Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 1 | /* |
| 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 | |
Elliott Hughes | abbe07d | 2012-06-05 17:42:23 -0700 | [diff] [blame] | 19 | #include <dirent.h> |
| 20 | #include <sys/types.h> |
Elliott Hughes | 038a806 | 2011-09-18 14:12:41 -0700 | [diff] [blame] | 21 | #include <unistd.h> |
| 22 | |
Ian Rogers | c7dd295 | 2014-10-21 23:31:19 -0700 | [diff] [blame] | 23 | #include <sstream> |
Christopher Ferris | b1b0ed2 | 2017-10-24 15:38:14 -0700 | [diff] [blame] | 24 | #include <vector> |
Ian Rogers | c7dd295 | 2014-10-21 23:31:19 -0700 | [diff] [blame] | 25 | |
Andreas Gampe | 46ee31b | 2016-12-14 10:11:49 -0800 | [diff] [blame] | 26 | #include "android-base/stringprintf.h" |
Andreas Gampe | 8cf9cb3 | 2017-07-19 09:28:38 -0700 | [diff] [blame] | 27 | #include "backtrace/BacktraceMap.h" |
Andreas Gampe | 373a9b5 | 2017-10-18 09:01:57 -0700 | [diff] [blame] | 28 | #include "nativehelper/scoped_local_ref.h" |
| 29 | #include "nativehelper/scoped_utf_chars.h" |
Andreas Gampe | 46ee31b | 2016-12-14 10:11:49 -0800 | [diff] [blame] | 30 | |
Andreas Gampe | 39b378c | 2017-12-07 15:44:13 -0800 | [diff] [blame] | 31 | #include "base/aborting.h" |
Mathieu Chartier | 70a596d | 2014-12-17 14:56:47 -0800 | [diff] [blame] | 32 | #include "base/histogram-inl.h" |
Hiroshi Yamauchi | 967a0ad | 2013-09-10 16:24:21 -0700 | [diff] [blame] | 33 | #include "base/mutex-inl.h" |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 34 | #include "base/systrace.h" |
Vladimir Marko | 80afd02 | 2015-05-19 18:08:00 +0100 | [diff] [blame] | 35 | #include "base/time_utils.h" |
Sameer Abu Asal | a843954 | 2013-02-14 16:06:42 -0800 | [diff] [blame] | 36 | #include "base/timing_logger.h" |
Elliott Hughes | 475fc23 | 2011-10-25 15:00:35 -0700 | [diff] [blame] | 37 | #include "debugger.h" |
Hiroshi Yamauchi | 0b71357 | 2015-06-16 18:29:23 -0700 | [diff] [blame] | 38 | #include "gc/collector/concurrent_copying.h" |
Andreas Gampe | 6e64445 | 2017-05-09 16:30:27 -0700 | [diff] [blame] | 39 | #include "gc/gc_pause_listener.h" |
Andreas Gampe | d490129 | 2017-05-30 18:41:34 -0700 | [diff] [blame] | 40 | #include "gc/heap.h" |
Hiroshi Yamauchi | 3049324 | 2016-11-03 13:06:52 -0700 | [diff] [blame] | 41 | #include "gc/reference_processor.h" |
Andreas Gampe | 513061a | 2017-06-01 09:17:34 -0700 | [diff] [blame] | 42 | #include "gc_root.h" |
Vladimir Marko | a3ad0cd | 2018-05-04 10:06:38 +0100 | [diff] [blame] | 43 | #include "jni/jni_internal.h" |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 44 | #include "lock_word.h" |
| 45 | #include "monitor.h" |
Andreas Gampe | 5dd44d0 | 2016-08-02 17:20:03 -0700 | [diff] [blame] | 46 | #include "native_stack_dump.h" |
Mathieu Chartier | 0795f23 | 2016-09-27 18:43:30 -0700 | [diff] [blame] | 47 | #include "scoped_thread_state_change-inl.h" |
Ian Rogers | 2dd0e2c | 2013-01-24 12:42:14 -0800 | [diff] [blame] | 48 | #include "thread.h" |
Jeff Hao | e094b87 | 2014-10-14 13:12:01 -0700 | [diff] [blame] | 49 | #include "trace.h" |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 50 | #include "well_known_classes.h" |
Elliott Hughes | 475fc23 | 2011-10-25 15:00:35 -0700 | [diff] [blame] | 51 | |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 52 | #if ART_USE_FUTEXES |
| 53 | #include "linux/futex.h" |
| 54 | #include "sys/syscall.h" |
| 55 | #ifndef SYS_futex |
| 56 | #define SYS_futex __NR_futex |
| 57 | #endif |
| 58 | #endif // ART_USE_FUTEXES |
| 59 | |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 60 | namespace art { |
| 61 | |
Andreas Gampe | 46ee31b | 2016-12-14 10:11:49 -0800 | [diff] [blame] | 62 | using android::base::StringPrintf; |
| 63 | |
Mathieu Chartier | 251755c | 2014-07-15 18:10:25 -0700 | [diff] [blame] | 64 | static constexpr uint64_t kLongThreadSuspendThreshold = MsToNs(5); |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 65 | // Use 0 since we want to yield to prevent blocking for an unpredictable amount of time. |
| 66 | static constexpr useconds_t kThreadSuspendInitialSleepUs = 0; |
| 67 | static constexpr useconds_t kThreadSuspendMaxYieldUs = 3000; |
| 68 | static constexpr useconds_t kThreadSuspendMaxSleepUs = 5000; |
Mathieu Chartier | 251755c | 2014-07-15 18:10:25 -0700 | [diff] [blame] | 69 | |
Andreas Gampe | 8d1594d | 2016-03-01 14:38:37 -0800 | [diff] [blame] | 70 | // Whether we should try to dump the native stack of unattached threads. See commit ed8b723 for |
| 71 | // some history. |
Christopher Ferris | 4531aae | 2017-11-02 14:01:14 -0700 | [diff] [blame] | 72 | static constexpr bool kDumpUnattachedThreadNativeStackForSigQuit = true; |
Andreas Gampe | 8d1594d | 2016-03-01 14:38:37 -0800 | [diff] [blame] | 73 | |
Mathieu Chartier | 3fceaf5 | 2017-01-22 13:33:40 -0800 | [diff] [blame] | 74 | ThreadList::ThreadList(uint64_t thread_suspend_timeout_ns) |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 75 | : suspend_all_count_(0), |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 76 | unregistering_count_(0), |
| 77 | suspend_all_historam_("suspend all histogram", 16, 64), |
Hiroshi Yamauchi | 3049324 | 2016-11-03 13:06:52 -0700 | [diff] [blame] | 78 | long_suspend_(false), |
Andreas Gampe | c4bed16 | 2017-05-01 13:46:24 -0700 | [diff] [blame] | 79 | shut_down_(false), |
Mathieu Chartier | 3fceaf5 | 2017-01-22 13:33:40 -0800 | [diff] [blame] | 80 | thread_suspend_timeout_ns_(thread_suspend_timeout_ns), |
Hiroshi Yamauchi | 3049324 | 2016-11-03 13:06:52 -0700 | [diff] [blame] | 81 | empty_checkpoint_barrier_(new Barrier(0)) { |
Hiroshi Yamauchi | e15ea08 | 2015-02-09 17:11:42 -0800 | [diff] [blame] | 82 | CHECK(Monitor::IsValidLockWord(LockWord::FromThinLockId(kMaxThreadId, 1, 0U))); |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 83 | } |
| 84 | |
| 85 | ThreadList::~ThreadList() { |
Andreas Gampe | c4bed16 | 2017-05-01 13:46:24 -0700 | [diff] [blame] | 86 | CHECK(shut_down_); |
| 87 | } |
| 88 | |
| 89 | void ThreadList::ShutDown() { |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 90 | ScopedTrace trace(__PRETTY_FUNCTION__); |
Elliott Hughes | e52e49b | 2012-04-02 16:05:44 -0700 | [diff] [blame] | 91 | // Detach the current thread if necessary. If we failed to start, there might not be any threads. |
Elliott Hughes | 6a14433 | 2012-04-03 13:07:11 -0700 | [diff] [blame] | 92 | // We need to detach the current thread here in case there's another thread waiting to join with |
| 93 | // us. |
Mathieu Chartier | fec72f4 | 2014-10-09 12:57:58 -0700 | [diff] [blame] | 94 | bool contains = false; |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 95 | Thread* self = Thread::Current(); |
Mathieu Chartier | fec72f4 | 2014-10-09 12:57:58 -0700 | [diff] [blame] | 96 | { |
Mathieu Chartier | fec72f4 | 2014-10-09 12:57:58 -0700 | [diff] [blame] | 97 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 98 | contains = Contains(self); |
| 99 | } |
| 100 | if (contains) { |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 101 | Runtime::Current()->DetachCurrentThread(); |
| 102 | } |
Elliott Hughes | 6a14433 | 2012-04-03 13:07:11 -0700 | [diff] [blame] | 103 | WaitForOtherNonDaemonThreadsToExit(); |
Mathieu Chartier | 5116837 | 2015-08-12 16:40:32 -0700 | [diff] [blame] | 104 | // Disable GC and wait for GC to complete in case there are still daemon threads doing |
| 105 | // allocations. |
| 106 | gc::Heap* const heap = Runtime::Current()->GetHeap(); |
| 107 | heap->DisableGCForShutdown(); |
| 108 | // In case a GC is in progress, wait for it to finish. |
| 109 | heap->WaitForGcToComplete(gc::kGcCauseBackground, Thread::Current()); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 110 | // TODO: there's an unaddressed race here where a thread may attach during shutdown, see |
| 111 | // Thread::Init. |
Mathieu Chartier | 4d87df6 | 2016-01-07 15:14:19 -0800 | [diff] [blame] | 112 | SuspendAllDaemonThreadsForShutdown(); |
Andreas Gampe | c4bed16 | 2017-05-01 13:46:24 -0700 | [diff] [blame] | 113 | |
| 114 | shut_down_ = true; |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 115 | } |
| 116 | |
| 117 | bool ThreadList::Contains(Thread* thread) { |
| 118 | return find(list_.begin(), list_.end(), thread) != list_.end(); |
| 119 | } |
| 120 | |
Elliott Hughes | abbe07d | 2012-06-05 17:42:23 -0700 | [diff] [blame] | 121 | bool ThreadList::Contains(pid_t tid) { |
Mathieu Chartier | 02e2511 | 2013-08-14 16:14:24 -0700 | [diff] [blame] | 122 | for (const auto& thread : list_) { |
Ian Rogers | dd7624d | 2014-03-14 17:43:00 -0700 | [diff] [blame] | 123 | if (thread->GetTid() == tid) { |
Elliott Hughes | abbe07d | 2012-06-05 17:42:23 -0700 | [diff] [blame] | 124 | return true; |
| 125 | } |
| 126 | } |
| 127 | return false; |
| 128 | } |
| 129 | |
Brian Carlstrom | 24a3c2e | 2011-10-17 18:07:52 -0700 | [diff] [blame] | 130 | pid_t ThreadList::GetLockOwner() { |
Ian Rogers | b726dcb | 2012-09-05 08:57:23 -0700 | [diff] [blame] | 131 | return Locks::thread_list_lock_->GetExclusiveOwnerTid(); |
Elliott Hughes | accd83d | 2011-10-17 14:25:58 -0700 | [diff] [blame] | 132 | } |
| 133 | |
Mathieu Chartier | 590fee9 | 2013-09-13 13:46:47 -0700 | [diff] [blame] | 134 | void ThreadList::DumpNativeStacks(std::ostream& os) { |
| 135 | MutexLock mu(Thread::Current(), *Locks::thread_list_lock_); |
Christopher Ferris | 6cff48f | 2014-01-26 21:36:13 -0800 | [diff] [blame] | 136 | std::unique_ptr<BacktraceMap> map(BacktraceMap::Create(getpid())); |
Mathieu Chartier | 590fee9 | 2013-09-13 13:46:47 -0700 | [diff] [blame] | 137 | for (const auto& thread : list_) { |
Ian Rogers | dd7624d | 2014-03-14 17:43:00 -0700 | [diff] [blame] | 138 | os << "DUMPING THREAD " << thread->GetTid() << "\n"; |
Christopher Ferris | 6cff48f | 2014-01-26 21:36:13 -0800 | [diff] [blame] | 139 | DumpNativeStack(os, thread->GetTid(), map.get(), "\t"); |
Mathieu Chartier | 590fee9 | 2013-09-13 13:46:47 -0700 | [diff] [blame] | 140 | os << "\n"; |
| 141 | } |
| 142 | } |
| 143 | |
Elliott Hughes | c967f78 | 2012-04-16 10:23:15 -0700 | [diff] [blame] | 144 | void ThreadList::DumpForSigQuit(std::ostream& os) { |
Mathieu Chartier | 70a596d | 2014-12-17 14:56:47 -0800 | [diff] [blame] | 145 | { |
| 146 | ScopedObjectAccess soa(Thread::Current()); |
Mathieu Chartier | 23f6e69 | 2014-12-18 18:24:39 -0800 | [diff] [blame] | 147 | // Only print if we have samples. |
| 148 | if (suspend_all_historam_.SampleSize() > 0) { |
| 149 | Histogram<uint64_t>::CumulativeData data; |
| 150 | suspend_all_historam_.CreateHistogram(&data); |
| 151 | suspend_all_historam_.PrintConfidenceIntervals(os, 0.99, data); // Dump time to suspend. |
| 152 | } |
Mathieu Chartier | 70a596d | 2014-12-17 14:56:47 -0800 | [diff] [blame] | 153 | } |
Nicolas Geoffray | 6ee4971 | 2018-03-30 14:39:05 +0000 | [diff] [blame] | 154 | bool dump_native_stack = Runtime::Current()->GetDumpNativeStackOnSigQuit(); |
| 155 | Dump(os, dump_native_stack); |
| 156 | DumpUnattachedThreads(os, dump_native_stack && kDumpUnattachedThreadNativeStackForSigQuit); |
Elliott Hughes | abbe07d | 2012-06-05 17:42:23 -0700 | [diff] [blame] | 157 | } |
| 158 | |
Nicolas Geoffray | d3c5965 | 2016-03-17 09:35:04 +0000 | [diff] [blame] | 159 | static void DumpUnattachedThread(std::ostream& os, pid_t tid, bool dump_native_stack) |
| 160 | NO_THREAD_SAFETY_ANALYSIS { |
Mathieu Chartier | 2cebb24 | 2015-04-21 16:50:40 -0700 | [diff] [blame] | 161 | // TODO: No thread safety analysis as DumpState with a null thread won't access fields, should |
Ian Rogers | cfaa455 | 2012-11-26 21:00:08 -0800 | [diff] [blame] | 162 | // refactor DumpState to avoid skipping analysis. |
Mathieu Chartier | 2cebb24 | 2015-04-21 16:50:40 -0700 | [diff] [blame] | 163 | Thread::DumpState(os, nullptr, tid); |
Mathieu Chartier | 3f386d5 | 2016-10-05 14:12:45 -0700 | [diff] [blame] | 164 | if (dump_native_stack) { |
Christopher Ferris | 6cff48f | 2014-01-26 21:36:13 -0800 | [diff] [blame] | 165 | DumpNativeStack(os, tid, nullptr, " native: "); |
Brian Carlstrom | ed8b723 | 2012-06-27 17:54:47 -0700 | [diff] [blame] | 166 | } |
Andreas Gampe | 5544e72 | 2017-06-05 17:01:27 -0700 | [diff] [blame] | 167 | os << std::endl; |
Elliott Hughes | abbe07d | 2012-06-05 17:42:23 -0700 | [diff] [blame] | 168 | } |
| 169 | |
Nicolas Geoffray | d3c5965 | 2016-03-17 09:35:04 +0000 | [diff] [blame] | 170 | void ThreadList::DumpUnattachedThreads(std::ostream& os, bool dump_native_stack) { |
Elliott Hughes | abbe07d | 2012-06-05 17:42:23 -0700 | [diff] [blame] | 171 | DIR* d = opendir("/proc/self/task"); |
| 172 | if (!d) { |
| 173 | return; |
| 174 | } |
| 175 | |
Ian Rogers | 50b35e2 | 2012-10-04 10:09:15 -0700 | [diff] [blame] | 176 | Thread* self = Thread::Current(); |
Elliott Hughes | 4696b5b | 2012-10-30 10:35:10 -0700 | [diff] [blame] | 177 | dirent* e; |
