Merge commit 'origin/master' into fedora/master

ChangeLog

@@ -1,3 +1,49 @@
+2009-07-29  Ulrich Drepper  <drepper@redhat.com>
+
+	* math/s_fma.c: Don't define alias if __fma is a macro.
+	* math/s_fmaf.c: Likewise.
+	* sysdeps/x86_64/multiarch/s_fma.c: New file.
+	* sysdeps/x86_64/multiarch/s_fmaf.c: New file.
+	Partially based on a patch by H.J. Lu <hongjiu.lu@intel.com>.
+
+	* sysdeps/x86_64/multiarch/init-arch.h (__get_cpu_features): Declare.
+	(HAS_POPCOUNT, HAS_SSE4_2): Add variants which work outside libc.
+	New macro HAS_FMA.
+	* sysdeps/x86_64/multiarch/init-arch.c (__get_cpu_features): New
+	function.
+	* include/libc-symbols.h (libm_ifunc): Define.
+	* sysdeps/x86_64/multiarch/Versions: New file.
+
+	* sysdeps/x86_64/dl-trampoline.S (_dl_runtime_profile): Improve CFI.
+
+2009-07-28  H.J. Lu  <hongjiu.lu@intel.com>
+
+	* sysdeps/x86_64/dl-trampoline.S: Properly restore AVX registers.
+
+2009-07-29  Ulrich Drepper  <drepper@redhat.com>
+
+	* elf/dl-runtime.c (_dl_fixup): Indicate before _dl_lookup_symbol_x
+	call that registers used in calling conventions need to be preserved.
+	* elf/dl-lookup.c (do_lookup_x): Use RTLD_*_FOREIGN_CALL macros
+	to preserve register content if necessary.
+	* sysdeps/x86_64/dl-trampoline.S (_dl_x86_64_save_sse): New function.
+	(_dl_x86_64_restore_sse): New function.
+	* sysdeps/x86_64/tst-xmmymm.sh: There is now one more function that
+	is allowed to modify xmm/ymm registers.
+
+	* stdio-common/scanf15.c: Undefine _LIBC.  We want to test from an
+	application's perspective.
+	* stdio-common/scanf17.c: Likewise.
+
+2009-07-28  Ulrich Drepper  <drepper@redhat.com>
+
+	* csu/libc-tls.c (__libc_setup_tls) [TLS_TCB_AT_TP]: Don't add TCB
+	size to memsz.
+	(init_static_tls) [TLS_TCB_AT_TP]: Add it to GL(dl_tls_static_size)
+	here.
+	* elf/dl-reloc.c (_dl_try_allocate_static_tls): Compute freebytes in
+	two steps to catch bugs.
+
 2009-07-27  Ulrich Drepper  <drepper@redhat.com>
 
 	* sysdeps/x86_64/tst-xmmymm.sh: Refine testing.  The script now

csu/libc-tls.c

@@ -1,5 +1,5 @@
 /* Initialization code for TLS in statically linked application.
-   Copyright (C) 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc.
+   Copyright (C) 2002-2006, 2009 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
@@ -99,6 +99,9 @@ init_static_tls (size_t memsz, size_t align)
      surplus that permits dynamic loading of modules with IE-model TLS.  */
   GL(dl_tls_static_size) = roundup (memsz + GL(dl_tls_static_size),
 				    TLS_TCB_ALIGN);
+#if TLS_TCB_AT_TP
+  GL(dl_tls_static_size) += TLS_TCB_SIZE;
+#endif
   GL(dl_tls_static_used) = memsz;
   /* The alignment requirement for the static TLS block.  */
   GL(dl_tls_static_align) = align;
@@ -211,9 +214,7 @@ __libc_setup_tls (size_t tcbsize, size_t tcbalign)
 
   memsz = roundup (memsz, align ?: 1);
 
-#if TLS_TCB_AT_TP
-  memsz += tcbsize;
-#elif TLS_DTV_AT_TP
+#if TLS_DTV_AT_TP
   memsz += tcb_offset;
 #endif
 

