 16:03:43 -0700950 Locks::mutator_lock_->ExclusiveUnlock(self);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700951 }
Ian Rogers66aee5c2012-08-15 17:17:47 -0700952#else
Ian Rogers81d425b2012-09-27 16:03:43 -0700953 Locks::mutator_lock_->ExclusiveLock(self);
954 Locks::mutator_lock_->ExclusiveUnlock(self);
Ian Rogers66aee5c2012-08-15 17:17:47 -0700955#endif
Mathieu Chartier9450c6c2015-11-07 11:55:23 -0800956 // Disabled for the following race condition:
957 // Thread 1 calls SuspendAllForDebugger, gets preempted after pulsing the mutator lock.
958 // Thread 2 calls SuspendAll and SetStateUnsafe (perhaps from Dbg::Disconnected).
959 // Thread 1 fails assertion that all threads are suspended due to thread 2 being in a runnable
960 // state (from SetStateUnsafe).
961 // AssertThreadsAreSuspended(self, self, debug_thread);
Ian Rogers00f7d0e2012-07-19 15:28:27 -0700962
Sebastien Hertzed2be172014-08-19 15:33:43 +0200963 VLOG(threads) << *self << " SuspendAllForDebugger complete";
Elliott Hughes01158d72011-09-19 19:47:10 -0700964}
965
Elliott Hughes475fc232011-10-25 15:00:35 -0700966void ThreadList::SuspendSelfForDebugger() {
Sebastien Hertz1558b572015-02-25 15:05:59 +0100967 Thread* const self = Thread::Current();
968 self->SetReadyForDebugInvoke(true);
Elliott Hughes01158d72011-09-19 19:47:10 -0700969
Elliott Hughes475fc232011-10-25 15:00:35 -0700970 // The debugger thread must not suspend itself due to debugger activity!
971 Thread* debug_thread = Dbg::GetDebugThread();
Elliott Hughes475fc232011-10-25 15:00:35 -0700972 CHECK(self != debug_thread);
jeffhaoa77f0f62012-12-05 17:19:31 -0800973 CHECK_NE(self->GetState(), kRunnable);
974 Locks::mutator_lock_->AssertNotHeld(self);
Elliott Hughes475fc232011-10-25 15:00:35 -0700975
Sebastien Hertzcbc50642015-06-01 17:33:12 +0200976 // The debugger may have detached while we were executing an invoke request. In that case, we
977 // must not suspend ourself.
978 DebugInvokeReq* pReq = self->GetInvokeReq();
979 const bool skip_thread_suspension = (pReq != nullptr && !Dbg::IsDebuggerActive());
980 if (!skip_thread_suspension) {
jeffhaoa77f0f62012-12-05 17:19:31 -0800981 // Collisions with other suspends aren't really interesting. We want
982 // to ensure that we're the only one fiddling with the suspend count
983 // though.
984 MutexLock mu(self, *Locks::thread_suspend_count_lock_);
Yu Lieac44242015-06-29 10:50:03 +0800985 self->ModifySuspendCount(self, +1, nullptr, true);
Ian Rogersdd7624d2014-03-14 17:43:00 -0700986 CHECK_GT(self->GetSuspendCount(), 0);
Sebastien Hertzcbc50642015-06-01 17:33:12 +0200987
988 VLOG(threads) << *self << " self-suspending (debugger)";
989 } else {
990 // We must no longer be subject to debugger suspension.
991 MutexLock mu(self, *Locks::thread_suspend_count_lock_);
992 CHECK_EQ(self->GetDebugSuspendCount(), 0) << "Debugger detached without resuming us";
993
994 VLOG(threads) << *self << " not self-suspending because debugger detached during invoke";
jeffhaoa77f0f62012-12-05 17:19:31 -0800995 }
Elliott Hughes475fc232011-10-25 15:00:35 -0700996
Sebastien Hertzcbc50642015-06-01 17:33:12 +0200997 // If the debugger requested an invoke, we need to send the reply and clear the request.
