[libc] avoid cmpxchg on the fastpath of callonce (#91748)

Avoid `cmpxchg` operation if the function has already been called.
The destination operand of `cmpxchg` may receive a write cycle without
regard to the result of the comparison

GitOrigin-RevId: b8f4f39d3d43d348e31fc5709b72e1f51392e52d
diff --git a/src/__support/threads/linux/callonce.cpp b/src/__support/threads/linux/callonce.cpp
index 1c29db5..b48a514 100644
--- a/src/__support/threads/linux/callonce.cpp
+++ b/src/__support/threads/linux/callonce.cpp
@@ -7,6 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "src/__support/threads/callonce.h"
+#include "src/__support/macros/optimization.h"
 #include "src/__support/threads/linux/futex_utils.h"
 
 namespace LIBC_NAMESPACE {
@@ -21,6 +22,12 @@
 
   FutexWordType not_called = NOT_CALLED;
 
+  // Avoid cmpxchg operation if the function has already been called.
+  // The destination operand of cmpxchg may receive a write cycle without
+  // regard to the result of the comparison
+  if (LIBC_LIKELY(futex_word->load(cpp::MemoryOrder::RELAXED) == FINISH))
+    return 0;
+
   // The call_once call can return only after the called function |func|
   // returns. So, we use futexes to synchronize calls with the same flag value.
   if (futex_word->compare_exchange_strong(not_called, START)) {