Update.

2003-04-22  Jakub Jelinek  <jakub@redhat.com>

	* include/link.h (NO_TLS_OFFSET): Define to 0 if not defined.
	* elf/dl-close.c (_dl_close): Use NO_TLS_OFFSET.
	* elf/dl-object.c (_dl_new_object): Initialize l_tls_offset to
	NO_TLS_OFFSET.
	* elf/rtld.c (_dl_start_final, _dl_start): Likewise.
	* elf/dl-reloc.c (CHECK_STATIC_TLS): Use NO_TLS_OFFSET.
	* sysdeps/generic/dl-tls.c (_dl_allocate_tls_init): Likewise.
	* sysdeps/powerpc/dl-tls.h (TLS_TPREL_VALUE): Don't subtract
	TLS_TCB_SIZE.

33 files changed, 225 insertions, 113 deletions

diff --git a/ChangeLog b/ChangeLog
index 584e9b5..e432c0f 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,15 @@
+2003-04-22  Jakub Jelinek  <jakub@redhat.com>
+
+	* include/link.h (NO_TLS_OFFSET): Define to 0 if not defined.
+	* elf/dl-close.c (_dl_close): Use NO_TLS_OFFSET.
+	* elf/dl-object.c (_dl_new_object): Initialize l_tls_offset to
+	NO_TLS_OFFSET.
+	* elf/rtld.c (_dl_start_final, _dl_start): Likewise.
+	* elf/dl-reloc.c (CHECK_STATIC_TLS): Use NO_TLS_OFFSET.
+	* sysdeps/generic/dl-tls.c (_dl_allocate_tls_init): Likewise.
+	* sysdeps/powerpc/dl-tls.h (TLS_TPREL_VALUE): Don't subtract
+	TLS_TCB_SIZE.
+
 2003-04-24  Ulrich Drepper  <drepper@redhat.com>
 
 	* nss/getent.c: Implement alternative host database lookup via
diff --git a/elf/dl-close.c b/elf/dl-close.c
index 5a61ad0..1e6dd12 100644
--- a/elf/dl-close.c
+++ b/elf/dl-close.c
@@ -365,7 +365,7 @@ _dl_close (void *_map)
 		/* All dynamically loaded modules with TLS are unloaded.  */
 		GL(dl_tls_max_dtv_idx) = GL(dl_tls_static_nelem);
 
-	      if (imap->l_tls_offset != 0)
+	      if (imap->l_tls_offset != NO_TLS_OFFSET)
 		{
 		  /* Collect a contiguous chunk built from the objects in
 		     this search list, going in either direction.  When the
diff --git a/elf/dl-object.c b/elf/dl-object.c
index 6196cd7..a33b40a 100644
--- a/elf/dl-object.c
+++ b/elf/dl-object.c
@@ -53,6 +53,10 @@ _dl_new_object (char *realname, const char *libname, int type,
   new->l_name = realname;
   new->l_type = type;
   new->l_loader = loader;
+#if defined USE_TLS && NO_TLS_OFFSET != 0
+  new->l_tls_offset = NO_TLS_OFFSET;
+#endif
+
   /* new->l_global = 0;	We use calloc therefore not necessary.  */
 
   /* Use the 'l_scope_mem' array by default for the the 'l_scope'
diff --git a/elf/dl-reloc.c b/elf/dl-reloc.c
index 5d8dc0d..c7f1f34 100644
--- a/elf/dl-reloc.c
+++ b/elf/dl-reloc.c
@@ -209,7 +209,7 @@ _dl_relocate_object (struct link_map *l, struct r_scope_elem *scope[],
        intended to produce.  */
 #define CHECK_STATIC_TLS(map, sym_map)					      \
     do {								      \
-      if (__builtin_expect ((sym_map)->l_tls_offset == 0, 0))		      \
+      if (__builtin_expect ((sym_map)->l_tls_offset == NO_TLS_OFFSET, 0))     \
 	_dl_allocate_static_tls (sym_map);				      \
     } while (0)
 
diff --git a/elf/rtld.c b/elf/rtld.c
index bd66149..9d2a94c 100644
--- a/elf/rtld.c
+++ b/elf/rtld.c
@@ -233,6 +233,9 @@ _dl_start_final (void *arg, struct dl_start_final_info *info)
   GL(dl_rtld_map).l_tls_modid = 1;
 # else
   assert (info->l.l_tls_modid == 0);
+#  if NO_TLS_OFFSET != 0
+  GL(dl_rtld_map).l_tls_offset = NO_TLS_OFFSET;
+#  endif
 # endif
 
 #endif
@@ -316,6 +319,10 @@ _dl_start (void *arg)
   bootstrap_map.l_ld = (void *) bootstrap_map.l_addr + elf_machine_dynamic ();
   elf_get_dynamic_info (&bootstrap_map);
 
