Enable annotalysis on clang ART builds.

Fix clang build errors aswell as restructure locking/mutex code for correct
thread safety analysis support.
Reorder make dependencies so that host builds build first as they should
provide better compilation errors than target.
Remove host's use of -fno-omit-frame-pointer as it has no value with correct
use of CFI, which we should have.

Change-Id: I72cea8da9a3757b1a0b3acb4081feccb7c6cef90
diff --git a/runtime/base/logging.h b/runtime/base/logging.h
index 075d571..0fcec1f 100644
--- a/runtime/base/logging.h
+++ b/runtime/base/logging.h
@@ -192,7 +192,7 @@
     : data_(new LogMessageData(file, line, severity, error)) {
   }
 
-  ~LogMessage() LOCKS_EXCLUDED(Locks::logging_lock_);
+  ~LogMessage();  // TODO: enable LOCKS_EXCLUDED(Locks::logging_lock_).
 
   std::ostream& stream() {
     return data_->buffer;
@@ -235,32 +235,6 @@
   return os;
 }
 
-template<typename T>
-class MutatorLockedDumpable {
- public:
-  explicit MutatorLockedDumpable(T& value)
-      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) : value_(value) {
-  }
-
-  void Dump(std::ostream& os) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    value_.Dump(os);
-  }
-
- private:
-  T& value_;
-
-  DISALLOW_COPY_AND_ASSIGN(MutatorLockedDumpable);
-};
-
-template<typename T>
-std::ostream& operator<<(std::ostream& os, const MutatorLockedDumpable<T>& rhs)
-// TODO: should be SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) however annotalysis
-//       currently fails for this.
-    NO_THREAD_SAFETY_ANALYSIS {
-  rhs.Dump(os);
-  return os;
-}
-
 // Helps you use operator<< in a const char*-like context such as our various 'F' methods with
 // format strings.
 template<typename T>
diff --git a/runtime/base/macros.h b/runtime/base/macros.h
index 6cc9396..b193ff1 100644
--- a/runtime/base/macros.h
+++ b/runtime/base/macros.h
@@ -178,48 +178,40 @@
 
 template<typename T> void UNUSED(const T&) {}
 
-#if defined(__SUPPORT_TS_ANNOTATION__)
-
-#define ACQUIRED_AFTER(...) __attribute__ ((acquired_after(__VA_ARGS__)))
-#define ACQUIRED_BEFORE(...) __attribute__ ((acquired_before(__VA_ARGS__)))
-#define EXCLUSIVE_LOCK_FUNCTION(...) __attribute__ ((exclusive_lock(__VA_ARGS__)))
-#define EXCLUSIVE_LOCKS_REQUIRED(...) __attribute__ ((exclusive_locks_required(__VA_ARGS__)))
-#define EXCLUSIVE_TRYLOCK_FUNCTION(...) __attribute__ ((exclusive_trylock(__VA_ARGS__)))
-#define GUARDED_BY(x) __attribute__ ((guarded_by(x)))
-#define GUARDED_VAR __attribute__ ((guarded))
-#define LOCKABLE __attribute__ ((lockable))
-#define LOCK_RETURNED(x) __attribute__ ((lock_returned(x)))
-#define LOCKS_EXCLUDED(...) __attribute__ ((locks_excluded(__VA_ARGS__)))
-#define NO_THREAD_SAFETY_ANALYSIS __attribute__ ((no_thread_safety_analysis))
-#define PT_GUARDED_BY(x) __attribute__ ((point_to_guarded_by(x)))
-#define PT_GUARDED_VAR __attribute__ ((point_to_guarded))
-#define SCOPED_LOCKABLE __attribute__ ((scoped_lockable))
-#define SHARED_LOCK_FUNCTION(...) __attribute__ ((shared_lock(__VA_ARGS__)))
-#define SHARED_LOCKS_REQUIRED(...) __attribute__ ((shared_locks_required(__VA_ARGS__)))
-#define SHARED_TRYLOCK_FUNCTION(...) __attribute__ ((shared_trylock(__VA_ARGS__)))
-#define UNLOCK_FUNCTION(...) __attribute__ ((unlock(__VA_ARGS__)))
-
+// Annotalysis thread-safety analysis support.
+#if defined(__SUPPORT_TS_ANNOTATION__) || defined(__clang__)
+#define THREAD_ANNOTATION_ATTRIBUTE__(x)   __attribute__((x))
 #else
+#define THREAD_ANNOTATION_ATTRIBUTE__(x)   // no-op
+#endif
 
