[PATCH] Make __get_cpu_features hidden

["H.J. Lu" <hongjiu dot lu at intel dot com>]

Hi,

As shown in

http://www.sourceware.org/bugzilla/show_bug.cgi?id=14955

we can't call the external __get_cpu_features while relocating a DSO.
This patch makes __get_cpu_features hidden and includes it all shared
objects which use it.  Currently only libc and libm use __get_cpu_features.
OK for 2.18?


H.J.
---
 include/libc-symbols.h                   | 12 +-----------
 sysdeps/i386/i686/fpu/multiarch/Makefile |  1 +
 sysdeps/i386/i686/multiarch/Makefile     |  3 +++
 sysdeps/i386/i686/multiarch/Versions     |  5 -----
 sysdeps/wordsize-32/symbol-hacks.h       |  5 +++--
 sysdeps/x86/Makefile                     | 10 ++++++++++
 sysdeps/x86/tst-ifunc-cpu1-main.c        | 10 ++++++++++
 sysdeps/x86/tst-ifunc-cpu1-mod.c         | 24 ++++++++++++++++++++++++
 sysdeps/x86_64/fpu/multiarch/Makefile    |  1 +
 sysdeps/x86_64/multiarch/Makefile        |  2 ++
 sysdeps/x86_64/multiarch/Versions        |  5 -----
 sysdeps/x86_64/multiarch/init-arch.c     | 14 +++-----------
 sysdeps/x86_64/multiarch/init-arch.h     | 25 ++++++++++++-------------
 14 files changed, 100 insertions(+), 47 deletions(-)
 create mode 100644 ChangeLog.bz14955
 delete mode 100644 sysdeps/i386/i686/multiarch/Versions
 create mode 100644 sysdeps/x86/tst-ifunc-cpu1-main.c
 create mode 100644 sysdeps/x86/tst-ifunc-cpu1-mod.c
 delete mode 100644 sysdeps/x86_64/multiarch/Versions

2012-12-14  H.J. Lu  <hongjiu.lu@intel.com>

	[BZ #14955]
	* include/libc-symbols.h (libc_ifunc): Remove INIT_ARCH.
	(libm_ifunc): Defined to libc_ifunc.
	* sysdeps/i386/i686/fpu/multiarch/Makefile (libm-sysdep_routines):
	Add init-arch.
	* sysdeps/x86_64/fpu/multiarch/Makefile (libm-sysdep_routines):
	Likewise.
	* sysdeps/i386/i686/multiarch/Makefile (CPPFLAGS-init-arch.c): New
	macro.  Defined to -Din_init_arch_c.
	($(objpfx)test-multiarch): New dependency.
	* sysdeps/i386/i686/multiarch/Versions: Removed.
	* sysdeps/x86_64/multiarch/Versions: Likewise.
	* sysdeps/wordsize-32/symbol-hacks.h: Skip if in_init_arch_c is
	defined.
	* sysdeps/x86/Makefile (tests): Add tst-ifunc-cpu1-main.
	(modules-names): Add tst-ifunc-cpu1-mod.
	(tst-ifunc-cpu1-mod.so-no-z-defs): New macro.
	($(objpfx)tst-ifunc-cpu1-mod.so): New dependency.
	($(objpfx)tst-ifunc-cpu1-main): Likewise.
	* sysdeps/x86/tst-ifunc-cpu1-main.c: New file.
	* sysdeps/x86/tst-ifunc-cpu1-mod.c: Likewise.
	* sysdeps/x86_64/multiarch/Makefile ($(objpfx)test-multiarch): New
	dependency.
	* sysdeps/x86_64/multiarch/init-arch.c (__always_inline): Defined
	to empty.
	(__get_cpu_features): Removed.
	* sysdeps/x86_64/multiarch/init-arch.h (INIT_ARCH): Removed.
	(__get_cpu_features): Defined as an inline function.

diff --git a/include/libc-symbols.h b/include/libc-symbols.h
index a626d59..1cecaca 100644
--- a/include/libc-symbols.h
+++ b/include/libc-symbols.h
@@ -751,22 +751,12 @@ for linking")
   extern void *name##_ifunc (void) __asm__ (#name);			\
   void *name##_ifunc (void)						\
   {									\
-    INIT_ARCH ();							\
     __typeof (name) *res = expr;					\
     return res;								\
   }									\
   __asm__ (".type " #name ", %gnu_indirect_function");
 
-/* The body of the function is supposed to use __get_cpu_features
-   which will, if necessary, initialize the data first.  */
-#define libm_ifunc(name, expr)						\
-  extern void *name##_ifunc (void) __asm__ (#name);			\
-  void *name##_ifunc (void)						\
-  {									\
-    __typeof (name) *res = expr;					\
-    return res;								\
-  }									\
-  __asm__ (".type " #name ", %gnu_indirect_function");
+#define libm_ifunc(name, expr) libc_ifunc (name, (expr))
 
