include/sys/cdefs.h: Add __attribute_optimization_barrier__

Add __attribute_optimization_barrier__ to disable inlining and cloning on a
function.  For Clang, expand it to

__attribute__ ((optnone))

Otherwise, expand it to

__attribute__ ((noinline, clone))

Co-Authored-By: H.J. Lu <hjl.tools@gmail.com>
Reviewed-by: Sam James <sam@gentoo.org>

45 files changed, 112 insertions, 101 deletions

diff --git a/benchtests/bench-hash-funcs-kernel.h b/benchtests/bench-hash-funcs-kernel.h
index 7eaa3665b6..3e58fde084 100644
--- a/benchtests/bench-hash-funcs-kernel.h
+++ b/benchtests/bench-hash-funcs-kernel.h
@@ -35,7 +35,7 @@
 #define PRIMITIVE_CAT(x, y) x ## y
 #define CAT(x, y) PRIMITIVE_CAT (x, y)
 
-static double __attribute__ ((noinline, noclone))
+static double __attribute_optimization_barrier__
 CAT (do_one_test_kernel, POSTFIX) (const char *s, size_t len)
 {
 
@@ -58,7 +58,7 @@ CAT (do_one_test_kernel, POSTFIX) (const char *s, size_t len)
   return (double) cur / (double) NFIXED_ITERS;
 }
 
-static double __attribute__ ((noinline, noclone))
+static double __attribute_optimization_barrier__
 CAT (do_rand_test_kernel, POSTFIX) (char const *bufs,
 				    unsigned int const *sizes)
 {
diff --git a/benchtests/bench-hash-funcs.c b/benchtests/bench-hash-funcs.c
index 8dd88c9e39..3d15029278 100644
--- a/benchtests/bench-hash-funcs.c
+++ b/benchtests/bench-hash-funcs.c
@@ -67,7 +67,7 @@ do_one_test (json_ctx_t *json_ctx, size_t len)
   json_element_object_end (json_ctx);
 }
 
-static void __attribute__ ((noinline, noclone))
+static void __attribute_optimization_barrier__
 do_rand_test (json_ctx_t *json_ctx)
 {
   size_t i, sz, offset;
diff --git a/benchtests/bench-strchr.c b/benchtests/bench-strchr.c
index eed6882085..fed3b9b223 100644
--- a/benchtests/bench-strchr.c
+++ b/benchtests/bench-strchr.c
@@ -92,7 +92,7 @@ IMPL (generic_strchrnul, 0)
    branch coming we want to test the case where a potential branch in
    strchr can be used to skip a later mispredict because of the
    relationship between the two branches. */
-static void __attribute__ ((noinline, noclone))
+static void __attribute_optimization_barrier__
 do_one_rand_plus_branch_test (json_ctx_t *json_ctx, impl_t *impl,
                               const CHAR *s, const CHAR *c)
 {
@@ -117,7 +117,7 @@ do_one_rand_plus_branch_test (json_ctx_t *json_ctx, impl_t *impl,
   json_element_double (json_ctx, (double)cur / (double)iters);
 }
 
-static void __attribute__ ((noinline, noclone))
+static void __attribute_optimization_barrier__
 do_one_rand_test (json_ctx_t *json_ctx, impl_t *impl, const CHAR *s,
                   const CHAR *c)
 {
diff --git a/debug/tst-backtrace.h b/debug/tst-backtrace.h
index 53239dad59..0ed38b6892 100644
--- a/debug/tst-backtrace.h
+++ b/debug/tst-backtrace.h
@@ -33,7 +33,7 @@ volatile int x;
 
 /* Use this attribute to prevent inlining, so that all expected frames
    are present.  */
-#define NO_INLINE __attribute__ ((noinline, noclone, weak))
+#define NO_INLINE __attribute__ ((weak)) __attribute_optimization_barrier__
 
 /* Look for a match in SYM from backtrace_symbols to NAME, a fragment
    of a function name.  Ignore the filename before '(', but presume
diff --git a/debug/tst-ssp-1.c b/debug/tst-ssp-1.c
index a1e4fb0630..2c8861ae59 100644
--- a/debug/tst-ssp-1.c
+++ b/debug/tst-ssp-1.c
@@ -21,7 +21,7 @@
 #include <signal.h>
 
 static void
-__attribute__ ((noinline, noclone))
+__attribute_optimization_barrier__
 test (char *foo)
 {
   int i;
diff --git a/elf/tst-p_alignmod-base.c b/elf/tst-p_alignmod-base.c
index d6930b9f16..23b20f12db 100644
--- a/elf/tst-p_alignmod-base.c
+++ b/elf/tst-p_alignmod-base.c
@@ -26,7 +26,7 @@
 int foo __attribute__ ((aligned (ALIGN))) = 1;
 
 bool
-__attribute__ ((weak, noclone, noinline))
+__attribute__ ((weak)) __attribute_optimization_barrier__
 is_aligned_p (void *p, int align)
 {
   return (((uintptr_t) p) & (align - 1)) == 0;
diff --git a/gmon/tst-gmon.c b/gmon/tst-gmon.c
index 9de4e7f85f..eca8246a5c 100644
--- a/gmon/tst-gmon.c
+++ b/gmon/tst-gmon.c
@@ -16,15 +16,17 @@
    License along with the GNU C Library; if not, see
    <https://www.gnu.org/licenses/>.  */
 