-#define ACQUIRED_AFTER(...)
-#define ACQUIRED_BEFORE(...)
-#define EXCLUSIVE_LOCK_FUNCTION(...)
-#define EXCLUSIVE_LOCKS_REQUIRED(...)
-#define EXCLUSIVE_TRYLOCK_FUNCTION(...)
-#define GUARDED_BY(x)
-#define GUARDED_VAR
-#define LOCKABLE
-#define LOCK_RETURNED(x)
-#define LOCKS_EXCLUDED(...)
-#define NO_THREAD_SAFETY_ANALYSIS
+#define ACQUIRED_AFTER(...) THREAD_ANNOTATION_ATTRIBUTE__(acquired_after(__VA_ARGS__))
+#define ACQUIRED_BEFORE(...) THREAD_ANNOTATION_ATTRIBUTE__(acquired_before(__VA_ARGS__))
+#define EXCLUSIVE_LOCKS_REQUIRED(...) THREAD_ANNOTATION_ATTRIBUTE__(exclusive_locks_required(__VA_ARGS__))
+#define GUARDED_BY(x) THREAD_ANNOTATION_ATTRIBUTE__(guarded_by(x))
+#define GUARDED_VAR THREAD_ANNOTATION_ATTRIBUTE__(guarded)
+#define LOCKABLE THREAD_ANNOTATION_ATTRIBUTE__(lockable)
+#define LOCK_RETURNED(x) THREAD_ANNOTATION_ATTRIBUTE__(lock_returned(x))
+#define LOCKS_EXCLUDED(...) THREAD_ANNOTATION_ATTRIBUTE__(locks_excluded(__VA_ARGS__))
+#define NO_THREAD_SAFETY_ANALYSIS THREAD_ANNOTATION_ATTRIBUTE__(no_thread_safety_analysis)
 #define PT_GUARDED_BY(x)
-#define PT_GUARDED_VAR
-#define SCOPED_LOCKABLE
-#define SHARED_LOCK_FUNCTION(...)
-#define SHARED_LOCKS_REQUIRED(...)
-#define SHARED_TRYLOCK_FUNCTION(...)
-#define UNLOCK_FUNCTION(...)
+// THREAD_ANNOTATION_ATTRIBUTE__(point_to_guarded_by(x))
+#define PT_GUARDED_VAR THREAD_ANNOTATION_ATTRIBUTE__(point_to_guarded)
+#define SCOPED_LOCKABLE THREAD_ANNOTATION_ATTRIBUTE__(scoped_lockable)
+#define SHARED_LOCKS_REQUIRED(...) THREAD_ANNOTATION_ATTRIBUTE__(shared_locks_required(__VA_ARGS__))
 
-#endif  // defined(__SUPPORT_TS_ANNOTATION__)
+#if defined(__clang__)
+#define EXCLUSIVE_LOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(exclusive_lock_function(__VA_ARGS__))
+#define EXCLUSIVE_TRYLOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(exclusive_trylock_function(__VA_ARGS__))
+#define SHARED_LOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(shared_lock_function(__VA_ARGS__))
+#define SHARED_TRYLOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(shared_trylock_function(__VA_ARGS__))
+#define UNLOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(unlock_function(__VA_ARGS__))
+#else
+#define EXCLUSIVE_LOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(exclusive_lock(__VA_ARGS__))
+#define EXCLUSIVE_TRYLOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(exclusive_trylock(__VA_ARGS__))
+#define SHARED_LOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(shared_lock(__VA_ARGS__))
+#define SHARED_TRYLOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(shared_trylock(__VA_ARGS__))
+#define UNLOCK_FUNCTION(...) THREAD_ANNOTATION_ATTRIBUTE__(unlock(__VA_ARGS__))
+#endif
 
