[X86][LLD] Handle R_X86_64_CODE_6_GOTTPOFF relocation type (#117675)

For

    add %reg1, name@GOTTPOFF(%rip), %reg2
    add name@GOTTPOFF(%rip), %reg1, %reg2
    {nf} add %reg1, name@GOTTPOFF(%rip), %reg2
    {nf} add name@GOTTPOFF(%rip), %reg1, %reg2
    {nf} add name@GOTTPOFF(%rip), %reg

add

    R_X86_64_CODE_6_GOTTPOFF = 50

in #117277.

Linker can treat R_X86_64_CODE_6_GOTTPOFF as R_X86_64_GOTTPOFF or
convert the instructions above to

    add $name@tpoff, %reg1, %reg2
    add $name@tpoff, %reg1, %reg2
    {nf} add $name@tpoff, %reg1, %reg2
    {nf} add $name@tpoff, %reg1, %reg2
    {nf} add $name@tpoff, %reg

if the first byte of the instruction at the relocation offset - 6 is
0x62 (namely, encoded w/EVEX prefix) when possible.

Binutils patch: bminor/binutils-gdb@5bc71c2
Binutils mailthread:
https://sourceware.org/pipermail/binutils/2024-February/132351.html
ABI discussion:
https://groups.google.com/g/x86-64-abi/c/FhEZjCtDLFw/m/VHDjN4orAgAJ
Blog: https://kanrobert.github.io/rfc/All-about-APX-relocation
This commit is contained in:
Feng Zou 2024-12-05 16:26:26 +08:00 committed by GitHub
parent a9a4a83b61
commit 636beb6a28
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
6 changed files with 270 additions and 21 deletions