Mathieu Chartier | 2cebb24 | 2015-04-21 16:50:40 -0700 | [diff] [blame] | 178 | while ((e = readdir(d)) != nullptr) { |
Elliott Hughes | abbe07d | 2012-06-05 17:42:23 -0700 | [diff] [blame] | 179 | char* end; |
Elliott Hughes | 4696b5b | 2012-10-30 10:35:10 -0700 | [diff] [blame] | 180 | pid_t tid = strtol(e->d_name, &end, 10); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 181 | if (!*end) { |
| 182 | bool contains; |
| 183 | { |
Ian Rogers | 50b35e2 | 2012-10-04 10:09:15 -0700 | [diff] [blame] | 184 | MutexLock mu(self, *Locks::thread_list_lock_); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 185 | contains = Contains(tid); |
| 186 | } |
| 187 | if (!contains) { |
Nicolas Geoffray | d3c5965 | 2016-03-17 09:35:04 +0000 | [diff] [blame] | 188 | DumpUnattachedThread(os, tid, dump_native_stack); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 189 | } |
Elliott Hughes | abbe07d | 2012-06-05 17:42:23 -0700 | [diff] [blame] | 190 | } |
| 191 | } |
| 192 | closedir(d); |
Elliott Hughes | ff73806 | 2012-02-03 15:00:42 -0800 | [diff] [blame] | 193 | } |
| 194 | |
Mathieu Chartier | 47c1959 | 2016-03-07 11:59:01 -0800 | [diff] [blame] | 195 | // Dump checkpoint timeout in milliseconds. Larger amount on the target, since the device could be |
| 196 | // overloaded with ANR dumps. |
| 197 | static constexpr uint32_t kDumpWaitTimeout = kIsTargetBuild ? 100000 : 20000; |
Andreas Gampe | 4a3d19b | 2015-01-09 17:54:51 -0800 | [diff] [blame] | 198 | |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 199 | // A closure used by Thread::Dump. |
Roland Levillain | bbc6e7e | 2018-08-24 16:58:47 +0100 | [diff] [blame] | 200 | class DumpCheckpoint final : public Closure { |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 201 | public: |
Nicolas Geoffray | 6ee4971 | 2018-03-30 14:39:05 +0000 | [diff] [blame] | 202 | DumpCheckpoint(std::ostream* os, bool dump_native_stack) |
Nicolas Geoffray | a73280d | 2016-02-15 13:05:16 +0000 | [diff] [blame] | 203 | : os_(os), |
Mathieu Chartier | 5a83225 | 2019-03-28 07:31:16 -0700 | [diff] [blame] | 204 | // Avoid verifying count in case a thread doesn't end up passing through the barrier. |
| 205 | // This avoids a SIGABRT that would otherwise happen in the destructor. |
Mathieu Chartier | 4f1e328 | 2019-03-27 14:41:32 -0700 | [diff] [blame] | 206 | barrier_(0, /*verify_count_on_shutdown=*/false), |
Nicolas Geoffray | 6ee4971 | 2018-03-30 14:39:05 +0000 | [diff] [blame] | 207 | backtrace_map_(dump_native_stack ? BacktraceMap::Create(getpid()) : nullptr), |
| 208 | dump_native_stack_(dump_native_stack) { |
Christopher Ferris | b1b0ed2 | 2017-10-24 15:38:14 -0700 | [diff] [blame] | 209 | if (backtrace_map_ != nullptr) { |
| 210 | backtrace_map_->SetSuffixesToIgnore(std::vector<std::string> { "oat", "odex" }); |
| 211 | } |
| 212 | } |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 213 | |
Roland Levillain | bbc6e7e | 2018-08-24 16:58:47 +0100 | [diff] [blame] | 214 | void Run(Thread* thread) override { |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 215 | // Note thread and self may not be equal if thread was already suspended at the point of the |
| 216 | // request. |
| 217 | Thread* self = Thread::Current(); |
Mathieu Chartier | 3f386d5 | 2016-10-05 14:12:45 -0700 | [diff] [blame] | 218 | CHECK(self != nullptr); |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 219 | std::ostringstream local_os; |
| 220 | { |
| 221 | ScopedObjectAccess soa(self); |
Nicolas Geoffray | 6ee4971 | 2018-03-30 14:39:05 +0000 | [diff] [blame] | 222 | thread->Dump(local_os, dump_native_stack_, backtrace_map_.get()); |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 223 | } |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 224 | { |
| 225 | // Use the logging lock to ensure serialization when writing to the common ostream. |
| 226 | MutexLock mu(self, *Locks::logging_lock_); |
Andreas Gampe | 5544e72 | 2017-06-05 17:01:27 -0700 | [diff] [blame] | 227 | *os_ << local_os.str() << std::endl; |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 228 | } |
Mathieu Chartier | 10d2508 | 2015-10-28 18:36:09 -0700 | [diff] [blame] | 229 | barrier_.Pass(self); |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 230 | } |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 231 | |
| 232 | void WaitForThreadsToRunThroughCheckpoint(size_t threads_running_checkpoint) { |
| 233 | Thread* self = Thread::Current(); |
| 234 | ScopedThreadStateChange tsc(self, kWaitingForCheckPointsToRun); |
Andreas Gampe | 1e4b0ca | 2015-01-14 09:06:32 -0800 | [diff] [blame] | 235 | bool timed_out = barrier_.Increment(self, threads_running_checkpoint, kDumpWaitTimeout); |
Ian Rogers | 2156ff1 | 2014-09-13 19:20:54 -0700 | [diff] [blame] | 236 | if (timed_out) { |
Nicolas Geoffray | db97871 | 2014-12-09 13:33:38 +0000 | [diff] [blame] | 237 | // Avoid a recursive abort. |
Andreas Gampe | 3fec9ac | 2016-09-13 10:47:28 -0700 | [diff] [blame] | 238 | LOG((kIsDebugBuild && (gAborting == 0)) ? ::android::base::FATAL : ::android::base::ERROR) |
Nicolas Geoffray | db97871 | 2014-12-09 13:33:38 +0000 | [diff] [blame] | 239 | << "Unexpected time out during dump checkpoint."; |
Ian Rogers | 2156ff1 | 2014-09-13 19:20:54 -0700 | [diff] [blame] | 240 | } |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 241 | } |
| 242 | |
| 243 | private: |
| 244 | // The common stream that will accumulate all the dumps. |
| 245 | std::ostream* const os_; |
| 246 | // The barrier to be passed through and for the requestor to wait upon. |
| 247 | Barrier barrier_; |
Christopher Ferris | 6cff48f | 2014-01-26 21:36:13 -0800 | [diff] [blame] | 248 | // A backtrace map, so that all threads use a shared info and don't reacquire/parse separately. |
| 249 | std::unique_ptr<BacktraceMap> backtrace_map_; |
Nicolas Geoffray | 6ee4971 | 2018-03-30 14:39:05 +0000 | [diff] [blame] | 250 | // Whether we should dump the native stack. |
| 251 | const bool dump_native_stack_; |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 252 | }; |
| 253 | |
Nicolas Geoffray | 6ee4971 | 2018-03-30 14:39:05 +0000 | [diff] [blame] | 254 | void ThreadList::Dump(std::ostream& os, bool dump_native_stack) { |
Mathieu Chartier | 3f386d5 | 2016-10-05 14:12:45 -0700 | [diff] [blame] | 255 | Thread* self = Thread::Current(); |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 256 | { |
Mathieu Chartier | 3f386d5 | 2016-10-05 14:12:45 -0700 | [diff] [blame] | 257 | MutexLock mu(self, *Locks::thread_list_lock_); |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 258 | os << "DALVIK THREADS (" << list_.size() << "):\n"; |
| 259 | } |
Mathieu Chartier | 3f386d5 | 2016-10-05 14:12:45 -0700 | [diff] [blame] | 260 | if (self != nullptr) { |
Nicolas Geoffray | 6ee4971 | 2018-03-30 14:39:05 +0000 | [diff] [blame] | 261 | DumpCheckpoint checkpoint(&os, dump_native_stack); |
Mathieu Chartier | 3f386d5 | 2016-10-05 14:12:45 -0700 | [diff] [blame] | 262 | size_t threads_running_checkpoint; |
| 263 | { |
| 264 | // Use SOA to prevent deadlocks if multiple threads are calling Dump() at the same time. |
| 265 | ScopedObjectAccess soa(self); |
| 266 | threads_running_checkpoint = RunCheckpoint(&checkpoint); |
| 267 | } |
| 268 | if (threads_running_checkpoint != 0) { |
| 269 | checkpoint.WaitForThreadsToRunThroughCheckpoint(threads_running_checkpoint); |
| 270 | } |
| 271 | } else { |
Nicolas Geoffray | 6ee4971 | 2018-03-30 14:39:05 +0000 | [diff] [blame] | 272 | DumpUnattachedThreads(os, dump_native_stack); |
Lei Li | dd9943d | 2015-02-02 14:24:44 +0800 | [diff] [blame] | 273 | } |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 274 | } |
| 275 | |
Ian Rogers | 50b35e2 | 2012-10-04 10:09:15 -0700 | [diff] [blame] | 276 | void ThreadList::AssertThreadsAreSuspended(Thread* self, Thread* ignore1, Thread* ignore2) { |
| 277 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 278 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
Mathieu Chartier | 02e2511 | 2013-08-14 16:14:24 -0700 | [diff] [blame] | 279 | for (const auto& thread : list_) { |
jeffhao | 725a957 | 2012-11-13 18:20:12 -0800 | [diff] [blame] | 280 | if (thread != ignore1 && thread != ignore2) { |
Ian Rogers | 01ae580 | 2012-09-28 16:14:01 -0700 | [diff] [blame] | 281 | CHECK(thread->IsSuspended()) |
| 282 | << "\nUnsuspended thread: <<" << *thread << "\n" |
| 283 | << "self: <<" << *Thread::Current(); |
| 284 | } |
Elliott Hughes | 8d768a9 | 2011-09-14 16:35:25 -0700 | [diff] [blame] | 285 | } |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 286 | } |
| 287 | |
Ian Rogers | 66aee5c | 2012-08-15 17:17:47 -0700 | [diff] [blame] | 288 | #if HAVE_TIMED_RWLOCK |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 289 | // Attempt to rectify locks so that we dump thread list with required locks before exiting. |
Andreas Gampe | 794ad76 | 2015-02-23 08:12:24 -0800 | [diff] [blame] | 290 | NO_RETURN static void UnsafeLogFatalForThreadSuspendAllTimeout() { |
Mathieu Chartier | 767d039 | 2017-11-09 14:06:43 -0800 | [diff] [blame] | 291 | // Increment gAborting before doing the thread list dump since we don't want any failures from |
| 292 | // AssertThreadSuspensionIsAllowable in cases where thread suspension is not allowed. |
| 293 | // See b/69044468. |
| 294 | ++gAborting; |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 295 | Runtime* runtime = Runtime::Current(); |
| 296 | std::ostringstream ss; |
| 297 | ss << "Thread suspend timeout\n"; |
Mathieu Chartier | 5869a2c | 2014-10-08 14:26:23 -0700 | [diff] [blame] | 298 | Locks::mutator_lock_->Dump(ss); |
| 299 | ss << "\n"; |
Ian Rogers | 7b078e8 | 2014-09-10 14:44:24 -0700 | [diff] [blame] | 300 | runtime->GetThreadList()->Dump(ss); |
Mathieu Chartier | 767d039 | 2017-11-09 14:06:43 -0800 | [diff] [blame] | 301 | --gAborting; |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 302 | LOG(FATAL) << ss.str(); |
Ian Rogers | 719d1a3 | 2014-03-06 12:13:39 -0800 | [diff] [blame] | 303 | exit(0); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 304 | } |
Ian Rogers | 66aee5c | 2012-08-15 17:17:47 -0700 | [diff] [blame] | 305 | #endif |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 306 | |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 307 | // Unlike suspending all threads where we can wait to acquire the mutator_lock_, suspending an |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 308 | // individual thread requires polling. delay_us is the requested sleep wait. If delay_us is 0 then |
| 309 | // we use sched_yield instead of calling usleep. |
David Sehr | 446c6b2 | 2017-11-09 14:08:19 -0800 | [diff] [blame] | 310 | // Although there is the possibility, here and elsewhere, that usleep could return -1 and |
| 311 | // errno = EINTR, there should be no problem if interrupted, so we do not check. |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 312 | static void ThreadSuspendSleep(useconds_t delay_us) { |
| 313 | if (delay_us == 0) { |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 314 | sched_yield(); |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 315 | } else { |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 316 | usleep(delay_us); |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 317 | } |
| 318 | } |
| 319 | |
Hiroshi Yamauchi | febd0cf | 2016-09-14 19:31:25 -0700 | [diff] [blame] | 320 | size_t ThreadList::RunCheckpoint(Closure* checkpoint_function, Closure* callback) { |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 321 | Thread* self = Thread::Current(); |
Mathieu Chartier | 6dda898 | 2014-03-06 11:11:48 -0800 | [diff] [blame] | 322 | Locks::mutator_lock_->AssertNotExclusiveHeld(self); |
| 323 | Locks::thread_list_lock_->AssertNotHeld(self); |
| 324 | Locks::thread_suspend_count_lock_->AssertNotHeld(self); |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 325 | |
| 326 | std::vector<Thread*> suspended_count_modified_threads; |
| 327 | size_t count = 0; |
| 328 | { |
| 329 | // Call a checkpoint function for each thread, threads which are suspend get their checkpoint |
| 330 | // manually called. |
| 331 | MutexLock mu(self, *Locks::thread_list_lock_); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 332 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
Mathieu Chartier | 10d2508 | 2015-10-28 18:36:09 -0700 | [diff] [blame] | 333 | count = list_.size(); |
Mathieu Chartier | 02e2511 | 2013-08-14 16:14:24 -0700 | [diff] [blame] | 334 | for (const auto& thread : list_) { |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 335 | if (thread != self) { |
Vladimir Marko | e41eec3 | 2019-10-18 14:41:54 +0100 | [diff] [blame] | 336 | bool requested_suspend = false; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 337 | while (true) { |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 338 | if (thread->RequestCheckpoint(checkpoint_function)) { |
