| OLD | NEW |
| 1 // Copyright (c) 2010 The Chromium Authors. All rights reserved. | 1 // Copyright (c) 2010 The Chromium Authors. All rights reserved. |
| 2 // Use of this source code is governed by a BSD-style license that can be | 2 // Use of this source code is governed by a BSD-style license that can be |
| 3 // found in the LICENSE file. | 3 // found in the LICENSE file. |
| 4 | 4 |
| 5 #if defined(OS_WIN) | 5 #if defined(OS_WIN) |
| 6 #include <windows.h> | 6 #include <windows.h> |
| 7 #endif | 7 #endif |
| 8 | 8 |
| 9 #include "chrome/gpu/gpu_watchdog_thread.h" | 9 #include "chrome/gpu/gpu_watchdog_thread.h" |
| 10 | 10 |
| 11 #include "base/compiler_specific.h" | 11 #include "base/compiler_specific.h" |
| 12 #include "build/build_config.h" | 12 #include "build/build_config.h" |
| 13 | 13 |
| 14 namespace { | 14 namespace { |
| 15 const int64 kCheckPeriod = 2000; | 15 const int64 kCheckPeriod = 2000; |
| 16 | 16 |
| 17 void DoNothing() { | 17 void DoNothing() { |
| 18 } | 18 } |
| 19 } | 19 } |
| 20 | 20 |
| 21 GpuWatchdogThread::GpuWatchdogThread(MessageLoop* watched_message_loop, | 21 GpuWatchdogThread::GpuWatchdogThread(int timeout) |
| 22 int timeout) | |
| 23 : base::Thread("Watchdog"), | 22 : base::Thread("Watchdog"), |
| 24 watched_message_loop_(watched_message_loop), | 23 watched_message_loop_(MessageLoop::current()), |
| 25 timeout_(timeout), | 24 timeout_(timeout), |
| 26 armed_(false), | 25 armed_(false), |
| 26 #if defined(OS_WIN) |
| 27 watched_thread_handle_(0), |
| 28 arm_time_(0), |
| 29 #endif |
| 27 ALLOW_THIS_IN_INITIALIZER_LIST(task_observer_(this)) { | 30 ALLOW_THIS_IN_INITIALIZER_LIST(task_observer_(this)) { |
| 28 DCHECK(watched_message_loop); | |
| 29 DCHECK(timeout >= 0); | 31 DCHECK(timeout >= 0); |
| 30 | 32 |
| 33 #if defined(OS_WIN) |
| 34 // GetCurrentThread returns a pseudo-handle that cannot be used by one thread |
| 35 // to identify another. DuplicateHandle creates a "real" handle that can be |
| 36 // used for this purpose. |
| 37 BOOL result = DuplicateHandle(GetCurrentProcess(), |
| 38 GetCurrentThread(), |
| 39 GetCurrentProcess(), |
| 40 &watched_thread_handle_, |
| 41 THREAD_QUERY_INFORMATION, |
| 42 FALSE, |
| 43 0); |
| 44 DCHECK(result); |
| 45 #endif |
| 46 |
| 31 watched_message_loop_->AddTaskObserver(&task_observer_); | 47 watched_message_loop_->AddTaskObserver(&task_observer_); |
| 32 } | 48 } |
| 33 | 49 |
| 34 GpuWatchdogThread::~GpuWatchdogThread() { | 50 GpuWatchdogThread::~GpuWatchdogThread() { |
| 35 // Verify that the thread was explicitly stopped. If the thread is stopped | 51 // Verify that the thread was explicitly stopped. If the thread is stopped |
| 36 // implicitly by the destructor, CleanUp() will not be called. | 52 // implicitly by the destructor, CleanUp() will not be called. |
| 37 DCHECK(!method_factory_.get()); | 53 DCHECK(!method_factory_.get()); |
| 38 | 54 |
| 55 #if defined(OS_WIN) |
| 56 CloseHandle(watched_thread_handle_); |
| 57 #endif |
| 58 |
| 39 watched_message_loop_->RemoveTaskObserver(&task_observer_); | 59 watched_message_loop_->RemoveTaskObserver(&task_observer_); |
| 40 } | 60 } |
| 41 | 61 |
| 42 void GpuWatchdogThread::PostAcknowledge() { | 62 void GpuWatchdogThread::PostAcknowledge() { |
| 43 // Called on the monitored thread. Responds with OnAcknowledge. Cannot use | 63 // Called on the monitored thread. Responds with OnAcknowledge. Cannot use |
| 44 // the method factory. Rely on reference counting instead. | 64 // the method factory. Rely on reference counting instead. |
| 45 message_loop()->PostTask( | 65 message_loop()->PostTask( |
| 46 FROM_HERE, | 66 FROM_HERE, |
| 47 NewRunnableMethod(this, &GpuWatchdogThread::OnAcknowledge)); | 67 NewRunnableMethod(this, &GpuWatchdogThread::OnAcknowledge)); |
| 48 } | 68 } |
| 49 | 69 |
| 50 void GpuWatchdogThread::Init() { | 70 void GpuWatchdogThread::Init() { |