View File

@ -401,6 +401,7 @@ RelExpr X86_64::getRelExpr(RelType type, const Symbol &s,
case R_X86_64_CODE_4_GOTPCRELX: case R_X86_64_CODE_4_GOTPCRELX:
case R_X86_64_GOTTPOFF: case R_X86_64_GOTTPOFF:
case R_X86_64_CODE_4_GOTTPOFF: case R_X86_64_CODE_4_GOTTPOFF:
case R_X86_64_CODE_6_GOTTPOFF:
return R_GOT_PC; return R_GOT_PC;
case R_X86_64_GOTOFF64: case R_X86_64_GOTOFF64:
return R_GOTPLTREL; return R_GOTPLTREL;
@ -562,8 +563,9 @@ void X86_64::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
} }
} }
// In some conditions, R_X86_64_GOTTPOFF/R_X86_64_CODE_4_GOTTPOFF relocation can // In some conditions,
// be optimized to R_X86_64_TPOFF32 so that it does not use GOT. // R_X86_64_GOTTPOFF/R_X86_64_CODE_4_GOTTPOFF/R_X86_64_CODE_6_GOTTPOFF
// relocation can be optimized to R_X86_64_TPOFF32 so that it does not use GOT.
void X86_64::relaxTlsIeToLe(uint8_t *loc, const Relocation &rel, void X86_64::relaxTlsIeToLe(uint8_t *loc, const Relocation &rel,
uint64_t val) const { uint64_t val) const {
uint8_t *inst = loc - 3; uint8_t *inst = loc - 3;
@ -605,7 +607,7 @@ void X86_64::relaxTlsIeToLe(uint8_t *loc, const Relocation &rel,
} else if (rel.type == R_X86_64_CODE_4_GOTTPOFF) { } else if (rel.type == R_X86_64_CODE_4_GOTTPOFF) {
if (loc[-4] != 0xd5) { if (loc[-4] != 0xd5) {
Err(ctx) << getErrorLoc(ctx, loc - 4) Err(ctx) << getErrorLoc(ctx, loc - 4)
<< "Invalid prefix with R_X86_64_CODE_4_GOTTPOFF!"; << "invalid prefix with R_X86_64_CODE_4_GOTTPOFF!";
return; return;
} }
const uint8_t rex = loc[-3]; const uint8_t rex = loc[-3];
@ -623,6 +625,41 @@ void X86_64::relaxTlsIeToLe(uint8_t *loc, const Relocation &rel,
<< "R_X86_64_CODE_4_GOTTPOFF must be used in MOVQ or ADDQ " << "R_X86_64_CODE_4_GOTTPOFF must be used in MOVQ or ADDQ "
"instructions only"; "instructions only";
} }
} else if (rel.type == R_X86_64_CODE_6_GOTTPOFF) {
if (loc[-6] != 0x62) {
Err(ctx) << getErrorLoc(ctx, loc - 6)
<< "invalid prefix with R_X86_64_CODE_6_GOTTPOFF!";
return;
}
// Check bits are satisfied:
// loc[-5]: X==1 (inverted polarity), (loc[-5] & 0x7) == 0x4
// loc[-4]: W==1, X2==1 (inverted polarity), pp==0b00(NP)
// loc[-3]: NF==1 or ND==1
// loc[-2]: opcode==0x1 or opcode==0x3
// loc[-1]: Mod==0b00, RM==0b101
if (((loc[-5] & 0x47) == 0x44) && ((loc[-4] & 0x87) == 0x84) &&
((loc[-3] & 0x14) != 0) && (loc[-2] == 0x1 || loc[-2] == 0x3) &&
((loc[-1] & 0xc7) == 0x5)) {
// "addq %reg1, foo@GOTTPOFF(%rip), %reg2" -> "addq $foo, %reg1, %reg2"
// "addq foo@GOTTPOFF(%rip), %reg1, %reg2" -> "addq $foo, %reg1, %reg2"
// "{nf} addq %reg1, foo@GOTTPOFF(%rip), %reg2"
// -> "{nf} addq $foo, %reg1, %reg2"
// "{nf} addq name@GOTTPOFF(%rip), %reg1, %reg2"
// -> "{nf} addq $foo, %reg1, %reg2"
// "{nf} addq name@GOTTPOFF(%rip), %reg" -> "{nf} addq $foo, %reg"
loc[-2] = 0x81;
// Move R bits to B bits in EVEX payloads and ModRM byte.
const uint8_t evexPayload0 = loc[-5];
if ((evexPayload0 & (1 << 7)) == 0)
loc[-5] = (evexPayload0 | (1 << 7)) & ~(1 << 5);
if ((evexPayload0 & (1 << 4)) == 0)
loc[-5] = evexPayload0 | (1 << 4) | (1 << 3);
*regSlot = 0xc0 | reg;
} else {
Err(ctx) << getErrorLoc(ctx, loc - 6)
<< "R_X86_64_CODE_6_GOTTPOFF must be used in ADDQ instructions "
"with NDD/NF/NDD+NF only";
}
} else { } else {
llvm_unreachable("Unsupported relocation type!"); llvm_unreachable("Unsupported relocation type!");
} }
@ -782,6 +819,7 @@ int64_t X86_64::getImplicitAddend(const uint8_t *buf, RelType type) const {
case R_X86_64_PC32: case R_X86_64_PC32:
case R_X86_64_GOTTPOFF: case R_X86_64_GOTTPOFF:
case R_X86_64_CODE_4_GOTTPOFF: case R_X86_64_CODE_4_GOTTPOFF:
case R_X86_64_CODE_6_GOTTPOFF:
case R_X86_64_PLT32: case R_X86_64_PLT32:
case R_X86_64_TLSGD: case R_X86_64_TLSGD:
case R_X86_64_TLSLD: case R_X86_64_TLSLD:
@ -893,6 +931,7 @@ void X86_64::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const {
break; break;
case R_X86_64_GOTTPOFF: case R_X86_64_GOTTPOFF:
case R_X86_64_CODE_4_GOTTPOFF: case R_X86_64_CODE_4_GOTTPOFF:
case R_X86_64_CODE_6_GOTTPOFF:
if (rel.expr == R_RELAX_TLS_IE_TO_LE) { if (rel.expr == R_RELAX_TLS_IE_TO_LE) {
relaxTlsIeToLe(loc, rel, val); relaxTlsIeToLe(loc, rel, val);
} else { } else {