elf/dl-lookup.c

@@ -380,6 +380,10 @@ do_lookup_x (const char *undef_name, uint_fast32_t new_hash,
 		  if (size * 3 <= tab->n_elements * 4)
 		    {
 		      /* Expand the table.  */
+#ifdef RTLD_CHECK_FOREIGN_CALL
+		      /* This must not happen during runtime relocations.  */
+		      assert (!RTLD_CHECK_FOREIGN_CALL);
+#endif
 		      size_t newsize = _dl_higher_prime_number (size + 1);
 		      struct unique_sym *newentries
 			= calloc (sizeof (struct unique_sym), newsize);
@@ -405,6 +409,11 @@ do_lookup_x (const char *undef_name, uint_fast32_t new_hash,
 		}
 	      else
 		{
+#ifdef RTLD_CHECK_FOREIGN_CALL
+		  /* This must not happen during runtime relocations.  */
+		  assert (!RTLD_CHECK_FOREIGN_CALL);
+#endif
+
 #define INITIAL_NUNIQUE_SYM_TABLE 31
 		  size = INITIAL_NUNIQUE_SYM_TABLE;
 		  entries = calloc (sizeof (struct unique_sym), size);
@@ -600,6 +609,10 @@ add_dependency (struct link_map *undef_map, struct link_map *map, int flags)
 	  unsigned int max
 	    = undef_map->l_reldepsmax ? undef_map->l_reldepsmax * 2 : 10;
 
+#ifdef RTLD_PREPARE_FOREIGN_CALL
+	  RTLD_PREPARE_FOREIGN_CALL;
+#endif
+
 	  newp = malloc (sizeof (*newp) + max * sizeof (struct link_map *));
 	  if (newp == NULL)
 	    {

elf/dl-reloc.c

@@ -61,7 +61,10 @@ _dl_try_allocate_static_tls (struct link_map *map)
   size_t n;
   size_t blsize;
 
-  freebytes = GL(dl_tls_static_size) - GL(dl_tls_static_used) - TLS_TCB_SIZE;
+  freebytes = GL(dl_tls_static_size) - GL(dl_tls_static_used);
+  if (freebytes < TLS_TCB_SIZE)
+    goto fail;
+  freebytes -= TLS_TCB_SIZE;
 
   blsize = map->l_tls_blocksize + map->l_tls_firstbyte_offset;
   if (freebytes < blsize)

elf/dl-runtime.c

@@ -111,13 +111,21 @@ _dl_fixup (
 	  flags |= DL_LOOKUP_GSCOPE_LOCK;
 	}
 
+#ifdef RTLD_ENABLE_FOREIGN_CALL
+      RTLD_ENABLE_FOREIGN_CALL;
+#endif
+
       result = _dl_lookup_symbol_x (strtab + sym->st_name, l, &sym, l->l_scope,
 				    version, ELF_RTYPE_CLASS_PLT, flags, NULL);
 
       /* We are done with the global scope.  */
       if (!RTLD_SINGLE_THREAD_P)
 	THREAD_GSCOPE_RESET_FLAG ();
 
+#ifdef RTLD_FINALIZE_FOREIGN_CALL
+      RTLD_FINALIZE_FOREIGN_CALL;
+#endif
+
       /* Currently result contains the base load address (or link map)
 	 of the object that defines sym.  Now add in the symbol
 	 offset.  */

include/libc-symbols.h

@@ -1,6 +1,6 @@
 /* Support macros for making weak and strong aliases for symbols,
    and for using symbol sets and linker warnings with GNU ld.
-   Copyright (C) 1995-1998, 2000-2006, 2008 Free Software Foundation, Inc.
+   Copyright (C) 1995-1998,2000-2006,2008,2009 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
@@ -845,6 +845,17 @@ for linking")
   }									\
   __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");
+
 #ifdef HAVE_ASM_SET_DIRECTIVE
 # define libc_ifunc_hidden_def1(local, name)				\
     __asm__ (declare_symbol_alias_1_stringify (ASM_GLOBAL_DIRECTIVE)	\

math/s_fma.c

@@ -1,5 +1,5 @@
 /* Compute x * y + z as ternary operation.
-   Copyright (C) 1997, 2001 Free Software Foundation, Inc.
+   Copyright (C) 1997, 2001, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
 
@@ -25,7 +25,9 @@ __fma (double x, double y, double z)
 {
   return (x * y) + z;
 }
+#ifndef __fma
 weak_alias (__fma, fma)
+#endif
 
 #ifdef NO_LONG_DOUBLE
 strong_alias (__fma, __fmal)

math/s_fmaf.c

@@ -1,5 +1,5 @@
 /* Compute x * y + z as ternary operation.
-   Copyright (C) 1997 Free Software Foundation, Inc.
+   Copyright (C) 1997, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
 