Sebastien Hertz1558b572015-02-25 15:05:59 +0100998 if (pReq != nullptr) {
Sebastien Hertzcbc50642015-06-01 17:33:12 +0200999 Dbg::FinishInvokeMethod(pReq);
Sebastien Hertz1558b572015-02-25 15:05:59 +01001000 self->ClearDebugInvokeReq();
Sebastien Hertzcbc50642015-06-01 17:33:12 +02001001 pReq = nullptr; // object has been deleted, clear it for safety.
Sebastien Hertz21e729c2014-02-18 14:16:00 +01001002 }
1003
Elliott Hughes475fc232011-10-25 15:00:35 -07001004 // Tell JDWP that we've completed suspension. The JDWP thread can't
1005 // tell us to resume before we're fully asleep because we hold the
1006 // suspend count lock.
1007 Dbg::ClearWaitForEventThread();
1008
jeffhaoa77f0f62012-12-05 17:19:31 -08001009 {
1010 MutexLock mu(self, *Locks::thread_suspend_count_lock_);
Ian Rogersdd7624d2014-03-14 17:43:00 -07001011 while (self->GetSuspendCount() != 0) {
jeffhaoa77f0f62012-12-05 17:19:31 -08001012 Thread::resume_cond_->Wait(self);
Ian Rogersdd7624d2014-03-14 17:43:00 -07001013 if (self->GetSuspendCount() != 0) {
jeffhaoa77f0f62012-12-05 17:19:31 -08001014 // The condition was signaled but we're still suspended. This
Sebastien Hertzf272af42014-09-18 10:20:42 +02001015 // can happen when we suspend then resume all threads to
1016 // update instrumentation or compute monitor info. This can
1017 // also happen if the debugger lets go while a SIGQUIT thread
jeffhaoa77f0f62012-12-05 17:19:31 -08001018 // dump event is pending (assuming SignalCatcher was resumed for
1019 // just long enough to try to grab the thread-suspend lock).
Sebastien Hertzf272af42014-09-18 10:20:42 +02001020 VLOG(jdwp) << *self << " still suspended after undo "
1021 << "(suspend count=" << self->GetSuspendCount() << ", "
1022 << "debug suspend count=" << self->GetDebugSuspendCount() << ")";
jeffhaoa77f0f62012-12-05 17:19:31 -08001023 }
Elliott Hughes475fc232011-10-25 15:00:35 -07001024 }
Ian Rogersdd7624d2014-03-14 17:43:00 -07001025 CHECK_EQ(self->GetSuspendCount(), 0);
Elliott Hughes475fc232011-10-25 15:00:35 -07001026 }
jeffhaoa77f0f62012-12-05 17:19:31 -08001027
Sebastien Hertz1558b572015-02-25 15:05:59 +01001028 self->SetReadyForDebugInvoke(false);
Elliott Hughes1f729aa2012-03-02 13:55:41 -08001029 VLOG(threads) << *self << " self-reviving (debugger)";
Elliott Hughes475fc232011-10-25 15:00:35 -07001030}
1031
Sebastien Hertz253fa552014-10-14 17:27:15 +02001032void ThreadList::ResumeAllForDebugger() {
1033 Thread* self = Thread::Current();
1034 Thread* debug_thread = Dbg::GetDebugThread();
Sebastien Hertz253fa552014-10-14 17:27:15 +02001035
1036 VLOG(threads) << *self << " ResumeAllForDebugger starting...";
1037
1038 // Threads can't resume if we exclusively hold the mutator lock.
1039 Locks::mutator_lock_->AssertNotExclusiveHeld(self);
1040
1041 {
Andreas Gampe277ccbd2014-11-03 21:36:10 -08001042 MutexLock thread_list_mu(self, *Locks::thread_list_lock_);
Sebastien Hertz253fa552014-10-14 17:27:15 +02001043 {
Andreas Gampe277ccbd2014-11-03 21:36:10 -08001044 MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_);
Sebastien Hertz253fa552014-10-14 17:27:15 +02001045 // Update global suspend all state for attaching threads.
