Update.

2003-07-24  Ulrich Drepper  <drepper@redhat.com>

	* include/link.h (struct link_map): Add l_tls_firstbyte_offset field.
	* sysdeps/generic/dl-tls.c [TLS_TCB_AT_TP] (_dl_determine_tlsoffset):
	Fix calculation of offsets to take misalignment of first byte in
	file into account.
	* elf/dl-load.c (_dl_map_object_from_fd): Initialize
	l_tls_firstbyte_offset field.
	* elf/rtld.c (_dl_start_final, _dl_start, dl_main): Likewise.
	* elf/dl-reloc.c (_dl_allocate_static_tls): Change return type to int.
	Take l_tls_firstbyte_offset information into account.
	(CHECK_STATIS_TLS): _dl_allocate_static_tls can fail now.
	* sysdeps/generic/ldsodefs.h: Adjust _dl_allocate_static_tls prototype.
	* elf/Makefile: Add rules to build and run tst-tls14.
	* elf/tst-tls14.c: New file.
	* elf/tst-tlsmod14a.c: New file.
	* elf/tst-tlsmod14b.c: New file.

ChangeLog

@@ -1,3 +1,21 @@
+2003-07-24  Ulrich Drepper  <drepper@redhat.com>
+
+	* include/link.h (struct link_map): Add l_tls_firstbyte_offset field.
+	* sysdeps/generic/dl-tls.c [TLS_TCB_AT_TP] (_dl_determine_tlsoffset):
+	Fix calculation of offsets to take misalignment of first byte in
+	file into account.
+	* elf/dl-load.c (_dl_map_object_from_fd): Initialize
+	l_tls_firstbyte_offset field.
+	* elf/rtld.c (_dl_start_final, _dl_start, dl_main): Likewise.
+	* elf/dl-reloc.c (_dl_allocate_static_tls): Change return type to int.
+	Take l_tls_firstbyte_offset information into account.
+	(CHECK_STATIS_TLS): _dl_allocate_static_tls can fail now.
+	* sysdeps/generic/ldsodefs.h: Adjust _dl_allocate_static_tls prototype.
+	* elf/Makefile: Add rules to build and run tst-tls14.
+	* elf/tst-tls14.c: New file.
+	* elf/tst-tlsmod14a.c: New file.
+	* elf/tst-tlsmod14b.c: New file.
+
 2003-07-23  Jakub Jelinek  <jakub@redhat.com>
 
 	* sysdeps/pthread/lio_listio.c (LIO_OPCODE_BASE): Define.

elf/Makefile

@@ -148,7 +148,7 @@ tests += loadtest restest1 preloadtest loadfail multiload origtest resolvfail \
 	 neededtest3 neededtest4 unload2 lateglobal initfirst global \
 	 restest2 next dblload dblunload reldep5 reldep6 reldep7 reldep8 \
 	 circleload1 tst-tls3 tst-tls4 tst-tls5 tst-tls6 tst-tls7 tst-tls8 \
-	 tst-tls10 tst-tls11 tst-tls12 tst-tls13
+	 tst-tls10 tst-tls11 tst-tls12 tst-tls13 tst-tls14
 #	 reldep9
 test-srcs = tst-pathopt
 tests-vis-yes = vismain
@@ -171,7 +171,7 @@ modules-names = testobj1 testobj2 testobj3 testobj4 testobj5 testobj6 \
 		tst-tlsmod1 tst-tlsmod2 tst-tlsmod3 tst-tlsmod4 \
 		tst-tlsmod5 tst-tlsmod6 tst-tlsmod7 tst-tlsmod8 \
 		tst-tlsmod9 tst-tlsmod10 tst-tlsmod11 tst-tlsmod12 \
-		tst-tlsmod13 tst-tlsmod13a \
+		tst-tlsmod13 tst-tlsmod13a tst-tlsmod14a tst-tlsmod14b \
 		circlemod1 circlemod1a circlemod2 circlemod2a \
 		circlemod3 circlemod3a \
 		reldep8mod1 reldep8mod2 reldep8mod3 \
@@ -428,6 +428,8 @@ tst-tlsmod8.so-no-z-defs = yes
 tst-tlsmod9.so-no-z-defs = yes
 tst-tlsmod10.so-no-z-defs = yes
 tst-tlsmod12.so-no-z-defs = yes
+tst-tlsmod14a.so-no-z-defs = yes
+tst-tlsmod14b.so-no-z-defs = yes
 circlemod2.so-no-z-defs = yes
 circlemod3.so-no-z-defs = yes
 circlemod3a.so-no-z-defs = yes
@@ -633,6 +635,9 @@ $(objpfx)tst-tls12: $(objpfx)tst-tlsmod12.so
 $(objpfx)tst-tls13: $(libdl)
 $(objpfx)tst-tls13.out: $(objpfx)tst-tlsmod13a.so
 