View File

@ -1,15 +1,15 @@
# REQUIRES: x86 # REQUIRES: x86
# RUN: yaml2obj %s -o %t.o # RUN: yaml2obj --docnum=1 %s -o %t1.o
# RUN: not ld.lld %t.o -o /dev/null 2>&1 | FileCheck --check-prefix=ERR %s # RUN: not ld.lld %t1.o -o /dev/null 2>&1 | FileCheck --check-prefix=ERR %s
# ERR: R_X86_64_GOTTPOFF must be used in MOVQ or ADDQ instructions only # ERR: error: {{.*}}: R_X86_64_GOTTPOFF must be used in MOVQ or ADDQ instructions only
# ERR: R_X86_64_GOTTPOFF must be used in MOVQ or ADDQ instructions only # ERR: error: {{.*}}: R_X86_64_GOTTPOFF must be used in MOVQ or ADDQ instructions only
## YAML below contains 2 relocations of type R_X86_64_GOTTPOFF, and a .text ## YAML below contains 2 relocations of type R_X86_64_GOTTPOFF, and a .text
## with fake content filled by 0xFF. That means instructions for relaxation are ## with fake content filled by 0xFF. That means instructions for relaxation are
## "broken", so they does not match any known valid relaxations. We also generate ## "broken", so they does not match any known valid relaxations. We also generate
## .tls section because we need it for correct processing of STT_TLS symbol. ## .tls section because we need it for correct processing of STT_TLS symbol.
!ELF --- !ELF
FileHeader: FileHeader:
Class: ELFCLASS64 Class: ELFCLASS64
Data: ELFDATA2LSB Data: ELFDATA2LSB
@ -44,4 +44,131 @@ Symbols:
Value: 0x12345 Value: 0x12345
Size: 4 Size: 4
Binding: STB_GLOBAL Binding: STB_GLOBAL
# RUN: yaml2obj --docnum=2 %s -o %t2.o
# RUN: not ld.lld %t2.o -o /dev/null 2>&1 | FileCheck --check-prefix=ERR2 %s
# ERR2: error: {{.*}}: invalid prefix with R_X86_64_CODE_4_GOTTPOFF!
# ERR2: error: {{.*}}: invalid prefix with R_X86_64_CODE_6_GOTTPOFF!
## YAML below contains 2 relocations of
## R_X86_64_CODE_4_GOTTPOFF/R_X86_64_CODE_6_GOTTPOFF type, and a .text with
## fake content filled by 0xFF. It's expected to get "invalid prefix" error
## message as above.
--- !ELF
FileHeader:
Class: ELFCLASS64
Data: ELFDATA2LSB
OSABI: ELFOSABI_FREEBSD
Type: ET_REL
Machine: EM_X86_64
Sections:
- Type: SHT_PROGBITS
Name: .text
Flags: [ SHF_ALLOC, SHF_EXECINSTR ]
AddressAlign: 0x04
Content: "FFFFFFFFFFFFFFFFFFFF"
- Type: SHT_PROGBITS
Name: .tls
Flags: [ SHF_ALLOC, SHF_TLS ]
- Type: SHT_REL
Name: .rel.text
Link: .symtab
Info: .text
AddressAlign: 0x04
Relocations:
- Offset: 4
Symbol: foo
Type: R_X86_64_CODE_4_GOTTPOFF
- Offset: 6
Symbol: foo
Type: R_X86_64_CODE_6_GOTTPOFF
Symbols:
- Name: foo
Type: STT_TLS
Section: .text
Value: 0x12345
Size: 4
Binding: STB_GLOBAL
# RUN: yaml2obj --docnum=3 %s -o %t3.o
# RUN: not ld.lld %t3.o -o /dev/null 2>&1 | FileCheck --check-prefix=ERR3 %s
# ERR3: error: {{.*}}: R_X86_64_CODE_4_GOTTPOFF must be used in MOVQ or ADDQ instructions only
## YAML below contains R_X86_64_CODE_4_GOTTPOFF relocation type, and a .text
## with fake content filled by 0xd5, 0xFF, ... and 0xFF. It's expected to get
## the error message as above.
--- !ELF
FileHeader:
Class: ELFCLASS64
Data: ELFDATA2LSB
OSABI: ELFOSABI_FREEBSD
Type: ET_REL
Machine: EM_X86_64
Sections:
- Type: SHT_PROGBITS
Name: .text
Flags: [ SHF_ALLOC, SHF_EXECINSTR ]
AddressAlign: 0x04
Content: "d5FFFFFFFFFFFFFFFFFF"
- Type: SHT_PROGBITS
Name: .tls
Flags: [ SHF_ALLOC, SHF_TLS ]
- Type: SHT_REL
Name: .rel.text
Link: .symtab
Info: .text
AddressAlign: 0x04
Relocations:
- Offset: 4
Symbol: foo
Type: R_X86_64_CODE_4_GOTTPOFF
Symbols:
- Name: foo
Type: STT_TLS
Section: .text
Value: 0x12345
Size: 4
Binding: STB_GLOBAL
# RUN: yaml2obj --docnum=4 %s -o %t4.o
# RUN: not ld.lld %t4.o -o /dev/null 2>&1 | FileCheck --check-prefix=ERR4 %s
# ERR4: error: {{.*}}: R_X86_64_CODE_6_GOTTPOFF must be used in ADDQ instructions with NDD/NF/NDD+NF only
## YAML below contains R_X86_64_CODE_6_GOTTPOFF relocation type, and a .text
## with fake content filled by 0x62, 0xFF, ... and 0xFF. It's expected to get
## the error message as above.
--- !ELF
FileHeader:
Class: ELFCLASS64
Data: ELFDATA2LSB
OSABI: ELFOSABI_FREEBSD
Type: ET_REL
Machine: EM_X86_64
Sections:
- Type: SHT_PROGBITS
Name: .text
Flags: [ SHF_ALLOC, SHF_EXECINSTR ]
AddressAlign: 0x04
Content: "62FFFFFFFFFFFFFFFFFF"
- Type: SHT_PROGBITS
Name: .tls
Flags: [ SHF_ALLOC, SHF_TLS ]
- Type: SHT_REL
Name: .rel.text
Link: .symtab
Info: .text
AddressAlign: 0x04
Relocations:
- Offset: 6
Symbol: foo
Type: R_X86_64_CODE_6_GOTTPOFF
Symbols:
- Name: foo
Type: STT_TLS
Section: .text
Value: 0x12345
Size: 4
Binding: STB_GLOBAL