 #endif  // ART_RUNTIME_BASE_MACROS_H_
diff --git a/runtime/base/mutex.cc b/runtime/base/mutex.cc
index ff72d16..fdf5763 100644
--- a/runtime/base/mutex.cc
+++ b/runtime/base/mutex.cc
@@ -29,6 +29,30 @@
 
 namespace art {
 
+Mutex* Locks::abort_lock_ = nullptr;
+Mutex* Locks::breakpoint_lock_ = nullptr;
+Mutex* Locks::deoptimization_lock_ = nullptr;
+ReaderWriterMutex* Locks::classlinker_classes_lock_ = nullptr;
+ReaderWriterMutex* Locks::heap_bitmap_lock_ = nullptr;
+Mutex* Locks::logging_lock_ = nullptr;
+ReaderWriterMutex* Locks::mutator_lock_ = nullptr;
+Mutex* Locks::runtime_shutdown_lock_ = nullptr;
+Mutex* Locks::thread_list_lock_ = nullptr;
+Mutex* Locks::thread_suspend_count_lock_ = nullptr;
+Mutex* Locks::trace_lock_ = nullptr;
+Mutex* Locks::profiler_lock_ = nullptr;
+Mutex* Locks::unexpected_signal_lock_ = nullptr;
+Mutex* Locks::intern_table_lock_ = nullptr;
+
+struct AllMutexData {
+  // A guard for all_mutexes_ that's not a mutex (Mutexes must CAS to acquire and busy wait).
+  Atomic<const BaseMutex*> all_mutexes_guard;
+  // All created mutexes guarded by all_mutexes_guard_.
+  std::set<BaseMutex*>* all_mutexes;
+  AllMutexData() : all_mutexes(NULL) {}
+};
+static struct AllMutexData gAllMutexData[kAllMutexDataSize];
+
 #if ART_USE_FUTEXES
 static bool ComputeRelativeTimeSpec(timespec* result_ts, const timespec& lhs, const timespec& rhs) {
   const int32_t one_sec = 1000 * 1000 * 1000;  // one second in nanoseconds.
@@ -45,15 +69,6 @@
 }
 #endif
 
-struct AllMutexData {
-  // A guard for all_mutexes_ that's not a mutex (Mutexes must CAS to acquire and busy wait).
-  Atomic<const BaseMutex*> all_mutexes_guard;
-  // All created mutexes guarded by all_mutexes_guard_.
-  std::set<BaseMutex*>* all_mutexes;
-  AllMutexData() : all_mutexes(NULL) {}
-};
-static struct AllMutexData gAllMutexData[kAllMutexDataSize];
-
 class ScopedAllMutexesLock {
  public:
   explicit ScopedAllMutexesLock(const BaseMutex* mutex) : mutex_(mutex) {
@@ -792,4 +807,53 @@
   guard_.recursion_count_ = old_recursion_count;
 }
 
