[Lldb-commits] [PATCH] D77295: [lldb/Core] Fix a race in the Communication class

[Pavel Labath via Phabricator via lldb-commits]

labath added a comment.

In D77295#1995077 <https://reviews.llvm.org/D77295#1995077>, @JDevlieghere wrote:

> Hi Pavel,
>
> After this landed I started seeing a bunch of reproducers fail with `SIGPIPE`. I ignored the signal and added a check for `EPIPE` to the `::write` call in `PipePosix.cpp`. Below is the thread state:
>
>   * thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGPIPE
>     * frame #0: 0x00007fff6876c55e libsystem_kernel.dylib`__ulock_wait + 10
>       frame #1: 0x00007fff6882f5c2 libsystem_pthread.dylib`_pthread_join + 347
>       frame #2: 0x000000010115ef62 liblldb.11.0.0git.dylib`lldb_private::HostThreadPosix::Join(this=0x000000011e7ca490, result=0x00007ffeefbfe290) at HostThreadPosix.cpp:28:15
>       frame #3: 0x000000010110b64f liblldb.11.0.0git.dylib`lldb_private::HostThread::Join(this=0x000000011e9f09a0, result=0x00007ffeefbfe290) at HostThread.cpp:21:27
>       frame #4: 0x00000001012ea5ba liblldb.11.0.0git.dylib`lldb_private::Process::ControlPrivateStateThread(this=0x000000011e9f0818, signal=1) at Process.cpp:3642:30
>       frame #5: 0x00000001012dbc81 liblldb.11.0.0git.dylib`lldb_private::Process::StopPrivateStateThread(this=0x000000011e9f0818) at Process.cpp:3593:5
>       frame #6: 0x00000001012dc514 liblldb.11.0.0git.dylib`lldb_private::Process::Destroy(this=0x000000011e9f0818, force_kill=true) at Process.cpp:3317:7
>       frame #7: 0x0000000100981cc1 liblldb.11.0.0git.dylib`lldb::SBProcess::Kill(this=0x000000011e767de0) at SBProcess.cpp:662:35
>  
>   * thread #2
>     * frame #0: 0x00007fff6876c4ce libsystem_kernel.dylib`__workq_kernreturn + 10
>       frame #1: 0x00007fff6882aaa1 libsystem_pthread.dylib`_pthread_wqthread + 390
>       frame #2: 0x00007fff68829b77 libsystem_pthread.dylib`start_wqthread + 15
>  
>   * thread #3, name = '<lldb.process.internal-state(pid=47347)>'
>     * frame #0: 0x00007fff6876c55e libsystem_kernel.dylib`__ulock_wait + 10
>       frame #1: 0x00007fff6882f5c2 libsystem_pthread.dylib`_pthread_join + 347
>       frame #2: 0x000000010115ef62 liblldb.11.0.0git.dylib`lldb_private::HostThreadPosix::Join(this=0x000000011e7ca510, result=0x0000000000000000) at HostThreadPosix.cpp:28:15
>       frame #3: 0x000000010110b64f liblldb.11.0.0git.dylib`lldb_private::HostThread::Join(this=0x000000011e9f0e28, result=0x0000000000000000) at HostThread.cpp:21:27
>       frame #4: 0x0000000100f46f4e liblldb.11.0.0git.dylib`lldb_private::Communication::StopReadThread(this=0x000000011e9f0de8, error_ptr=0x0000000000000000) at Communication.cpp:239:32
>       frame #5: 0x00000001012e9d95 liblldb.11.0.0git.dylib`lldb_private::Process::ShouldBroadcastEvent(this=0x000000011e9f0818, event_ptr=0x000000011e76aa40) at Process.cpp:3390:27
>       frame #6: 0x00000001012e5e4f liblldb.11.0.0git.dylib`lldb_private::Process::HandlePrivateEvent(this=0x000000011e9f0818, event_sp=std::__1::shared_ptr<lldb_private::Event>::element_type @ 0x000000011e76aa40 strong=2 weak=1) at Process.cpp:3697:33
>       frame #7: 0x00000001012eb147 liblldb.11.0.0git.dylib`lldb_private::Process::RunPrivateStateThread(this=0x000000011e9f0818, is_secondary_thread=false) at Process.cpp:3891:7
>       frame #8: 0x00000001012ea2fa liblldb.11.0.0git.dylib`lldb_private::Process::PrivateStateThread(arg=0x000000011e7c9a60) at Process.cpp:3789:25
>       frame #9: 0x000000010110b45b liblldb.11.0.0git.dylib`lldb_private::HostNativeThreadBase::ThreadCreateTrampoline(arg=0x000000011e7ca4e0) at HostNativeThreadBase.cpp:68:10
>       frame #10: 0x000000010726ab83 liblldb.11.0.0git.dylib`lldb_private::HostThreadMacOSX::ThreadCreateTrampoline(arg=0x000000011e7ca4e0) at HostThreadMacOSX.mm:68:10
>  
>   * thread #4, name = '<lldb.comm.process.stdio>', stop reason = breakpoint 1.1
>     * frame #0: 0x000000010116071a liblldb.11.0.0git.dylib`lldb_private::PipePosix::Write(this=0x000000011e7ca2b0, buf=43698118819080247, size=1, bytes_written=0x00007000076908b8) at PipePosix.cpp:312:7
>       frame #1: 0x0000000101156d15 liblldb.11.0.0git.dylib`lldb_private::ConnectionFileDescriptor::Disconnect(this=0x000000011e7ca210, error_ptr=0x0000000000000000) at ConnectionFileDescriptorPosix.cpp:321:30
>       frame #2: 0x0000000100f454e6 liblldb.11.0.0git.dylib`lldb_private::Communication::Disconnect(this=0x000000011e9f0de8, error_ptr=0x0000000000000000) at Communication.cpp:98:46
>       frame #3: 0x0000000100f46d4e liblldb.11.0.0git.dylib`lldb_private::Communication::ReadThread(p=0x000000011e9f0de8) at Communication.cpp:377:13
>       frame #4: 0x000000010110b45b liblldb.11.0.0git.dylib`lldb_private::HostNativeThreadBase::ThreadCreateTrampoline(arg=0x000000011e7ca4e0) at HostNativeThreadBase.cpp:68:10
>       frame #5: 0x000000010726ab83 liblldb.11.0.0git.dylib`lldb_private::HostThreadMacOSX::ThreadCreateTrampoline(arg=0x000000011e7ca4e0) at HostThreadMacOSX.mm:68:10
>       frame #6: 0x00007fff6882e109 libsystem_pthread.dylib`_pthread_start + 148
>       frame #7: 0x00007fff68829b8b libsystem_pthread.dylib`thread_start + 15
>
>
> Any chance you might know what's going on here?


Hm... this probably didn't introduce the bug, but I can certainly see how it could cause an existing issue to start to appear more often.

I think I have an idea for what could be problem here -- a race between the Disconnect call in `Communication::ReadThread` and the one in `Process::ShouldBroadcastEvent`. Can you check if swapping the order of these two lines in `Process::ShouldBroadcastEvent`:

  m_stdio_communication.Disconnect();
  m_stdio_communication.StopReadThread();

makes the problem go away?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D77295/new/

https://reviews.llvm.org/D77295