@@ -25,4 +25,6 @@ __fmaf (float x, float y, float z)
 {
   return (x * y) + z;
 }
+#ifndef __fmaf
 weak_alias (__fmaf, fmaf)
+#endif

nptl/ChangeLog

@@ -1,3 +1,32 @@
+2009-07-29  Ulrich Drepper  <drepper@redhat.com>
+
+	* sysdeps/x86_64/tls.h (TLS_TCB_ALIGN): Define explicitly to 32.
+
+	* sysdeps/x86_64/tls.h (tcbhead_t): Add room for SSE registers the
+	dynamic linker might have to save.
+	Define RTLD_CHECK_FOREIGN_CALL, RTLD_ENABLE_FOREIGN_CALL,
+	RTLD_PREPARE_FOREIGN_CALL, and RTLD_FINALIZE_FOREIGN_CALL.  Pretty
+	printing.
+
+	* sysdeps/x86_64/tcb-offsets.sym: Add RTLD_SAVESPACE_SSE.
+
+2009-07-28  Ulrich Drepper  <drepper@redhat.com>
+
+	* pthread_mutex_lock.c [NO_INCR] (__pthread_mutex_cond_lock_adjust):
+	New function.
+	* pthreadP.h: Declare __pthread_mutex_cond_lock_adjust.
+	* sysdeps/unix/sysv/linux/pthread-pi-defines.sym: Add ROBUST_BIT.
+	* sysdeps/unix/sysv/linux/x86_64/pthread_cond_broadcast.S: Don't use
+	requeue_pi for robust mutexes.
+	* sysdeps/unix/sysv/linux/x86_64/pthread_cond_signal.S: Likewise.
+	* sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S: Likewise.
+	Don't only skip __pthread_mutex_cond_lock.  Call instead
+	__pthread_mutex_cond_lock_adjust.
+	* sysdeps/unix/sysv/linux/x86_64/pthread_cond_timedwait.S: Likewise.
+
+	* pthread_mutex_unlock.c (__pthread_mutex_unlock_full): Minor
+	optimization of PI mutex handling.
+
 2009-07-27  Ulrich Drepper  <drepper@redhat.com>
 
 	[BZ #10418]

nptl/pthreadP.h

@@ -418,6 +418,8 @@ extern int __pthread_mutex_lock_internal (pthread_mutex_t *__mutex)
      attribute_hidden;
 extern int __pthread_mutex_cond_lock (pthread_mutex_t *__mutex)
      attribute_hidden internal_function;
+extern void __pthread_mutex_cond_lock_adjust (pthread_mutex_t *__mutex)
+     attribute_hidden internal_function;
 extern int __pthread_mutex_unlock (pthread_mutex_t *__mutex);
 extern int __pthread_mutex_unlock_internal (pthread_mutex_t *__mutex)
      attribute_hidden;

nptl/pthread_mutex_lock.c

@@ -1,4 +1,4 @@
-/* Copyright (C) 2002-2007, 2008 Free Software Foundation, Inc.
+/* Copyright (C) 2002-2007, 2008, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
 
@@ -473,3 +473,22 @@ __pthread_mutex_lock_full (pthread_mutex_t *mutex)
 strong_alias (__pthread_mutex_lock, pthread_mutex_lock)
 strong_alias (__pthread_mutex_lock, __pthread_mutex_lock_internal)
 #endif
+
+
+#ifdef NO_INCR
+void
+__pthread_mutex_cond_lock_adjust (mutex)
+     pthread_mutex_t *mutex;
+{
+  assert ((mutex->__data.__kind & PTHREAD_MUTEX_PRIO_INHERIT_NP) != 0);
+  assert ((mutex->__data.__kind & PTHREAD_MUTEX_ROBUST_NORMAL_NP) == 0);
+  assert ((mutex->__data.__kind & PTHREAD_MUTEX_PSHARED_BIT) == 0);
+
+  /* Record the ownership.  */
+  pid_t id = THREAD_GETMEM (THREAD_SELF, tid);
+  mutex->__data.__owner = id;
+
+  if (mutex->__data.__kind == PTHREAD_MUTEX_PI_RECURSIVE_NP)
+    ++mutex->__data.__count;
+}
+#endif

nptl/pthread_mutex_unlock.c

@@ -150,7 +150,7 @@ __pthread_mutex_unlock_full (pthread_mutex_t *mutex, int decr)
       if (--mutex->__data.__count != 0)
 	/* We still hold the mutex.  */
 	return 0;
-      goto continue_pi;
+      goto continue_pi_non_robust;
 
     case PTHREAD_MUTEX_PI_ROBUST_RECURSIVE_NP:
       /* Recursive mutex.  */
@@ -173,7 +173,7 @@ __pthread_mutex_unlock_full (pthread_mutex_t *mutex, int decr)
 	/* We still hold the mutex.  */
 	return 0;
 
-      goto continue_pi;
+      goto continue_pi_robust;
 
     case PTHREAD_MUTEX_PI_ERRORCHECK_NP:
     case PTHREAD_MUTEX_PI_NORMAL_NP:
@@ -195,9 +195,9 @@ __pthread_mutex_unlock_full (pthread_mutex_t *mutex, int decr)
       pi_notrecoverable:
        newowner = PTHREAD_MUTEX_NOTRECOVERABLE;
 