1046 DCHECK_GE(suspend_all_count_, debug_suspend_all_count_);
Sebastien Hertzf9d233d2015-01-09 14:51:41 +01001047 if (debug_suspend_all_count_ > 0) {
Sebastien Hertz253fa552014-10-14 17:27:15 +02001048 --suspend_all_count_;
1049 --debug_suspend_all_count_;
Sebastien Hertz253fa552014-10-14 17:27:15 +02001050 } else {
1051 // We've been asked to resume all threads without being asked to
Sebastien Hertzf9d233d2015-01-09 14:51:41 +01001052 // suspend them all before. That may happen if a debugger tries
1053 // to resume some suspended threads (with suspend count == 1)
1054 // at once with a VirtualMachine.Resume command. Let's print a
1055 // warning.
Sebastien Hertz253fa552014-10-14 17:27:15 +02001056 LOG(WARNING) << "Debugger attempted to resume all threads without "
1057 << "having suspended them all before.";
1058 }
Sebastien Hertzf9d233d2015-01-09 14:51:41 +01001059 // Decrement everybody's suspend count (except our own).
1060 for (const auto& thread : list_) {
1061 if (thread == self || thread == debug_thread) {
1062 continue;
1063 }
1064 if (thread->GetDebugSuspendCount() == 0) {
1065 // This thread may have been individually resumed with ThreadReference.Resume.
1066 continue;
1067 }
1068 VLOG(threads) << "requesting thread resume: " << *thread;
Yu Lieac44242015-06-29 10:50:03 +08001069 thread->ModifySuspendCount(self, -1, nullptr, true);
Sebastien Hertzf9d233d2015-01-09 14:51:41 +01001070 }
Sebastien Hertz253fa552014-10-14 17:27:15 +02001071 }
1072 }
1073
Sebastien Hertzf9d233d2015-01-09 14:51:41 +01001074 {
Sebastien Hertz253fa552014-10-14 17:27:15 +02001075 MutexLock mu(self, *Locks::thread_suspend_count_lock_);
1076 Thread::resume_cond_->Broadcast(self);
1077 }
1078
1079 VLOG(threads) << *self << " ResumeAllForDebugger complete";
1080}
1081
Elliott Hughes234ab152011-10-26 14:02:26 -07001082void ThreadList::UndoDebuggerSuspensions() {
1083 Thread* self = Thread::Current();
1084
Elliott Hughes4dd9b4d2011-12-12 18:29:24 -08001085 VLOG(threads) << *self << " UndoDebuggerSuspensions starting";
Elliott Hughes234ab152011-10-26 14:02:26 -07001086
1087 {
Ian Rogers81d425b2012-09-27 16:03:43 -07001088 MutexLock mu(self, *Locks::thread_list_lock_);
1089 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001090 // Update global suspend all state for attaching threads.
1091 suspend_all_count_ -= debug_suspend_all_count_;
1092 debug_suspend_all_count_ = 0;
1093 // Update running threads.
Mathieu Chartier02e25112013-08-14 16:14:24 -07001094 for (const auto& thread : list_) {
Ian Rogersdd7624d2014-03-14 17:43:00 -07001095 if (thread == self || thread->GetDebugSuspendCount() == 0) {
Elliott Hughes234ab152011-10-26 14:02:26 -07001096 continue;
1097 }
Yu Lieac44242015-06-29 10:50:03 +08001098 thread->ModifySuspendCount(self, -thread->GetDebugSuspendCount(), nullptr, true);
Elliott Hughes234ab152011-10-26 14:02:26 -07001099 }
1100 }
1101
1102 {
Ian Rogers81d425b2012-09-27 16:03:43 -07001103 MutexLock mu(self, *Locks::thread_suspend_count_lock_);
Ian Rogersc604d732012-10-14 16:09:54 -07001104 Thread::resume_cond_->Broadcast(self);
Elliott Hughes234ab152011-10-26 14:02:26 -07001105 }
1106
Elliott Hughes4dd9b4d2011-12-12 18:29:24 -08001107 VLOG(threads) << "UndoDebuggerSuspensions(" << *self << ") complete";
Elliott Hughes234ab152011-10-26 14:02:26 -07001108}
1109
Elliott Hughese52e49b2012-04-02 16:05:44 -07001110void ThreadList::WaitForOtherNonDaemonThreadsToExit() {
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -08001111 ScopedTrace trace(__PRETTY_FUNCTION__);
Ian Rogers81d425b2012-09-27 16:03:43 -07001112 Thread* self = Thread::Current();
1113 Locks::mutator_lock_->AssertNotHeld(self);
Mathieu Chartier91e56692015-03-03 13:51:04 -08001114 while (true) {
Ian Rogers120f1c72012-09-28 17:17:10 -07001115 {
1116 // No more threads can be born after we start to shutdown.