+#if defined USE_TLS && NO_TLS_OFFSET != 0
+  bootstrap_map.l_tls_offset = NO_TLS_OFFSET;
+#endif
+
 #if USE___THREAD
   /* Get the dynamic linker's own program header.  First we need the ELF
      file header.  The `_begin' symbol created by the linker script points
diff --git a/include/link.h b/include/link.h
index 5113ab7..cc2387b 100644
--- a/include/link.h
+++ b/include/link.h
@@ -268,6 +268,9 @@ struct link_map
     size_t l_tls_blocksize;
     /* Alignment requirement of the TLS block.  */
     size_t l_tls_align;
+# ifndef NO_TLS_OFFSET
+#  define NO_TLS_OFFSET	0
+# endif
     /* For objects present at startup time: offset in the static TLS block.  */
     ptrdiff_t l_tls_offset;
     /* Index of the module in the dtv array.  */
diff --git a/linuxthreads/ChangeLog b/linuxthreads/ChangeLog
index 59204cb..b8cd228 100644
--- a/linuxthreads/ChangeLog
+++ b/linuxthreads/ChangeLog
@@ -1,3 +1,38 @@
+2003-04-22  Jakub Jelinek  <jakub@redhat.com>
+
+	* descr.h (p_multiple_threads): Define as function-like macro.
+	(struct _pthread_descr_struct) [TLS_MULTIPLE_THREADS_IN_TCB]:
+	Move multiple_threads to last int in the structure.
+	* pthread.c (__pthread_initialize_manager): Use p_multiple_threads
+	macro.  Subtract TLS_PRE_TCB_SIZE bytes from tcbp to get to descr.
+	* manager.c (pthread_handle_create): Use p_multiple_threads macro.
+	Subtract or add TLS_PRE_TCB_SIZE instead of sizeof (pthread_descr).
+	(pthread_free): Add TLS_PRE_TCB_SIZE instead of
+	sizeof (pthread_descr).
+	* sysdeps/powerpc/tls.h: Don't include tcb-offsets.h.
+	(TLS_INIT_TCB_SIZE, TLS_TCB_SIZE): Define to 0.
+	(TLS_INIT_TCB_ALIGN, TLS_TCB_ALIGN): Define to alignment of
+	pthread_descr.
+	(TLS_PRE_TCB_SIZE): Increase to cover tcbhead_t preceeded by pad
+	to TLS_TCB_ALIGN.
+	(INSTALL_DTV, GET_DTV, THREAD_DTV): tcbhead_t is immediately before
+	tcbp.
+	(TLS_INIT_TP, THREAD_SELF, INIT_THREAD_SELF): Don't add TLS_TCB_SIZE
+	unneccessarily.
+	(NO_TLS_OFFSET): Define.
+	* sysdeps/powerpc/powerpc32/tcb-offsets.sym: New file.
+	* sysdeps/powerpc/tcb-offsets.sym: Removed.
+	* sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h: Include
+	tcb-offsets.h if __ASSEMBLER__.
+	(SINGLE_THREAD_P): Use p_multiple_threads macro.
+	* sysdeps/ia64/tcb-offsets.sym (MULTIPLE_THREADS): Define to
+	-sizeof(int).
+	* sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h (SINGLE_THREAD_P):
+	Use p_multiple_threads macro.
+	* sysdeps/unix/sysv/linux/sh/sysdep-cancel.h (SINGLE_THREAD_P):
+	Likewise.
+	* sysdeps/sh/tcb-offsets.sym (MULTIPLE_THREADS): Likewise.
+
 2003-04-22  Roland McGrath  <roland@redhat.com>
 
 	* Makeconfig (shared-thread-library): Reverse link order to work
diff --git a/linuxthreads/descr.h b/linuxthreads/descr.h
index 2dcb376..dd496f6 100644
--- a/linuxthreads/descr.h
+++ b/linuxthreads/descr.h
@@ -120,17 +120,14 @@ struct _pthread_descr_struct
       union dtv *dtvp;
       pthread_descr self;	/* Pointer to this structure */
       int multiple_threads;
-#ifdef NEED_DL_SYSINFO
+# define p_multiple_threads(descr) (descr)->p_header.data.multiple_threads
+# ifdef NEED_DL_SYSINFO
       uintptr_t sysinfo;
-#endif
+# endif
     } data;
     void *__padding[16];
   } p_header;
-# define p_multiple_threads p_header.data.multiple_threads
-#elif TLS_MULTIPLE_THREADS_IN_TCB
-  int p_multiple_threads;
 #endif
-
   pthread_descr p_nextlive, p_prevlive;
                                 /* Double chaining of active threads */
   pthread_descr p_nextwaiting;  /* Next element in the queue holding the thr */
@@ -189,7 +186,22 @@ struct _pthread_descr_struct
 #endif
   size_t p_alloca_cutoff;	/* Maximum size which should be allocated
 				   using alloca() instead of malloc().  */
-  /* New elements must be added at the end.  */
+  /* New elements must be added at the end before __multiple_threads.  */
+#if TLS_MULTIPLE_THREADS_IN_TCB
+  /* This field here isn't necessarily multiple_threads, which is really
+     the last integer in struct _pthread_descr_struct.  */
+  int __multiple_threads;
+# define p_multiple_threads(descr) \
+  (((union								      \
+     {									      \
+       struct _pthread_descr_struct s;					      \
+       struct								      \
+       {								      \
+	 char dummy[sizeof (struct _pthread_descr_struct) - sizeof (int)];    \
+	 int multiple_threads;						      \
+       } m;								      \
+     } *)(descr))->m.multiple_threads)
+#endif
 } __attribute__ ((aligned(32))); /* We need to align the structure so that
 				    doubles are aligned properly.  This is 8
 				    bytes on MIPS and 16 bytes on MIPS64.
diff --git a/linuxthreads/manager.c b/linuxthreads/manager.c
index 3ecf7d1..b4893c1 100644
--- a/linuxthreads/manager.c
+++ b/linuxthreads/manager.c
@@ -591,8 +591,8 @@ static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
   if (new_thread == NULL)
     return EAGAIN;
 # if TLS_DTV_AT_TP
-  /* pthread_descr is right below TP.  */
-  --new_thread;
+  /* pthread_descr is below TP.  */
+  new_thread = (pthread_descr) ((char *) new_thread - TLS_PRE_TCB_SIZE);
 # endif
 #else
   /* Prevent warnings.  */
@@ -612,7 +612,7 @@ static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
 	{
 #ifdef USE_TLS
 # if TLS_DTV_AT_TP
-	  ++new_thread;
+	  new_thread = (pthread_descr) ((char *) new_thread + TLS_PRE_TCB_SIZE);
 # endif
 	  _dl_deallocate_tls (new_thread, true);
 #endif
@@ -643,7 +643,7 @@ static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
   new_thread->p_header.data.self = new_thread;
 #endif
 #if TLS_MULTIPLE_THREADS_IN_TCB || !defined USE_TLS || !TLS_DTV_AT_TP
-  new_thread->p_multiple_threads = 1;
+  p_multiple_threads (new_thread) = 1;
 #endif
   new_thread->p_tid = new_thread_id;
   new_thread->p_lock = &(__pthread_handles[sseg].h_lock);
@@ -806,7 +806,7 @@ static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
       }
 #ifdef USE_TLS
 # if TLS_DTV_AT_TP
-    ++new_thread;
+    new_thread = (pthread_descr) ((char *) new_thread + TLS_PRE_TCB_SIZE);
 # endif
     _dl_deallocate_tls (new_thread, true);
 #endif
@@ -896,7 +896,7 @@ static void pthread_free(pthread_descr th)
 
