Support TLS x32 GD->IE, GD->LE and LD->LE transitions.

bfd/

2011-03-12  H.J. Lu  <hongjiu.lu@intel.com>

	* elf64-x86-64.c (elf_x86_64_check_tls_transition): Supprt TLS
	x32 GD->IE and GD->LE transitions.
	(elf_x86_64_relocate_section): Supprt TLS x32 GD->IE, GD->LE
	and LD->LE transitions.

ld/testsuite/

2011-03-12  H.J. Lu  <hongjiu.lu@intel.com>

	* ld-x86-64/tlsgd4.dd: New.
	* ld-x86-64/tlsgd4.s: Likewise.
	* ld-x86-64/tlsgd5.dd: Likewise.
	* ld-x86-64/tlsgd5a.s: Likewise.
	* ld-x86-64/tlsgd5b.s: Likewise.
	* ld-x86-64/tlsgd6.dd: Likewise.
	* ld-x86-64/tlsgd6a.s: Likewise.
	* ld-x86-64/tlsgd6b.s: Likewise.
	* ld-x86-64/tlsld2.dd: Likewise.
	* ld-x86-64/tlsld2.s: Likewise.

	* ld-x86-64/x86-64.exp (x86_64tests): Add tlsgd4, libtlsgd5.so,
	tlsgd5, libtlsgd6.so, tlsgd6 and tlsld2.

bfd/ChangeLog

@@ -1,3 +1,10 @@
+2011-03-12  H.J. Lu  <hongjiu.lu@intel.com>
+
+	* elf64-x86-64.c (elf_x86_64_check_tls_transition): Supprt TLS
+	x32 GD->IE and GD->LE transitions.
+	(elf_x86_64_relocate_section): Supprt TLS x32 GD->IE, GD->LE
+	and LD->LE transitions.
+
 2011-03-10  Alan Modra  <amodra@gmail.com>
 
 	* elf64-ppc.c (ppc64_elf_relocate_section): Provide section/offset

bfd/elf64-x86-64.c

@@ -865,18 +865,34 @@ elf_x86_64_check_tls_transition (bfd *abfd,
 
       if (r_type == R_X86_64_TLSGD)
 	{
-	  /* Check transition from GD access model.  Only
+	  /* Check transition from GD access model.  For 64bit, only
 		.byte 0x66; leaq foo@tlsgd(%rip), %rdi
 		.word 0x6666; rex64; call __tls_get_addr
+	     can transit to different access model.  For 32bit, only
+		leaq foo@tlsgd(%rip), %rdi
+		.word 0x6666; rex64; call __tls_get_addr
 	     can transit to different access model.  */
 
-	  static x86_64_opcode32 leaq = { { 0x66, 0x48, 0x8d, 0x3d } },
+	  static x86_64_opcode32 call = { { 0x66, 0x66, 0x48, 0xe8 } };
-				 call = { { 0x66, 0x66, 0x48, 0xe8 } };
+	  if ((offset + 12) > sec->size
-	  if (offset < 4
-	      || (offset + 12) > sec->size
-	      || bfd_get_32 (abfd, contents + offset - 4) != leaq.i
 	      || bfd_get_32 (abfd, contents + offset + 4) != call.i)
 	    return FALSE;
+
+	  if (ABI_64_P (abfd))
+	    {
+	      static x86_64_opcode32 leaq = { { 0x66, 0x48, 0x8d, 0x3d } };
+	      if (offset < 4
+		  || bfd_get_32 (abfd, contents + offset - 4) != leaq.i)
+		return FALSE;
+	    }
+	  else
+	    {
+	      static x86_64_opcode16 lea = { { 0x8d, 0x3d } };
+	      if (offset < 3
+		  || bfd_get_8 (abfd, contents + offset - 3) != 0x48
+		  || bfd_get_16 (abfd, contents + offset - 2) != lea.i)
+		return FALSE;
+	    }
 	}
       else
 	{
@@ -3434,15 +3450,26 @@ elf_x86_64_relocate_section (bfd *output_bfd,
 