| 51 // The method factory must be created on the watchdog thread. | 71 // The method factory must be created on the watchdog thread. |
| 52 method_factory_.reset(new MethodFactory(this)); | 72 method_factory_.reset(new MethodFactory(this)); |
| 53 | 73 |
| 54 // Schedule the first check. | 74 // Schedule the first check. |
| 55 OnCheck(); | 75 OnCheck(); |
| 56 } | 76 } |
| 57 | 77 |
| 58 void GpuWatchdogThread::CleanUp() { | 78 void GpuWatchdogThread::CleanUp() { |
| 59 // The method factory must be destroyed on the watchdog thread. | 79 // The method factory must be destroyed on the watchdog thread. |
| 60 method_factory_->RevokeAll(); | 80 method_factory_->RevokeAll(); |
| 61 method_factory_.reset(); | 81 method_factory_.reset(); |
| 62 | |
| 63 // Prevent any more delayed tasks from being posted. | |
| 64 watched_message_loop_ = NULL; | |
| 65 } | 82 } |
| 66 | 83 |
| 67 GpuWatchdogThread::GpuWatchdogTaskObserver::GpuWatchdogTaskObserver( | 84 GpuWatchdogThread::GpuWatchdogTaskObserver::GpuWatchdogTaskObserver( |
| 68 GpuWatchdogThread* watchdog) | 85 GpuWatchdogThread* watchdog) |
| 69 : watchdog_(watchdog) { | 86 : watchdog_(watchdog) { |
| 70 } | 87 } |
| 71 | 88 |
| 72 GpuWatchdogThread::GpuWatchdogTaskObserver::~GpuWatchdogTaskObserver() { | 89 GpuWatchdogThread::GpuWatchdogTaskObserver::~GpuWatchdogTaskObserver() { |
| 73 } | 90 } |
| 74 | 91 |
| (...skipping 24 matching lines...) Expand all Loading... |
| 99 // watched thread to see armed_ being true multiple times before | 116 // watched thread to see armed_ being true multiple times before |
| 100 // the OnAcknowledge task is run on the watchdog thread. | 117 // the OnAcknowledge task is run on the watchdog thread. |
| 101 if (!armed_) | 118 if (!armed_) |
| 102 return; | 119 return; |
| 103 | 120 |
| 104 // Revoke any pending OnExit. | 121 // Revoke any pending OnExit. |
| 105 method_factory_->RevokeAll(); | 122 method_factory_->RevokeAll(); |
| 106 armed_ = false; | 123 armed_ = false; |
| 107 | 124 |
| 108 // The monitored thread has responded. Post a task to check it again. | 125 // The monitored thread has responded. Post a task to check it again. |
| 109 if (watched_message_loop_) { | 126 message_loop()->PostDelayedTask( |
| 110 message_loop()->PostDelayedTask( | 127 FROM_HERE, |
| 111 FROM_HERE, | 128 method_factory_->NewRunnableMethod(&GpuWatchdogThread::OnCheck), |
| 112 method_factory_->NewRunnableMethod(&GpuWatchdogThread::OnCheck), | 129 kCheckPeriod); |
| 113 kCheckPeriod); | |
| 114 } | |
| 115 } | 130 } |
| 116 | 131 |
| 132 #if defined(OS_WIN) |
| 133 int64 GpuWatchdogThread::GetWatchedThreadTime() { |
| 134 FILETIME creation_time; |
| 135 FILETIME exit_time; |
| 136 FILETIME user_time; |
| 137 FILETIME kernel_time; |
| 138 BOOL result = GetThreadTimes(watched_thread_handle_, |
| 139 &creation_time, |
| 140 &exit_time, |
| 141 &kernel_time, |
| 142 &user_time); |
| 143 DCHECK(result); |
| 144 |
| 145 ULARGE_INTEGER user_time64; |
| 146 user_time64.HighPart = user_time.dwHighDateTime; |
| 147 user_time64.LowPart = user_time.dwLowDateTime; |
| 148 |
| 149 ULARGE_INTEGER kernel_time64; |
| 150 kernel_time64.HighPart = kernel_time.dwHighDateTime; |
| 151 kernel_time64.LowPart = kernel_time.dwLowDateTime; |
| 152 |
| 153 // Time is reported in units of 100 nanoseconds. Kernel and user time are |
| 154 // summed to deal with to kinds of hangs. One is where the GPU process is |
| 155 // stuck in user level, never calling into the kernel and kernel time is |
| 156 // not increasing. The other is where either the kernel hangs and never |
| 157 // returns to user level or where user level code |
| 158 // calls into kernel level repeatedly, giving up its quanta before it is |
| 159 // tracked, for example a loop that repeatedly Sleeps. |
| 160 return static_cast<int64>( |
| 161 (user_time64.QuadPart + kernel_time64.QuadPart) / 10000); |
| 162 } |