 #ifdef HAVE_ASM_SET_DIRECTIVE
 # define libc_ifunc_hidden_def1(local, name)				\
diff --git a/sysdeps/i386/i686/fpu/multiarch/Makefile b/sysdeps/i386/i686/fpu/multiarch/Makefile
index 7d90892..92ecb04 100644
--- a/sysdeps/i386/i686/fpu/multiarch/Makefile
+++ b/sysdeps/i386/i686/fpu/multiarch/Makefile
@@ -1,4 +1,5 @@
 ifeq ($(subdir),math)
+libm-sysdep_routines += init-arch
 libm-sysdep_routines += e_expf-sse2 e_expf-ia32 s_sinf-sse2 s_cosf-sse2 \
                         s_sincosf-sse2
 endif
diff --git a/sysdeps/i386/i686/multiarch/Makefile b/sysdeps/i386/i686/multiarch/Makefile
index 8946bfa..7126e23 100644
--- a/sysdeps/i386/i686/multiarch/Makefile
+++ b/sysdeps/i386/i686/multiarch/Makefile
@@ -1,7 +1,10 @@
 ifeq ($(subdir),csu)
 aux += init-arch
+CPPFLAGS-init-arch.c = -Din_init_arch_c
 tests += test-multiarch
 gen-as-const-headers += ifunc-defines.sym
+
+$(objpfx)test-multiarch: $(common-objpfx)csu/init-arch.os
 endif
 
 ifeq ($(subdir),string)
diff --git a/sysdeps/i386/i686/multiarch/Versions b/sysdeps/i386/i686/multiarch/Versions
deleted file mode 100644
index 59b185a..0000000
--- a/sysdeps/i386/i686/multiarch/Versions
+++ /dev/null
@@ -1,5 +0,0 @@
-libc {
-  GLIBC_PRIVATE {
-    __get_cpu_features;
-  }
-}
diff --git a/sysdeps/wordsize-32/symbol-hacks.h b/sysdeps/wordsize-32/symbol-hacks.h
index 52a14fc..9814486 100644
--- a/sysdeps/wordsize-32/symbol-hacks.h
+++ b/sysdeps/wordsize-32/symbol-hacks.h
@@ -1,5 +1,5 @@
 /* Hacks needed for symbol manipulation.
-   Copyright (C) 2004 Free Software Foundation, Inc.
+   Copyright (C) 2004-2012 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -21,7 +21,8 @@
    therefore we get PLTs.  Unnecessarily so.  Changing gcc is a big
    task which might not be worth it so we play tricks with the
    assembler.  */
-#if !defined __ASSEMBLER__ && !defined in_divdi3_c && !defined NOT_IN_libc && defined SHARED
+#if !defined __ASSEMBLER__ && !defined in_divdi3_c \
+    && !defined in_init_arch_c && !defined NOT_IN_libc && defined SHARED
 asm ("__divdi3 = __divdi3_internal");
 asm ("__udivdi3 = __udivdi3_internal");
 asm ("__moddi3 = __moddi3_internal");
diff --git a/sysdeps/x86/Makefile b/sysdeps/x86/Makefile
index 3e87a77..e722a34 100644
--- a/sysdeps/x86/Makefile
+++ b/sysdeps/x86/Makefile
@@ -6,4 +6,14 @@ tests: $(objpfx)tst-xmmymm.out
 $(objpfx)tst-xmmymm.out: ../sysdeps/x86/tst-xmmymm.sh $(objpfx)ld.so
 	@echo "Checking ld.so for SSE register use.  This will take a few seconds..."
 	$(SHELL) $< $(objpfx) '$(NM)' '$(OBJDUMP)' '$(READELF)' > $@
+
+ifeq (yesyes,$(build-shared)$(multi-arch))
+tests += tst-ifunc-cpu1-main
+modules-names += tst-ifunc-cpu1-mod
+
+tst-ifunc-cpu1-mod.so-no-z-defs = yes
+
+$(objpfx)tst-ifunc-cpu1-mod.so: $(common-objpfx)csu/init-arch.os
+$(objpfx)tst-ifunc-cpu1-main: $(objpfx)tst-ifunc-cpu1-mod.so
+endif
 endif
diff --git a/sysdeps/x86/tst-ifunc-cpu1-main.c b/sysdeps/x86/tst-ifunc-cpu1-main.c
new file mode 100644
index 0000000..37c16e2
--- /dev/null
+++ b/sysdeps/x86/tst-ifunc-cpu1-main.c
@@ -0,0 +1,10 @@
+/* Test function pointer to local STT_GNU_IFUNC function.  */
+
+extern void (*foo_ptr) (void);
+
+int
+main (void)
+{
+  foo_ptr ();
+  return 0;
+}
diff --git a/sysdeps/x86/tst-ifunc-cpu1-mod.c b/sysdeps/x86/tst-ifunc-cpu1-mod.c
new file mode 100644
index 0000000..08ff30f
--- /dev/null
+++ b/sysdeps/x86/tst-ifunc-cpu1-mod.c
@@ -0,0 +1,24 @@
+#include <init-arch.h>
+
+static void
+one (void)
+{
+}
+
+static void
+two (void)
+{
+}
+
+void * foo_ifunc (void) __asm__ ("foo") attribute_hidden;
+__asm__(".type foo, %gnu_indirect_function");
+
+void *
+foo_ifunc (void)
+{
+  const struct cpu_features * cpu = __get_cpu_features ();
+  return cpu->max_cpuid > 1 ? two : one;
+}
+
+extern void foo (void) attribute_hidden;
+void (*foo_ptr) (void) = foo;
diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile
index 12b0526..2cf07bf 100644
--- a/sysdeps/x86_64/fpu/multiarch/Makefile
+++ b/sysdeps/x86_64/fpu/multiarch/Makefile
@@ -1,4 +1,5 @@
 ifeq ($(subdir),math)
+libm-sysdep_routines += init-arch
 libm-sysdep_routines += s_floor-c s_ceil-c s_floorf-c s_ceilf-c \
 			s_rint-c s_rintf-c s_nearbyint-c s_nearbyintf-c
 