 	      if (ELF32_R_TYPE (rel->r_info) == R_X86_64_TLSGD)
 		{
-		  /* GD->LE transition.
+		  /* GD->LE transition.  For 64bit, change
 		     .byte 0x66; leaq foo@tlsgd(%rip), %rdi
 		     .word 0x6666; rex64; call __tls_get_addr
-		     Change it into:
+		     into:
 		     movq %fs:0, %rax
+		     leaq foo@tpoff(%rax), %rax
+		     For 32bit, change
+		     leaq foo@tlsgd(%rip), %rdi
+		     .word 0x6666; rex64; call __tls_get_addr
+		     into:
+		     movl %fs:0, %eax
 		     leaq foo@tpoff(%rax), %rax */
-		  memcpy (contents + roff - 4,
+		  if (ABI_64_P (output_bfd))
-			  "\x64\x48\x8b\x04\x25\0\0\0\0\x48\x8d\x80\0\0\0",
+		    memcpy (contents + roff - 4,
-			  16);
+			    "\x64\x48\x8b\x04\x25\0\0\0\0\x48\x8d\x80\0\0\0",
+			    16);
+		  else
+		    memcpy (contents + roff - 3,
+			    "\x64\x8b\x04\x25\0\0\0\0\x48\x8d\x80\0\0\0",
+			    15);
 		  bfd_put_32 (output_bfd,
 			      elf_x86_64_tpoff (info, relocation),
 			      contents + roff + 8);
@@ -3670,15 +3697,26 @@ elf_x86_64_relocate_section (bfd *output_bfd,
 
 	      if (ELF32_R_TYPE (rel->r_info) == R_X86_64_TLSGD)
 		{
-		  /* GD->IE transition.
+		  /* GD->IE transition.  For 64bit, change
 		     .byte 0x66; leaq foo@tlsgd(%rip), %rdi
 		     .word 0x6666; rex64; call __tls_get_addr@plt
-		     Change it into:
+		     into:
 		     movq %fs:0, %rax
+		     addq foo@gottpoff(%rip), %rax
+		     For 32bit, change
+		     leaq foo@tlsgd(%rip), %rdi
+		     .word 0x6666; rex64; call __tls_get_addr@plt
+		     into:
+		     movl %fs:0, %eax
 		     addq foo@gottpoff(%rip), %rax */
-		  memcpy (contents + roff - 4,
+		  if (ABI_64_P (output_bfd))
-			  "\x64\x48\x8b\x04\x25\0\0\0\0\x48\x03\x05\0\0\0",
+		    memcpy (contents + roff - 4,
-			  16);
+			    "\x64\x48\x8b\x04\x25\0\0\0\0\x48\x03\x05\0\0\0",
+			    16);
+		  else
+		    memcpy (contents + roff - 3,
+			    "\x64\x8b\x04\x25\0\0\0\0\x48\x03\x05\0\0\0",
+			    15);
 
 		  relocation = (htab->elf.sgot->output_section->vma
 				+ htab->elf.sgot->output_offset + off
@@ -3747,12 +3785,18 @@ elf_x86_64_relocate_section (bfd *output_bfd,
 	    {
 	      /* LD->LE transition:
 		 leaq foo@tlsld(%rip), %rdi; call __tls_get_addr.
-		 We change it into:
+		 For 64bit, we change it into:
-		 .word 0x6666; .byte 0x66; movl %fs:0, %rax.  */
+		 .word 0x6666; .byte 0x66; movq %fs:0, %rax.
+		 For 32bit, we change it into:
+		 nopl 0x0(%rax); movl %fs:0, %eax.  */
 
