Revert "Remove stack protector from some interpreter functions"

A performance regression was noticed in go/lem banchmarks for ART.  This
is likely caused by the no_stack_protector attribute preventing inlining
into hot functions.

Reason for revert: performance regression
Reverted Changes:
Ie0c81f97f:Remove stack protector from some interpreter funct...
Iaf565a374:Reset stack guard after fork from Zygote

Bug: 168258494
Change-Id: I018d58759c3a5bb624ea8e7fe09b6b49c9b19b6a
diff --git a/libartbase/base/macros.h b/libartbase/base/macros.h
index c9ad68d..eec73cb 100644
--- a/libartbase/base/macros.h
+++ b/libartbase/base/macros.h
@@ -71,12 +71,10 @@
 #define ALWAYS_INLINE
 #define FLATTEN
 #else
-#define ALWAYS_INLINE  __attribute__ ((always_inline, no_stack_protector))
+#define ALWAYS_INLINE  __attribute__ ((always_inline))
 #define FLATTEN  __attribute__ ((flatten))
 #endif
 
-#define NO_STACK_PROTECTOR __attribute__ ((no_stack_protector))
-
 // clang doesn't like attributes on lambda functions. It would be nice to say:
 //   #define ALWAYS_INLINE_LAMBDA ALWAYS_INLINE
 #define ALWAYS_INLINE_LAMBDA