1117 MutexLock mu(self, *Locks::runtime_shutdown_lock_);
Mathieu Chartier590fee92013-09-13 13:46:47 -07001118 CHECK(Runtime::Current()->IsShuttingDownLocked());
Ian Rogers120f1c72012-09-28 17:17:10 -07001119 CHECK_EQ(Runtime::Current()->NumberOfThreadsBeingBorn(), 0U);
1120 }
Ian Rogers120f1c72012-09-28 17:17:10 -07001121 MutexLock mu(self, *Locks::thread_list_lock_);
Mathieu Chartier91e56692015-03-03 13:51:04 -08001122 // Also wait for any threads that are unregistering to finish. This is required so that no
1123 // threads access the thread list after it is deleted. TODO: This may not work for user daemon
1124 // threads since they could unregister at the wrong time.
1125 bool done = unregistering_count_ == 0;
1126 if (done) {
1127 for (const auto& thread : list_) {
1128 if (thread != self && !thread->IsDaemon()) {
1129 done = false;
1130 break;
1131 }
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001132 }
1133 }
Mathieu Chartier91e56692015-03-03 13:51:04 -08001134 if (done) {
1135 break;
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001136 }
Mathieu Chartier91e56692015-03-03 13:51:04 -08001137 // Wait for another thread to exit before re-checking.
1138 Locks::thread_exit_cond_->Wait(self);
1139 }
Elliott Hughes038a8062011-09-18 14:12:41 -07001140}
1141
Mathieu Chartier4d87df62016-01-07 15:14:19 -08001142void ThreadList::SuspendAllDaemonThreadsForShutdown() {
Mathieu Chartier32ce2ad2016-03-04 14:58:03 -08001143 ScopedTrace trace(__PRETTY_FUNCTION__);
Ian Rogers81d425b2012-09-27 16:03:43 -07001144 Thread* self = Thread::Current();
1145 MutexLock mu(self, *Locks::thread_list_lock_);
Mathieu Chartier62597d12016-01-11 10:19:06 -08001146 size_t daemons_left = 0;
Brian Carlstrom7934ac22013-07-26 10:54:15 -07001147 { // Tell all the daemons it's time to suspend.
Ian Rogers81d425b2012-09-27 16:03:43 -07001148 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Mathieu Chartier02e25112013-08-14 16:14:24 -07001149 for (const auto& thread : list_) {
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001150 // This is only run after all non-daemon threads have exited, so the remainder should all be
1151 // daemons.
Ian Rogers7e762862012-10-22 15:45:08 -07001152 CHECK(thread->IsDaemon()) << *thread;
Ian Rogers81d425b2012-09-27 16:03:43 -07001153 if (thread != self) {
Yu Lieac44242015-06-29 10:50:03 +08001154 thread->ModifySuspendCount(self, +1, nullptr, false);
Mathieu Chartier62597d12016-01-11 10:19:06 -08001155 ++daemons_left;
Elliott Hughese52e49b2012-04-02 16:05:44 -07001156 }
Mathieu Chartier4d87df62016-01-07 15:14:19 -08001157 // We are shutting down the runtime, set the JNI functions of all the JNIEnvs to be
1158 // the sleep forever one.
1159 thread->GetJniEnv()->SetFunctionsToRuntimeShutdownFunctions();
Elliott Hughes038a8062011-09-18 14:12:41 -07001160 }
1161 }
Mathieu Chartier62597d12016-01-11 10:19:06 -08001162 // If we have any daemons left, wait 200ms to ensure they are not stuck in a place where they
1163 // are about to access runtime state and are not in a runnable state. Examples: Monitor code
1164 // or waking up from a condition variable. TODO: Try and see if there is a better way to wait
1165 // for daemon threads to be in a blocked state.