+#include <sys/cdefs.h>
+
 /* This program does not use the test harness because we want tight
    control over the call graph.  */
 
-__attribute__ ((noinline, noclone, weak)) void
+__attribute__ ((weak)) __attribute_optimization_barrier__ void
 f1 (void)
 {
 }
 
-__attribute__ ((noinline, noclone, weak)) void
+__attribute__ ((weak)) __attribute_optimization_barrier__ void
 f2 (void)
 {
   f1 ();
@@ -32,7 +34,7 @@ f2 (void)
   asm volatile ("");
 }
 
-__attribute__ ((noinline, noclone, weak)) void
+__attribute__ ((weak)) __attribute_optimization_barrier__ void
 f3 (int count)
 {
   for (int i = 0; i < count; ++i)
diff --git a/gmon/tst-mcount-overflow.c b/gmon/tst-mcount-overflow.c
index 0e60f7e2e6..a681872bf7 100644
--- a/gmon/tst-mcount-overflow.c
+++ b/gmon/tst-mcount-overflow.c
@@ -16,6 +16,8 @@
    License along with the GNU C Library; if not, see
    <https://www.gnu.org/licenses/>.  */
 
+#include <sys/cdefs.h>
+
 /* Program with sufficiently complex, yet pointless, call graph
    that it will trigger an mcount overflow, when you set the
    minarcs/maxarcs tunables to very low values. */
@@ -31,12 +33,12 @@
 
 /* Defines 16 leaf functions named f1_0 to f1_15 */
 #define REP(n) \
-  __attribute__ ((noinline, noclone, weak)) void f1_##n (void) {};
+  __attribute__ ((weak)) __attribute_optimization_barrier__ void f1_##n (void) {};
 REPS
 #undef REP
 
 /* Calls all 16 leaf functions f1_* in succession */
-__attribute__ ((noinline, noclone, weak)) void
+__attribute__ ((weak)) __attribute_optimization_barrier__ void
 f2 (void)
 {
 # define REP(n) f1_##n();
@@ -47,12 +49,12 @@ f2 (void)
 
 /* Defines 16 functions named f2_0 to f2_15, which all just call f2 */
 #define REP(n) \
-  __attribute__ ((noinline, noclone, weak)) void \
+  __attribute__ ((weak)) __attribute_optimization_barrier__ void \
   f2_##n (void) { f2(); PREVENT_TAIL_CALL; };
 REPS
 #undef REP
 
-__attribute__ ((noinline, noclone, weak)) void
+__attribute__ ((weak)) __attribute_optimization_barrier__ void
 f3 (int count)
 {
   for (int i = 0; i < count; ++i)
diff --git a/include/sys/cdefs.h b/include/sys/cdefs.h
index b84ad34a70..a676f75f62 100644
--- a/include/sys/cdefs.h
+++ b/include/sys/cdefs.h
@@ -56,4 +56,11 @@ rtld_hidden_proto (__chk_fail)
 
 #endif /* !defined _ISOMAC */
 
+/*  Prevents a function from being considered for inlining and cloning.  */
+#ifdef __clang__
+# define __attribute_optimization_barrier__ __attribute__ ((optnone))
+#else
+# define __attribute_optimization_barrier__ __attribute__ ((noinline, noclone))
+#endif
+
 #endif
diff --git a/malloc/tst-malloc-thread-exit.c b/malloc/tst-malloc-thread-exit.c
index aec3897087..269abcea8c 100644
--- a/malloc/tst-malloc-thread-exit.c
+++ b/malloc/tst-malloc-thread-exit.c
@@ -42,7 +42,7 @@ static int inner_thread_count = 4;
 static size_t malloc_size = 32;
 
 static void
-__attribute__ ((noinline, noclone))
+__attribute_optimization_barrier__
 unoptimized_free (void *ptr)
 {
   free (ptr);
diff --git a/malloc/tst-malloc-thread-fail.c b/malloc/tst-malloc-thread-fail.c
index f698871876..56553dd299 100644
--- a/malloc/tst-malloc-thread-fail.c
+++ b/malloc/tst-malloc-thread-fail.c
@@ -33,7 +33,7 @@
 