Dave Allison | 0aded08 | 2013-11-07 13:15:11 -0800 | [diff] [blame] | 339 | // This thread will run its checkpoint some time in the near future. |
Vladimir Marko | e41eec3 | 2019-10-18 14:41:54 +0100 | [diff] [blame] | 340 | if (requested_suspend) { |
| 341 | // The suspend request is now unnecessary. |
| 342 | bool updated = |
| 343 | thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal); |
| 344 | DCHECK(updated); |
| 345 | requested_suspend = false; |
| 346 | } |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 347 | break; |
| 348 | } else { |
Vladimir Marko | e41eec3 | 2019-10-18 14:41:54 +0100 | [diff] [blame] | 349 | // The thread is probably suspended, try to make sure that it stays suspended. |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 350 | if (thread->GetState() == kRunnable) { |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 351 | // Spurious fail, try again. |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 352 | continue; |
| 353 | } |
Vladimir Marko | e41eec3 | 2019-10-18 14:41:54 +0100 | [diff] [blame] | 354 | if (!requested_suspend) { |
| 355 | bool updated = |
| 356 | thread->ModifySuspendCount(self, +1, nullptr, SuspendReason::kInternal); |
| 357 | DCHECK(updated); |
| 358 | requested_suspend = true; |
| 359 | if (thread->IsSuspended()) { |
| 360 | break; |
| 361 | } |
| 362 | // The thread raced us to become Runnable. Try to RequestCheckpoint() again. |
| 363 | } else { |
| 364 | // The thread previously raced our suspend request to become Runnable but |
| 365 | // since it is suspended again, it must honor that suspend request now. |
| 366 | DCHECK(thread->IsSuspended()); |
| 367 | break; |
| 368 | } |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 369 | } |
| 370 | } |
Vladimir Marko | e41eec3 | 2019-10-18 14:41:54 +0100 | [diff] [blame] | 371 | if (requested_suspend) { |
| 372 | suspended_count_modified_threads.push_back(thread); |
| 373 | } |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 374 | } |
| 375 | } |
Hiroshi Yamauchi | febd0cf | 2016-09-14 19:31:25 -0700 | [diff] [blame] | 376 | // Run the callback to be called inside this critical section. |
| 377 | if (callback != nullptr) { |
| 378 | callback->Run(self); |
| 379 | } |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 380 | } |
| 381 | |
| 382 | // Run the checkpoint on ourself while we wait for threads to suspend. |
| 383 | checkpoint_function->Run(self); |
| 384 | |
| 385 | // Run the checkpoint on the suspended threads. |
Mathieu Chartier | 02e2511 | 2013-08-14 16:14:24 -0700 | [diff] [blame] | 386 | for (const auto& thread : suspended_count_modified_threads) { |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 387 | // We know for sure that the thread is suspended at this point. |
Vladimir Marko | e41eec3 | 2019-10-18 14:41:54 +0100 | [diff] [blame] | 388 | DCHECK(thread->IsSuspended()); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 389 | checkpoint_function->Run(thread); |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 390 | { |
| 391 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 392 | bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 393 | DCHECK(updated); |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 394 | } |
| 395 | } |
| 396 | |
Mathieu Chartier | 664bebf | 2012-11-12 16:54:11 -0800 | [diff] [blame] | 397 | { |
| 398 | // Imitate ResumeAll, threads may be waiting on Thread::resume_cond_ since we raised their |
| 399 | // suspend count. Now the suspend_count_ is lowered so we must do the broadcast. |
| 400 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 401 | Thread::resume_cond_->Broadcast(self); |
| 402 | } |
| 403 | |
Lei Li | dd9943d | 2015-02-02 14:24:44 +0800 | [diff] [blame] | 404 | return count; |
Mathieu Chartier | 858f1c5 | 2012-10-17 17:45:55 -0700 | [diff] [blame] | 405 | } |
| 406 | |
Hiroshi Yamauchi | a222404 | 2017-02-08 16:35:45 -0800 | [diff] [blame] | 407 | void ThreadList::RunEmptyCheckpoint() { |
Hiroshi Yamauchi | 3049324 | 2016-11-03 13:06:52 -0700 | [diff] [blame] | 408 | Thread* self = Thread::Current(); |
| 409 | Locks::mutator_lock_->AssertNotExclusiveHeld(self); |
| 410 | Locks::thread_list_lock_->AssertNotHeld(self); |
| 411 | Locks::thread_suspend_count_lock_->AssertNotHeld(self); |
Hiroshi Yamauchi | a222404 | 2017-02-08 16:35:45 -0800 | [diff] [blame] | 412 | std::vector<uint32_t> runnable_thread_ids; |
Hiroshi Yamauchi | 3049324 | 2016-11-03 13:06:52 -0700 | [diff] [blame] | 413 | size_t count = 0; |
Hiroshi Yamauchi | a222404 | 2017-02-08 16:35:45 -0800 | [diff] [blame] | 414 | Barrier* barrier = empty_checkpoint_barrier_.get(); |
| 415 | barrier->Init(self, 0); |
Hiroshi Yamauchi | 3049324 | 2016-11-03 13:06:52 -0700 | [diff] [blame] | 416 | { |
| 417 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 418 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 419 | for (Thread* thread : list_) { |
| 420 | if (thread != self) { |
| 421 | while (true) { |
| 422 | if (thread->RequestEmptyCheckpoint()) { |
| 423 | // This thread will run an empty checkpoint (decrement the empty checkpoint barrier) |
| 424 | // some time in the near future. |
| 425 | ++count; |
Hiroshi Yamauchi | a82769c | 2016-12-02 17:01:51 -0800 | [diff] [blame] | 426 | if (kIsDebugBuild) { |
| 427 | runnable_thread_ids.push_back(thread->GetThreadId()); |
| 428 | } |
Hiroshi Yamauchi | 3049324 | 2016-11-03 13:06:52 -0700 | [diff] [blame] | 429 | break; |
| 430 | } |
| 431 | if (thread->GetState() != kRunnable) { |
| 432 | // It's seen suspended, we are done because it must not be in the middle of a mutator |
| 433 | // heap access. |
| 434 | break; |
| 435 | } |
| 436 | } |
| 437 | } |
| 438 | } |
| 439 | } |
| 440 | |
| 441 | // Wake up the threads blocking for weak ref access so that they will respond to the empty |
| 442 | // checkpoint request. Otherwise we will hang as they are blocking in the kRunnable state. |
| 443 | Runtime::Current()->GetHeap()->GetReferenceProcessor()->BroadcastForSlowPath(self); |
Andreas Gampe | 98ea9d9 | 2018-10-19 14:06:15 -0700 | [diff] [blame] | 444 | Runtime::Current()->BroadcastForNewSystemWeaks(/*broadcast_for_checkpoint=*/true); |
Hiroshi Yamauchi | a222404 | 2017-02-08 16:35:45 -0800 | [diff] [blame] | 445 | { |
| 446 | ScopedThreadStateChange tsc(self, kWaitingForCheckPointsToRun); |
| 447 | uint64_t total_wait_time = 0; |
| 448 | bool first_iter = true; |
| 449 | while (true) { |
| 450 | // Wake up the runnable threads blocked on the mutexes that another thread, which is blocked |
| 451 | // on a weak ref access, holds (indirectly blocking for weak ref access through another thread |
| 452 | // and a mutex.) This needs to be done periodically because the thread may be preempted |
| 453 | // between the CheckEmptyCheckpointFromMutex call and the subsequent futex wait in |
| 454 | // Mutex::ExclusiveLock, etc. when the wakeup via WakeupToRespondToEmptyCheckpoint |
| 455 | // arrives. This could cause a *very rare* deadlock, if not repeated. Most of the cases are |
| 456 | // handled in the first iteration. |
| 457 | for (BaseMutex* mutex : Locks::expected_mutexes_on_weak_ref_access_) { |
| 458 | mutex->WakeupToRespondToEmptyCheckpoint(); |
| 459 | } |
| 460 | static constexpr uint64_t kEmptyCheckpointPeriodicTimeoutMs = 100; // 100ms |
| 461 | static constexpr uint64_t kEmptyCheckpointTotalTimeoutMs = 600 * 1000; // 10 minutes. |
| 462 | size_t barrier_count = first_iter ? count : 0; |
| 463 | first_iter = false; // Don't add to the barrier count from the second iteration on. |
| 464 | bool timed_out = barrier->Increment(self, barrier_count, kEmptyCheckpointPeriodicTimeoutMs); |
| 465 | if (!timed_out) { |
| 466 | break; // Success |
| 467 | } |
| 468 | // This is a very rare case. |
| 469 | total_wait_time += kEmptyCheckpointPeriodicTimeoutMs; |
| 470 | if (kIsDebugBuild && total_wait_time > kEmptyCheckpointTotalTimeoutMs) { |
| 471 | std::ostringstream ss; |
| 472 | ss << "Empty checkpoint timeout\n"; |
| 473 | ss << "Barrier count " << barrier->GetCount(self) << "\n"; |
| 474 | ss << "Runnable thread IDs"; |
| 475 | for (uint32_t tid : runnable_thread_ids) { |
| 476 | ss << " " << tid; |
| 477 | } |
| 478 | ss << "\n"; |
| 479 | Locks::mutator_lock_->Dump(ss); |
| 480 | ss << "\n"; |
| 481 | LOG(FATAL_WITHOUT_ABORT) << ss.str(); |
| 482 | // Some threads in 'runnable_thread_ids' are probably stuck. Try to dump their stacks. |
| 483 | // Avoid using ThreadList::Dump() initially because it is likely to get stuck as well. |
| 484 | { |
| 485 | ScopedObjectAccess soa(self); |
| 486 | MutexLock mu1(self, *Locks::thread_list_lock_); |
| 487 | for (Thread* thread : GetList()) { |
| 488 | uint32_t tid = thread->GetThreadId(); |
| 489 | bool is_in_runnable_thread_ids = |
| 490 | std::find(runnable_thread_ids.begin(), runnable_thread_ids.end(), tid) != |
| 491 | runnable_thread_ids.end(); |
| 492 | if (is_in_runnable_thread_ids && |
| 493 | thread->ReadFlag(kEmptyCheckpointRequest)) { |
| 494 | // Found a runnable thread that hasn't responded to the empty checkpoint request. |
| 495 | // Assume it's stuck and safe to dump its stack. |
| 496 | thread->Dump(LOG_STREAM(FATAL_WITHOUT_ABORT), |
Andreas Gampe | 98ea9d9 | 2018-10-19 14:06:15 -0700 | [diff] [blame] | 497 | /*dump_native_stack=*/ true, |
| 498 | /*backtrace_map=*/ nullptr, |
| 499 | /*force_dump_stack=*/ true); |
Hiroshi Yamauchi | a222404 | 2017-02-08 16:35:45 -0800 | [diff] [blame] | 500 | } |
| 501 | } |
| 502 | } |
| 503 | LOG(FATAL_WITHOUT_ABORT) |
| 504 | << "Dumped runnable threads that haven't responded to empty checkpoint."; |
| 505 | // Now use ThreadList::Dump() to dump more threads, noting it may get stuck. |
| 506 | Dump(LOG_STREAM(FATAL_WITHOUT_ABORT)); |
| 507 | LOG(FATAL) << "Dumped all threads."; |
| 508 | } |
| 509 | } |
| 510 | } |
Hiroshi Yamauchi | 3049324 | 2016-11-03 13:06:52 -0700 | [diff] [blame] | 511 | } |
| 512 | |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 513 | // A checkpoint/suspend-all hybrid to switch thread roots from |
| 514 | // from-space to to-space refs. Used to synchronize threads at a point |
| 515 | // to mark the initiation of marking while maintaining the to-space |
| 516 | // invariant. |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 517 | size_t ThreadList::FlipThreadRoots(Closure* thread_flip_visitor, |
| 518 | Closure* flip_callback, |
Andreas Gampe | 6e64445 | 2017-05-09 16:30:27 -0700 | [diff] [blame] | 519 | gc::collector::GarbageCollector* collector, |
| 520 | gc::GcPauseListener* pause_listener) { |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 521 | TimingLogger::ScopedTiming split("ThreadListFlip", collector->GetTimings()); |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 522 | Thread* self = Thread::Current(); |
| 523 | Locks::mutator_lock_->AssertNotHeld(self); |
| 524 | Locks::thread_list_lock_->AssertNotHeld(self); |
| 525 | Locks::thread_suspend_count_lock_->AssertNotHeld(self); |
| 526 | CHECK_NE(self->GetState(), kRunnable); |
| 527 | |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 528 | collector->GetHeap()->ThreadFlipBegin(self); // Sync with JNI critical calls. |
| 529 | |
Mathieu Chartier | e9429c8 | 2017-01-27 15:22:56 -0800 | [diff] [blame] | 530 | // ThreadFlipBegin happens before we suspend all the threads, so it does not count towards the |
| 531 | // pause. |
| 532 | const uint64_t suspend_start_time = NanoTime(); |
Mathieu Chartier | b19ccb1 | 2015-07-15 10:24:16 -0700 | [diff] [blame] | 533 | SuspendAllInternal(self, self, nullptr); |
Andreas Gampe | 6e64445 | 2017-05-09 16:30:27 -0700 | [diff] [blame] | 534 | if (pause_listener != nullptr) { |
| 535 | pause_listener->StartPause(); |
| 536 | } |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 537 | |
| 538 | // Run the flip callback for the collector. |
| 539 | Locks::mutator_lock_->ExclusiveLock(self); |
Mathieu Chartier | e9429c8 | 2017-01-27 15:22:56 -0800 | [diff] [blame] | 540 | suspend_all_historam_.AdjustAndAddValue(NanoTime() - suspend_start_time); |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 541 | flip_callback->Run(self); |
| 542 | Locks::mutator_lock_->ExclusiveUnlock(self); |
Mathieu Chartier | e9429c8 | 2017-01-27 15:22:56 -0800 | [diff] [blame] | 543 | collector->RegisterPause(NanoTime() - suspend_start_time); |
Andreas Gampe | 6e64445 | 2017-05-09 16:30:27 -0700 | [diff] [blame] | 544 | if (pause_listener != nullptr) { |
| 545 | pause_listener->EndPause(); |
| 546 | } |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 547 | |
| 548 | // Resume runnable threads. |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 549 | size_t runnable_thread_count = 0; |
