[compiler-rt] 437b760 - [NFC][lsan] Use LowLevelAllocator to allocate ThreadContext

[Vitaly Buka via llvm-commits]

Author: Vitaly Buka
Date: 2023-04-17T20:05:26-07:00
New Revision: 437b7602e4a998220871de78afcb020b9c14a661

URL: https://github.com/llvm/llvm-project/commit/437b7602e4a998220871de78afcb020b9c14a661
DIFF: https://github.com/llvm/llvm-project/commit/437b7602e4a998220871de78afcb020b9c14a661.diff

LOG: [NFC][lsan] Use LowLevelAllocator to allocate ThreadContext

This is more RAM and CPU efficient than allocating entire page per
context, and this approach is used by other sanitizers already.

With the patch "create_thread_loop2.cpp.tmp 5000" is 30% faster.

Added: 
    

Modified: 
    compiler-rt/lib/lsan/lsan_thread.cpp

Removed: 
    


################################################################################
diff  --git a/compiler-rt/lib/lsan/lsan_thread.cpp b/compiler-rt/lib/lsan/lsan_thread.cpp
index f1075afe9ef6..9da42f32e06b 100644
--- a/compiler-rt/lib/lsan/lsan_thread.cpp
+++ b/compiler-rt/lib/lsan/lsan_thread.cpp
@@ -25,9 +25,12 @@ namespace __lsan {
 
 static ThreadRegistry *thread_registry;
 
+static Mutex mu_for_thread_context;
+static LowLevelAllocator allocator_for_thread_context;
+
 static ThreadContextBase *CreateThreadContext(u32 tid) {
-  void *mem = MmapOrDie(sizeof(ThreadContext), "ThreadContext");
-  return new (mem) ThreadContext(tid);
+  Lock lock(&mu_for_thread_context);
+  return new (allocator_for_thread_context) ThreadContext(tid);
 }
 
 void InitializeThreadRegistry() {