[lldb-dev] Resuming traced process on Linux and SIGSTOP

[Andrew MacPherson]

Currently when attaching to a running process on Linux, a SIGSTOP signal is
(incorrectly I think) injected into the inferior on resume. This can be
reproduced by simply launching any process and then in a separate terminal
doing:

sudo lldb -p <pid>
c
process interrupt
c

On the second continue the SIGSTOP that was used to stop the process is
injected into the inferior by being passed to PTRACE_CONT, resulting in the
process going into a stop state outside the control of LLDB. The SIGSTOP
comes from the SetResumeSignal() call in POSIXThread and StopInfo.

I can't think of any reason why a SIGSTOP should ever be injected into the
inferior so as a test I simply prevented it from ever happening:

diff --git a/source/Plugins/Process/Linux/ProcessMonitor.cpp
b/source/Plugins/Process/Linux/ProcessMonitor.cpp
index 3dec6de..3079379 100644
--- a/source/Plugins/Process/Linux/ProcessMonitor.cpp
+++ b/source/Plugins/Process/Linux/ProcessMonitor.cpp
@@ -2209,6 +2209,9 @@ ProcessMonitor::Resume(lldb::tid_t tid, uint32_t
signo)
     bool result;
     Log *log (ProcessPOSIXLog::GetLogIfAllCategoriesSet
(POSIX_LOG_PROCESS));

+    if (signo == SIGSTOP)
+      signo = eResumeSignalNone;
+
     if (log)
         log->Printf ("ProcessMonitor::%s() resuming thread = %"  PRIu64 "
with signal %s", __FUNCTION__, tid,

m_process->GetUnixSignals().GetSignalAsCString (signo));

This resolves the issue and doesn't cause any other problems that I can
find but almost certainly isn't the proper fix. My main concern is that all
of the resume signal code is shared with other OSes which I'm guessing
treat this differently.

Any thoughts as to what the proper fix here might be?

Thanks,
Andrew
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20140321/05a28a8e/attachment.html>