Mathieu Chartier | b19ccb1 | 2015-07-15 10:24:16 -0700 | [diff] [blame] | 550 | std::vector<Thread*> other_threads; |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 551 | { |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 552 | TimingLogger::ScopedTiming split2("ResumeRunnableThreads", collector->GetTimings()); |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 553 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 554 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 555 | --suspend_all_count_; |
| 556 | for (const auto& thread : list_) { |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 557 | // Set the flip function for all threads because Thread::DumpState/DumpJavaStack() (invoked by |
| 558 | // a checkpoint) may cause the flip function to be run for a runnable/suspended thread before |
| 559 | // a runnable thread runs it for itself or we run it for a suspended thread below. |
| 560 | thread->SetFlipFunction(thread_flip_visitor); |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 561 | if (thread == self) { |
| 562 | continue; |
| 563 | } |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 564 | // Resume early the threads that were runnable but are suspended just for this thread flip or |
| 565 | // about to transition from non-runnable (eg. kNative at the SOA entry in a JNI function) to |
| 566 | // runnable (both cases waiting inside Thread::TransitionFromSuspendedToRunnable), or waiting |
| 567 | // for the thread flip to end at the JNI critical section entry (kWaitingForGcThreadFlip), |
| 568 | ThreadState state = thread->GetState(); |
Hiroshi Yamauchi | 15af34c | 2016-09-26 16:56:24 -0700 | [diff] [blame] | 569 | if ((state == kWaitingForGcThreadFlip || thread->IsTransitioningToRunnable()) && |
| 570 | thread->GetSuspendCount() == 1) { |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 571 | // The thread will resume right after the broadcast. |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 572 | bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 573 | DCHECK(updated); |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 574 | ++runnable_thread_count; |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 575 | } else { |
| 576 | other_threads.push_back(thread); |
| 577 | } |
| 578 | } |
| 579 | Thread::resume_cond_->Broadcast(self); |
| 580 | } |
| 581 | |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 582 | collector->GetHeap()->ThreadFlipEnd(self); |
| 583 | |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 584 | // Run the closure on the other threads and let them resume. |
| 585 | { |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 586 | TimingLogger::ScopedTiming split3("FlipOtherThreads", collector->GetTimings()); |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 587 | ReaderMutexLock mu(self, *Locks::mutator_lock_); |
| 588 | for (const auto& thread : other_threads) { |
| 589 | Closure* flip_func = thread->GetFlipFunction(); |
| 590 | if (flip_func != nullptr) { |
| 591 | flip_func->Run(thread); |
| 592 | } |
| 593 | } |
| 594 | // Run it for self. |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 595 | Closure* flip_func = self->GetFlipFunction(); |
| 596 | if (flip_func != nullptr) { |
| 597 | flip_func->Run(self); |
| 598 | } |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 599 | } |
| 600 | |
| 601 | // Resume other threads. |
| 602 | { |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 603 | TimingLogger::ScopedTiming split4("ResumeOtherThreads", collector->GetTimings()); |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 604 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 605 | for (const auto& thread : other_threads) { |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 606 | bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 607 | DCHECK(updated); |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 608 | } |
| 609 | Thread::resume_cond_->Broadcast(self); |
| 610 | } |
| 611 | |
Hiroshi Yamauchi | ee23582 | 2016-08-19 17:03:27 -0700 | [diff] [blame] | 612 | return runnable_thread_count + other_threads.size() + 1; // +1 for self. |
Hiroshi Yamauchi | 2cd334a | 2015-01-09 14:03:35 -0800 | [diff] [blame] | 613 | } |
| 614 | |
Mathieu Chartier | bf44d42 | 2015-06-02 11:42:18 -0700 | [diff] [blame] | 615 | void ThreadList::SuspendAll(const char* cause, bool long_suspend) { |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 616 | Thread* self = Thread::Current(); |
| 617 | |
Jeff Hao | c5d824a | 2014-07-28 18:35:38 -0700 | [diff] [blame] | 618 | if (self != nullptr) { |
Mathieu Chartier | bf9fc58 | 2015-03-13 17:21:25 -0700 | [diff] [blame] | 619 | VLOG(threads) << *self << " SuspendAll for " << cause << " starting..."; |
Jeff Hao | c5d824a | 2014-07-28 18:35:38 -0700 | [diff] [blame] | 620 | } else { |
Mathieu Chartier | bf9fc58 | 2015-03-13 17:21:25 -0700 | [diff] [blame] | 621 | VLOG(threads) << "Thread[null] SuspendAll for " << cause << " starting..."; |
Jeff Hao | c5d824a | 2014-07-28 18:35:38 -0700 | [diff] [blame] | 622 | } |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 623 | { |
| 624 | ScopedTrace trace("Suspending mutator threads"); |
| 625 | const uint64_t start_time = NanoTime(); |
Mathieu Chartier | 6f365cc | 2014-04-23 12:42:27 -0700 | [diff] [blame] | 626 | |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 627 | SuspendAllInternal(self, self); |
| 628 | // All threads are known to have suspended (but a thread may still own the mutator lock) |
| 629 | // Make sure this thread grabs exclusive access to the mutator lock and its protected data. |
Ian Rogers | 66aee5c | 2012-08-15 17:17:47 -0700 | [diff] [blame] | 630 | #if HAVE_TIMED_RWLOCK |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 631 | while (true) { |
Mathieu Chartier | 3fceaf5 | 2017-01-22 13:33:40 -0800 | [diff] [blame] | 632 | if (Locks::mutator_lock_->ExclusiveLockWithTimeout(self, |
| 633 | NsToMs(thread_suspend_timeout_ns_), |
| 634 | 0)) { |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 635 | break; |
| 636 | } else if (!long_suspend_) { |
| 637 | // Reading long_suspend without the mutator lock is slightly racy, in some rare cases, this |
| 638 | // could result in a thread suspend timeout. |
Mathieu Chartier | 3fceaf5 | 2017-01-22 13:33:40 -0800 | [diff] [blame] | 639 | // Timeout if we wait more than thread_suspend_timeout_ns_ nanoseconds. |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 640 | UnsafeLogFatalForThreadSuspendAllTimeout(); |
| 641 | } |
Mathieu Chartier | bf44d42 | 2015-06-02 11:42:18 -0700 | [diff] [blame] | 642 | } |
Ian Rogers | 66aee5c | 2012-08-15 17:17:47 -0700 | [diff] [blame] | 643 | #else |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 644 | Locks::mutator_lock_->ExclusiveLock(self); |
Ian Rogers | 66aee5c | 2012-08-15 17:17:47 -0700 | [diff] [blame] | 645 | #endif |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 646 | |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 647 | long_suspend_ = long_suspend; |
Mathieu Chartier | bf44d42 | 2015-06-02 11:42:18 -0700 | [diff] [blame] | 648 | |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 649 | const uint64_t end_time = NanoTime(); |
| 650 | const uint64_t suspend_time = end_time - start_time; |
| 651 | suspend_all_historam_.AdjustAndAddValue(suspend_time); |
| 652 | if (suspend_time > kLongThreadSuspendThreshold) { |
| 653 | LOG(WARNING) << "Suspending all threads took: " << PrettyDuration(suspend_time); |
| 654 | } |
| 655 | |
| 656 | if (kDebugLocking) { |
| 657 | // Debug check that all threads are suspended. |
| 658 | AssertThreadsAreSuspended(self, self); |
| 659 | } |
Mathieu Chartier | 251755c | 2014-07-15 18:10:25 -0700 | [diff] [blame] | 660 | } |
Orion Hodson | 119733d | 2019-01-30 15:14:41 +0000 | [diff] [blame] | 661 | ATraceBegin((std::string("Mutator threads suspended for ") + cause).c_str()); |
Mathieu Chartier | 6f365cc | 2014-04-23 12:42:27 -0700 | [diff] [blame] | 662 | |
Jeff Hao | c5d824a | 2014-07-28 18:35:38 -0700 | [diff] [blame] | 663 | if (self != nullptr) { |
| 664 | VLOG(threads) << *self << " SuspendAll complete"; |
| 665 | } else { |
| 666 | VLOG(threads) << "Thread[null] SuspendAll complete"; |
| 667 | } |
Elliott Hughes | 8d768a9 | 2011-09-14 16:35:25 -0700 | [diff] [blame] | 668 | } |
| 669 | |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 670 | // Ensures all threads running Java suspend and that those not running Java don't start. |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 671 | void ThreadList::SuspendAllInternal(Thread* self, |
| 672 | Thread* ignore1, |
| 673 | Thread* ignore2, |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 674 | SuspendReason reason) { |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 675 | Locks::mutator_lock_->AssertNotExclusiveHeld(self); |
| 676 | Locks::thread_list_lock_->AssertNotHeld(self); |
| 677 | Locks::thread_suspend_count_lock_->AssertNotHeld(self); |
| 678 | if (kDebugLocking && self != nullptr) { |
| 679 | CHECK_NE(self->GetState(), kRunnable); |
| 680 | } |
| 681 | |
| 682 | // First request that all threads suspend, then wait for them to suspend before |
| 683 | // returning. This suspension scheme also relies on other behaviour: |
| 684 | // 1. Threads cannot be deleted while they are suspended or have a suspend- |
| 685 | // request flag set - (see Unregister() below). |
| 686 | // 2. When threads are created, they are created in a suspended state (actually |
| 687 | // kNative) and will never begin executing Java code without first checking |
| 688 | // the suspend-request flag. |
| 689 | |
| 690 | // The atomic counter for number of threads that need to pass the barrier. |
| 691 | AtomicInteger pending_threads; |
| 692 | uint32_t num_ignored = 0; |
| 693 | if (ignore1 != nullptr) { |
| 694 | ++num_ignored; |
| 695 | } |
| 696 | if (ignore2 != nullptr && ignore1 != ignore2) { |
| 697 | ++num_ignored; |
| 698 | } |
| 699 | { |
| 700 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 701 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 702 | // Update global suspend all state for attaching threads. |
| 703 | ++suspend_all_count_; |
Orion Hodson | 88591fe | 2018-03-06 13:35:43 +0000 | [diff] [blame] | 704 | pending_threads.store(list_.size() - num_ignored, std::memory_order_relaxed); |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 705 | // Increment everybody's suspend count (except those that should be ignored). |
| 706 | for (const auto& thread : list_) { |
| 707 | if (thread == ignore1 || thread == ignore2) { |
| 708 | continue; |
| 709 | } |
| 710 | VLOG(threads) << "requesting thread suspend: " << *thread; |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 711 | bool updated = thread->ModifySuspendCount(self, +1, &pending_threads, reason); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 712 | DCHECK(updated); |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 713 | |
| 714 | // Must install the pending_threads counter first, then check thread->IsSuspend() and clear |
| 715 | // the counter. Otherwise there's a race with Thread::TransitionFromRunnableToSuspended() |
| 716 | // that can lead a thread to miss a call to PassActiveSuspendBarriers(). |
| 717 | if (thread->IsSuspended()) { |
| 718 | // Only clear the counter for the current thread. |
| 719 | thread->ClearSuspendBarrier(&pending_threads); |
Orion Hodson | 88591fe | 2018-03-06 13:35:43 +0000 | [diff] [blame] | 720 | pending_threads.fetch_sub(1, std::memory_order_seq_cst); |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 721 | } |
| 722 | } |
| 723 | } |
| 724 | |
| 725 | // Wait for the barrier to be passed by all runnable threads. This wait |
| 726 | // is done with a timeout so that we can detect problems. |
Mathieu Chartier | 19af117 | 2015-07-14 10:05:45 -0700 | [diff] [blame] | 727 | #if ART_USE_FUTEXES |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 728 | timespec wait_timeout; |
Mathieu Chartier | 3fceaf5 | 2017-01-22 13:33:40 -0800 | [diff] [blame] | 729 | InitTimeSpec(false, CLOCK_MONOTONIC, NsToMs(thread_suspend_timeout_ns_), 0, &wait_timeout); |
Mathieu Chartier | 19af117 | 2015-07-14 10:05:45 -0700 | [diff] [blame] | 730 | #endif |
Mathieu Chartier | 32c8337 | 2017-01-11 10:09:30 -0800 | [diff] [blame] | 731 | const uint64_t start_time = NanoTime(); |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 732 | while (true) { |
Orion Hodson | 88591fe | 2018-03-06 13:35:43 +0000 | [diff] [blame] | 733 | int32_t cur_val = pending_threads.load(std::memory_order_relaxed); |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 734 | if (LIKELY(cur_val > 0)) { |
| 735 | #if ART_USE_FUTEXES |
Charles Munger | 7530bae | 2018-10-29 20:03:51 -0700 | [diff] [blame] | 736 | if (futex(pending_threads.Address(), FUTEX_WAIT_PRIVATE, cur_val, &wait_timeout, nullptr, 0) |
