This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
[PATCH] Change x86-64 TLSGD sequences
- From: Jakub Jelinek <jakub at redhat dot com>
- To: binutils at sources dot redhat dot com
- Cc: aj at suse dot de, jh at suse dot cz
- Date: Tue, 1 Oct 2002 11:04:13 +0200
- Subject: [PATCH] Change x86-64 TLSGD sequences
- Reply-to: Jakub Jelinek <jakub at redhat dot com>
Hi!
x86-64 prefers to have at most 3 prefixes for an instruction, so TLSGD
sequence was changed from
.long 0x666666; leaq foo@tlsgd(%rip), %rdi
call __tls_get_addr@plt
to
.byte 0x66; leaq foo@tlsgd(%rip), %rdi
.word 0x6666; rex64; call __tls_get_addr@plt
Ok to commit?
2002-10-01 Jakub Jelinek <jakub@redhat.com>
bfd/
* elf64-x86-64.c (elf64_x86_64_relocate_section): Change TLSGD
sequence and its transitions.
ld/testsuite/
* ld-x86-64/tlspic1.s: Change TLSGD sequences.
* ld-x86-64/tlsbinpic.s: Likewise.
* ld-x86-64/tlspic.dd: Adjust.
--- bfd/elf64-x86-64.c.jj 2002-10-01 10:29:06.000000000 +0200
+++ bfd/elf64-x86-64.c 2002-10-01 10:30:25.000000000 +0200
@@ -2224,31 +2224,32 @@ elf64_x86_64_relocate_section (output_bf
if (ELF64_R_TYPE (rel->r_info) == R_X86_64_TLSGD)
{
unsigned int i;
- static unsigned char tlsgd[7]
- = { 0x66, 0x66, 0x66, 0x66, 0x48, 0x8d, 0x3d };
+ static unsigned char tlsgd[8]
+ = { 0x66, 0x48, 0x8d, 0x3d, 0x66, 0x66, 0x48, 0xe8 };
/* GD->LE transition.
- .long 0x66666666; leaq foo@tlsgd(%rip), %rdi
- callq __tls_get_addr@plt
+ .byte 0x66; leaq foo@tlsgd(%rip), %rdi
+ .word 0x6666; rex64; call __tls_get_addr@plt
Change it into:
movq %fs:0, %rax
leaq foo@tpoff(%rax), %rax */
- BFD_ASSERT (rel->r_offset >= 7);
- for (i = 0; i < 7; i++)
+ BFD_ASSERT (rel->r_offset >= 4);
+ for (i = 0; i < 4; i++)
BFD_ASSERT (bfd_get_8 (input_bfd,
- contents + rel->r_offset - 7 + i)
+ contents + rel->r_offset - 4 + i)
== tlsgd[i]);
- BFD_ASSERT (rel->r_offset + 9 <= input_section->_raw_size);
- BFD_ASSERT (bfd_get_8 (input_bfd,
- contents + rel->r_offset + 4)
- == 0xe8);
+ BFD_ASSERT (rel->r_offset + 12 <= input_section->_raw_size);
+ for (i = 0; i < 4; i++)
+ BFD_ASSERT (bfd_get_8 (input_bfd,
+ contents + rel->r_offset + 4 + i)
+ == tlsgd[i+4]);
BFD_ASSERT (rel + 1 < relend);
BFD_ASSERT (ELF64_R_TYPE (rel[1].r_info) == R_X86_64_PLT32);
- memcpy (contents + rel->r_offset - 7,
+ memcpy (contents + rel->r_offset - 4,
"\x64\x48\x8b\x04\x25\0\0\0\0\x48\x8d\x80\0\0\0",
16);
bfd_put_32 (output_bfd, tpoff (info, relocation),
- contents + rel->r_offset + 5);
+ contents + rel->r_offset + 8);
/* Skip R_X86_64_PLT32. */
rel++;
continue;
@@ -2397,27 +2398,28 @@ elf64_x86_64_relocate_section (output_bf
else
{
unsigned int i;
- static unsigned char tlsgd[7]
- = { 0x66, 0x66, 0x66, 0x66, 0x48, 0x8d, 0x3d };
+ static unsigned char tlsgd[8]
+ = { 0x66, 0x48, 0x8d, 0x3d, 0x66, 0x66, 0x48, 0xe8 };
/* GD->IE transition.
