tsan: split thread into logical and physical state

Currently ThreadState holds both logical state (required for race-detection algorithm, user-visible)
and physical state (various caches, most notably malloc cache). Move physical state in a new
Process entity. Besides just being the right thing from abstraction point of view, this solves several
problems:
1. Cache everything on P level in Go. Currently we cache on a mix of goroutine and OS thread levels.
This unnecessary increases memory consumption.
2. Properly handle free operations in Go. Frees are issue by GC which don't have goroutine context.
As the result we could not do anything more than just clearing shadow. For example, we leaked
sync objects and heap block descriptors.
3. This will allow to get rid of libc malloc in Go (now we have Processor context for internal allocator cache).
This in turn will allow to get rid of dependency on libc entirely.
4. Potentially we can make Processor per-CPU in C++ mode instead of per-thread, which will
reduce resource consumption.
The distinction between Thread and Processor is currently used only by Go, C++ creates Processor per OS thread,
which is equivalent to the current scheme.



git-svn-id: https://llvm.org/svn/llvm-project/compiler-rt/trunk@262037 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 28 insertions, 7 deletions

diff --git a/lib/tsan/rtl/tsan_rtl.h b/lib/tsan/rtl/tsan_rtl.h
index 0b59a507c..6e812c496 100644
--- a/lib/tsan/rtl/tsan_rtl.h
+++ b/lib/tsan/rtl/tsan_rtl.h
@@ -325,6 +325,26 @@ struct JmpBuf {
   uptr *shadow_stack_pos;
 };
 
+// A Processor represents a physical thread, or a P for Go.
+// It is used to store internal resources like allocate cache, and does not
+// participate in race-detection logic (invisible to end user).
+// In C++ it is tied to an OS thread just like ThreadState, however ideally
+// it should be tied to a CPU (this way we will have fewer allocator caches).
+// In Go it is tied to a P, so there are significantly fewer Processor's than
+// ThreadState's (which are tied to Gs).
+// A ThreadState must be wired with a Processor to handle events.
+struct Processor {
+  ThreadState *thr; // currently wired thread, or nullptr
+#ifndef SANITIZER_GO
+  AllocatorCache alloc_cache;
+  InternalAllocatorCache internal_alloc_cache;
+#endif
+  DenseSlabAllocCache block_cache;
+  DenseSlabAllocCache sync_cache;
+  DenseSlabAllocCache clock_cache;
+  DDPhysicalThread *dd_pt;
+};
+
 // This struct is stored in TLS.
 struct ThreadState {
   FastState fast_state;
@@ -360,8 +380,6 @@ struct ThreadState {
   MutexSet mset;
   ThreadClock clock;
 #ifndef SANITIZER_GO
-  AllocatorCache alloc_cache;
-  InternalAllocatorCache internal_alloc_cache;
   Vector<JmpBuf> jmp_bufs;
   int ignore_interceptors;
 #endif
@@ -385,16 +403,14 @@ struct ThreadState {
 #if SANITIZER_DEBUG && !SANITIZER_GO
   InternalDeadlockDetector internal_deadlock_detector;
 #endif
-  DDPhysicalThread *dd_pt;
   DDLogicalThread *dd_lt;
 
+  // Current wired Processor, or nullptr. Required to handle any events.
+  Processor *proc;
+
   atomic_uintptr_t in_signal_handler;
   ThreadSignalContext *signal_ctx;
 
-  DenseSlabAllocCache block_cache;
-  DenseSlabAllocCache sync_cache;
-  DenseSlabAllocCache clock_cache;
-
 #ifndef SANITIZER_GO
   u32 last_sleep_stack_id;
   ThreadClock last_sleep_clock;
@@ -683,6 +699,11 @@ void ThreadSetName(ThreadState *thr, const char *name);
 int ThreadCount(ThreadState *thr);
 void ProcessPendingSignals(ThreadState *thr);
 
+Processor *ProcCreate();
+void ProcDestroy(Processor *proc);
+void ProcWire(Processor *proc, ThreadState *thr);
+void ProcUnwire(Processor *proc, ThreadState *thr);
+
 void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
                  bool rw, bool recursive, bool linker_init);
 void MutexDestroy(ThreadState *thr, uptr pc, uptr addr);