| 737 | != 0) { |
Yuntao.Xiao | a1358e6 | 2018-11-26 14:28:51 +0800 | [diff] [blame] | 738 | if ((errno == EAGAIN) || (errno == EINTR)) { |
| 739 | // EAGAIN and EINTR both indicate a spurious failure, try again from the beginning. |
| 740 | continue; |
| 741 | } |
| 742 | if (errno == ETIMEDOUT) { |
| 743 | const uint64_t wait_time = NanoTime() - start_time; |
| 744 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 745 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 746 | std::ostringstream oss; |
| 747 | for (const auto& thread : list_) { |
| 748 | if (thread == ignore1 || thread == ignore2) { |
| 749 | continue; |
| 750 | } |
| 751 | if (!thread->IsSuspended()) { |
| 752 | oss << std::endl << "Thread not suspended: " << *thread; |
| 753 | } |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 754 | } |
Yuntao.Xiao | a1358e6 | 2018-11-26 14:28:51 +0800 | [diff] [blame] | 755 | LOG(kIsDebugBuild ? ::android::base::FATAL : ::android::base::ERROR) |
| 756 | << "Timed out waiting for threads to suspend, waited for " |
| 757 | << PrettyDuration(wait_time) |
| 758 | << oss.str(); |
| 759 | } else { |
| 760 | PLOG(FATAL) << "futex wait failed for SuspendAllInternal()"; |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 761 | } |
Vladimir Marko | d778cd6 | 2016-07-05 17:29:55 +0100 | [diff] [blame] | 762 | } // else re-check pending_threads in the next iteration (this may be a spurious wake-up). |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 763 | #else |
| 764 | // Spin wait. This is likely to be slow, but on most architecture ART_USE_FUTEXES is set. |
Mathieu Chartier | 32c8337 | 2017-01-11 10:09:30 -0800 | [diff] [blame] | 765 | UNUSED(start_time); |
Yu Li | eac4424 | 2015-06-29 10:50:03 +0800 | [diff] [blame] | 766 | #endif |
| 767 | } else { |
| 768 | CHECK_EQ(cur_val, 0); |
| 769 | break; |
| 770 | } |
| 771 | } |
| 772 | } |
| 773 | |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 774 | void ThreadList::ResumeAll() { |
| 775 | Thread* self = Thread::Current(); |
| 776 | |
Jeff Hao | c5d824a | 2014-07-28 18:35:38 -0700 | [diff] [blame] | 777 | if (self != nullptr) { |
| 778 | VLOG(threads) << *self << " ResumeAll starting"; |
| 779 | } else { |
| 780 | VLOG(threads) << "Thread[null] ResumeAll starting"; |
| 781 | } |
Ian Rogers | 01ae580 | 2012-09-28 16:14:01 -0700 | [diff] [blame] | 782 | |
Orion Hodson | 119733d | 2019-01-30 15:14:41 +0000 | [diff] [blame] | 783 | ATraceEnd(); |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 784 | |
| 785 | ScopedTrace trace("Resuming mutator threads"); |
Mathieu Chartier | 6f365cc | 2014-04-23 12:42:27 -0700 | [diff] [blame] | 786 | |
Mathieu Chartier | 6dda898 | 2014-03-06 11:11:48 -0800 | [diff] [blame] | 787 | if (kDebugLocking) { |
| 788 | // Debug check that all threads are suspended. |
| 789 | AssertThreadsAreSuspended(self, self); |
| 790 | } |
Ian Rogers | 01ae580 | 2012-09-28 16:14:01 -0700 | [diff] [blame] | 791 | |
Mathieu Chartier | bf44d42 | 2015-06-02 11:42:18 -0700 | [diff] [blame] | 792 | long_suspend_ = false; |
| 793 | |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 794 | Locks::mutator_lock_->ExclusiveUnlock(self); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 795 | { |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 796 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 797 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 798 | // Update global suspend all state for attaching threads. |
| 799 | --suspend_all_count_; |
| 800 | // Decrement the suspend counts for all threads. |
Mathieu Chartier | 02e2511 | 2013-08-14 16:14:24 -0700 | [diff] [blame] | 801 | for (const auto& thread : list_) { |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 802 | if (thread == self) { |
| 803 | continue; |
| 804 | } |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 805 | bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 806 | DCHECK(updated); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 807 | } |
| 808 | |
| 809 | // Broadcast a notification to all suspended threads, some or all of |
| 810 | // which may choose to wake up. No need to wait for them. |
Jeff Hao | c5d824a | 2014-07-28 18:35:38 -0700 | [diff] [blame] | 811 | if (self != nullptr) { |
| 812 | VLOG(threads) << *self << " ResumeAll waking others"; |
| 813 | } else { |
| 814 | VLOG(threads) << "Thread[null] ResumeAll waking others"; |
| 815 | } |
Ian Rogers | c604d73 | 2012-10-14 16:09:54 -0700 | [diff] [blame] | 816 | Thread::resume_cond_->Broadcast(self); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 817 | } |
Jeff Hao | c5d824a | 2014-07-28 18:35:38 -0700 | [diff] [blame] | 818 | |
| 819 | if (self != nullptr) { |
| 820 | VLOG(threads) << *self << " ResumeAll complete"; |
| 821 | } else { |
| 822 | VLOG(threads) << "Thread[null] ResumeAll complete"; |
| 823 | } |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 824 | } |
| 825 | |
Alex Light | 88fd720 | 2017-06-30 08:31:59 -0700 | [diff] [blame] | 826 | bool ThreadList::Resume(Thread* thread, SuspendReason reason) { |
Orion Hodson | 119733d | 2019-01-30 15:14:41 +0000 | [diff] [blame] | 827 | // This assumes there was an ATraceBegin when we suspended the thread. |
| 828 | ATraceEnd(); |
Mathieu Chartier | f0dc8b5 | 2014-12-17 10:13:30 -0800 | [diff] [blame] | 829 | |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 830 | Thread* self = Thread::Current(); |
| 831 | DCHECK_NE(thread, self); |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 832 | VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") starting..." << reason; |
Elliott Hughes | 01158d7 | 2011-09-19 19:47:10 -0700 | [diff] [blame] | 833 | |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 834 | { |
| 835 | // To check Contains. |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 836 | MutexLock mu(self, *Locks::thread_list_lock_); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 837 | // To check IsSuspended. |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 838 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
Alex Light | 88fd720 | 2017-06-30 08:31:59 -0700 | [diff] [blame] | 839 | if (UNLIKELY(!thread->IsSuspended())) { |
| 840 | LOG(ERROR) << "Resume(" << reinterpret_cast<void*>(thread) |
| 841 | << ") thread not suspended"; |
| 842 | return false; |
| 843 | } |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 844 | if (!Contains(thread)) { |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 845 | // We only expect threads within the thread-list to have been suspended otherwise we can't |
| 846 | // stop such threads from delete-ing themselves. |
| 847 | LOG(ERROR) << "Resume(" << reinterpret_cast<void*>(thread) |
| 848 | << ") thread not within thread list"; |
Alex Light | 88fd720 | 2017-06-30 08:31:59 -0700 | [diff] [blame] | 849 | return false; |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 850 | } |
Alex Light | 88fd720 | 2017-06-30 08:31:59 -0700 | [diff] [blame] | 851 | if (UNLIKELY(!thread->ModifySuspendCount(self, -1, nullptr, reason))) { |
| 852 | LOG(ERROR) << "Resume(" << reinterpret_cast<void*>(thread) |
| 853 | << ") could not modify suspend count."; |
| 854 | return false; |
| 855 | } |
Elliott Hughes | 01158d7 | 2011-09-19 19:47:10 -0700 | [diff] [blame] | 856 | } |
| 857 | |
| 858 | { |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 859 | VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") waking others"; |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 860 | MutexLock mu(self, *Locks::thread_suspend_count_lock_); |
Ian Rogers | c604d73 | 2012-10-14 16:09:54 -0700 | [diff] [blame] | 861 | Thread::resume_cond_->Broadcast(self); |
Elliott Hughes | 01158d7 | 2011-09-19 19:47:10 -0700 | [diff] [blame] | 862 | } |
| 863 | |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 864 | VLOG(threads) << "Resume(" << reinterpret_cast<void*>(thread) << ") complete"; |
Alex Light | 88fd720 | 2017-06-30 08:31:59 -0700 | [diff] [blame] | 865 | return true; |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 866 | } |
Elliott Hughes | 01158d7 | 2011-09-19 19:47:10 -0700 | [diff] [blame] | 867 | |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 868 | static void ThreadSuspendByPeerWarning(Thread* self, |
| 869 | LogSeverity severity, |
| 870 | const char* message, |
Ian Rogers | c7dd295 | 2014-10-21 23:31:19 -0700 | [diff] [blame] | 871 | jobject peer) { |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 872 | JNIEnvExt* env = self->GetJniEnv(); |
| 873 | ScopedLocalRef<jstring> |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 874 | scoped_name_string(env, static_cast<jstring>(env->GetObjectField( |
| 875 | peer, WellKnownClasses::java_lang_Thread_name))); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 876 | ScopedUtfChars scoped_name_chars(env, scoped_name_string.get()); |
Mathieu Chartier | 2cebb24 | 2015-04-21 16:50:40 -0700 | [diff] [blame] | 877 | if (scoped_name_chars.c_str() == nullptr) { |
Ian Rogers | c7dd295 | 2014-10-21 23:31:19 -0700 | [diff] [blame] | 878 | LOG(severity) << message << ": " << peer; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 879 | env->ExceptionClear(); |
| 880 | } else { |
Ian Rogers | c7dd295 | 2014-10-21 23:31:19 -0700 | [diff] [blame] | 881 | LOG(severity) << message << ": " << peer << ":" << scoped_name_chars.c_str(); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 882 | } |
| 883 | } |
| 884 | |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 885 | Thread* ThreadList::SuspendThreadByPeer(jobject peer, |
| 886 | bool request_suspension, |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 887 | SuspendReason reason, |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 888 | bool* timed_out) { |
Mathieu Chartier | 3a958aa | 2015-02-04 12:52:34 -0800 | [diff] [blame] | 889 | const uint64_t start_time = NanoTime(); |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 890 | useconds_t sleep_us = kThreadSuspendInitialSleepUs; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 891 | *timed_out = false; |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 892 | Thread* const self = Thread::Current(); |
Mathieu Chartier | 82a800d | 2014-12-15 15:59:49 -0800 | [diff] [blame] | 893 | Thread* suspended_thread = nullptr; |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 894 | VLOG(threads) << "SuspendThreadByPeer starting"; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 895 | while (true) { |
| 896 | Thread* thread; |
| 897 | { |
Ian Rogers | f3d874c | 2014-07-17 18:52:42 -0700 | [diff] [blame] | 898 | // Note: this will transition to runnable and potentially suspend. We ensure only one thread |
| 899 | // is requesting another suspend, to avoid deadlock, by requiring this function be called |
| 900 | // holding Locks::thread_list_suspend_thread_lock_. Its important this thread suspend rather |
| 901 | // than request thread suspension, to avoid potential cycles in threads requesting each other |
| 902 | // suspend. |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 903 | ScopedObjectAccess soa(self); |
Andreas Gampe | 277ccbd | 2014-11-03 21:36:10 -0800 | [diff] [blame] | 904 | MutexLock thread_list_mu(self, *Locks::thread_list_lock_); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 905 | thread = Thread::FromManagedThread(soa, peer); |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 906 | if (thread == nullptr) { |
Mathieu Chartier | 82a800d | 2014-12-15 15:59:49 -0800 | [diff] [blame] | 907 | if (suspended_thread != nullptr) { |
| 908 | MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_); |
| 909 | // If we incremented the suspend count but the thread reset its peer, we need to |
| 910 | // re-decrement it since it is shutting down and may deadlock the runtime in |
| 911 | // ThreadList::WaitForOtherNonDaemonThreadsToExit. |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 912 | bool updated = suspended_thread->ModifySuspendCount(soa.Self(), |
| 913 | -1, |
| 914 | nullptr, |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 915 | reason); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 916 | DCHECK(updated); |
Mathieu Chartier | 82a800d | 2014-12-15 15:59:49 -0800 | [diff] [blame] | 917 | } |
Andreas Gampe | 3fec9ac | 2016-09-13 10:47:28 -0700 | [diff] [blame] | 918 | ThreadSuspendByPeerWarning(self, |
| 919 | ::android::base::WARNING, |
| 920 | "No such thread for suspend", |
| 921 | peer); |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 922 | return nullptr; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 923 | } |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 924 | if (!Contains(thread)) { |