- .long 0x66666666; leaq foo@tlsgd(%rip), %rdi
- callq __tls_get_addr@plt
+ .byte 0x66; leaq foo@tlsgd(%rip), %rdi
+ .word 0x6666; rex64; call __tls_get_addr@plt
Change it into:
movq %fs:0, %rax
addq foo@gottpoff(%rip), %rax */
- BFD_ASSERT (rel->r_offset >= 7);
- for (i = 0; i < 7; i++)
+ BFD_ASSERT (rel->r_offset >= 4);
+ for (i = 0; i < 4; i++)
BFD_ASSERT (bfd_get_8 (input_bfd,
- contents + rel->r_offset - 7 + i)
+ contents + rel->r_offset - 4 + i)
== tlsgd[i]);
- BFD_ASSERT (rel->r_offset + 9 <= input_section->_raw_size);
- BFD_ASSERT (bfd_get_8 (input_bfd,
- contents + rel->r_offset + 4)
- == 0xe8);
+ BFD_ASSERT (rel->r_offset + 12 <= input_section->_raw_size);
+ for (i = 0; i < 4; i++)
+ BFD_ASSERT (bfd_get_8 (input_bfd,
+ contents + rel->r_offset + 4 + i)
+ == tlsgd[i+4]);
BFD_ASSERT (rel + 1 < relend);
BFD_ASSERT (ELF64_R_TYPE (rel[1].r_info) == R_X86_64_PLT32);
- memcpy (contents + rel->r_offset - 7,
+ memcpy (contents + rel->r_offset - 4,
"\x64\x48\x8b\x04\x25\0\0\0\0\x48\x03\x05\0\0\0",
16);
@@ -2426,9 +2428,9 @@ elf64_x86_64_relocate_section (output_bf
- rel->r_offset
- input_section->output_section->vma
- input_section->output_offset
- - 9);
+ - 12);
bfd_put_32 (output_bfd, relocation,
- contents + rel->r_offset + 5);
+ contents + rel->r_offset + 8);
/* Skip R_X86_64_PLT32. */
rel++;
continue;
--- ld/testsuite/ld-x86-64/tlspic1.s.jj 2002-09-26 16:11:31.000000000 +0200
+++ ld/testsuite/ld-x86-64/tlspic1.s 2002-10-01 10:30:25.000000000 +0200
@@ -41,52 +41,68 @@ fn1:
nop;nop;nop;nop
/* GD */
- .long 0x66666666
+ .byte 0x66
leaq sg1@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD -> IE because variable is referenced through IE too */
- .long 0x66666666
+ .byte 0x66
leaq sg2@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD against local variable */
- .long 0x66666666
+ .byte 0x66
leaq sl1@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD -> IE against local variable referenced through IE too */
- .long 0x66666666
+ .byte 0x66
leaq sl2@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD against hidden and local variable */
- .long 0x66666666
+ .byte 0x66
leaq sh1@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD -> IE against hidden and local variable referenced through
IE too */
- .long 0x66666666
+ .byte 0x66
leaq sh2@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD against hidden but not local variable */
- .long 0x66666666
+ .byte 0x66
leaq sH1@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD -> IE against hidden but not local variable referenced through
IE too */
- .long 0x66666666
+ .byte 0x66
leaq sH2@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
--- ld/testsuite/ld-x86-64/tlsbinpic.s.jj 2002-10-01 10:29:06.000000000 +0200
+++ ld/testsuite/ld-x86-64/tlsbinpic.s 2002-10-01 10:30:25.000000000 +0200
@@ -40,33 +40,43 @@ fn2:
movq %rsp, %rbp
/* GD -> IE because variable is not defined in executable */
- .long 0x66666666
+ .byte 0x66
leaq sG1@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD -> IE because variable is not defined in executable where
the variable is referenced through IE too */
- .long 0x66666666
+ .byte 0x66