+void Locks::Init() {
+  if (logging_lock_ != nullptr) {
+    // Already initialized.
+    DCHECK(abort_lock_ != nullptr);
+    DCHECK(breakpoint_lock_ != nullptr);
+    DCHECK(deoptimization_lock_ != nullptr);
+    DCHECK(classlinker_classes_lock_ != nullptr);
+    DCHECK(heap_bitmap_lock_ != nullptr);
+    DCHECK(logging_lock_ != nullptr);
+    DCHECK(mutator_lock_ != nullptr);
+    DCHECK(thread_list_lock_ != nullptr);
+    DCHECK(thread_suspend_count_lock_ != nullptr);
+    DCHECK(trace_lock_ != nullptr);
+    DCHECK(profiler_lock_ != nullptr);
+    DCHECK(unexpected_signal_lock_ != nullptr);
+    DCHECK(intern_table_lock_ != nullptr);
+  } else {
+    logging_lock_ = new Mutex("logging lock", kLoggingLock, true);
+    abort_lock_ = new Mutex("abort lock", kAbortLock, true);
+
+    DCHECK(breakpoint_lock_ == nullptr);
+    breakpoint_lock_ = new Mutex("breakpoint lock", kBreakpointLock);
+    DCHECK(deoptimization_lock_ == nullptr);
+    deoptimization_lock_ = new Mutex("deoptimization lock", kDeoptimizationLock);
+    DCHECK(classlinker_classes_lock_ == nullptr);
+    classlinker_classes_lock_ = new ReaderWriterMutex("ClassLinker classes lock",
+                                                      kClassLinkerClassesLock);
+    DCHECK(heap_bitmap_lock_ == nullptr);
+    heap_bitmap_lock_ = new ReaderWriterMutex("heap bitmap lock", kHeapBitmapLock);
+    DCHECK(mutator_lock_ == nullptr);
+    mutator_lock_ = new ReaderWriterMutex("mutator lock", kMutatorLock);
+    DCHECK(runtime_shutdown_lock_ == nullptr);
+    runtime_shutdown_lock_ = new Mutex("runtime shutdown lock", kRuntimeShutdownLock);
+    DCHECK(thread_list_lock_ == nullptr);
+    thread_list_lock_ = new Mutex("thread list lock", kThreadListLock);
+    DCHECK(thread_suspend_count_lock_ == nullptr);
+    thread_suspend_count_lock_ = new Mutex("thread suspend count lock", kThreadSuspendCountLock);
+    DCHECK(trace_lock_ == nullptr);
+    trace_lock_ = new Mutex("trace lock", kTraceLock);
+    DCHECK(profiler_lock_ == nullptr);
+    profiler_lock_ = new Mutex("profiler lock", kProfilerLock);
+    DCHECK(unexpected_signal_lock_ == nullptr);
+    unexpected_signal_lock_ = new Mutex("unexpected signal lock", kUnexpectedSignalLock, true);
+    DCHECK(intern_table_lock_ == nullptr);
+    intern_table_lock_ = new Mutex("InternTable lock", kInternTableLock);
+  }
+}
+
+
 }  // namespace art
diff --git a/runtime/base/mutex.h b/runtime/base/mutex.h
index 63ed6cb..55ec1c3 100644
--- a/runtime/base/mutex.h
+++ b/runtime/base/mutex.h
@@ -27,7 +27,6 @@
 #include "base/logging.h"
 #include "base/macros.h"
 #include "globals.h"
-#include "locks.h"
 
 #if defined(__APPLE__)
 #define ART_USE_FUTEXES 0
@@ -44,9 +43,56 @@
 
 namespace art {
 
+class LOCKABLE ReaderWriterMutex;
 class ScopedContentionRecorder;
 class Thread;
 
+// LockLevel is used to impose a lock hierarchy [1] where acquisition of a Mutex at a higher or
+// equal level to a lock a thread holds is invalid. The lock hierarchy achieves a cycle free
+// partial ordering and thereby cause deadlock situations to fail checks.
+//
+// [1] http://www.drdobbs.com/parallel/use-lock-hierarchies-to-avoid-deadlock/204801163
+enum LockLevel {
+  kLoggingLock = 0,
+  kUnexpectedSignalLock,
+  kThreadSuspendCountLock,
+  kAbortLock,
+  kJdwpSocketLock,
+  kRosAllocGlobalLock,
+  kRosAllocBracketLock,
+  kRosAllocBulkFreeLock,
+  kAllocSpaceLock,
+  kDexFileMethodInlinerLock,
+  kDexFileToMethodInlinerMapLock,
+  kMarkSweepMarkStackLock,
+  kTransactionLogLock,
+  kInternTableLock,
+  kMonitorPoolLock,
+  kDefaultMutexLevel,
+  kMarkSweepLargeObjectLock,
+  kPinTableLock,
+  kLoadLibraryLock,
+  kJdwpObjectRegistryLock,
+  kClassLinkerClassesLock,
+  kBreakpointLock,
+  kMonitorLock,
+  kThreadListLock,
+  kBreakpointInvokeLock,
+  kDeoptimizationLock,
+  kTraceLock,
+  kProfilerLock,
+  kJdwpEventListLock,
+  kJdwpAttachLock,
+  kJdwpStartLock,
+  kRuntimeShutdownLock,
+  kHeapBitmapLock,
+  kMutatorLock,
+  kZygoteCreationLock,
+
+  kLockLevelCount  // Must come last.
+};
+std::ostream& operator<<(std::ostream& os, const LockLevel& rhs);
+
 const bool kDebugLocking = kIsDebugBuild;
 