View File

@ -13,6 +13,7 @@
foo: foo:
movq tlsvar@GOTTPOFF(%rip), %rcx movq tlsvar@GOTTPOFF(%rip), %rcx
movq tlsvar2@GOTTPOFF(%rip), %r31 movq tlsvar2@GOTTPOFF(%rip), %r31
addq tlsvar3@GOTTPOFF(%rip), %rcx, %r16
.section .tdata,"awT",@progbits .section .tdata,"awT",@progbits
@ -21,7 +22,11 @@ tlsvar:
.word 42 .word 42
tlsvar2: tlsvar2:
.word 42 .word 42
tlsvar3:
.word 42
// CHECK: Section ({{.+}}) .rela.dyn { // CHECK: Section ({{.+}}) .rela.dyn {
// CHECK-NEXT: R_X86_64_TPOFF64 - 0x1234 // CHECK-NEXT: R_X86_64_TPOFF64 - 0x1234
// CHECK-NEXT: R_X86_64_TPOFF64 - 0x1236 // CHECK-NEXT: R_X86_64_TPOFF64 - 0x1236
// CHECK-NEXT: R_X86_64_TPOFF64 - 0x1238
// CHECK-NEXT: } // CHECK-NEXT: }

View File

@ -20,12 +20,25 @@
// DISASM-NEXT: leaq -4(%r15), %r15 // DISASM-NEXT: leaq -4(%r15), %r15
// DISASM-NEXT: addq $-4, %rsp // DISASM-NEXT: addq $-4, %rsp
// DISASM-NEXT: addq $-4, %r12 // DISASM-NEXT: addq $-4, %r12
# EGPR
// DISASM-NEXT: movq $-8, %r16 // DISASM-NEXT: movq $-8, %r16
// DISASM-NEXT: movq $-8, %r20 // DISASM-NEXT: movq $-8, %r20
// DISASM-NEXT: movq $-4, %r16 // DISASM-NEXT: movq $-4, %r16
// DISASM-NEXT: addq $-8, %r16 // DISASM-NEXT: addq $-8, %r16
// DISASM-NEXT: addq $-8, %r28 // DISASM-NEXT: addq $-8, %r28
// DISASM-NEXT: addq $-4, %r16 // DISASM-NEXT: addq $-4, %r16
# NDD
// DISASM-NEXT: addq $-10, %r16, %r16
// DISASM-NEXT: addq $-10, %r16, %r20
// DISASM-NEXT: addq $-10, %r16, %rax
// DISASM-NEXT: addq $-10, %rax, %r16
// DISASM-NEXT: addq $-10, %r8, %r16
// DISASM-NEXT: addq $-10, %rax, %r12
# NDD + NF
// DISASM-NEXT: {nf} addq $-10, %r8, %r16
// DISASM-NEXT: {nf} addq $-10, %rax, %r12
# NF
// DISASM-NEXT: {nf} addq $-10, %r12
// LD to LE: // LD to LE:
// DISASM-NEXT: movq %fs:0, %rax // DISASM-NEXT: movq %fs:0, %rax
@ -82,6 +95,18 @@ _start:
addq tls0@GOTTPOFF(%rip), %r16 addq tls0@GOTTPOFF(%rip), %r16
addq tls0@GOTTPOFF(%rip), %r28 addq tls0@GOTTPOFF(%rip), %r28
addq tls1@GOTTPOFF(%rip), %r16 addq tls1@GOTTPOFF(%rip), %r16
# NDD
addq tls0@GOTTPOFF(%rip), %r16, %r16
addq tls0@GOTTPOFF(%rip), %r16, %r20
addq tls0@GOTTPOFF(%rip), %r16, %rax
addq tls0@GOTTPOFF(%rip), %rax, %r16
addq %r8, tls0@GOTTPOFF(%rip), %r16
addq tls0@GOTTPOFF(%rip), %rax, %r12
# NDD + NF
{nf} addq %r8, tls0@GOTTPOFF(%rip), %r16
{nf} addq tls0@GOTTPOFF(%rip), %rax, %r12
# NF
{nf} addq tls0@GOTTPOFF(%rip), %r12
// LD to LE // LD to LE
leaq tls0@tlsld(%rip), %rdi leaq tls0@tlsld(%rip), %rdi

