* Makerules (cpp-srcs-left): When setting this to run

	cppflags-iterator.mk, must append .c to $(tests) and $(xtests)
	words.  Combine the two loops into one on the concatenated list,
	including those as well as $(test-srcs).

	* elf/dl-minimal.c (__libc_memalign): Guts of malloc moved here,
	since we align here with optimally minimal waste anyway.
	(malloc): Just call that.

	* sysdeps/generic/libc-tls.c (__libc_setup_tls): Set l_tls_offset to
	the right variable.

	* elf/dl-load.c (_dl_map_object_from_fd): Use p_vaddr, not p_offset,
	to compute memory location for l_tls_initimage.
	* elf/rtld.c (_dl_start): Likewise.
	* sysdeps/generic/libc-tls.c (__libc_setup_tls): Likewise.

	* libio/oldiopopen.c: Move #include's before #if SHLIB_COMPAT,
	because the .d file is generated in the non-shared case and
	so fails to catch them otherwise.

	* sysdeps/generic/dl-tls.c (_dl_allocate_tls_storage): New function,
	split out of _dl_allocate_tls.
	(_dl_allocate_tls_init): Likewise.
	(_dl_allocate_tls): Call those.
	* sysdeps/generic/ldsodefs.h: Declare them with attribute_hidden.
	* elf/rtld.c (dl_main): Call them separately instead of calling
	_dl_allocate_tls.  Delay _dl_allocate_tls_init until after relocation
	is finished, so that the initializer data has been relocated before we
	copy it into the main thread's TLS block.

	* sysdeps/generic/dl-tls.c (_dl_allocate_tls): Fix off-by-one error in
	loop conditions, prevented the last used module from being initialized.

3 files changed, 28 insertions, 18 deletions

diff --git a/elf/dl-load.c b/elf/dl-load.c
index 2955bc5173..aabbf21128 100644
--- a/elf/dl-load.c
+++ b/elf/dl-load.c
@@ -949,7 +949,7 @@ _dl_map_object_from_fd (const char *name, int fd, struct filebuf *fbp,
 	      l->l_tls_initimage_size = ph->p_filesz;
 	      /* Since we don't know the load address yet only store the
 		 offset.  We will adjust it later.  */
-	      l->l_tls_initimage = (void *) ph->p_offset;
+	      l->l_tls_initimage = (void *) ph->p_vaddr;
 
 	      /* Assign the next available module ID.  */
 	      l->l_tls_modid = _dl_next_tls_modid ();
diff --git a/elf/dl-minimal.c b/elf/dl-minimal.c
index 2e45fc0a65..893a3b1f2e 100644
--- a/elf/dl-minimal.c
+++ b/elf/dl-minimal.c
@@ -46,8 +46,9 @@ extern unsigned long int weak_function strtoul (const char *nptr,
 						char **endptr, int base);
 
 
+/* Allocate an aligned memory block.  */
 void * weak_function
-malloc (size_t n)
+__libc_memalign (size_t align, size_t n)
 {
 #ifdef MAP_ANON
 #define	_dl_zerofd (-1)
@@ -70,8 +71,8 @@ malloc (size_t n)
     }
 
   /* Make sure the allocation pointer is ideally aligned.  */
-  alloc_ptr = (void *) 0 + (((alloc_ptr - (void *) 0) + sizeof (double) - 1)
-			    & ~(sizeof (double) - 1));
+  alloc_ptr = (void *) 0 + (((alloc_ptr - (void *) 0) + align - 1)
+			    & ~(align - 1));
 
   if (alloc_ptr + n >= alloc_end)
     {
@@ -91,6 +92,12 @@ malloc (size_t n)
   return alloc_last_block;
 }
 
+void * weak_function
+malloc (size_t n)
+{
+  return __libc_memalign (sizeof (double), n);
+}
+
 /* We use this function occasionally since the real implementation may
    be optimized when it can assume the memory it returns already is
    set to NUL.  */
@@ -124,15 +131,6 @@ realloc (void *ptr, size_t n)
   assert (new == ptr);
   return new;
 }
-
-/* Return alligned memory block.  */
-void * weak_function
-__libc_memalign (size_t align, size_t n)
-{
-  void *newp = malloc (n + align - 1);
-
-  return (void *) roundup ((uintptr_t) newp, align);
-}
 
 /* Avoid signal frobnication in setjmp/longjmp.  Keeps things smaller.  */
 
diff --git a/elf/rtld.c b/elf/rtld.c
index 82fe809f70..1472c18235 100644
--- a/elf/rtld.c
+++ b/elf/rtld.c
@@ -191,7 +191,7 @@ _dl_start (void *arg)
 	assert (bootstrap_map.l_tls_blocksize != 0);
 	bootstrap_map.l_tls_initimage_size = phdr[cnt].p_filesz;
 	bootstrap_map.l_tls_initimage = (void *) (bootstrap_map.l_addr
-						  + phdr[cnt].p_offset);
+						  + phdr[cnt].p_vaddr);
 
 	/* We can now allocate the initial TLS block.  This can happen
 	   on the stack.  We'll get the final memory later when we
@@ -1087,13 +1087,10 @@ of this helper program; chances are you did not intend to run this program.\n\
 	 for the thread descriptor.  The memory for the TLS block will
 	 never be freed.  It should be allocated accordingly.  The dtv
 	 array can be changed if dynamic loading requires it.  */
-      tcbp = INTUSE(_dl_allocate_tls) ();
+      tcbp = _dl_allocate_tls_storage ();
       if (tcbp == NULL)
 	_dl_fatal_printf ("\
 cannot allocate TLS data structures for initial thread");
-
-      /* And finally install it for the main thread.  */
-      TLS_INIT_TP (tcbp);
     }
 #endif
 
@@ -1445,6 +1442,21 @@ cannot allocate TLS data structures for initial thread");
      we need it in the memory handling later.  */
   GL(dl_initial_searchlist) = *GL(dl_main_searchlist);
 
+#ifdef USE_TLS
+# ifndef SHARED
+  if (GL(dl_tls_max_dtv_idx) > 0)
+# endif
+    {
+      /* Now that we have completed relocation, the initializer data
+	 for the TLS blocks has its final values and we can copy them
+	 into the main thread's TLS area, which we allocated above.  */
+      _dl_allocate_tls_init (tcbp);
+
+      /* And finally install it for the main thread.  */
+      TLS_INIT_TP (tcbp);
+    }
+#endif
+
   {
     /* Initialize _r_debug.  */
     struct r_debug *r = _dl_debug_initialize (GL(dl_rtld_map).l_addr);