Mathieu Chartier | 82a800d | 2014-12-15 15:59:49 -0800 | [diff] [blame] | 925 | CHECK(suspended_thread == nullptr); |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 926 | VLOG(threads) << "SuspendThreadByPeer failed for unattached thread: " |
| 927 | << reinterpret_cast<void*>(thread); |
| 928 | return nullptr; |
| 929 | } |
| 930 | VLOG(threads) << "SuspendThreadByPeer found thread: " << *thread; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 931 | { |
Andreas Gampe | 277ccbd | 2014-11-03 21:36:10 -0800 | [diff] [blame] | 932 | MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 933 | if (request_suspension) { |
Ian Rogers | 4ad5cd3 | 2014-11-11 23:08:07 -0800 | [diff] [blame] | 934 | if (self->GetSuspendCount() > 0) { |
| 935 | // We hold the suspend count lock but another thread is trying to suspend us. Its not |
| 936 | // safe to try to suspend another thread in case we get a cycle. Start the loop again |
| 937 | // which will allow this thread to be suspended. |
| 938 | continue; |
| 939 | } |
Mathieu Chartier | 82a800d | 2014-12-15 15:59:49 -0800 | [diff] [blame] | 940 | CHECK(suspended_thread == nullptr); |
| 941 | suspended_thread = thread; |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 942 | bool updated = suspended_thread->ModifySuspendCount(self, +1, nullptr, reason); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 943 | DCHECK(updated); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 944 | request_suspension = false; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 945 | } else { |
| 946 | // If the caller isn't requesting suspension, a suspension should have already occurred. |
| 947 | CHECK_GT(thread->GetSuspendCount(), 0); |
| 948 | } |
| 949 | // IsSuspended on the current thread will fail as the current thread is changed into |
| 950 | // Runnable above. As the suspend count is now raised if this is the current thread |
| 951 | // it will self suspend on transition to Runnable, making it hard to work with. It's simpler |
| 952 | // to just explicitly handle the current thread in the callers to this code. |
| 953 | CHECK_NE(thread, self) << "Attempt to suspend the current thread for the debugger"; |
| 954 | // If thread is suspended (perhaps it was already not Runnable but didn't have a suspend |
| 955 | // count, or else we've waited and it has self suspended) or is the current thread, we're |
| 956 | // done. |
| 957 | if (thread->IsSuspended()) { |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 958 | VLOG(threads) << "SuspendThreadByPeer thread suspended: " << *thread; |
Orion Hodson | 119733d | 2019-01-30 15:14:41 +0000 | [diff] [blame] | 959 | if (ATraceEnabled()) { |
Mathieu Chartier | f0dc8b5 | 2014-12-17 10:13:30 -0800 | [diff] [blame] | 960 | std::string name; |
| 961 | thread->GetThreadName(name); |
Orion Hodson | 119733d | 2019-01-30 15:14:41 +0000 | [diff] [blame] | 962 | ATraceBegin(StringPrintf("SuspendThreadByPeer suspended %s for peer=%p", name.c_str(), |
Mathieu Chartier | f0dc8b5 | 2014-12-17 10:13:30 -0800 | [diff] [blame] | 963 | peer).c_str()); |
| 964 | } |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 965 | return thread; |
| 966 | } |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 967 | const uint64_t total_delay = NanoTime() - start_time; |
Mathieu Chartier | 3fceaf5 | 2017-01-22 13:33:40 -0800 | [diff] [blame] | 968 | if (total_delay >= thread_suspend_timeout_ns_) { |
Andreas Gampe | 3fec9ac | 2016-09-13 10:47:28 -0700 | [diff] [blame] | 969 | ThreadSuspendByPeerWarning(self, |
| 970 | ::android::base::FATAL, |
Andreas Gampe | d6e54bb | 2016-09-26 14:07:57 -0700 | [diff] [blame] | 971 | "Thread suspension timed out", |
| 972 | peer); |
Mathieu Chartier | 82a800d | 2014-12-15 15:59:49 -0800 | [diff] [blame] | 973 | if (suspended_thread != nullptr) { |
| 974 | CHECK_EQ(suspended_thread, thread); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 975 | bool updated = suspended_thread->ModifySuspendCount(soa.Self(), |
| 976 | -1, |
| 977 | nullptr, |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 978 | reason); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 979 | DCHECK(updated); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 980 | } |
| 981 | *timed_out = true; |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 982 | return nullptr; |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 983 | } else if (sleep_us == 0 && |
| 984 | total_delay > static_cast<uint64_t>(kThreadSuspendMaxYieldUs) * 1000) { |
| 985 | // We have spun for kThreadSuspendMaxYieldUs time, switch to sleeps to prevent |
| 986 | // excessive CPU usage. |
| 987 | sleep_us = kThreadSuspendMaxYieldUs / 2; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 988 | } |
| 989 | } |
| 990 | // Release locks and come out of runnable state. |
| 991 | } |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 992 | VLOG(threads) << "SuspendThreadByPeer waiting to allow thread chance to suspend"; |
| 993 | ThreadSuspendSleep(sleep_us); |
| 994 | // This may stay at 0 if sleep_us == 0, but this is WAI since we want to avoid using usleep at |
| 995 | // all if possible. This shouldn't be an issue since time to suspend should always be small. |
| 996 | sleep_us = std::min(sleep_us * 2, kThreadSuspendMaxSleepUs); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 997 | } |
| 998 | } |
| 999 | |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 1000 | static void ThreadSuspendByThreadIdWarning(LogSeverity severity, |
| 1001 | const char* message, |
Ian Rogers | c7dd295 | 2014-10-21 23:31:19 -0700 | [diff] [blame] | 1002 | uint32_t thread_id) { |
| 1003 | LOG(severity) << StringPrintf("%s: %d", message, thread_id); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1004 | } |
| 1005 | |
Mathieu Chartier | b56200b | 2015-10-29 10:41:51 -0700 | [diff] [blame] | 1006 | Thread* ThreadList::SuspendThreadByThreadId(uint32_t thread_id, |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 1007 | SuspendReason reason, |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1008 | bool* timed_out) { |
Mathieu Chartier | 3a958aa | 2015-02-04 12:52:34 -0800 | [diff] [blame] | 1009 | const uint64_t start_time = NanoTime(); |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 1010 | useconds_t sleep_us = kThreadSuspendInitialSleepUs; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1011 | *timed_out = false; |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 1012 | Thread* suspended_thread = nullptr; |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 1013 | Thread* const self = Thread::Current(); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1014 | CHECK_NE(thread_id, kInvalidThreadId); |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 1015 | VLOG(threads) << "SuspendThreadByThreadId starting"; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1016 | while (true) { |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1017 | { |
Ian Rogers | f3d874c | 2014-07-17 18:52:42 -0700 | [diff] [blame] | 1018 | // Note: this will transition to runnable and potentially suspend. We ensure only one thread |
| 1019 | // is requesting another suspend, to avoid deadlock, by requiring this function be called |
| 1020 | // holding Locks::thread_list_suspend_thread_lock_. Its important this thread suspend rather |
| 1021 | // than request thread suspension, to avoid potential cycles in threads requesting each other |
| 1022 | // suspend. |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1023 | ScopedObjectAccess soa(self); |
Andreas Gampe | 277ccbd | 2014-11-03 21:36:10 -0800 | [diff] [blame] | 1024 | MutexLock thread_list_mu(self, *Locks::thread_list_lock_); |
Ian Rogers | f3d874c | 2014-07-17 18:52:42 -0700 | [diff] [blame] | 1025 | Thread* thread = nullptr; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1026 | for (const auto& it : list_) { |
| 1027 | if (it->GetThreadId() == thread_id) { |
| 1028 | thread = it; |
| 1029 | break; |
| 1030 | } |
| 1031 | } |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 1032 | if (thread == nullptr) { |
| 1033 | CHECK(suspended_thread == nullptr) << "Suspended thread " << suspended_thread |
| 1034 | << " no longer in thread list"; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1035 | // There's a race in inflating a lock and the owner giving up ownership and then dying. |
Andreas Gampe | 3fec9ac | 2016-09-13 10:47:28 -0700 | [diff] [blame] | 1036 | ThreadSuspendByThreadIdWarning(::android::base::WARNING, |
| 1037 | "No such thread id for suspend", |
| 1038 | thread_id); |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 1039 | return nullptr; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1040 | } |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 1041 | VLOG(threads) << "SuspendThreadByThreadId found thread: " << *thread; |
| 1042 | DCHECK(Contains(thread)); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1043 | { |
Andreas Gampe | 277ccbd | 2014-11-03 21:36:10 -0800 | [diff] [blame] | 1044 | MutexLock suspend_count_mu(self, *Locks::thread_suspend_count_lock_); |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 1045 | if (suspended_thread == nullptr) { |
Ian Rogers | 4ad5cd3 | 2014-11-11 23:08:07 -0800 | [diff] [blame] | 1046 | if (self->GetSuspendCount() > 0) { |
| 1047 | // We hold the suspend count lock but another thread is trying to suspend us. Its not |
| 1048 | // safe to try to suspend another thread in case we get a cycle. Start the loop again |
| 1049 | // which will allow this thread to be suspended. |
| 1050 | continue; |
| 1051 | } |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 1052 | bool updated = thread->ModifySuspendCount(self, +1, nullptr, reason); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 1053 | DCHECK(updated); |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 1054 | suspended_thread = thread; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1055 | } else { |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 1056 | CHECK_EQ(suspended_thread, thread); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1057 | // If the caller isn't requesting suspension, a suspension should have already occurred. |
| 1058 | CHECK_GT(thread->GetSuspendCount(), 0); |
| 1059 | } |
| 1060 | // IsSuspended on the current thread will fail as the current thread is changed into |
| 1061 | // Runnable above. As the suspend count is now raised if this is the current thread |
| 1062 | // it will self suspend on transition to Runnable, making it hard to work with. It's simpler |
| 1063 | // to just explicitly handle the current thread in the callers to this code. |
| 1064 | CHECK_NE(thread, self) << "Attempt to suspend the current thread for the debugger"; |
| 1065 | // If thread is suspended (perhaps it was already not Runnable but didn't have a suspend |
| 1066 | // count, or else we've waited and it has self suspended) or is the current thread, we're |
| 1067 | // done. |
| 1068 | if (thread->IsSuspended()) { |
Orion Hodson | 119733d | 2019-01-30 15:14:41 +0000 | [diff] [blame] | 1069 | if (ATraceEnabled()) { |
Mathieu Chartier | f0dc8b5 | 2014-12-17 10:13:30 -0800 | [diff] [blame] | 1070 | std::string name; |
| 1071 | thread->GetThreadName(name); |
Orion Hodson | 119733d | 2019-01-30 15:14:41 +0000 | [diff] [blame] | 1072 | ATraceBegin(StringPrintf("SuspendThreadByThreadId suspended %s id=%d", |
Mathieu Chartier | f0dc8b5 | 2014-12-17 10:13:30 -0800 | [diff] [blame] | 1073 | name.c_str(), thread_id).c_str()); |
| 1074 | } |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 1075 | VLOG(threads) << "SuspendThreadByThreadId thread suspended: " << *thread; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1076 | return thread; |
| 1077 | } |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 1078 | const uint64_t total_delay = NanoTime() - start_time; |
Mathieu Chartier | 3fceaf5 | 2017-01-22 13:33:40 -0800 | [diff] [blame] | 1079 | if (total_delay >= thread_suspend_timeout_ns_) { |
Andreas Gampe | 3fec9ac | 2016-09-13 10:47:28 -0700 | [diff] [blame] | 1080 | ThreadSuspendByThreadIdWarning(::android::base::WARNING, |
| 1081 | "Thread suspension timed out", |
| 1082 | thread_id); |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 1083 | if (suspended_thread != nullptr) { |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 1084 | bool updated = thread->ModifySuspendCount(soa.Self(), -1, nullptr, reason); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 1085 | DCHECK(updated); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1086 | } |
| 1087 | *timed_out = true; |
Brian Carlstrom | ba32de4 | 2014-08-27 23:43:46 -0700 | [diff] [blame] | 1088 | return nullptr; |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 1089 | } else if (sleep_us == 0 && |
| 1090 | total_delay > static_cast<uint64_t>(kThreadSuspendMaxYieldUs) * 1000) { |