| 163 #endif |
| 164 |
| 117 void GpuWatchdogThread::OnCheck() { | 165 void GpuWatchdogThread::OnCheck() { |
| 118 if (watched_message_loop_) { | 166 if (armed_) |
| 119 // Must set armed before posting the task. This task might be the only task | 167 return; |
| 120 // that will activate the TaskObserver on the watched thread and it must not | |
| 121 // miss the false -> true transition. | |
| 122 armed_ = true; | |
| 123 | 168 |
| 124 // Post a task to the monitored thread that does nothing but wake up the | 169 // Must set armed before posting the task. This task might be the only task |
| 125 // TaskObserver. Any other tasks that are pending on the watched thread will | 170 // that will activate the TaskObserver on the watched thread and it must not |
| 126 // also wake up the observer. This simply ensures there is at least one. | 171 // miss the false -> true transition. |
| 127 watched_message_loop_->PostTask( | 172 armed_ = true; |
| 128 FROM_HERE, | |
| 129 NewRunnableFunction(DoNothing)); | |
| 130 | 173 |
| 131 // Post a task to the watchdog thread to exit if the monitored thread does | 174 #if defined(OS_WIN) |
| 132 // not respond in time. | 175 arm_time_ = GetWatchedThreadTime(); |
| 133 message_loop()->PostDelayedTask( | 176 #endif |
| 134 FROM_HERE, | 177 |
| 135 method_factory_->NewRunnableMethod(&GpuWatchdogThread::OnExit), | 178 // Post a task to the monitored thread that does nothing but wake up the |
| 136 timeout_); | 179 // TaskObserver. Any other tasks that are pending on the watched thread will |
| 137 } | 180 // also wake up the observer. This simply ensures there is at least one. |
| 181 watched_message_loop_->PostTask( |
| 182 FROM_HERE, |
| 183 NewRunnableFunction(DoNothing)); |
| 184 |
| 185 // Post a task to the watchdog thread to exit if the monitored thread does |
| 186 // not respond in time. |
| 187 message_loop()->PostDelayedTask( |
| 188 FROM_HERE, |
| 189 method_factory_->NewRunnableMethod(&GpuWatchdogThread::OnExit), |
| 190 timeout_); |
| 138 } | 191 } |
| 139 | 192 |
| 140 // Use the --disable-gpu-watchdog command line switch to disable this. | 193 // Use the --disable-gpu-watchdog command line switch to disable this. |
| 141 void GpuWatchdogThread::OnExit() { | 194 void GpuWatchdogThread::OnExit() { |
| 195 #if defined(OS_WIN) |
| 196 // Defer termination until a certain amount of user time has elapsed. |
| 197 int64 time_since_arm = GetWatchedThreadTime() - arm_time_; |
| 198 if (time_since_arm < timeout_) { |
| 199 message_loop()->PostDelayedTask( |
| 200 FROM_HERE, |
| 201 method_factory_->NewRunnableMethod(&GpuWatchdogThread::OnExit), |
| 202 timeout_ - time_since_arm); |
| 203 return; |
| 204 } |
| 205 #endif |
| 206 |
| 142 // Make sure the timeout period is on the stack before crashing. | 207 // Make sure the timeout period is on the stack before crashing. |
| 143 volatile int timeout = timeout_; | 208 volatile int timeout = timeout_; |
| 144 | 209 |
| 145 // For minimal developer annoyance, don't keep crashing. | 210 // For minimal developer annoyance, don't keep crashing. |
| 146 static bool crashed = false; | 211 static bool crashed = false; |
| 147 if (crashed) | 212 if (crashed) |
| 148 return; | 213 return; |
| 149 | 214 |
| 150 #if defined(OS_WIN) | 215 #if defined(OS_WIN) |
| 151 if (IsDebuggerPresent()) | 216 if (IsDebuggerPresent()) |
| 152 return; | 217 return; |
| 153 #endif | 218 #endif |
| 154 | 219 |
| 155 LOG(ERROR) << "The GPU process hung. Restarting after " | 220 LOG(ERROR) << "The GPU process hung. Restarting after " |
| 156 << timeout_ << " seconds."; | 221 << timeout_ << " seconds."; |
| 157 | 222 |
| 158 volatile int* null_pointer = NULL; | 223 volatile int* null_pointer = NULL; |
| 159 *null_pointer = timeout; | 224 *null_pointer = timeout; |
| 160 | 225 |
| 161 crashed = true; | 226 crashed = true; |
| 162 } | 227 } |
| OLD | NEW |