-    continue_pi:
       if ((mutex->__data.__kind & PTHREAD_MUTEX_ROBUST_NORMAL_NP) != 0)
 	{
+	continue_pi_robust:
 	  /* Remove mutex from the list.
 	     Note: robust PI futexes are signaled by setting bit 0.  */
 	  THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
@@ -206,6 +206,7 @@ __pthread_mutex_unlock_full (pthread_mutex_t *mutex, int decr)
 	  DEQUEUE_MUTEX (mutex);
 	}
 
+    continue_pi_non_robust:
       mutex->__data.__owner = newowner;
       if (decr)
 	/* One less user.  */

nptl/sysdeps/unix/sysv/linux/pthread-pi-defines.sym

@@ -3,5 +3,6 @@
 -- These PI macros are used by assembly code.
 
 MUTEX_KIND	offsetof (pthread_mutex_t, __data.__kind)
+ROBUST_BIT	PTHREAD_MUTEX_ROBUST_NORMAL_NP
 PI_BIT		PTHREAD_MUTEX_PRIO_INHERIT_NP
 PS_BIT		PTHREAD_MUTEX_PSHARED_BIT

nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_broadcast.S

@@ -75,8 +75,10 @@ __pthread_cond_broadcast:
 	jne	9f
 
 	/* Requeue to a PI mutex if the PI bit is set.  */
-	testl	$PI_BIT, MUTEX_KIND(%r8)
-	jne	81f
+	movl	MUTEX_KIND(%r8), %eax
+	andl	$(ROBUST_BIT|PI_BIT), %eax
+	cmpl	$PI_BIT, %eax
+	je	81f
 
 	/* Wake up all threads.  */
 #ifdef __ASSUME_PRIVATE_FUTEX

nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_signal.S

@@ -64,8 +64,10 @@ __pthread_cond_signal:
 
 	/* Get the address of the mutex used.  */
 	movq    dep_mutex(%r8), %rcx
-	testl	$PI_BIT, MUTEX_KIND(%rcx)
-	jne	9f
+	movl	MUTEX_KIND(%rcx), %eax
+	andl	$(ROBUST_BIT|PI_BIT), %eax
+	cmpl	$PI_BIT, %eax
+	je	9f
 
 #ifdef __ASSUME_PRIVATE_FUTEX
 	movl	$(FUTEX_WAKE_OP|FUTEX_PRIVATE_FLAG), %esi

nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_timedwait.S

@@ -165,9 +165,12 @@ __pthread_cond_timedwait:
 	je	60f
 
 	movq	dep_mutex(%rdi), %r8
-	/* Requeue to a PI mutex if the PI bit is set.  */
-	testl	$PI_BIT, MUTEX_KIND(%r8)
-	je	61f
+	/* Requeue to a non-robust PI mutex if the PI bit is set and
+	the robust bit is not set.  */
+	movl	MUTEX_KIND(%r8), %eax
+	andl	$(ROBUST_BIT|PI_BIT), %eax
+	cmpl	$PI_BIT, %eax
+	jne	61f
 
 	movl	$(FUTEX_WAIT_REQUEUE_PI|FUTEX_PRIVATE_FLAG), %esi
 	xorl	%eax, %eax
@@ -289,11 +292,10 @@ __pthread_cond_timedwait:
 
 	/* If requeue_pi is used the kernel performs the locking of the
 	   mutex. */
-41:	xorl	%eax, %eax
+41:	movq	16(%rsp), %rdi
 	testl	%r15d, %r15d
-	jnz	63f
+	jnz	64f
 
-	movq	16(%rsp), %rdi
 	callq	__pthread_mutex_cond_lock
 
 63:	testq	%rax, %rax
@@ -316,12 +318,18 @@ __pthread_cond_timedwait:
 
 	retq
 
-	/* Initial locking failed.  */
-31:	cfi_adjust_cfa_offset(4 * 8 + FRAME_SIZE)
+	cfi_adjust_cfa_offset(4 * 8 + FRAME_SIZE)
 	cfi_rel_offset(%r12, FRAME_SIZE + 24)
 	cfi_rel_offset(%r13, FRAME_SIZE + 16)
 	cfi_rel_offset(%r14, FRAME_SIZE + 8)
 	cfi_rel_offset(%r15, FRAME_SIZE)
+
+64:	callq	__pthread_mutex_cond_lock_adjust
+	movq	%r14, %rax
+	jmp	48b
+
+	/* Initial locking failed.  */
+31:
 #if cond_lock != 0
 	addq	$cond_lock, %rdi
 #endif