 /* Wrapper for calloc with an optimization barrier.  */
 static void *
-__attribute__ ((noinline, noclone))
+__attribute_optimization_barrier__
 allocate_zeroed (size_t a, size_t b)
 {
   return calloc (a, b);
diff --git a/malloc/tst-mallocstate.c b/malloc/tst-mallocstate.c
index f72e1a1b28..3664fcda01 100644
--- a/malloc/tst-mallocstate.c
+++ b/malloc/tst-mallocstate.c
@@ -366,7 +366,7 @@ full_heap_check (void)
 }
 
 /* Used as an optimization barrier to force a heap allocation.  */
-__attribute__ ((noinline, noclone))
+__attribute_optimization_barrier__
 static void
 my_free (void *ptr)
 {
diff --git a/misc/tst-syscalls.c b/misc/tst-syscalls.c
index d060b125c9..ac1e1014e1 100644
--- a/misc/tst-syscalls.c
+++ b/misc/tst-syscalls.c
@@ -37,7 +37,7 @@ struct Array
 
 static int error_count;
 
-__attribute__ ((noclone, noinline))
+__attribute_optimization_barrier__
 struct Array
 allocate (size_t bytes)
 {
@@ -52,7 +52,7 @@ allocate (size_t bytes)
   return __extension__ (struct Array) {bytes, p};
 }
 
-__attribute__ ((noclone, noinline))
+__attribute_optimization_barrier__
 void
 deallocate (struct Array b)
 {
@@ -66,7 +66,7 @@ deallocate (struct Array b)
     }
 }
 
-__attribute__ ((noclone, noinline))
+__attribute_optimization_barrier__
 void *
 do_mmap (void *addr, size_t length)
 {
@@ -74,7 +74,7 @@ do_mmap (void *addr, size_t length)
 	       MAP_PRIVATE | MAP_ANON, -1, 0);
 }
 
-__attribute__ ((noclone, noinline))
+__attribute_optimization_barrier__
 void *
 reallocate (struct Array b)
 {
@@ -86,7 +86,7 @@ reallocate (struct Array b)
   return NULL;
 }
 
-__attribute__ ((noclone, noinline))
+__attribute_optimization_barrier__
 void
 protect (struct Array b)
 {
@@ -104,7 +104,7 @@ protect (struct Array b)
     }
 }
 
-__attribute__ ((noclone, noinline))
+__attribute_optimization_barrier__
 ssize_t
 do_read (int fd, void *ptr, struct Array b)
 {
@@ -116,7 +116,7 @@ do_read (int fd, void *ptr, struct Array b)
   return 0;
 }
 
-__attribute__ ((noclone, noinline))
+__attribute_optimization_barrier__
 ssize_t
 do_write (int fd, void *ptr, struct Array b)
 {
diff --git a/nptl/tst-minstack-throw.cc b/nptl/tst-minstack-throw.cc
index b729091ec4..ff0d3d39da 100644
--- a/nptl/tst-minstack-throw.cc
+++ b/nptl/tst-minstack-throw.cc
@@ -24,7 +24,7 @@
 #include <support/xthread.h>
 
 /* Throw a std::runtime_exception.  */
-__attribute__ ((noinline, noclone, weak))
+__attribute__ ((weak)) __attribute_optimization_barrier__
 void
 do_throw_exception ()
 {
@@ -38,17 +38,17 @@ struct class_with_destructor
   ~class_with_destructor ();
 };
 
-__attribute__ ((noinline, noclone, weak))
+__attribute__ ((weak)) __attribute_optimization_barrier__
 class_with_destructor::class_with_destructor ()
 {
 }
 
-__attribute__ ((noinline, noclone, weak))
+__attribute__ ((weak)) __attribute_optimization_barrier__
 class_with_destructor::~class_with_destructor ()
 {
 }
 
-__attribute__ ((noinline, noclone, weak))
+__attribute__ ((weak)) __attribute_optimization_barrier__
 void
 function_with_destructed_object ()
 {
diff --git a/nptl/tst-thread-exit-clobber.cc b/nptl/tst-thread-exit-clobber.cc
index b3b6989268..9a50848ef8 100644
--- a/nptl/tst-thread-exit-clobber.cc
+++ b/nptl/tst-thread-exit-clobber.cc
@@ -73,7 +73,7 @@ enum { no_check = -1 };
 
 /* Check that VALUE is the magic value for INDEX, behind a compiler
    barrier.  */
-__attribute__ ((noinline, noclone, weak))
+__attribute__ ((weak)) __attribute_optimization_barrier__
 void
 check_magic (int index, unsigned int value)
 {
@@ -103,7 +103,7 @@ check_magic (int index, unsigned int value)
 
 /* Check that VALUE is the magic value for INDEX, behind a compiler
    barrier.  Double variant.  */
-__attribute__ ((noinline, noclone, weak))
+__attribute__ ((weak)) __attribute_optimization_barrier__
 void
 check_magic (int index, double value)
 {
@@ -