[NativeProcessLinux] Fix a bug in instruction-stepping over thread creation

Summary:
There was a bug in NativeProcessLinux, where doing an instruction-level single-step over the
thread-creation syscall resulted in loss of control over the inferior. This happened because
after the inferior entered the thread-creation maintenance stop, we unconditionally performed a
PTRACE_CONT, even though the original intention was to do a PTRACE_SINGLESTEP. This is fixed by
storing the original state of the thread before the stop (stepping or running) and then
performing the appropriate action when resuming.

I also get rid of the callback in the ThreadContext structure, which stored the lambda used to
resume the thread, but which was not used consistently.

A test verifying the correctness of the new behavior is included.

Reviewers: ovyalov, tberghammer

Subscribers: lldb-commits

Differential Revision: http://reviews.llvm.org/D12104

llvm-svn: 245545

3 files changed, 129 insertions, 0 deletions

diff --git a/lldb/test/functionalities/thread/create_during_instruction_step/Makefile b/lldb/test/functionalities/thread/create_during_instruction_step/Makefile
new file mode 100644
index 00000000000..67aa16625bf
--- /dev/null
+++ b/lldb/test/functionalities/thread/create_during_instruction_step/Makefile
@@ -0,0 +1,5 @@
+LEVEL = ../../../make
+
+CXX_SOURCES := main.cpp
+ENABLE_THREADS := YES
+include $(LEVEL)/Makefile.rules
diff --git a/lldb/test/functionalities/thread/create_during_instruction_step/TestCreateDuringInstructionStep.py b/lldb/test/functionalities/thread/create_during_instruction_step/TestCreateDuringInstructionStep.py
new file mode 100644
index 00000000000..116cd334ff0
--- /dev/null
+++ b/lldb/test/functionalities/thread/create_during_instruction_step/TestCreateDuringInstructionStep.py
@@ -0,0 +1,88 @@
+"""
+This tests that we do not lose control of the inferior, while doing an instruction-level step
+over a thread creation instruction.
+"""
+
+import os
+import unittest2
+import lldb
+from lldbtest import *
+import lldbutil
+
+class CreateDuringInstructionStepTestCase(TestBase):
+
+    mydir = TestBase.compute_mydir(__file__)
+
+    def setUp(self):
+        # Call super's setUp().
+        TestBase.setUp(self)
+        # Find the line numbers to break and continue.
+        self.breakpoint = line_number('main.cpp', '// Set breakpoint here')
+
+    @dsym_test
+    def test_step_inst_with_dsym(self):
+        self.buildDsym(dictionary=self.getBuildFlags())
+        self.create_during_step_inst_test()
+
+    @dwarf_test
+    def test_step_inst_with_dwarf(self):
+        self.buildDwarf(dictionary=self.getBuildFlags())
+        self.create_during_step_inst_test()
+
+    def create_during_step_inst_test(self):
+        exe = os.path.join(os.getcwd(), "a.out")
+        target = self.dbg.CreateTarget(exe)
+        self.assertTrue(target and target.IsValid(), "Target is valid")
+
+        # This should create a breakpoint in the stepping thread.
+        self.bp_num = lldbutil.run_break_set_by_file_and_line (self, "main.cpp", self.breakpoint, num_expected_locations=-1)
+
+        # Run the program.
+        process = target.LaunchSimple(None, None, self.get_process_working_directory())
+        self.assertTrue(process and process.IsValid(), PROCESS_IS_VALID)
+
+        # The stop reason of the thread should be breakpoint.
+        self.assertEqual(process.GetState(), lldb.eStateStopped, PROCESS_STOPPED)
+        self.assertEqual(lldbutil.get_stopped_thread(process, lldb.eStopReasonBreakpoint).IsValid(), 1,
+                STOPPED_DUE_TO_BREAKPOINT)
+
+        # Get the number of threads
+        num_threads = process.GetNumThreads()
+
+        # Make sure we see only one threads
+        self.assertEqual(num_threads, 1, 'Number of expected threads and actual threads do not match.')
+
+        thread = process.GetThreadAtIndex(0)
+        self.assertTrue(thread and thread.IsValid(), "Thread is valid")
+
+        # Keep stepping until we see the thread creation
+        while process.GetNumThreads() < 2:
+            # This skips some functions we have trouble stepping into. Testing stepping
+            # through these is not the purpose of this test. We just want to find the
+            # instruction, which creates the thread.
+            if thread.GetFrameAtIndex(0).GetFunctionName() in [
+                    '__sync_fetch_and_add_4', # Android arm: unable to set a breakpoint for software single-step
+                    'pthread_mutex_lock'      # Android arm: function contains atomic instruction sequences
+                    ]:
+                thread.StepOut()
+            else:
+                thread.StepInstruction(False)
+            self.assertEqual(process.GetState(), lldb.eStateStopped, PROCESS_STOPPED)
+            self.assertEqual(thread.GetStopReason(), lldb.eStopReasonPlanComplete, "Step operation succeeded")
+            if self.TraceOn():
+                self.runCmd("disassemble --pc")
+
+        if self.TraceOn():
+            self.runCmd("thread list")
+
+        # We have successfully caught thread creation. Now just run to completion
+        process.Continue()
+
+        # At this point, the inferior process should have exited.
+        self.assertEqual(process.GetState(), lldb.eStateExited, PROCESS_EXITED)
+
+if __name__ == '__main__':
+    import atexit
+    lldb.SBDebugger.Initialize()
+    atexit.register(lambda: lldb.SBDebugger.Terminate())
+    unittest2.main()
diff --git a/lldb/test/functionalities/thread/create_during_instruction_step/main.cpp b/lldb/test/functionalities/thread/create_during_instruction_step/main.cpp
new file mode 100644
index 00000000000..c50ce479079
--- /dev/null
+++ b/lldb/test/functionalities/thread/create_during_instruction_step/main.cpp
@@ -0,0 +1,36 @@
+//===-- main.cpp ------------------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include <atomic>
+#include <thread>
+
+std::atomic<bool> flag(false);
+
+void do_nothing()
+{
+    while (flag)
+        ;
+}
+
+int main ()
+{
+    // Instruction-level stepping over a creation of the first thread takes a very long time, so
+    // we give the threading machinery a chance to initialize all its data structures.
+    // This way, stepping over the second thread will be much faster.
+    std::thread dummy(do_nothing);
+    dummy.join();
+
+    // Make sure the new thread does not exit before we get a chance to notice the main thread stopped
+    flag = true;
+
+    std::thread thread(do_nothing); // Set breakpoint here
+    flag = false; // Release the new thread.
+    thread.join();
+    return 0;
+}