1166 if (daemons_left > 0) {
1167 static constexpr size_t kDaemonSleepTime = 200 * 1000;
1168 usleep(kDaemonSleepTime);
1169 }
Elliott Hughes038a8062011-09-18 14:12:41 -07001170 // Give the threads a chance to suspend, complaining if they're slow.
1171 bool have_complained = false;
Mathieu Chartierba098ba2016-01-07 09:31:33 -08001172 static constexpr size_t kTimeoutMicroseconds = 2000 * 1000;
1173 static constexpr size_t kSleepMicroseconds = 1000;
1174 for (size_t i = 0; i < kTimeoutMicroseconds / kSleepMicroseconds; ++i) {
Elliott Hughes038a8062011-09-18 14:12:41 -07001175 bool all_suspended = true;
Mathieu Chartier02e25112013-08-14 16:14:24 -07001176 for (const auto& thread : list_) {
Ian Rogers81d425b2012-09-27 16:03:43 -07001177 if (thread != self && thread->GetState() == kRunnable) {
Elliott Hughes038a8062011-09-18 14:12:41 -07001178 if (!have_complained) {
1179 LOG(WARNING) << "daemon thread not yet suspended: " << *thread;
1180 have_complained = true;
1181 }
1182 all_suspended = false;
1183 }
1184 }
1185 if (all_suspended) {
1186 return;
1187 }
Mathieu Chartierba098ba2016-01-07 09:31:33 -08001188 usleep(kSleepMicroseconds);
Elliott Hughes038a8062011-09-18 14:12:41 -07001189 }
Mathieu Chartierba098ba2016-01-07 09:31:33 -08001190 LOG(WARNING) << "timed out suspending all daemon threads";
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001191}
Mathieu Chartierfbc31082016-01-24 11:59:56 -08001192
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001193void ThreadList::Register(Thread* self) {
1194 DCHECK_EQ(self, Thread::Current());
1195
1196 if (VLOG_IS_ON(threads)) {
1197 std::ostringstream oss;
1198 self->ShortDump(oss); // We don't hold the mutator_lock_ yet and so cannot call Dump.
Ian Rogers5a9ba012014-05-19 13:28:52 -07001199 LOG(INFO) << "ThreadList::Register() " << *self << "\n" << oss.str();
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001200 }
1201
1202 // Atomically add self to the thread list and make its thread_suspend_count_ reflect ongoing
1203 // SuspendAll requests.
Ian Rogers81d425b2012-09-27 16:03:43 -07001204 MutexLock mu(self, *Locks::thread_list_lock_);
1205 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Ian Rogersdd7624d2014-03-14 17:43:00 -07001206 CHECK_GE(suspend_all_count_, debug_suspend_all_count_);
Ian Rogers2966e132014-04-02 08:34:36 -07001207 // Modify suspend count in increments of 1 to maintain invariants in ModifySuspendCount. While
1208 // this isn't particularly efficient the suspend counts are most commonly 0 or 1.
1209 for (int delta = debug_suspend_all_count_; delta > 0; delta--) {
Yu Lieac44242015-06-29 10:50:03 +08001210 self->ModifySuspendCount(self, +1, nullptr, true);
Ian Rogersdd7624d2014-03-14 17:43:00 -07001211 }
Ian Rogers2966e132014-04-02 08:34:36 -07001212 for (int delta = suspend_all_count_ - debug_suspend_all_count_; delta > 0; delta--) {
Yu Lieac44242015-06-29 10:50:03 +08001213 self->ModifySuspendCount(self, +1, nullptr, false);
Ian Rogers01ae5802012-09-28 16:14:01 -07001214 }
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001215 CHECK(!Contains(self));
1216 list_.push_back(self);
Hiroshi Yamauchi0b713572015-06-16 18:29:23 -07001217 if (kUseReadBarrier) {
Hiroshi Yamauchi00370822015-08-18 14:47:25 -07001218 // Initialize according to the state of the CC collector.