 // Record Log contention information, dumpable via SIGQUIT.
@@ -413,6 +459,117 @@
 // "WriterMutexLock mu(lock)".
 #define WriterMutexLock(x) COMPILE_ASSERT(0, writer_mutex_lock_declaration_missing_variable_name)
 
+// Global mutexes corresponding to the levels above.
+class Locks {
+ public:
+  static void Init();
+
+  // The mutator_lock_ is used to allow mutators to execute in a shared (reader) mode or to block
+  // mutators by having an exclusive (writer) owner. In normal execution each mutator thread holds
+  // a share on the mutator_lock_. The garbage collector may also execute with shared access but
+  // at times requires exclusive access to the heap (not to be confused with the heap meta-data
+  // guarded by the heap_lock_ below). When the garbage collector requires exclusive access it asks
+  // the mutators to suspend themselves which also involves usage of the thread_suspend_count_lock_
+  // to cover weaknesses in using ReaderWriterMutexes with ConditionVariables. We use a condition
+  // variable to wait upon in the suspension logic as releasing and then re-acquiring a share on
+  // the mutator lock doesn't necessarily allow the exclusive user (e.g the garbage collector)
+  // chance to acquire the lock.
+  //
+  // Thread suspension:
+  // Shared users                                  | Exclusive user
+  // (holding mutator lock and in kRunnable state) |   .. running ..
+  //   .. running ..                               | Request thread suspension by:
+  //   .. running ..                               |   - acquiring thread_suspend_count_lock_
+  //   .. running ..                               |   - incrementing Thread::suspend_count_ on
+  //   .. running ..                               |     all mutator threads
+  //   .. running ..                               |   - releasing thread_suspend_count_lock_
+  //   .. running ..                               | Block trying to acquire exclusive mutator lock
+  // Poll Thread::suspend_count_ and enter full    |   .. blocked ..
+  // suspend code.                                 |   .. blocked ..
+  // Change state to kSuspended                    |   .. blocked ..
+  // x: Release share on mutator_lock_             | Carry out exclusive access
+  // Acquire thread_suspend_count_lock_            |   .. exclusive ..
+  // while Thread::suspend_count_ > 0              |   .. exclusive ..
+  //   - wait on Thread::resume_cond_              |   .. exclusive ..
+  //     (releases thread_suspend_count_lock_)     |   .. exclusive ..
+  //   .. waiting ..                               | Release mutator_lock_
+  //   .. waiting ..                               | Request thread resumption by:
+  //   .. waiting ..                               |   - acquiring thread_suspend_count_lock_
+  //   .. waiting ..                               |   - decrementing Thread::suspend_count_ on
+  //   .. waiting ..                               |     all mutator threads
+  //   .. waiting ..                               |   - notifying on Thread::resume_cond_
+  //    - re-acquire thread_suspend_count_lock_    |   - releasing thread_suspend_count_lock_
+  // Release thread_suspend_count_lock_            |  .. running ..
+  // Acquire share on mutator_lock_                |  .. running ..
+  //  - This could block but the thread still      |  .. running ..