 #ifdef USE_TLS
 # if TLS_DTV_AT_TP
-      ++th;
+      th = (pthread_descr) ((char *) th + TLS_PRE_TCB_SIZE);
 # endif
       _dl_deallocate_tls (th, true);
 #endif
diff --git a/linuxthreads/pthread.c b/linuxthreads/pthread.c
index 8b1b14f..38a9f7d 100644
--- a/linuxthreads/pthread.c
+++ b/linuxthreads/pthread.c
@@ -571,7 +571,7 @@ int __pthread_initialize_manager(void)
 
   __pthread_multiple_threads = 1;
 #if TLS_MULTIPLE_THREADS_IN_TCB || !defined USE_TLS || !TLS_DTV_AT_TP
-  __pthread_main_thread->p_multiple_threads = 1;
+  p_multiple_threads (__pthread_main_thread) = 1;
 #endif
   *__libc_multiple_threads_ptr = 1;
 
@@ -612,7 +612,7 @@ int __pthread_initialize_manager(void)
 # elif TLS_DTV_AT_TP
   /* pthread_descr is located right below tcbhead_t which _dl_allocate_tls
      returns.  */
-  mgr = (pthread_descr) tcbp - 1;
+  mgr = (pthread_descr) ((char *) tcbp - TLS_PRE_TCB_SIZE);
 # endif
   __pthread_handles[1].h_descr = manager_thread = mgr;
 
@@ -620,9 +620,9 @@ int __pthread_initialize_manager(void)
 #if !defined USE_TLS || !TLS_DTV_AT_TP
   mgr->p_header.data.tcb = tcbp;
   mgr->p_header.data.self = mgr;
-  mgr->p_header.data.multiple_threads = 1;
+  p_multiple_threads (mgr) = 1;
 #elif TLS_MULTIPLE_THREADS_IN_TCB
-  mgr->p_multiple_threads = 1;
+  p_multiple_threads (mgr) = 1;
 #endif
   mgr->p_lock = &__pthread_handles[1].h_lock;
 # ifndef HAVE___THREAD
diff --git a/linuxthreads/sysdeps/ia64/tcb-offsets.sym b/linuxthreads/sysdeps/ia64/tcb-offsets.sym
index f7793f7..c1d307d 100644
--- a/linuxthreads/sysdeps/ia64/tcb-offsets.sym
+++ b/linuxthreads/sysdeps/ia64/tcb-offsets.sym
@@ -3,7 +3,7 @@
 
 --
 #ifdef USE_TLS
-MULTIPLE_THREADS_OFFSET offsetof (struct _pthread_descr_struct, p_multiple_threads) - sizeof (struct _pthread_descr_struct)
+MULTIPLE_THREADS_OFFSET -sizeof(int)
 #else
 MULTIPLE_THREADS_OFFSET offsetof (tcbhead_t, multiple_threads)
 #endif
diff --git a/linuxthreads/sysdeps/powerpc/powerpc32/tcb-offsets.sym b/linuxthreads/sysdeps/powerpc/powerpc32/tcb-offsets.sym
new file mode 100644
index 0000000..8c6bddb
--- /dev/null
+++ b/linuxthreads/sysdeps/powerpc/powerpc32/tcb-offsets.sym
@@ -0,0 +1,9 @@
+#include <sysdep.h>
+#include <tls.h>
+
+--
+#ifdef USE_TLS
+MULTIPLE_THREADS_OFFSET	((void *) &p_multiple_threads ((pthread_descr) ((void *) 0 - TLS_TCB_OFFSET - TLS_PRE_TCB_SIZE)) - (void *) 0)
+#else
+MULTIPLE_THREADS_OFFSET	offsetof (tcbhead_t, multiple_threads)
+#endif
diff --git a/linuxthreads/sysdeps/powerpc/tcb-offsets.sym b/linuxthreads/sysdeps/powerpc/tcb-offsets.sym
deleted file mode 100644
index bb4226f..0000000
--- a/linuxthreads/sysdeps/powerpc/tcb-offsets.sym
+++ /dev/null
@@ -1,24 +0,0 @@
-#include <sysdep.h>
-#include <tls.h>
-
---
-
--- This could go into powerpc32/ instead and conditionalize #include of it.
-#ifndef __powerpc64__
-
-# ifdef USE_TLS
-
--- Abuse tls.h macros to derive offsets relative to the thread register.
-#  undef __thread_register
-#  define __thread_register	((void *) 0)
-#  define thread_offsetof(mem)	((void *) &THREAD_SELF->p_##mem - (void *) 0)
-
-# else
-
-#  define thread_offsetof(mem)	offsetof (tcbhead_t, mem)
-
-# endif
-
-MULTIPLE_THREADS_OFFSET		thread_offsetof (multiple_threads)
-
-#endif
diff --git a/linuxthreads/sysdeps/powerpc/tls.h b/linuxthreads/sysdeps/powerpc/tls.h
index d946d06..55e915e 100644
--- a/linuxthreads/sysdeps/powerpc/tls.h
+++ b/linuxthreads/sysdeps/powerpc/tls.h
@@ -32,8 +32,6 @@ typedef union dtv
   void *pointer;
 } dtv_t;
 
-#else /* __ASSEMBLER__ */
-# include <tcb-offsets.h>
 #endif /* __ASSEMBLER__ */
 
 #ifdef HAVE_TLS_SUPPORT
@@ -52,27 +50,29 @@ typedef struct
 } tcbhead_t;
 
 /* This is the size of the initial TCB.  */
-#  define TLS_INIT_TCB_SIZE	sizeof (tcbhead_t)
+#  define TLS_INIT_TCB_SIZE	0
 