1219 bool is_gc_marking =
1220 Runtime::Current()->GetHeap()->ConcurrentCopyingCollector()->IsMarking();
1221 self->SetIsGcMarking(is_gc_marking);
Hiroshi Yamauchi0b713572015-06-16 18:29:23 -07001222 bool weak_ref_access_enabled =
1223 Runtime::Current()->GetHeap()->ConcurrentCopyingCollector()->IsWeakRefAccessEnabled();
1224 self->SetWeakRefAccessEnabled(weak_ref_access_enabled);
1225 }
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001226}
1227
1228void ThreadList::Unregister(Thread* self) {
1229 DCHECK_EQ(self, Thread::Current());
Ian Rogers68d8b422014-07-17 11:09:10 -07001230 CHECK_NE(self->GetState(), kRunnable);
1231 Locks::mutator_lock_->AssertNotHeld(self);
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001232
1233 VLOG(threads) << "ThreadList::Unregister() " << *self;
1234
Mathieu Chartier91e56692015-03-03 13:51:04 -08001235 {
1236 MutexLock mu(self, *Locks::thread_list_lock_);
1237 ++unregistering_count_;
1238 }
1239
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001240 // Any time-consuming destruction, plus anything that can call back into managed code or
Mathieu Chartier91e56692015-03-03 13:51:04 -08001241 // suspend and so on, must happen at this point, and not in ~Thread. The self->Destroy is what
1242 // causes the threads to join. It is important to do this after incrementing unregistering_count_
1243 // since we want the runtime to wait for the daemon threads to exit before deleting the thread
1244 // list.
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001245 self->Destroy();
1246
Jeff Haoe094b872014-10-14 13:12:01 -07001247 // If tracing, remember thread id and name before thread exits.
1248 Trace::StoreExitingThreadInfo(self);
1249
Ian Rogersdd7624d2014-03-14 17:43:00 -07001250 uint32_t thin_lock_id = self->GetThreadId();
Mathieu Chartier91e56692015-03-03 13:51:04 -08001251 while (true) {
Ian Rogerscfaa4552012-11-26 21:00:08 -08001252 // Remove and delete the Thread* while holding the thread_list_lock_ and
1253 // thread_suspend_count_lock_ so that the unregistering thread cannot be suspended.
Ian Rogers0878d652013-04-18 17:38:35 -07001254 // Note: deliberately not using MutexLock that could hold a stale self pointer.
Mathieu Chartier91e56692015-03-03 13:51:04 -08001255 MutexLock mu(self, *Locks::thread_list_lock_);
Ian Rogersa2af5c72014-09-15 15:17:07 -07001256 if (!Contains(self)) {
Mathieu Chartier9db831a2015-02-24 17:20:30 -08001257 std::string thread_name;
1258 self->GetThreadName(thread_name);
Ian Rogersa2af5c72014-09-15 15:17:07 -07001259 std::ostringstream os;
Christopher Ferris6cff48f2014-01-26 21:36:13 -08001260 DumpNativeStack(os, GetTid(), nullptr, " native: ", nullptr);
Mathieu Chartier9db831a2015-02-24 17:20:30 -08001261 LOG(ERROR) << "Request to unregister unattached thread " << thread_name << "\n" << os.str();
Mathieu Chartier91e56692015-03-03 13:51:04 -08001262 break;
Ian Rogersa2af5c72014-09-15 15:17:07 -07001263 } else {
Mathieu Chartier91e56692015-03-03 13:51:04 -08001264 MutexLock mu2(self, *Locks::thread_suspend_count_lock_);
Ian Rogersa2af5c72014-09-15 15:17:07 -07001265 if (!self->IsSuspended()) {
1266 list_.remove(self);
Mathieu Chartier91e56692015-03-03 13:51:04 -08001267 break;
Ian Rogersa2af5c72014-09-15 15:17:07 -07001268 }
Ian Rogers68d8b422014-07-17 11:09:10 -07001269 }
Mathieu Chartier91e56692015-03-03 13:51:04 -08001270 // We failed to remove the thread due to a suspend request, loop and try again.
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001271 }
Mathieu Chartier91e56692015-03-03 13:51:04 -08001272 delete self;
1273
Mathieu Chartier5f51d4b2013-12-03 14:24:05 -08001274 // Release the thread ID after the thread is finished and deleted to avoid cases where we can
1275 // temporarily have multiple threads with the same thread id. When this occurs, it causes
1276 // problems in FindThreadByThreadId / SuspendThreadByThreadId.