| 1091 | // We have spun for kThreadSuspendMaxYieldUs time, switch to sleeps to prevent |
| 1092 | // excessive CPU usage. |
| 1093 | sleep_us = kThreadSuspendMaxYieldUs / 2; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1094 | } |
| 1095 | } |
| 1096 | // Release locks and come out of runnable state. |
| 1097 | } |
Mathieu Chartier | 9914386 | 2015-02-03 14:26:46 -0800 | [diff] [blame] | 1098 | VLOG(threads) << "SuspendThreadByThreadId waiting to allow thread chance to suspend"; |
| 1099 | ThreadSuspendSleep(sleep_us); |
| 1100 | sleep_us = std::min(sleep_us * 2, kThreadSuspendMaxSleepUs); |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1101 | } |
| 1102 | } |
| 1103 | |
Mathieu Chartier | 61b3cd4 | 2016-04-18 11:43:29 -0700 | [diff] [blame] | 1104 | Thread* ThreadList::FindThreadByThreadId(uint32_t thread_id) { |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1105 | for (const auto& thread : list_) { |
Mathieu Chartier | 61b3cd4 | 2016-04-18 11:43:29 -0700 | [diff] [blame] | 1106 | if (thread->GetThreadId() == thread_id) { |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1107 | return thread; |
| 1108 | } |
| 1109 | } |
Mathieu Chartier | 2cebb24 | 2015-04-21 16:50:40 -0700 | [diff] [blame] | 1110 | return nullptr; |
Ian Rogers | d9c4fc9 | 2013-10-01 19:45:43 -0700 | [diff] [blame] | 1111 | } |
| 1112 | |
Hans Boehm | dc77ca3 | 2020-01-31 16:29:35 -0800 | [diff] [blame] | 1113 | void ThreadList::WaitForOtherNonDaemonThreadsToExit(bool check_no_birth) { |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 1114 | ScopedTrace trace(__PRETTY_FUNCTION__); |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 1115 | Thread* self = Thread::Current(); |
| 1116 | Locks::mutator_lock_->AssertNotHeld(self); |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1117 | while (true) { |
Hans Boehm | dc77ca3 | 2020-01-31 16:29:35 -0800 | [diff] [blame] | 1118 | Locks::runtime_shutdown_lock_->Lock(self); |
| 1119 | if (check_no_birth) { |
Ian Rogers | 120f1c7 | 2012-09-28 17:17:10 -0700 | [diff] [blame] | 1120 | // No more threads can be born after we start to shutdown. |
Mathieu Chartier | 590fee9 | 2013-09-13 13:46:47 -0700 | [diff] [blame] | 1121 | CHECK(Runtime::Current()->IsShuttingDownLocked()); |
Ian Rogers | 120f1c7 | 2012-09-28 17:17:10 -0700 | [diff] [blame] | 1122 | CHECK_EQ(Runtime::Current()->NumberOfThreadsBeingBorn(), 0U); |
Hans Boehm | dc77ca3 | 2020-01-31 16:29:35 -0800 | [diff] [blame] | 1123 | } else { |
| 1124 | if (Runtime::Current()->NumberOfThreadsBeingBorn() != 0U) { |
| 1125 | // Awkward. Shutdown_cond_ is private, but the only live thread may not be registered yet. |
| 1126 | // Fortunately, this is used mostly for testing, and not performance-critical. |
| 1127 | Locks::runtime_shutdown_lock_->Unlock(self); |
| 1128 | usleep(1000); |
| 1129 | continue; |
| 1130 | } |
Ian Rogers | 120f1c7 | 2012-09-28 17:17:10 -0700 | [diff] [blame] | 1131 | } |
Ian Rogers | 120f1c7 | 2012-09-28 17:17:10 -0700 | [diff] [blame] | 1132 | MutexLock mu(self, *Locks::thread_list_lock_); |
Hans Boehm | dc77ca3 | 2020-01-31 16:29:35 -0800 | [diff] [blame] | 1133 | Locks::runtime_shutdown_lock_->Unlock(self); |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1134 | // Also wait for any threads that are unregistering to finish. This is required so that no |
| 1135 | // threads access the thread list after it is deleted. TODO: This may not work for user daemon |
| 1136 | // threads since they could unregister at the wrong time. |
| 1137 | bool done = unregistering_count_ == 0; |
| 1138 | if (done) { |
| 1139 | for (const auto& thread : list_) { |
| 1140 | if (thread != self && !thread->IsDaemon()) { |
| 1141 | done = false; |
| 1142 | break; |
| 1143 | } |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1144 | } |
| 1145 | } |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1146 | if (done) { |
| 1147 | break; |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1148 | } |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1149 | // Wait for another thread to exit before re-checking. |
| 1150 | Locks::thread_exit_cond_->Wait(self); |
| 1151 | } |
Elliott Hughes | 038a806 | 2011-09-18 14:12:41 -0700 | [diff] [blame] | 1152 | } |
| 1153 | |
Mathieu Chartier | 4d87df6 | 2016-01-07 15:14:19 -0800 | [diff] [blame] | 1154 | void ThreadList::SuspendAllDaemonThreadsForShutdown() { |
Mathieu Chartier | 32ce2ad | 2016-03-04 14:58:03 -0800 | [diff] [blame] | 1155 | ScopedTrace trace(__PRETTY_FUNCTION__); |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 1156 | Thread* self = Thread::Current(); |
Mathieu Chartier | 62597d1 | 2016-01-11 10:19:06 -0800 | [diff] [blame] | 1157 | size_t daemons_left = 0; |
Nicolas Geoffray | aa45daa | 2016-06-20 15:58:32 +0100 | [diff] [blame] | 1158 | { |
| 1159 | // Tell all the daemons it's time to suspend. |
| 1160 | MutexLock mu(self, *Locks::thread_list_lock_); |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 1161 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
Mathieu Chartier | 02e2511 | 2013-08-14 16:14:24 -0700 | [diff] [blame] | 1162 | for (const auto& thread : list_) { |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1163 | // This is only run after all non-daemon threads have exited, so the remainder should all be |
| 1164 | // daemons. |
Ian Rogers | 7e76286 | 2012-10-22 15:45:08 -0700 | [diff] [blame] | 1165 | CHECK(thread->IsDaemon()) << *thread; |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 1166 | if (thread != self) { |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 1167 | bool updated = thread->ModifySuspendCount(self, +1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 1168 | DCHECK(updated); |
Mathieu Chartier | 62597d1 | 2016-01-11 10:19:06 -0800 | [diff] [blame] | 1169 | ++daemons_left; |
Elliott Hughes | e52e49b | 2012-04-02 16:05:44 -0700 | [diff] [blame] | 1170 | } |
Mathieu Chartier | 4d87df6 | 2016-01-07 15:14:19 -0800 | [diff] [blame] | 1171 | // We are shutting down the runtime, set the JNI functions of all the JNIEnvs to be |
| 1172 | // the sleep forever one. |
| 1173 | thread->GetJniEnv()->SetFunctionsToRuntimeShutdownFunctions(); |
Elliott Hughes | 038a806 | 2011-09-18 14:12:41 -0700 | [diff] [blame] | 1174 | } |
| 1175 | } |
Hans Boehm | 65c18a2 | 2020-01-03 23:37:13 +0000 | [diff] [blame] | 1176 | if (daemons_left == 0) { |
| 1177 | // No threads left; safe to shut down. |
| 1178 | return; |
Mathieu Chartier | 62597d1 | 2016-01-11 10:19:06 -0800 | [diff] [blame] | 1179 | } |
Hans Boehm | 65c18a2 | 2020-01-03 23:37:13 +0000 | [diff] [blame] | 1180 | // There is not a clean way to shut down if we have daemons left. We have no mechanism for |
| 1181 | // killing them and reclaiming thread stacks. We also have no mechanism for waiting until they |
| 1182 | // have truly finished touching the memory we are about to deallocate. We do the best we can with |
| 1183 | // timeouts. |
| 1184 | // |
| 1185 | // If we have any daemons left, wait until they are (a) suspended and (b) they are not stuck |
| 1186 | // in a place where they are about to access runtime state and are not in a runnable state. |
| 1187 | // We attempt to do the latter by just waiting long enough for things to |
| 1188 | // quiesce. Examples: Monitor code or waking up from a condition variable. |
| 1189 | // |
| 1190 | // Give the threads a chance to suspend, complaining if they're slow. (a) |
Elliott Hughes | 038a806 | 2011-09-18 14:12:41 -0700 | [diff] [blame] | 1191 | bool have_complained = false; |
Mathieu Chartier | ba098ba | 2016-01-07 09:31:33 -0800 | [diff] [blame] | 1192 | static constexpr size_t kTimeoutMicroseconds = 2000 * 1000; |
| 1193 | static constexpr size_t kSleepMicroseconds = 1000; |
Hans Boehm | 28aaf24 | 2020-04-17 17:46:48 -0700 | [diff] [blame] | 1194 | bool all_suspended = false; |
| 1195 | for (size_t i = 0; !all_suspended && i < kTimeoutMicroseconds / kSleepMicroseconds; ++i) { |
| 1196 | bool found_running = false; |
Nicolas Geoffray | aa45daa | 2016-06-20 15:58:32 +0100 | [diff] [blame] | 1197 | { |
| 1198 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 1199 | for (const auto& thread : list_) { |
| 1200 | if (thread != self && thread->GetState() == kRunnable) { |
| 1201 | if (!have_complained) { |
| 1202 | LOG(WARNING) << "daemon thread not yet suspended: " << *thread; |
| 1203 | have_complained = true; |
| 1204 | } |
Hans Boehm | 28aaf24 | 2020-04-17 17:46:48 -0700 | [diff] [blame] | 1205 | found_running = true; |
Elliott Hughes | 038a806 | 2011-09-18 14:12:41 -0700 | [diff] [blame] | 1206 | } |
Elliott Hughes | 038a806 | 2011-09-18 14:12:41 -0700 | [diff] [blame] | 1207 | } |
| 1208 | } |
Hans Boehm | 28aaf24 | 2020-04-17 17:46:48 -0700 | [diff] [blame] | 1209 | if (found_running) { |
| 1210 | // Sleep briefly before checking again. Max total sleep time is kTimeoutMicroseconds. |
| 1211 | usleep(kSleepMicroseconds); |
| 1212 | } else { |
| 1213 | all_suspended = true; |
| 1214 | } |
| 1215 | } |
| 1216 | if (!all_suspended) { |
| 1217 | // We can get here if a daemon thread executed a fastnative native call, so that it |
| 1218 | // remained in runnable state, and then made a JNI call after we called |
| 1219 | // SetFunctionsToRuntimeShutdownFunctions(), causing it to permanently stay in a harmless |
| 1220 | // but runnable state. See b/147804269 . |
| 1221 | LOG(WARNING) << "timed out suspending all daemon threads"; |
| 1222 | } |
| 1223 | // Assume all threads are either suspended or somehow wedged. |
| 1224 | // Wait again for all the now "suspended" threads to actually quiesce. (b) |
Hans Boehm | 6127ace | 2020-04-21 20:38:14 -0700 | [diff] [blame] | 1225 | static constexpr size_t kDaemonSleepTime = 400'000; |
Hans Boehm | 28aaf24 | 2020-04-17 17:46:48 -0700 | [diff] [blame] | 1226 | usleep(kDaemonSleepTime); |
| 1227 | std::list<Thread*> list_copy; |
| 1228 | { |
| 1229 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 1230 | // Half-way through the wait, set the "runtime deleted" flag, causing any newly awoken |
| 1231 | // threads to immediately go back to sleep without touching memory. This prevents us from |
| 1232 | // touching deallocated memory, but it also prevents mutexes from getting released. Thus we |
| 1233 | // only do this once we're reasonably sure that no system mutexes are still held. |
| 1234 | for (const auto& thread : list_) { |
| 1235 | DCHECK(thread == self || !all_suspended || thread->GetState() != kRunnable); |
| 1236 | // In the !all_suspended case, the target is probably sleeping. |
| 1237 | thread->GetJniEnv()->SetRuntimeDeleted(); |
| 1238 | // Possibly contended Mutex acquisitions are unsafe after this. |
| 1239 | // Releasing thread_list_lock_ is OK, since it can't block. |
| 1240 | } |
| 1241 | } |
| 1242 | // Finally wait for any threads woken before we set the "runtime deleted" flags to finish |
| 1243 | // touching memory. |
| 1244 | usleep(kDaemonSleepTime); |
Hans Boehm | ff5ce16 | 2020-01-17 18:06:41 -0800 | [diff] [blame] | 1245 | #if defined(__has_feature) |
| 1246 | #if __has_feature(address_sanitizer) || __has_feature(hwaddress_sanitizer) |
Hans Boehm | 28aaf24 | 2020-04-17 17:46:48 -0700 | [diff] [blame] | 1247 | // Sleep a bit longer with -fsanitize=address, since everything is slower. |
| 1248 | usleep(2 * kDaemonSleepTime); |
Hans Boehm | ff5ce16 | 2020-01-17 18:06:41 -0800 | [diff] [blame] | 1249 | #endif |
| 1250 | #endif |
Hans Boehm | 28aaf24 | 2020-04-17 17:46:48 -0700 | [diff] [blame] | 1251 | // At this point no threads should be touching our data structures anymore. |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1252 | } |
Mathieu Chartier | fbc3108 | 2016-01-24 11:59:56 -0800 | [diff] [blame] | 1253 | |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1254 | void ThreadList::Register(Thread* self) { |
| 1255 | DCHECK_EQ(self, Thread::Current()); |
Andreas Gampe | c4bed16 | 2017-05-01 13:46:24 -0700 | [diff] [blame] | 1256 | CHECK(!shut_down_); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1257 | |
| 1258 | if (VLOG_IS_ON(threads)) { |
| 1259 | std::ostringstream oss; |
| 1260 | self->ShortDump(oss); // We don't hold the mutator_lock_ yet and so cannot call Dump. |
Ian Rogers | 5a9ba01 | 2014-05-19 13:28:52 -0700 | [diff] [blame] | 1261 | LOG(INFO) << "ThreadList::Register() " << *self << "\n" << oss.str(); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1262 | } |
| 1263 | |
| 1264 | // Atomically add self to the thread list and make its thread_suspend_count_ reflect ongoing |
| 1265 | // SuspendAll requests. |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 1266 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 1267 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