 /* Alignment requirements for the initial TCB.  */
-#  define TLS_INIT_TCB_ALIGN	__alignof__ (tcbhead_t)
+#  define TLS_INIT_TCB_ALIGN	__alignof__ (struct _pthread_descr_struct)
 
 /* This is the size of the TCB.  */
-#  define TLS_TCB_SIZE		sizeof (tcbhead_t)
+#  define TLS_TCB_SIZE		0
 
 /* Alignment requirements for the TCB.  */
-#  define TLS_TCB_ALIGN		__alignof__ (tcbhead_t)
+#  define TLS_TCB_ALIGN		__alignof__ (struct _pthread_descr_struct)
 
 /* This is the size we need before TCB.  */
-#  define TLS_PRE_TCB_SIZE	sizeof (struct _pthread_descr_struct)
+#  define TLS_PRE_TCB_SIZE \
+  (sizeof (struct _pthread_descr_struct)				      \
+   + ((sizeof (tcbhead_t) + TLS_TCB_ALIGN - 1) & ~(TLS_TCB_ALIGN - 1)))
 
 /* The following assumes that TP (R2 or R13) is points to the end of the
    TCB + 0x7000 (per the ABI).  This implies that TCB address is
-   TP-(TLS_TCB_SIZE + 0x7000).  As we define TLS_DTV_AT_TP we can
+   TP - 0x7000.  As we define TLS_DTV_AT_TP we can
    assume that the pthread_descr is allocated immediately ahead of the
    TCB.  This implies that the pthread_descr address is
-   TP-(TLS_PRE_TCB_SIZE + TLS_TCB_SIZE + 0x7000).  */
-#  define TLS_TCB_OFFSET	0x7000
+   TP - (TLS_PRE_TCB_SIZE + 0x7000).  */
+#define TLS_TCB_OFFSET		0x7000
 
 /* The DTV is allocated at the TP; the TCB is placed elsewhere.  */
 /* This is not really true for powerpc64.  We are following alpha
@@ -82,13 +82,13 @@ typedef struct
 /* Install the dtv pointer.  The pointer passed is to the element with
    index -1 which contain the length.  */
 #  define INSTALL_DTV(TCBP, DTVP) \
-  (((tcbhead_t *) (TCBP))->dtv = (DTVP) + 1)
+  (((tcbhead_t *) (TCBP))[-1].dtv = (DTVP) + 1)
 
 /* Install new dtv for current thread.  */
 #  define INSTALL_NEW_DTV(DTV) (THREAD_DTV() = (DTV))
 
 /* Return dtv of given thread descriptor.  */
-#  define GET_DTV(TCBP)	(((tcbhead_t *) (TCBP))->dtv)
+#  define GET_DTV(TCBP)	(((tcbhead_t *) (TCBP))[-1].dtv)
 
 /* The global register variable is declared in pt-machine.h with
    the wrong type, but the compiler doesn't like us declaring another.  */
@@ -98,22 +98,22 @@ typedef struct
    special attention since 'errno' is not yet available and if the
    operation can cause a failure 'errno' must not be touched.  */
 # define TLS_INIT_TP(TCBP, SECONDCALL) \
-    (__thread_register = (void *) (TCBP) + TLS_TCB_OFFSET + TLS_TCB_SIZE, NULL)
+    (__thread_register = (void *) (TCBP) + TLS_TCB_OFFSET, NULL)
 
 /* Return the address of the dtv for the current thread.  */
 #  define THREAD_DTV() \
-     (((tcbhead_t *) (__thread_register - TLS_TCB_OFFSET - TLS_TCB_SIZE))->dtv)
+     (((tcbhead_t *) (__thread_register - TLS_TCB_OFFSET))[-1].dtv)
 
 /* Return the thread descriptor for the current thread.  */
 #  undef THREAD_SELF
 #  define THREAD_SELF \
     ((pthread_descr) (__thread_register \
-		      - TLS_TCB_OFFSET - TLS_TCB_SIZE - TLS_PRE_TCB_SIZE))
+		      - TLS_TCB_OFFSET - TLS_PRE_TCB_SIZE))
 
 #  undef INIT_THREAD_SELF
 #  define INIT_THREAD_SELF(DESCR, NR) \
      (__thread_register = ((void *) (DESCR) \
-		           + TLS_TCB_OFFSET + TLS_TCB_SIZE + TLS_PRE_TCB_SIZE))
+		           + TLS_TCB_OFFSET + TLS_PRE_TCB_SIZE))
 
 /* Make sure we have the p_multiple_threads member in the thread structure.
    See below.  */
@@ -124,6 +124,10 @@ typedef struct
 /* Get the thread descriptor definition.  */
 #  include <linuxthreads/descr.h>
 
+/* l_tls_offset == 0 is perfectly valid on PPC, so we have to use some
+   different value to mean unset l_tls_offset.  */
+#  define NO_TLS_OFFSET	-1
+
 # endif /* __ASSEMBLER__ */
 
 #elif !defined __ASSEMBLER__ && !defined __powerpc64__
diff --git a/linuxthreads/sysdeps/sh/tcb-offsets.sym b/linuxthreads/sysdeps/sh/tcb-offsets.sym
index 328eb05..7537daa 100644
--- a/linuxthreads/sysdeps/sh/tcb-offsets.sym
+++ b/linuxthreads/sysdeps/sh/tcb-offsets.sym
@@ -3,7 +3,7 @@
 
 --
 #ifdef USE_TLS
-MULTIPLE_THREADS_OFFSET offsetof (struct _pthread_descr_struct, p_multiple_threads)
+MULTIPLE_THREADS_OFFSET ((char *) &p_multiple_threads ((struct _pthread_descr_struct *)0) - (char *) 0)
 TLS_PRE_TCB_SIZE	sizeof (struct _pthread_descr_struct)
 #else
 MULTIPLE_THREADS_OFFSET offsetof (tcbhead_t, multiple_threads)
diff --git a/linuxthreads/sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h b/linuxthreads/sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h
index af45b19..7982d27 100644
--- a/linuxthreads/sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h
+++ b/linuxthreads/sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h
@@ -101,7 +101,7 @@ __syscall_error_##args:							      \
 