+$(objpfx)tst-tls14:  $(objpfx)tst-tlsmod14a.so $(libdl)
+$(objpfx)tst-tls14.out:$(objpfx)tst-tlsmod14b.so
+
 ifdef libdl
 $(objpfx)tst-tls9-static: $(common-objpfx)dlfcn/libdl.a
 $(objpfx)tst-tls9-static.out: $(objpfx)tst-tlsmod5.so $(objpfx)tst-tlsmod6.so

elf/dl-load.c

@@ -987,6 +987,10 @@ _dl_map_object_from_fd (const char *name, int fd, struct filebuf *fbp,
 
 	  l->l_tls_blocksize = ph->p_memsz;
 	  l->l_tls_align = ph->p_align;
+	  if (ph->p_align == 0)
+	    l->l_tls_firstbyte_offset = 0;
+	  else
+	    l->l_tls_firstbyte_offset = ph->p_vaddr & (ph->p_align - 1);
 	  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.  */

elf/dl-reloc.c

@@ -41,27 +41,39 @@
    the static TLS area already allocated for each running thread.  If this
    object's TLS segment is too big to fit, we fail.  If it fits,
    we set MAP->l_tls_offset and return.  */
-void
+int
 internal_function __attribute_noinline__
 _dl_allocate_static_tls (struct link_map *map)
 {
   size_t offset;
   size_t used;
   size_t check;
+  size_t freebytes;
+  size_t n;
+  size_t blsize;
+
+  /* If the alignment requirements are too high fail.  */
+  if (map->l_tls_align > GL(dl_tls_static_align))
+    return 1;
 
 # if TLS_TCB_AT_TP
-  offset = roundup (GL(dl_tls_static_used) + map->l_tls_blocksize,
-		    map->l_tls_align);
-  used = offset;
-  check = offset + TLS_TCB_SIZE;
+  freebytes = GL(dl_tls_static_size) - GL(dl_tls_static_used) - TLS_TCB_SIZE;
+
+  blsize = map->l_tls_blocksize + map->l_tls_firstbyte_offset;
+  if (freebytes < blsize)
+    return 1;
+
+  n = (freebytes - blsize) / map->l_tls_align;
+
+  offset = GL(dl_tls_static_used) + (freebytes - n * map->l_tls_align
+				     - map->l_tls_firstbyte_offset);
+
+  map->l_tls_offset = GL(dl_tls_static_used) = offset;
 # elif TLS_DTV_AT_TP
   offset = roundup (GL(dl_tls_static_used), map->l_tls_align);
   used = offset + map->l_tls_blocksize;
   check = used;
   /* dl_tls_static_used includes the TCB at the beginning.  */
-# else
-#  error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
-# endif
 
   if (check > GL(dl_tls_static_size))
     {
@@ -72,6 +84,11 @@ shared object cannot be dlopen()ed: static TLS memory too small");
 
   map->l_tls_offset = offset;
   GL(dl_tls_static_used) = used;
+# else
+#  error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
+# endif
+
+  return 0;
 }
 #endif
 