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index dd6c27d..948f32f 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -2,6 +2,8 @@ ifeq ($(subdir),csu)
 aux += init-arch
 tests += test-multiarch
 gen-as-const-headers += ifunc-defines.sym
+
+$(objpfx)test-multiarch: $(common-objpfx)csu/init-arch.os
 endif
 
 ifeq ($(subdir),string)
diff --git a/sysdeps/x86_64/multiarch/Versions b/sysdeps/x86_64/multiarch/Versions
deleted file mode 100644
index 59b185a..0000000
--- a/sysdeps/x86_64/multiarch/Versions
+++ /dev/null
@@ -1,5 +0,0 @@
-libc {
-  GLIBC_PRIVATE {
-    __get_cpu_features;
-  }
-}
diff --git a/sysdeps/x86_64/multiarch/init-arch.c b/sysdeps/x86_64/multiarch/init-arch.c
index fed5ab8..ff2331d 100644
--- a/sysdeps/x86_64/multiarch/init-arch.c
+++ b/sysdeps/x86_64/multiarch/init-arch.c
@@ -19,6 +19,9 @@
 
 #include <atomic.h>
 #include <cpuid.h>
+/* Define __get_cpu_features.  */
+#undef __always_inline
+#define __always_inline
 #include "init-arch.h"
 
 
@@ -170,14 +173,3 @@ __init_cpu_features (void)
   atomic_write_barrier ();
   __cpu_features.kind = kind;
 }
-
-#undef __get_cpu_features
-
-const struct cpu_features *
-__get_cpu_features (void)
-{
-  if (__cpu_features.kind == arch_kind_unknown)
-    __init_cpu_features ();
-
-  return &__cpu_features;
-}
diff --git a/sysdeps/x86_64/multiarch/init-arch.h b/sysdeps/x86_64/multiarch/init-arch.h
index f33f1c8..3ed3236 100644
--- a/sysdeps/x86_64/multiarch/init-arch.h
+++ b/sysdeps/x86_64/multiarch/init-arch.h
@@ -103,19 +103,18 @@ extern struct cpu_features
 
 
 extern void __init_cpu_features (void) attribute_hidden;
-# define INIT_ARCH() \
-  do							\
-    if (__cpu_features.kind == arch_kind_unknown)	\
-      __init_cpu_features ();				\
-  while (0)
-
-/* Used from outside libc.so to get access to the CPU features structure.  */
-extern const struct cpu_features *__get_cpu_features (void)
-     __attribute__ ((const));
-
-# ifndef NOT_IN_libc
-#  define __get_cpu_features()	(&__cpu_features)
-# endif
+
+/* Implicitly call __init_cpu_features before accessing the CPU features
+   structure.  */
+
+extern __always_inline attribute_hidden
+const struct cpu_features *
+__get_cpu_features (void)
+{
+  if (__cpu_features.kind == arch_kind_unknown)
+    __init_cpu_features ();
+  return &__cpu_features;
+}
 
 # define HAS_CPU_FEATURE(idx, reg, bit) \
   ((__get_cpu_features ()->cpuid[idx].reg & (bit)) != 0)
-- 
1.7.11.7