 # ifndef __ASSEMBLER__
 #  define SINGLE_THREAD_P \
-  __builtin_expect (THREAD_GETMEM (THREAD_SELF, p_multiple_threads) == 0, 1)
+  __builtin_expect (p_multiple_threads (THREAD_SELF) == 0, 1)
 # else
 #  define SINGLE_THREAD_P \
   adds r14 = MULTIPLE_THREADS_OFFSET, r13 ;; ld4 r14 = [r14] ;; cmp4.ne p6, p7 = 0, r14
diff --git a/linuxthreads/sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h b/linuxthreads/sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h
index 2d191d1..fb6ca06 100644
--- a/linuxthreads/sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h
+++ b/linuxthreads/sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h
@@ -21,6 +21,8 @@
 #include <tls.h>
 #ifndef __ASSEMBLER__
 # include <linuxthreads/internals.h>
+#else
+# include <tcb-offsets.h>
 #endif
 
 #if !defined NOT_IN_libc || defined IS_IN_libpthread
@@ -85,7 +87,7 @@
 
 # ifndef __ASSEMBLER__
 #  define SINGLE_THREAD_P						\
-  __builtin_expect (THREAD_GETMEM (THREAD_SELF, p_multiple_threads) == 0, 1)
+  __builtin_expect (p_multiple_threads (THREAD_SELF) == 0, 1)
 # else
 #  define SINGLE_THREAD_P						\
   lwz 10,MULTIPLE_THREADS_OFFSET(2);					\
diff --git a/linuxthreads/sysdeps/unix/sysv/linux/sh/sysdep-cancel.h b/linuxthreads/sysdeps/unix/sysv/linux/sh/sysdep-cancel.h
index e6d0cca..57cced9 100644
--- a/linuxthreads/sysdeps/unix/sysv/linux/sh/sysdep-cancel.h
+++ b/linuxthreads/sysdeps/unix/sysv/linux/sh/sysdep-cancel.h
@@ -121,7 +121,7 @@
 # ifndef __ASSEMBLER__
 #  if defined FLOATING_STACKS && USE___THREAD && defined PIC
 #   define SINGLE_THREAD_P \
-  __builtin_expect (THREAD_GETMEM (THREAD_SELF, p_multiple_threads) == 0, 1)
+  __builtin_expect (p_multiple_threads (THREAD_SELF) == 0, 1)
 #  else
 extern int __local_multiple_threads attribute_hidden;
 #   define SINGLE_THREAD_P __builtin_expect (__local_multiple_threads == 0, 1)
diff --git a/nptl/ChangeLog b/nptl/ChangeLog
index b1b68b4..8e76d9e 100644
--- a/nptl/ChangeLog
+++ b/nptl/ChangeLog
@@ -1,3 +1,41 @@
+2003-04-22  Jakub Jelinek  <jakub@redhat.com>
+
+	* descr.h (p_multiple_threads): Define.
+	(struct pthread) [TLS_MULTIPLE_THREADS_IN_TCB]: Move
+	multiple_threads to last int in the structure.
+	* allocatestack.c (TLS_TPADJ): Add TLS_PRE_TCB_SIZE instead of
+	sizeof (struct pthread).
+	(allocate_stack): Subtract TLS_PRE_TCB_SIZE bytes instead of
+	1 struct pthread.  Use p_multiple_threads macro.
+	* sysdeps/pthread/createthread.c (create_thread): Use
+	p_multiple_threads macro if TLS_DTV_AT_TP.
+	* sysdeps/powerpc/tls.h (TLS_INIT_TCB_SIZE, TLS_TCB_SIZE): Define
+	to 0.
+	(TLS_INIT_TCB_ALIGN, TLS_TCB_ALIGN): Define to alignment of
+	struct pthread.
+	(TLS_PRE_TCB_SIZE): Increase to cover tcbhead_t preceeded by pad
+	to 32-bit bytes.
+	(INSTALL_DTV, GET_DTV, THREAD_DTV): tcbhead_t is immediately before
+	tcbp.
+	(TLS_INIT_TP, THREAD_SELF, INIT_THREAD_SELF): Don't add TLS_TCB_SIZE
+	unneccessarily.
+	(NO_TLS_OFFSET): Define.
+	* sysdeps/powerpc/tcb-offsets.sym (MULTIPLE_THREADS): Use
+	p_multiple_threads macro.
+	* sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h
+	(SINGLE_THREAD_P): Likewise.
+	* sysdeps/unix/sysv/linux/powerpc/powerpc64/sysdep-cancel.h
+	(SINGLE_THREAD_P): Likewise.
+	* sysdeps/unix/sysv/linux/powerpc/createthread.c (TLS_VALUE): Don't
+	add TLS_TCB_SIZE unnecessarily.
+	* sysdeps/ia64/tcb-offsets.sym (MULTIPLE_THREADS): Define to
+	-sizeof(int).
+	* sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h (SINGLE_THREAD_P):
+	Use p_multiple_threads macro.
+	* sysdeps/sh/tcb-offsets.sym (MULTIPLE_THREADS): Likewise.
+	* sysdeps/unix/sysv/linux/sh/sysdep-cancel.h (SINGLE_THREAD_P):
+	Likewise.
+
 2003-04-22  Roland McGrath  <roland@redhat.com>
 
 	* Makeconfig (shared-thread-library): Reverse link order to work
diff --git a/nptl/allocatestack.c b/nptl/allocatestack.c
index c199cf2..c6b89d8 100644
--- a/nptl/allocatestack.c
+++ b/nptl/allocatestack.c
@@ -81,7 +81,7 @@
 #if TLS_TCB_AT_TP
 # define TLS_TPADJ(pd) (pd)
 #elif TLS_DTV_AT_TP
-# define TLS_TPADJ(pd) ((pd) + 1)
+# define TLS_TPADJ(pd) ((struct pthread *)((char *) (pd) + TLS_PRE_TCB_SIZE))
 #endif
 