+  //    has a state of kSuspended and so this      |  .. running ..
+  //    isn't an issue.                            |  .. running ..
+  // Acquire thread_suspend_count_lock_            |  .. running ..
+  //  - we poll here as we're transitioning into   |  .. running ..
+  //    kRunnable and an individual thread suspend |  .. running ..
+  //    request (e.g for debugging) won't try      |  .. running ..
+  //    to acquire the mutator lock (which would   |  .. running ..
+  //    block as we hold the mutator lock). This   |  .. running ..
+  //    poll ensures that if the suspender thought |  .. running ..
+  //    we were suspended by incrementing our      |  .. running ..
+  //    Thread::suspend_count_ and then reading    |  .. running ..
+  //    our state we go back to waiting on         |  .. running ..
+  //    Thread::resume_cond_.                      |  .. running ..
+  // can_go_runnable = Thread::suspend_count_ == 0 |  .. running ..
+  // Release thread_suspend_count_lock_            |  .. running ..
+  // if can_go_runnable                            |  .. running ..
+  //   Change state to kRunnable                   |  .. running ..
+  // else                                          |  .. running ..
+  //   Goto x                                      |  .. running ..
+  //  .. running ..                                |  .. running ..
+  static ReaderWriterMutex* mutator_lock_;
+
+  // Allow reader-writer mutual exclusion on the mark and live bitmaps of the heap.
+  static ReaderWriterMutex* heap_bitmap_lock_ ACQUIRED_AFTER(mutator_lock_);
+
+  // Guards shutdown of the runtime.
+  static Mutex* runtime_shutdown_lock_ ACQUIRED_AFTER(heap_bitmap_lock_);
+
+  // The thread_list_lock_ guards ThreadList::list_. It is also commonly held to stop threads
+  // attaching and detaching.
+  static Mutex* thread_list_lock_ ACQUIRED_AFTER(runtime_shutdown_lock_);
+
+  // Guards breakpoints.
+  static Mutex* breakpoint_lock_ ACQUIRED_AFTER(thread_list_lock_);
+
+  // Guards deoptimization requests.
+  static Mutex* deoptimization_lock_ ACQUIRED_AFTER(breakpoint_lock_);
+
+  // Guards trace requests.
+  static Mutex* trace_lock_ ACQUIRED_AFTER(deoptimization_lock_);
+
+  // Guards profile objects.
+  static Mutex* profiler_lock_ ACQUIRED_AFTER(trace_lock_);
+
+  // Guards lists of classes within the class linker.
+  static ReaderWriterMutex* classlinker_classes_lock_ ACQUIRED_AFTER(profiler_lock_);
+
+  // When declaring any Mutex add DEFAULT_MUTEX_ACQUIRED_AFTER to use annotalysis to check the code
+  // doesn't try to hold a higher level Mutex.
+  #define DEFAULT_MUTEX_ACQUIRED_AFTER ACQUIRED_AFTER(Locks::classlinker_classes_lock_)
+
+  // Guards intern table.
+  static Mutex* intern_table_lock_ ACQUIRED_AFTER(classlinker_classes_lock_);
+
+  // Have an exclusive aborting thread.
+  static Mutex* abort_lock_ ACQUIRED_AFTER(classlinker_classes_lock_);
+
+  // Allow mutual exclusion when manipulating Thread::suspend_count_.
+  // TODO: Does the trade-off of a per-thread lock make sense?
+  static Mutex* thread_suspend_count_lock_ ACQUIRED_AFTER(abort_lock_);
+
+  // One unexpected signal at a time lock.
+  static Mutex* unexpected_signal_lock_ ACQUIRED_AFTER(thread_suspend_count_lock_);
+
+  // Have an exclusive logging thread.
+  static Mutex* logging_lock_ ACQUIRED_AFTER(unexpected_signal_lock_);
+};
+
 }  // namespace art
 
 #endif  // ART_RUNTIME_BASE_MUTEX_H_