Ian Rogers | 2966e13 | 2014-04-02 08:34:36 -0700 | [diff] [blame] | 1268 | // Modify suspend count in increments of 1 to maintain invariants in ModifySuspendCount. While |
| 1269 | // this isn't particularly efficient the suspend counts are most commonly 0 or 1. |
Alex Light | fc58809 | 2020-01-23 15:39:08 -0800 | [diff] [blame] | 1270 | for (int delta = suspend_all_count_; delta > 0; delta--) { |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 1271 | bool updated = self->ModifySuspendCount(self, +1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 1272 | DCHECK(updated); |
Ian Rogers | 01ae580 | 2012-09-28 16:14:01 -0700 | [diff] [blame] | 1273 | } |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1274 | CHECK(!Contains(self)); |
| 1275 | list_.push_back(self); |
Hiroshi Yamauchi | 0b71357 | 2015-06-16 18:29:23 -0700 | [diff] [blame] | 1276 | if (kUseReadBarrier) { |
Mathieu Chartier | 3768ade | 2017-05-02 14:04:39 -0700 | [diff] [blame] | 1277 | gc::collector::ConcurrentCopying* const cc = |
| 1278 | Runtime::Current()->GetHeap()->ConcurrentCopyingCollector(); |
Hiroshi Yamauchi | 0037082 | 2015-08-18 14:47:25 -0700 | [diff] [blame] | 1279 | // Initialize according to the state of the CC collector. |
Mathieu Chartier | 3768ade | 2017-05-02 14:04:39 -0700 | [diff] [blame] | 1280 | self->SetIsGcMarkingAndUpdateEntrypoints(cc->IsMarking()); |
| 1281 | if (cc->IsUsingReadBarrierEntrypoints()) { |
| 1282 | self->SetReadBarrierEntrypoints(); |
| 1283 | } |
| 1284 | self->SetWeakRefAccessEnabled(cc->IsWeakRefAccessEnabled()); |
Hiroshi Yamauchi | 0b71357 | 2015-06-16 18:29:23 -0700 | [diff] [blame] | 1285 | } |
David Srbecky | 28f6cff | 2018-10-16 15:07:28 +0100 | [diff] [blame] | 1286 | self->NotifyInTheadList(); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1287 | } |
| 1288 | |
| 1289 | void ThreadList::Unregister(Thread* self) { |
| 1290 | DCHECK_EQ(self, Thread::Current()); |
Ian Rogers | 68d8b42 | 2014-07-17 11:09:10 -0700 | [diff] [blame] | 1291 | CHECK_NE(self->GetState(), kRunnable); |
| 1292 | Locks::mutator_lock_->AssertNotHeld(self); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1293 | |
| 1294 | VLOG(threads) << "ThreadList::Unregister() " << *self; |
| 1295 | |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1296 | { |
| 1297 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 1298 | ++unregistering_count_; |
| 1299 | } |
| 1300 | |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1301 | // Any time-consuming destruction, plus anything that can call back into managed code or |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1302 | // suspend and so on, must happen at this point, and not in ~Thread. The self->Destroy is what |
| 1303 | // causes the threads to join. It is important to do this after incrementing unregistering_count_ |
| 1304 | // since we want the runtime to wait for the daemon threads to exit before deleting the thread |
| 1305 | // list. |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1306 | self->Destroy(); |
| 1307 | |
Jeff Hao | e094b87 | 2014-10-14 13:12:01 -0700 | [diff] [blame] | 1308 | // If tracing, remember thread id and name before thread exits. |
| 1309 | Trace::StoreExitingThreadInfo(self); |
| 1310 | |
Ian Rogers | dd7624d | 2014-03-14 17:43:00 -0700 | [diff] [blame] | 1311 | uint32_t thin_lock_id = self->GetThreadId(); |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1312 | while (true) { |
Ian Rogers | cfaa455 | 2012-11-26 21:00:08 -0800 | [diff] [blame] | 1313 | // Remove and delete the Thread* while holding the thread_list_lock_ and |
| 1314 | // thread_suspend_count_lock_ so that the unregistering thread cannot be suspended. |
Ian Rogers | 0878d65 | 2013-04-18 17:38:35 -0700 | [diff] [blame] | 1315 | // Note: deliberately not using MutexLock that could hold a stale self pointer. |
Juju Sung | 3fef44a | 2019-01-07 18:18:16 +0800 | [diff] [blame] | 1316 | { |
| 1317 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 1318 | if (!Contains(self)) { |
| 1319 | std::string thread_name; |
| 1320 | self->GetThreadName(thread_name); |
| 1321 | std::ostringstream os; |
| 1322 | DumpNativeStack(os, GetTid(), nullptr, " native: ", nullptr); |
| 1323 | LOG(ERROR) << "Request to unregister unattached thread " << thread_name << "\n" << os.str(); |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1324 | break; |
Juju Sung | 3fef44a | 2019-01-07 18:18:16 +0800 | [diff] [blame] | 1325 | } else { |
| 1326 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 1327 | if (!self->IsSuspended()) { |
| 1328 | list_.remove(self); |
| 1329 | break; |
| 1330 | } |
Ian Rogers | a2af5c7 | 2014-09-15 15:17:07 -0700 | [diff] [blame] | 1331 | } |
Ian Rogers | 68d8b42 | 2014-07-17 11:09:10 -0700 | [diff] [blame] | 1332 | } |
Juju Sung | 3fef44a | 2019-01-07 18:18:16 +0800 | [diff] [blame] | 1333 | // In the case where we are not suspended yet, sleep to leave other threads time to execute. |
| 1334 | // This is important if there are realtime threads. b/111277984 |
| 1335 | usleep(1); |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1336 | // We failed to remove the thread due to a suspend request, loop and try again. |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1337 | } |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1338 | delete self; |
| 1339 | |
Mathieu Chartier | 5f51d4b | 2013-12-03 14:24:05 -0800 | [diff] [blame] | 1340 | // Release the thread ID after the thread is finished and deleted to avoid cases where we can |
| 1341 | // temporarily have multiple threads with the same thread id. When this occurs, it causes |
| 1342 | // problems in FindThreadByThreadId / SuspendThreadByThreadId. |
| 1343 | ReleaseThreadId(nullptr, thin_lock_id); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1344 | |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1345 | // Clear the TLS data, so that the underlying native thread is recognizably detached. |
| 1346 | // (It may wish to reattach later.) |
Andreas Gampe | a47a6e8 | 2019-07-24 09:46:16 -0700 | [diff] [blame] | 1347 | #ifdef __BIONIC__ |
Andreas Gampe | 4382f1e | 2015-08-05 01:08:53 +0000 | [diff] [blame] | 1348 | __get_tls()[TLS_SLOT_ART_THREAD_SELF] = nullptr; |
| 1349 | #else |
Mathieu Chartier | 2cebb24 | 2015-04-21 16:50:40 -0700 | [diff] [blame] | 1350 | CHECK_PTHREAD_CALL(pthread_setspecific, (Thread::pthread_key_self_, nullptr), "detach self"); |
Andreas Gampe | 8237200 | 2019-07-24 15:42:09 -0700 | [diff] [blame] | 1351 | Thread::self_tls_ = nullptr; |
Andreas Gampe | 4382f1e | 2015-08-05 01:08:53 +0000 | [diff] [blame] | 1352 | #endif |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1353 | |
| 1354 | // Signal that a thread just detached. |
Mathieu Chartier | 91e5669 | 2015-03-03 13:51:04 -0800 | [diff] [blame] | 1355 | MutexLock mu(nullptr, *Locks::thread_list_lock_); |
| 1356 | --unregistering_count_; |
| 1357 | Locks::thread_exit_cond_->Broadcast(nullptr); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1358 | } |
| 1359 | |
| 1360 | void ThreadList::ForEach(void (*callback)(Thread*, void*), void* context) { |
Mathieu Chartier | 02e2511 | 2013-08-14 16:14:24 -0700 | [diff] [blame] | 1361 | for (const auto& thread : list_) { |
| 1362 | callback(thread, context); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1363 | } |
| 1364 | } |
| 1365 | |
Mathieu Chartier | f8a86b9 | 2016-06-14 17:08:47 -0700 | [diff] [blame] | 1366 | void ThreadList::VisitRootsForSuspendedThreads(RootVisitor* visitor) { |
| 1367 | Thread* const self = Thread::Current(); |
| 1368 | std::vector<Thread*> threads_to_visit; |
| 1369 | |
| 1370 | // Tell threads to suspend and copy them into list. |
| 1371 | { |
| 1372 | MutexLock mu(self, *Locks::thread_list_lock_); |
| 1373 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 1374 | for (Thread* thread : list_) { |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 1375 | bool suspended = thread->ModifySuspendCount(self, +1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 1376 | DCHECK(suspended); |
Mathieu Chartier | f8a86b9 | 2016-06-14 17:08:47 -0700 | [diff] [blame] | 1377 | if (thread == self || thread->IsSuspended()) { |
| 1378 | threads_to_visit.push_back(thread); |
| 1379 | } else { |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 1380 | bool resumed = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 1381 | DCHECK(resumed); |
Mathieu Chartier | f8a86b9 | 2016-06-14 17:08:47 -0700 | [diff] [blame] | 1382 | } |
| 1383 | } |
| 1384 | } |
| 1385 | |
| 1386 | // Visit roots without holding thread_list_lock_ and thread_suspend_count_lock_ to prevent lock |
| 1387 | // order violations. |
| 1388 | for (Thread* thread : threads_to_visit) { |
Andreas Gampe | 513061a | 2017-06-01 09:17:34 -0700 | [diff] [blame] | 1389 | thread->VisitRoots(visitor, kVisitRootFlagAllRoots); |
Mathieu Chartier | f8a86b9 | 2016-06-14 17:08:47 -0700 | [diff] [blame] | 1390 | } |
| 1391 | |
| 1392 | // Restore suspend counts. |
| 1393 | { |
| 1394 | MutexLock mu2(self, *Locks::thread_suspend_count_lock_); |
| 1395 | for (Thread* thread : threads_to_visit) { |
Alex Light | 46f9340 | 2017-06-29 11:59:50 -0700 | [diff] [blame] | 1396 | bool updated = thread->ModifySuspendCount(self, -1, nullptr, SuspendReason::kInternal); |
Sebastien Hertz | 1c8f4ff | 2017-04-14 15:05:12 +0200 | [diff] [blame] | 1397 | DCHECK(updated); |
Mathieu Chartier | f8a86b9 | 2016-06-14 17:08:47 -0700 | [diff] [blame] | 1398 | } |
| 1399 | } |
| 1400 | } |
| 1401 | |
Andreas Gampe | 585da95 | 2016-12-02 14:52:29 -0800 | [diff] [blame] | 1402 | void ThreadList::VisitRoots(RootVisitor* visitor, VisitRootFlags flags) const { |
Ian Rogers | 81d425b | 2012-09-27 16:03:43 -0700 | [diff] [blame] | 1403 | MutexLock mu(Thread::Current(), *Locks::thread_list_lock_); |
Mathieu Chartier | 02e2511 | 2013-08-14 16:14:24 -0700 | [diff] [blame] | 1404 | for (const auto& thread : list_) { |
Andreas Gampe | 585da95 | 2016-12-02 14:52:29 -0800 | [diff] [blame] | 1405 | thread->VisitRoots(visitor, flags); |
Ian Rogers | 00f7d0e | 2012-07-19 15:28:27 -0700 | [diff] [blame] | 1406 | } |
Elliott Hughes | 038a806 | 2011-09-18 14:12:41 -0700 | [diff] [blame] | 1407 | } |
| 1408 | |
Nicolas Geoffray | e3f775b | 2019-12-04 14:41:52 +0000 | [diff] [blame] | 1409 | void ThreadList::SweepInterpreterCaches(IsMarkedVisitor* visitor) const { |
| 1410 | MutexLock mu(Thread::Current(), *Locks::thread_list_lock_); |
| 1411 | for (const auto& thread : list_) { |
| 1412 | thread->SweepInterpreterCache(visitor); |
| 1413 | } |
| 1414 | } |
| 1415 | |
Alex Light | 55eccdf | 2019-10-07 13:51:13 +0000 | [diff] [blame] | 1416 | void ThreadList::VisitReflectiveTargets(ReflectiveValueVisitor *visitor) const { |
| 1417 | MutexLock mu(Thread::Current(), *Locks::thread_list_lock_); |
| 1418 | for (const auto& thread : list_) { |
| 1419 | thread->VisitReflectiveTargets(visitor); |
| 1420 | } |
| 1421 | } |
| 1422 | |
Ian Rogers | cfaa455 | 2012-11-26 21:00:08 -0800 | [diff] [blame] | 1423 | uint32_t ThreadList::AllocThreadId(Thread* self) { |
Chao-ying Fu | 9e36931 | 2014-05-21 11:20:52 -0700 | [diff] [blame] | 1424 | MutexLock mu(self, *Locks::allocated_thread_ids_lock_); |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 1425 | for (size_t i = 0; i < allocated_ids_.size(); ++i) { |
| 1426 | if (!allocated_ids_[i]) { |
| 1427 | allocated_ids_.set(i); |
Brian Carlstrom | 7934ac2 | 2013-07-26 10:54:15 -0700 | [diff] [blame] | 1428 | return i + 1; // Zero is reserved to mean "invalid". |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 1429 | } |
| 1430 | } |
| 1431 | LOG(FATAL) << "Out of internal thread ids"; |
Elliott Hughes | c1896c9 | 2018-11-29 11:33:18 -0800 | [diff] [blame] | 1432 | UNREACHABLE(); |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 1433 | } |
| 1434 | |
Ian Rogers | cfaa455 | 2012-11-26 21:00:08 -0800 | [diff] [blame] | 1435 | void ThreadList::ReleaseThreadId(Thread* self, uint32_t id) { |
Chao-ying Fu | 9e36931 | 2014-05-21 11:20:52 -0700 | [diff] [blame] | 1436 | MutexLock mu(self, *Locks::allocated_thread_ids_lock_); |
Brian Carlstrom | 7934ac2 | 2013-07-26 10:54:15 -0700 | [diff] [blame] | 1437 | --id; // Zero is reserved to mean "invalid". |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 1438 | DCHECK(allocated_ids_[id]) << id; |
| 1439 | allocated_ids_.reset(id); |
| 1440 | } |
| 1441 | |
Mathieu Chartier | 4f55e22 | 2015-09-04 13:26:21 -0700 | [diff] [blame] | 1442 | ScopedSuspendAll::ScopedSuspendAll(const char* cause, bool long_suspend) { |
| 1443 | Runtime::Current()->GetThreadList()->SuspendAll(cause, long_suspend); |
| 1444 | } |
| 1445 | |
| 1446 | ScopedSuspendAll::~ScopedSuspendAll() { |
| 1447 | Runtime::Current()->GetThreadList()->ResumeAll(); |
| 1448 | } |
| 1449 | |
Elliott Hughes | 8daa092 | 2011-09-11 13:46:25 -0700 | [diff] [blame] | 1450 | } // namespace art |