 /* Cache handling for not-yet free stacks.  */
@@ -296,8 +296,9 @@ allocate_stack (const struct pthread_attr *attr, struct pthread **pdp,
       pd = (struct pthread *) ((uintptr_t) attr->stackaddr
 			       - TLS_TCB_SIZE - adj);
 #elif TLS_DTV_AT_TP
-      pd = (struct pthread *) ((uintptr_t) attr->stackaddr
-			       - __static_tls_size - adj) - 1;
+      pd = (struct pthread *) (((uintptr_t) attr->stackaddr
+			        - __static_tls_size - adj)
+			       - TLS_PRE_TCB_SIZE);
 #endif
 
       /* The user provided stack memory needs to be cleared.  */
@@ -321,7 +322,7 @@ allocate_stack (const struct pthread_attr *attr, struct pthread **pdp,
 
 #ifdef TLS_MULTIPLE_THREADS_IN_TCB
       /* This is at least the second thread.  */
-      pd->header.multiple_threads = 1;
+      p_multiple_threads (pd) = 1;
 #else
       __pthread_multiple_threads = *__libc_multiple_threads_ptr = 1;
 #endif
@@ -426,9 +427,10 @@ allocate_stack (const struct pthread_attr *attr, struct pthread **pdp,
 #if TLS_TCB_AT_TP
 	  pd = (struct pthread *) ((char *) mem + size - coloring) - 1;
 #elif TLS_DTV_AT_TP
-	  pd = (struct pthread *) (((uintptr_t) mem + size - coloring
+	  pd = (struct pthread *) ((((uintptr_t) mem + size - coloring
 				    - __static_tls_size)
-				   & ~__static_tls_align_m1) - 1;
+				    & ~__static_tls_align_m1)
+				   - TLS_PRE_TCB_SIZE);
 #endif
 
 	  /* Remember the stack-related values.  */
@@ -447,7 +449,7 @@ allocate_stack (const struct pthread_attr *attr, struct pthread **pdp,
 
 #ifdef TLS_MULTIPLE_THREADS_IN_TCB
 	  /* This is at least the second thread.  */
-	  pd->header.multiple_threads = 1;
+	  p_multiple_threads (pd) = 1;
 #else
 	  __pthread_multiple_threads = *__libc_multiple_threads_ptr = 1;
 #endif
diff --git a/nptl/descr.h b/nptl/descr.h
index efb25c7..7fd64ec 100644
--- a/nptl/descr.h
+++ b/nptl/descr.h
@@ -99,11 +99,7 @@ struct pthread
 #if !TLS_DTV_AT_TP
     /* This overlaps the TCB as used for TLS without threads (see tls.h).  */
     tcbhead_t header;
-#elif TLS_MULTIPLE_THREADS_IN_TCB
-    struct
-    {
-      int multiple_threads;
-    } header;
+# define p_multiple_threads(descr) (descr)->header.multiple_threads
 #endif
 
     /* This extra padding has no special purpose, and this structure layout
@@ -232,6 +228,21 @@ struct pthread
   size_t stackblock_size;
   /* Size of the included guard area.  */
   size_t guardsize;
+
+#if TLS_DTV_AT_TP && TLS_MULTIPLE_THREADS_IN_TCB
+  /* Must come last.  */
+  int __multiple_threads;
+# define p_multiple_threads(descr) \
+  ((union							\
+    {								\
+      struct pthread s;						\
+      struct							\
+	{							\
+	  char dummy[sizeof (struct pthread) - sizeof (int)];   \
+	  int multiple_threads;					\
+	} m;							\
+    } *)(descr)->m.multiple_threads)
+#endif
 } __attribute ((aligned (TCB_ALIGNMENT)));
 
 
diff --git a/nptl/sysdeps/ia64/tcb-offsets.sym b/nptl/sysdeps/ia64/tcb-offsets.sym
index 11cc06a..9f92bb6 100644
--- a/nptl/sysdeps/ia64/tcb-offsets.sym
+++ b/nptl/sysdeps/ia64/tcb-offsets.sym
@@ -1,4 +1,4 @@
 #include <sysdep.h>
 #include <tls.h>
 
-MULTIPLE_THREADS_OFFSET offsetof (struct pthread, header.multiple_threads) - sizeof (struct pthread)
+MULTIPLE_THREADS_OFFSET -sizeof(int)
diff --git a/nptl/sysdeps/powerpc/tcb-offsets.sym b/nptl/sysdeps/powerpc/tcb-offsets.sym
index d6b7560..58ee030 100644
--- a/nptl/sysdeps/powerpc/tcb-offsets.sym
+++ b/nptl/sysdeps/powerpc/tcb-offsets.sym
@@ -1,13 +1,4 @@
 #include <sysdep.h>
 #include <tls.h>
 
---
-
--- Abuse tls.h macros to derive offsets relative to the thread register.
-# undef __thread_register
-# define __thread_register	((void *) 0)
-# define thread_offsetof(mem)	((void *) &THREAD_SELF->mem - (void *) 0)
-
-#if TLS_MULTIPLE_THREADS_IN_TCB
-MULTIPLE_THREADS_OFFSET		thread_offsetof (header.multiple_threads)
-#endif
+MULTIPLE_THREADS_OFFSET	((void *) &p_multiple_threads ((struct pthread) ((void *) 0 - TLS_TCB_OFFSET - TLS_PRE_TCB_SIZE)) - (void *) 0)
diff --git a/nptl/sysdeps/powerpc/tls.h b/nptl/sysdeps/powerpc/tls.h
index 0ef5655..e3e0424 100644
--- a/nptl/sysdeps/powerpc/tls.h
+++ b/nptl/sysdeps/powerpc/tls.h
@@ -69,22 +69,19 @@ typedef struct
 } tcbhead_t;
 
 /* This is the size of the initial TCB.  */
-# define TLS_INIT_TCB_SIZE	sizeof (tcbhead_t)
+# define TLS_INIT_TCB_SIZE	0
 