 	      BFD_ASSERT (r_type == R_X86_64_TPOFF32);
-	      memcpy (contents + rel->r_offset - 3,
+	      if (ABI_64_P (output_bfd))
-		      "\x66\x66\x66\x64\x48\x8b\x04\x25\0\0\0", 12);
+		memcpy (contents + rel->r_offset - 3,
+			"\x66\x66\x66\x64\x48\x8b\x04\x25\0\0\0", 12);
+	      else
+		memcpy (contents + rel->r_offset - 3,
+			"\x0f\x1f\x40\x00\x64\x8b\x04\x25\0\0\0", 12);
 	      /* Skip R_X86_64_PC32/R_X86_64_PLT32.  */
 	      rel++;
 	      continue;

ld/testsuite/ChangeLog

@@ -1,3 +1,19 @@
+2011-03-12  H.J. Lu  <hongjiu.lu@intel.com>
+
+	* ld-x86-64/tlsgd4.dd: New.
+	* ld-x86-64/tlsgd4.s: Likewise.
+	* ld-x86-64/tlsgd5.dd: Likewise.
+	* ld-x86-64/tlsgd5a.s: Likewise.
+	* ld-x86-64/tlsgd5b.s: Likewise.
+	* ld-x86-64/tlsgd6.dd: Likewise.
+	* ld-x86-64/tlsgd6a.s: Likewise.
+	* ld-x86-64/tlsgd6b.s: Likewise.
+	* ld-x86-64/tlsld2.dd: Likewise.
+	* ld-x86-64/tlsld2.s: Likewise.
+
+	* ld-x86-64/x86-64.exp (x86_64tests): Add tlsgd4, libtlsgd5.so,
+	tlsgd5, libtlsgd6.so, tlsgd6 and tlsld2.
+
 2011-03-03  H.J. Lu  <hongjiu.lu@intel.com>
 
 	* ld-x86-64/tlsie4.dd: New.

ld/testsuite/ld-x86-64/tlsgd4.dd

@@ -0,0 +1,10 @@
+#target: x86_64-*-linux*
+
+.*: +file format .*
+
+Disassembly of section .text:
+
+[a-f0-9]+ <_start>:
+[ 	]*[a-f0-9]+:	64 8b 04 25 00 00 00 00 	mov    %fs:0x0,%eax
+[ 	]*[a-f0-9]+:	48 8d 80 fc ff ff ff 	lea    -0x4\(%rax\),%rax
+#pass

ld/testsuite/ld-x86-64/tlsgd4.s

@@ -0,0 +1,14 @@
+	.text
+	.globl _start
+_start:
+	leaq	foo@TLSGD(%rip), %rdi
+	.word	0x6666
+	rex64
+	call	__tls_get_addr
+	.globl foo
+	.section	.tdata,"awT",@progbits
+	.align 4
+	.type	foo, @object
+	.size	foo, 4
+foo:
+	.long	100

ld/testsuite/ld-x86-64/tlsgd5.dd

@@ -0,0 +1,14 @@
+#source: tlsgd5a.s
+#as: --64
+#ld: -melf_x86_64 tmpdir/tlsgd5
+#objdump: -drw
+#target: x86_64-*-linux*
+
+.*: +file format .*
+
+Disassembly of section .text:
+
+[a-f0-9]+ <_start>:
+[ 	]*[a-f0-9]+:	64 48 8b 04 25 00 00 00 00 	mov    %fs:0x0,%rax
+[ 	]*[a-f0-9]+:	48 03 05 00 01 20 00 	add    0x200100\(%rip\),%rax        # 600368 <_DYNAMIC\+0x100>
+#pass

ld/testsuite/ld-x86-64/tlsgd5a.s

@@ -0,0 +1,8 @@
+	.text
+	.globl _start
+_start:
+	.byte	0x66
+	leaq	foo@TLSGD(%rip), %rdi
+	.word	0x6666
+	rex64
+	call	__tls_get_addr

ld/testsuite/ld-x86-64/tlsgd5b.s

@@ -0,0 +1,7 @@
+	.globl foo
+	.section	.tdata,"awT",@progbits
+	.align 4
+	.type	foo, @object
+	.size	foo, 4
+foo:
+	.long	100