leaq sG2@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD -> LE with global variable defined in executable */
- .long 0x66666666
+ .byte 0x66
leaq sg1@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD -> LE with local variable defined in executable */
- .long 0x66666666
+ .byte 0x66
leaq sl1@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
/* GD -> LE with hidden variable defined in executable */
- .long 0x66666666
+ .byte 0x66
leaq sh1@tlsgd(%rip), %rdi
+ .word 0x6666
+ rex64
call __tls_get_addr@plt
nop;nop;nop;nop
--- ld/testsuite/ld-x86-64/tlspic.dd.jj 2002-09-26 16:20:08.000000000 +0200
+++ ld/testsuite/ld-x86-64/tlspic.dd 2002-10-01 10:30:25.000000000 +0200
@@ -17,10 +17,12 @@ Disassembly of section .text:
+1006: 90[ ]+nop *
+1007: 90[ ]+nop *
# GD
- +1008: 66 66 66 66 48 8d 3d[ ]+lea 1053165\(%rip\),%rdi +# 102200 <_GLOBAL_OFFSET_TABLE_\+0x70>
- +100f: ed 11 10 00 *
+ +1008: 66 48 8d 3d f0 11 10[ ]+lea 1053168\(%rip\),%rdi +# 102200 <_GLOBAL_OFFSET_TABLE_\+0x70>
+ +100f: 00 *
# -> R_X86_64_DTPMOD64 sg1
- +1013: e8 68 f6 ff ff[ ]+callq [0-9a-f]+ <.*>
+ +1010: 66[ ]+data16
+ +1011: 66[ ]+data16
+ +1012: 48 e8 68 f6 ff ff[ ]+rex64 callq [0-9a-f]+ <.*>
# -> R_X86_64_JUMP_SLOT __tls_get_addr
+1018: 90[ ]+nop *
+1019: 90[ ]+nop *
@@ -36,10 +38,12 @@ Disassembly of section .text:
+102e: 90[ ]+nop *
+102f: 90[ ]+nop *
# GD against local variable
- +1030: 66 66 66 66 48 8d 3d[ ]+lea 1053045\(%rip\),%rdi +# 1021b0 <_GLOBAL_OFFSET_TABLE_\+0x20>
- +1037: 75 11 10 00 *
+ +1030: 66 48 8d 3d 78 11 10[ ]+lea 1053048\(%rip\),%rdi +# 1021b0 <_GLOBAL_OFFSET_TABLE_\+0x20>
+ +1037: 00 *
# -> R_X86_64_DTPMOD64 [0 0x2000000000000000]
- +103b: e8 40 f6 ff ff[ ]+callq [0-9a-f]+ <.*>
+ +1038: 66[ ]+data16
+ +1039: 66[ ]+data16
+ +103a: 48 e8 40 f6 ff ff[ ]+rex64 callq [0-9a-f]+ <.*>
# -> R_X86_64_JUMP_SLOT __tls_get_addr
+1040: 90[ ]+nop *
+1041: 90[ ]+nop *
@@ -55,10 +59,12 @@ Disassembly of section .text:
+1056: 90[ ]+nop *
+1057: 90[ ]+nop *
# GD against hidden and local variable
- +1058: 66 66 66 66 48 8d 3d[ ]+lea 1053125\(%rip\),%rdi +# 102228 <_GLOBAL_OFFSET_TABLE_\+0x98>
- +105f: c5 11 10 00 *
+ +1058: 66 48 8d 3d c8 11 10[ ]+lea 1053128\(%rip\),%rdi +# 102228 <_GLOBAL_OFFSET_TABLE_\+0x98>
+ +105f: 00 *
# -> R_X86_64_DTPMOD64 [0 0x4000000000000000]
- +1063: e8 18 f6 ff ff[ ]+callq [0-9a-f]+ <.*>
+ +1060: 66[ ]+data16
+ +1061: 66[ ]+data16
+ +1062: 48 e8 18 f6 ff ff[ ]+rex64 callq [0-9a-f]+ <.*>
# -> R_X86_64_JUMP_SLOT __tls_get_addr
+1068: 90[ ]+nop *
+1069: 90[ ]+nop *
@@ -74,10 +80,12 @@ Disassembly of section .text:
+107e: 90[ ]+nop *
+107f: 90[ ]+nop *
# GD against hidden but not local variable
- +1080: 66 66 66 66 48 8d 3d[ ]+lea 1053013\(%rip\),%rdi +# 1021e0 <_GLOBAL_OFFSET_TABLE_\+0x50>
- +1087: 55 11 10 00 *
+ +1080: 66 48 8d 3d 58 11 10[ ]+lea 1053016\(%rip\),%rdi +# 1021e0 <_GLOBAL_OFFSET_TABLE_\+0x50>
+ +1087: 00 *
# -> R_X86_64_DTPMOD64 [0 0x6000000000000000]
- +108b: e8 f0 f5 ff ff[ ]+callq [0-9a-f]+ <.*>
+ +1088: 66[ ]+data16
+ +1089: 66[ ]+data16
+ +108a: 48 e8 f0 f5 ff ff[ ]+rex64 callq [0-9a-f]+ <.*>
# -> R_X86_64_JUMP_SLOT __tls_get_addr
+1090: 90[ ]+nop *
+1091: 90[ ]+nop *
Jakub