 /* Alignment requirements for the initial TCB.  */
-# define TLS_INIT_TCB_ALIGN	__alignof__ (tcbhead_t)
+# define TLS_INIT_TCB_ALIGN	__alignof__ (struct pthread)
 
 /* This is the size of the TCB.  */
-# define TLS_TCB_SIZE		sizeof (tcbhead_t)
+# define TLS_TCB_SIZE		0
 
 /* Alignment requirements for the TCB.  */
-# define TLS_TCB_ALIGN		__alignof__ (tcbhead_t)
+# define TLS_TCB_ALIGN		__alignof__ (struct pthread)
 
 /* This is the size we need before TCB.  */
-# define TLS_PRE_TCB_SIZE	sizeof (struct pthread)
-
-/* XXX if __alignof__ (struct pthread) > __alignof (tcbhead_t)
-   we could be in trouble.  -- paulus */
+# define TLS_PRE_TCB_SIZE	(sizeof (struct pthread) + 32)
 
 # ifndef __powerpc64__
 /* Register r2 (tp) is reserved by the ABI as "thread pointer". */
@@ -98,37 +95,37 @@ register void *__thread_register __asm__ ("r13");
 
 /* The following assumes that TP (R2 or R13) points to the end of the
    TCB + 0x7000 (per the ABI).  This implies that TCB address is
-   TP-(TLS_TCB_SIZE + 0x7000).  As we define TLS_DTV_AT_TP we can
+   TP - 0x7000.  As we define TLS_DTV_AT_TP we can
    assume that the pthread struct is allocated immediately ahead of the
    TCB.  This implies that the pthread_descr address is
-   TP-(TLS_PRE_TCB_SIZE + TLS_TCB_SIZE + 0x7000).  */
+   TP - (TLS_PRE_TCB_SIZE + 0x7000).  */
 # define TLS_TCB_OFFSET	0x7000
 
 /* Install the dtv pointer.  The pointer passed is to the element with
    index -1 which contain the length.  */
 # define INSTALL_DTV(tcbp, dtvp) \
-  ((tcbhead_t *) (tcbp))->dtv = dtvp + 1
+  ((tcbhead_t *) (tcbp))[-1].dtv = dtvp + 1
 
 /* Install new dtv for current thread.  */
 # define INSTALL_NEW_DTV(dtv) (THREAD_DTV() = (dtv))
 
 /* Return dtv of given thread descriptor.  */
-# define GET_DTV(tcbp)	(((tcbhead_t *) (tcbp))->dtv)
+# define GET_DTV(tcbp)	(((tcbhead_t *) (tcbp))[-1].dtv)
 
 /* Code to initially initialize the thread pointer.  This might need
    special attention since 'errno' is not yet available and if the
    operation can cause a failure 'errno' must not be touched.  */
 # define TLS_INIT_TP(tcbp, secondcall) \
-    (__thread_register = (void *) (tcbp) + TLS_TCB_OFFSET + TLS_TCB_SIZE, NULL)
+    (__thread_register = (void *) (tcbp) + TLS_TCB_OFFSET, NULL)
 
 /* Return the address of the dtv for the current thread.  */
 # define THREAD_DTV() \
-     (((tcbhead_t *) (__thread_register - TLS_TCB_OFFSET - TLS_TCB_SIZE))->dtv)
+     (((tcbhead_t *) (__thread_register - TLS_TCB_OFFSET))[-1].dtv)
 
 /* Return the thread descriptor for the current thread.  */
 # define THREAD_SELF \
     ((struct pthread *) (__thread_register \
-			 - TLS_TCB_OFFSET - TLS_TCB_SIZE - TLS_PRE_TCB_SIZE))
+			 - TLS_TCB_OFFSET - TLS_PRE_TCB_SIZE))
 
 /* Read member of the thread descriptor directly.  */
 # define THREAD_GETMEM(descr, member) ((void)(descr), (THREAD_SELF)->member)
@@ -145,6 +142,10 @@ register void *__thread_register __asm__ ("r13");
 # define THREAD_SETMEM_NC(descr, member, idx, value) \
     ((void)(descr), (THREAD_SELF)->member[idx] = (value))
 
+/* l_tls_offset == 0 is perfectly valid on PPC, so we have to use some
+   different value to mean unset l_tls_offset.  */
+# define NO_TLS_OFFSET		-1
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/pthread/createthread.c b/nptl/sysdeps/pthread/createthread.c
index 9d00e4e..f5c6406 100644
--- a/nptl/sysdeps/pthread/createthread.c
+++ b/nptl/sysdeps/pthread/createthread.c
@@ -87,7 +87,11 @@ create_thread (struct pthread *pd, STACK_VARIABLES_PARMS)
 	     thread might not yet have the flag set.  No need to set
 	     the global variable again if this is what we use.  */
 #ifdef TLS_MULTIPLE_THREADS_IN_TCB
+# if TLS_DTV_AT_TP
+	  p_multiple_threads (THREAD_SELF) = 1;
+# else
 	  THREAD_SETMEM (THREAD_SELF, header.multiple_threads, 1);
+# endif
 #endif
 
 	  /* Now fill in the information about the new thread in
@@ -159,7 +163,11 @@ create_thread (struct pthread *pd, STACK_VARIABLES_PARMS)
      not yet have the flag set.  No need to set the global variable
      again if this is what we use.  */
 #ifdef TLS_MULTIPLE_THREADS_IN_TCB
+# if TLS_DTV_AT_TP
+  p_multiple_threads (THREAD_SELF) = 1;
+# else
   THREAD_SETMEM (THREAD_SELF, header.multiple_threads, 1);
+# endif
 #endif
 
   return 0;
diff --git a/nptl/sysdeps/sh/tcb-offsets.sym b/nptl/sysdeps/sh/tcb-offsets.sym
index 3386f1d..940c933 100644
--- a/nptl/sysdeps/sh/tcb-offsets.sym
+++ b/nptl/sysdeps/sh/tcb-offsets.sym
@@ -1,5 +1,5 @@
 #include <sysdep.h>
 #include <tls.h>
 