1277 ReleaseThreadId(nullptr, thin_lock_id);
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001278
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001279 // Clear the TLS data, so that the underlying native thread is recognizably detached.
1280 // (It may wish to reattach later.)
Andreas Gampe4382f1e2015-08-05 01:08:53 +00001281#ifdef __ANDROID__
1282 __get_tls()[TLS_SLOT_ART_THREAD_SELF] = nullptr;
1283#else
Mathieu Chartier2cebb242015-04-21 16:50:40 -07001284 CHECK_PTHREAD_CALL(pthread_setspecific, (Thread::pthread_key_self_, nullptr), "detach self");
Andreas Gampe4382f1e2015-08-05 01:08:53 +00001285#endif
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001286
1287 // Signal that a thread just detached.
Mathieu Chartier91e56692015-03-03 13:51:04 -08001288 MutexLock mu(nullptr, *Locks::thread_list_lock_);
1289 --unregistering_count_;
1290 Locks::thread_exit_cond_->Broadcast(nullptr);
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001291}
1292
1293void ThreadList::ForEach(void (*callback)(Thread*, void*), void* context) {
Mathieu Chartier02e25112013-08-14 16:14:24 -07001294 for (const auto& thread : list_) {
1295 callback(thread, context);
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001296 }
1297}
1298
Mathieu Chartierbb87e0f2015-04-03 11:21:55 -07001299void ThreadList::VisitRoots(RootVisitor* visitor) const {
Ian Rogers81d425b2012-09-27 16:03:43 -07001300 MutexLock mu(Thread::Current(), *Locks::thread_list_lock_);
Mathieu Chartier02e25112013-08-14 16:14:24 -07001301 for (const auto& thread : list_) {
Mathieu Chartierbb87e0f2015-04-03 11:21:55 -07001302 thread->VisitRoots(visitor);
Ian Rogers00f7d0e2012-07-19 15:28:27 -07001303 }
Elliott Hughes038a8062011-09-18 14:12:41 -07001304}
1305
Ian Rogerscfaa4552012-11-26 21:00:08 -08001306uint32_t ThreadList::AllocThreadId(Thread* self) {
Chao-ying Fu9e369312014-05-21 11:20:52 -07001307 MutexLock mu(self, *Locks::allocated_thread_ids_lock_);
Elliott Hughes8daa0922011-09-11 13:46:25 -07001308 for (size_t i = 0; i < allocated_ids_.size(); ++i) {
1309 if (!allocated_ids_[i]) {
1310 allocated_ids_.set(i);
Brian Carlstrom7934ac22013-07-26 10:54:15 -07001311 return i + 1; // Zero is reserved to mean "invalid".
Elliott Hughes8daa0922011-09-11 13:46:25 -07001312 }
1313 }
1314 LOG(FATAL) << "Out of internal thread ids";
1315 return 0;
1316}
1317
Ian Rogerscfaa4552012-11-26 21:00:08 -08001318void ThreadList::ReleaseThreadId(Thread* self, uint32_t id) {
Chao-ying Fu9e369312014-05-21 11:20:52 -07001319 MutexLock mu(self, *Locks::allocated_thread_ids_lock_);
Brian Carlstrom7934ac22013-07-26 10:54:15 -07001320 --id; // Zero is reserved to mean "invalid".
Elliott Hughes8daa0922011-09-11 13:46:25 -07001321 DCHECK(allocated_ids_[id]) << id;
1322 allocated_ids_.reset(id);
1323}
1324
Mathieu Chartier4f55e222015-09-04 13:26:21 -07001325ScopedSuspendAll::ScopedSuspendAll(const char* cause, bool long_suspend) {
1326 Runtime::Current()->GetThreadList()->SuspendAll(cause, long_suspend);
1327}
1328
1329ScopedSuspendAll::~ScopedSuspendAll() {
1330 Runtime::Current()->GetThreadList()->ResumeAll();
1331}
1332
Elliott Hughes8daa0922011-09-11 13:46:25 -07001333} // namespace art