@@ -212,7 +229,9 @@ _dl_relocate_object (struct link_map *l, struct r_scope_elem *scope[],
 #define CHECK_STATIC_TLS(map, sym_map)					      \
     do {								      \
       if (__builtin_expect ((sym_map)->l_tls_offset == NO_TLS_OFFSET, 0))     \
-	_dl_allocate_static_tls (sym_map);				      \
+	if (_dl_allocate_static_tls (sym_map) != 0)			      \
+ 	  INTUSE(_dl_signal_error) (0, sym_map->l_name, NULL, N_("\
+cannot allocate memory in static TLS block"));				      \
     } while (0)
 
 #include "dynamic-link.h"

elf/rtld.c

@@ -227,6 +227,7 @@ _dl_start_final (void *arg, struct dl_start_final_info *info)
   assert (info->l.l_tls_modid != 0);
   GL(dl_rtld_map).l_tls_blocksize = info->l.l_tls_blocksize;
   GL(dl_rtld_map).l_tls_align = info->l.l_tls_align;
+  GL(dl_rtld_map).l_tls_firstbyte_offset = info->l.l_tls_firstbyte_offset;
   GL(dl_rtld_map).l_tls_initimage_size = info->l.l_tls_initimage_size;
   GL(dl_rtld_map).l_tls_initimage = info->l.l_tls_initimage;
   GL(dl_rtld_map).l_tls_offset = info->l.l_tls_offset;
@@ -347,6 +348,11 @@ _dl_start (void *arg)
 
 	bootstrap_map.l_tls_blocksize = phdr[cnt].p_memsz;
 	bootstrap_map.l_tls_align = phdr[cnt].p_align;
+	if (phdr[cnt].p_align == 0)
+	  bootstrap_map.l_tls_firstbyte_offset = 0;
+	else
+	  bootstrap_map.l_tls_firstbyte_offset = (phdr[cnt].p_vaddr
+						  & (phdr[cnt].p_align - 1));
 	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
@@ -860,6 +866,11 @@ of this helper program; chances are you did not intend to run this program.\n\
 	       check for this special but unimportant case.  */
 	    GL(dl_loaded)->l_tls_blocksize = ph->p_memsz;
 	    GL(dl_loaded)->l_tls_align = ph->p_align;
+	    if (ph->p_align == 0)
+	      GL(dl_loaded)->l_tls_firstbyte_offset = 0;
+	    else
+	      GL(dl_loaded)->l_tls_firstbyte_offset = (ph->p_vaddr
+						       & (ph->p_align - 1));
 	    GL(dl_loaded)->l_tls_initimage_size = ph->p_filesz;
 	    GL(dl_loaded)->l_tls_initimage = (void *) ph->p_vaddr;
 

elf/tst-tls14.c

@@ -0,0 +1,56 @@
+/* Check alignment of TLS variable.  */
+#include <dlfcn.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+
+#define AL 4096
+struct foo
+{
+  int i;
+} __attribute ((aligned (AL)));
+
+static __thread struct foo f;
+static struct foo g;
+
+
+extern int in_dso1 (void);
+
+
+static int
+do_test (void)
+{
+  int result = 0;
+
+  int fail = (((uintptr_t) &f) & (AL - 1)) != 0;
+  printf ("&f = %p %s\n", &f, fail ? "FAIL" : "OK");
+  result |= fail;
+
+  fail = (((uintptr_t) &g) & (AL - 1)) != 0;
+  printf ("&g = %p %s\n", &g, fail ? "FAIL" : "OK");
+  result |= fail;
+
+  result |= in_dso1 ();
+
+  void *h = dlopen ("tst-tlsmod14b.so", RTLD_LAZY);
+  if (h == NULL)
+    {
+      printf ("cannot open tst-tlsmod14b.so: %m\n");
+      exit (1);
+    }
+
+  int (*fp) (void) = (int (*) (void)) dlsym (h, "in_dso2");
+  if (fp == NULL)
+    {
+      puts ("cannot find in_dso2");
+      exit (1);
+    }
+
+  result |= fp ();
+
+  return result;
+}
+
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"