-MULTIPLE_THREADS_OFFSET		offsetof (struct pthread, header.multiple_threads)
+MULTIPLE_THREADS_OFFSET		((char *) &p_multiple_threads ((struct pthread *)0) - (char *) 0)
 TLS_PRE_TCB_SIZE		sizeof (struct pthread)
diff --git a/nptl/sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h b/nptl/sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h
index 667abce..930cc14 100644
--- a/nptl/sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h
+++ b/nptl/sysdeps/unix/sysv/linux/ia64/sysdep-cancel.h
@@ -101,7 +101,7 @@ __syscall_error_##args:							      \
 
 # ifndef __ASSEMBLER__
 #  define SINGLE_THREAD_P \
-  __builtin_expect (THREAD_GETMEM (THREAD_SELF, header.multiple_threads) == 0, 1)
+  __builtin_expect (p_multiple_threads (THREAD_SELF) == 0, 1)
 # else
 #  define SINGLE_THREAD_P \
   adds r14 = MULTIPLE_THREADS_OFFSET, r13 ;; ld4 r14 = [r14] ;; cmp4.ne p6, p7 = 0, r14
diff --git a/nptl/sysdeps/unix/sysv/linux/powerpc/createthread.c b/nptl/sysdeps/unix/sysv/linux/powerpc/createthread.c
index bcbcdd7..e811ad7 100644
--- a/nptl/sysdeps/unix/sysv/linux/powerpc/createthread.c
+++ b/nptl/sysdeps/unix/sysv/linux/powerpc/createthread.c
@@ -19,7 +19,7 @@
 
 /* Value passed to 'clone' for initialization of the thread register.  */
 #define TLS_VALUE ((void *) (pd) \
-		   + TLS_TCB_OFFSET + TLS_TCB_SIZE + TLS_PRE_TCB_SIZE)
+		   + TLS_TCB_OFFSET + TLS_PRE_TCB_SIZE)
 
 /* Get the real implementation.	 */
 #include <nptl/sysdeps/pthread/createthread.c>
diff --git a/nptl/sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h b/nptl/sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h
index d256f8d..f82addb 100644
--- a/nptl/sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h
+++ b/nptl/sysdeps/unix/sysv/linux/powerpc/powerpc32/sysdep-cancel.h
@@ -86,8 +86,7 @@
 
 # ifndef __ASSEMBLER__
 #  define SINGLE_THREAD_P						\
-  __builtin_expect (THREAD_GETMEM (THREAD_SELF,				\
-				   header.multiple_threads) == 0, 1)
+  __builtin_expect (p_multiple_threads (THREAD_SELF) == 0, 1)
 # else
 #  define SINGLE_THREAD_P						\
   lwz 10,MULTIPLE_THREADS_OFFSET(2);					\
diff --git a/nptl/sysdeps/unix/sysv/linux/powerpc/powerpc64/sysdep-cancel.h b/nptl/sysdeps/unix/sysv/linux/powerpc/powerpc64/sysdep-cancel.h
index 5483586..00dc3a2 100644
--- a/nptl/sysdeps/unix/sysv/linux/powerpc/powerpc64/sysdep-cancel.h
+++ b/nptl/sysdeps/unix/sysv/linux/powerpc/powerpc64/sysdep-cancel.h
@@ -86,8 +86,7 @@
 
 # ifndef __ASSEMBLER__
 #  define SINGLE_THREAD_P						\
-  __builtin_expect (THREAD_GETMEM (THREAD_SELF,				\
-				   header.multiple_threads) == 0, 1)
+  __builtin_expect (p_multiple_threads (THREAD_SELF) == 0, 1)
 # else
 #   define SINGLE_THREAD_P						\
   lwz   10,MULTIPLE_THREADS_OFFSET(13);				\
diff --git a/nptl/sysdeps/unix/sysv/linux/sh/sysdep-cancel.h b/nptl/sysdeps/unix/sysv/linux/sh/sysdep-cancel.h
index 16f8ad5..10189a4 100644
--- a/nptl/sysdeps/unix/sysv/linux/sh/sysdep-cancel.h
+++ b/nptl/sysdeps/unix/sysv/linux/sh/sysdep-cancel.h
@@ -117,8 +117,7 @@
 
 # ifndef __ASSEMBLER__
 #  define SINGLE_THREAD_P \
-  __builtin_expect (THREAD_GETMEM (THREAD_SELF, \
-				   header.multiple_threads) == 0, 1)
+  __builtin_expect (p_multiple_threads (THREAD_SELF) == 0, 1)
 # else
 #  define SINGLE_THREAD_P \
 	stc gbr,r0; \
diff --git a/sysdeps/generic/dl-tls.c b/sysdeps/generic/dl-tls.c
index 746d9a2..a51e3f7 100644
--- a/sysdeps/generic/dl-tls.c
+++ b/sysdeps/generic/dl-tls.c
@@ -359,7 +359,7 @@ _dl_allocate_tls_init (void *result)
 	    /* Unused entry.  */
 	    continue;
 
-	  if (map->l_tls_offset == 0)
+	  if (map->l_tls_offset == NO_TLS_OFFSET)
 	    {
 	      /* For dynamically loaded modules we simply store
 		 the value indicating deferred allocation.  */
diff --git a/sysdeps/powerpc/dl-tls.h b/sysdeps/powerpc/dl-tls.h
index 2f84a57..957d4b4 100644
--- a/sysdeps/powerpc/dl-tls.h
+++ b/sysdeps/powerpc/dl-tls.h
@@ -35,7 +35,7 @@ typedef struct
 /* Compute the value for a @tprel reloc.  */
 #define TLS_TPREL_VALUE(sym_map, sym, reloc) \
   ((sym_map)->l_tls_offset + (sym)->st_value + (reloc)->r_addend \
-   - TLS_TCB_SIZE - TLS_TP_OFFSET)
+   - TLS_TP_OFFSET)
 
 /* Compute the value for a @dtprel reloc.  */
 #define TLS_DTPREL_VALUE(sym, reloc) \