ld/testsuite/ld-x86-64/tlsgd6.dd

@@ -0,0 +1,14 @@
+#source: tlsgd6a.s
+#as: --x32
+#ld: -melf32_x86_64 tmpdir/tlsgd6
+#objdump: -drw
+#target: x86_64-*-linux*
+
+.*: +file format .*
+
+Disassembly of section .text:
+
+[a-f0-9]+ <_start>:
+[ 	]*[a-f0-9]+:	64 8b 04 25 00 00 00 00 	mov    %fs:0x0,%eax
+[ 	]*[a-f0-9]+:	48 03 05 81 00 20 00 	add    0x200081\(%rip\),%rax        # 60022c <_DYNAMIC\+0x80>
+#pass

ld/testsuite/ld-x86-64/tlsgd6a.s

@@ -0,0 +1,7 @@
+	.text
+	.globl _start
+_start:
+	leaq	foo@TLSGD(%rip), %rdi
+	.word	0x6666
+	rex64
+	call	__tls_get_addr

ld/testsuite/ld-x86-64/tlsgd6b.s

@@ -0,0 +1,7 @@
+	.globl foo
+	.section	.tdata,"awT",@progbits
+	.align 4
+	.type	foo, @object
+	.size	foo, 4
+foo:
+	.long	100

ld/testsuite/ld-x86-64/tlsld2.dd

@@ -0,0 +1,14 @@
+#source: tlsld2.s
+#as: --x32
+#ld: -melf32_x86_64 tmpdir/tlsld2
+#objdump: -drw
+#target: x86_64-*-linux*
+
+.*: +file format .*
+
+Disassembly of section .text:
+
+[a-f0-9]+ <_start>:
+[ 	]*[a-f0-9]+:	0f 1f 40 00          	nopl   0x0\(%rax\)
+[ 	]*[a-f0-9]+:	64 8b 04 25 00 00 00 00 	mov    %fs:0x0,%eax
+#pass

ld/testsuite/ld-x86-64/tlsld2.s

@@ -0,0 +1,12 @@
+	.text
+	.globl _start
+_start:
+	leaq    foo@TLSLD(%rip), %rdi
+	call    __tls_get_addr
+	.globl foo
+	.section	.tdata,"awT",@progbits
+	.align 4
+	.type	foo, @object
+	.size	foo, 4
+foo:
+	.long	100

ld/testsuite/ld-x86-64/x86-64.exp

@@ -90,6 +90,22 @@ set x86_64tests {
     {"TLS X32 IE->LE transition" "-melf32_x86_64"
      "--x32" {tlsie4.s}
      {{objdump -dwr tlsie4.dd}} "tlsie4"}
+    {"TLS X32 GD->LE transition" "-melf32_x86_64"
+     "--x32" {tlsgd4.s}
+     {{objdump -dwr tlsgd4.dd}} "tlsgd4"}
+    {"Helper TLS GD->IE transition DSO" "-shared -melf_x86_64"
+     "--64" {tlsgd5b.s} {} "libtlsgd5.so"}
+    {"TLS GD->IE transition" "-melf_x86_64 tmpdir/libtlsgd5.so"
+     "--64" {tlsgd5a.s}
+     {{objdump -dwr tlsgd5.dd}} "tlsgd5"}
+    {"Helper TLS X32 GD->IE transition DSO" "-shared -melf32_x86_64"
+     "--x32" {tlsgd6b.s} {} "libtlsgd6.so"}
+    {"TLS X32 GD->IE transition" "-melf32_x86_64 tmpdir/libtlsgd6.so"
+     "--x32" {tlsgd6a.s}
+     {{objdump -dwr tlsgd6.dd}} "tlsgd6"}
+    {"TLS X32 LD->LE transition" "-melf32_x86_64"
+     "--x32" {tlsld2.s}
+     {{objdump -dwr tlsld2.dd}} "tlsld2"}
 }
 
 run_ld_link_tests $x86_64tests