elf/tst-tlsmod14a.c

@@ -0,0 +1,36 @@
+#include <stdint.h>
+#include <stdio.h>
+
+
+#define AL 4096
+struct foo
+{
+  int i;
+} __attribute ((aligned (AL)));
+
+static __thread struct foo f;
+static struct foo g;
+
+
+#ifndef FCT
+# define FCT in_dso1
+#endif
+
+
+int
+FCT (void)
+{
+  puts (__func__);
+
+  int result = 0;
+
+  int fail = (((uintptr_t) &f) & (AL - 1)) != 0;
+  printf ("&f = %p %s\n", &f, fail ? "FAIL" : "OK");
+  result |= fail;
+
+  fail = (((uintptr_t) &g) & (AL - 1)) != 0;
+  printf ("&g = %p %s\n", &g, fail ? "FAIL" : "OK");
+  result |= fail;
+
+  return result;
+}

elf/tst-tlsmod14b.c

@@ -0,0 +1,2 @@
+#define FCT in_dso2
+#include "tst-tlsmod14a.c"

include/link.h

@@ -1,6 +1,6 @@
 /* Data structure for communication from the run-time dynamic linker for
    loaded ELF shared objects.
-   Copyright (C) 1995-1999,2000,01,02 Free Software Foundation, Inc.
+   Copyright (C) 1995-2002, 2003 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
@@ -268,6 +268,8 @@ struct link_map
     size_t l_tls_blocksize;
     /* Alignment requirement of the TLS block.  */
     size_t l_tls_align;
+    /* Offset of first byte module alignment.  */
+    size_t l_tls_firstbyte_offset;
 # ifndef NO_TLS_OFFSET
 #  define NO_TLS_OFFSET	0
 # endif

sysdeps/generic/dl-tls.c

@@ -117,9 +117,10 @@ internal_function
 _dl_determine_tlsoffset (void)
 {
   struct dtv_slotinfo *slotinfo;
-  size_t max_align = __alignof__ (void *);
+  size_t max_align;
   size_t offset;
   size_t cnt;
+  size_t freebytes;
 
   /* The first element of the dtv slot info list is allocated.  */
   assert (GL(dl_tls_dtv_slotinfo_list) != NULL);
@@ -127,24 +128,78 @@ _dl_determine_tlsoffset (void)
      dl_tls_dtv_slotinfo_list list.  */
   assert (GL(dl_tls_dtv_slotinfo_list)->next == NULL);
 
+  /* Determining the offset of the various parts of the static TLS
+     block has several dependencies.  In addition we have to work
+     around bugs in some toolchains.
+
+     Each TLS block from the objects available at link time has a size
+     and an alignment requirement.  The GNU ld computes the alignment
+     requirements for the data at the positions *in the file*, though.
+     I.e, it is not simply possible to allocate a block with the size
+     of the TLS program header entry.  The data is layed out assuming
+     that the first byte of the TLS block fulfills
+
+       p_vaddr mod p_align == &TLS_BLOCK mod p_align
+
+     This means we have to add artificial padding at the beginning of
+     the TLS block.  These bytes are never used for the TLS data in
+     this module but the first byte allocated must be aligned
+     according to mod p_align == 0 so that the first byte of the TLS
+     block is aligned according to p_vaddr mod p_align.  This is ugly
+     and the linker can help by computing the offsets in the TLS block
+     assuming the first byte of the TLS block is aligned according to
+     p_align.
+
+     We can handle this wrong behavior because of another bug in GNU
+     ld.  The p_vaddr field of the TLS segment must be zero (according
+     to the spec) since the linker does not know the address or offset
+     where it will end up at.  Once a linker is available which
+     handles the alignment correctly it should set p_addr to zero and
+     all will automatically fall into place.
+
+     The extra space which might be allocated before the first byte of
+     the TLS block need not go unused.  The code below tries to use
+     that memory for the next TLS block.  This can work if the total
+     memory requirement for the next TLS block is smaller than the
+     gap.  */
+
 # if TLS_TCB_AT_TP
   /* We simply start with zero.  */
+  max_align = __alignof (void *);
   offset = 0;
+  freebytes = 0;
 
