Worker: Protect a running debugger task from forcible worker termination
A running debugger task must be protected from forcible worker termination, but
in the current implementation, there is a corner case where the debugger task
can suddenly be killed as follows:
1) performDebuggerTaskOnWorkerThread() is called almost at the same time as
terminateInternal() on the main thread. terminateInternal() acquires
|m_threadStateMutex| before performDebuggerTaskOnWorkerThread().
2) terminateInternal() checks a flag |m_runningDebuggerTask| before
performDebuggerTaskOnWorkerThread() sets it, schedules a force termination
task, and releases the lock.
3) performDebuggerTaskOnWorkerThread() acquires the lock, sets the flag, and
runs the debugger task.
4) The scheduled task is fired while the debugger task is running and attempts
to terminate it. This may cause crashes.
To avoid the case, this CL makes a scheduled task check the flag before
terminating the worker execution. In addition, this ensures that postTask() and
appendDebuggerTask() don't post new tasks after terminateInternal() or
prepareForShutdownOnWorkerThread() is called.
BUG=616775
Review-Url: https://codereview.chromium.org/2041753002
Cr-Commit-Position: refs/heads/master@{#398837}
5 files changed