   slotinfo = GL(dl_tls_dtv_slotinfo_list)->slotinfo;
   for (cnt = 1; slotinfo[cnt].map != NULL; ++cnt)
     {
       assert (cnt < GL(dl_tls_dtv_slotinfo_list)->len);
 
-      max_align = MAX (max_align, slotinfo[cnt].map->l_tls_align);
+      size_t blsize = (slotinfo[cnt].map->l_tls_blocksize
+		       + slotinfo[cnt].map->l_tls_firstbyte_offset);
 
-      /* Compute the offset of the next TLS block.  */
-      offset = roundup (offset + slotinfo[cnt].map->l_tls_blocksize,
-			slotinfo[cnt].map->l_tls_align);
+      if (blsize <= freebytes)
+	{
+	  /* When we come here the amount of memory we was "wasted"
+             for the correct alignment of the previous block is larger
+             than what we need for this module.  So use it.  */
+	  size_t n = (freebytes - blsize) / slotinfo[cnt].map->l_tls_align;
+	  freebytes = (n * slotinfo[cnt].map->l_tls_align
+		       + slotinfo[cnt].map->l_tls_firstbyte_offset);
+	}
+      else
+	{
+	  /* There is either no gap from the bottom of the static TLS
+             block to the first used byte or the gap is too small.
+             Extend the static TLS block.  */
+	  offset += roundup (blsize, max_align);
+	  freebytes = slotinfo[cnt].map->l_tls_firstbyte_offset;
+	}
+
+      max_align = MAX (max_align, slotinfo[cnt].map->l_tls_align);
 
       /* XXX For some architectures we perhaps should store the
 	 negative offset.  */
-      slotinfo[cnt].map->l_tls_offset = offset;
+      slotinfo[cnt].map->l_tls_offset = offset - freebytes;
     }
 
   /* The thread descriptor (pointed to by the thread pointer) has its
@@ -156,11 +211,12 @@ _dl_determine_tlsoffset (void)
   // XXX model.
 
   GL(dl_tls_static_used) = offset;
-  GL(dl_tls_static_size) = roundup (offset + TLS_STATIC_SURPLUS + TLS_TCB_SIZE,
-				    TLS_TCB_ALIGN);
+  GL(dl_tls_static_size) = (offset + roundup (TLS_STATIC_SURPLUS, max_align)
+			    + TLS_TCB_SIZE);
 # elif TLS_DTV_AT_TP
   /* The TLS blocks start right after the TCB.  */
   offset = TLS_TCB_SIZE;
+  max_align = __alignof (void *);
 
   /* The first block starts right after the TCB.  */
   slotinfo = GL(dl_tls_dtv_slotinfo_list)->slotinfo;
@@ -201,7 +257,7 @@ _dl_determine_tlsoffset (void)
 # endif
 
   /* The alignment requirement for the static TLS block.  */
-  GL(dl_tls_static_align) = MAX (TLS_TCB_ALIGN, max_align);
+  GL(dl_tls_static_align) = max_align;
 }
 
 

sysdeps/generic/ldsodefs.h

@@ -797,7 +797,7 @@ rtld_hidden_proto (_dl_allocate_tls)
 extern void _dl_get_tls_static_info (size_t *sizep, size_t *alignp)
      internal_function;
 
-extern void _dl_allocate_static_tls (struct link_map *map)
+extern int _dl_allocate_static_tls (struct link_map *map)
      internal_function attribute_hidden;
 
 /* These are internal entry points to the two halves of _dl_allocate_tls,