View File

@ -0,0 +1,35 @@
# REQUIRES: x86
# RUN: llvm-mc -filetype=obj -triple=x86_64 %s -o %t.o
# RUN: not ld.lld %t.o -o /dev/null 2>&1 | FileCheck -DFILE=%t.o %s
# CHECK: error: [[FILE]]:(.text+0x2): invalid prefix with R_X86_64_CODE_4_GOTTPOFF!
# CHECK-NEXT: error: [[FILE]]:(.text+0x8): invalid prefix with R_X86_64_CODE_6_GOTTPOFF!
# CHECK-NEXT: error: [[FILE]]:(.text+0x12): R_X86_64_CODE_4_GOTTPOFF must be used in MOVQ or ADDQ instructions only
# CHECK-NEXT: error: [[FILE]]:(.text+0x1a): R_X86_64_CODE_6_GOTTPOFF must be used in ADDQ instructions with NDD/NF/NDD+NF only
## These negative tests are to check if the invalid prefix and unsupported
## instructions for TLS relocation types with APX instructions are handled as
## errors.
.type tls0,@object
.section .tbss,"awT",@nobits
.globl tls0
.align 4
tls0:
.long 0
.size tls0, 4
.text
.globl _start
_start:
addq 0(%rip), %rax, %r16
.reloc .-4, R_X86_64_CODE_4_GOTTPOFF, tls0-4
movq 0(%rip), %r16
.reloc .-4, R_X86_64_CODE_6_GOTTPOFF, tls0-4
andq 0(%rip), %r16
.reloc .-4, R_X86_64_CODE_4_GOTTPOFF, tls0-4
andq 0(%rip), %rax, %r16
.reloc .-4, R_X86_64_CODE_6_GOTTPOFF, tls0-4

View File

@ -5,29 +5,47 @@
# RUN: llvm-readobj -r %t.so | FileCheck --check-prefix=REL %s # RUN: llvm-readobj -r %t.so | FileCheck --check-prefix=REL %s
# RUN: llvm-objdump --no-print-imm-hex -d --no-show-raw-insn %t.so | FileCheck %s # RUN: llvm-objdump --no-print-imm-hex -d --no-show-raw-insn %t.so | FileCheck %s
# SEC: .got PROGBITS 0000000000002348 000348 000010 00 WA 0 0 8 # SEC: .got PROGBITS 0000000000002378 000378 000010 00 WA 0 0 8
## Dynamic relocations for non-preemptable symbols in a shared object have section index 0. ## Dynamic relocations for non-preemptable symbols in a shared object have section index 0.
# REL: .rela.dyn { # REL: .rela.dyn {
# REL-NEXT: 0x2348 R_X86_64_TPOFF64 - 0x0 # REL-NEXT: 0x2378 R_X86_64_TPOFF64 - 0x0
# REL-NEXT: 0x2350 R_X86_64_TPOFF64 - 0x4 # REL-NEXT: 0x2380 R_X86_64_TPOFF64 - 0x4
# REL-NEXT: } # REL-NEXT: }
## &.got[0] - 0x127f = 0x2348 - 0x127f = 4297 ## &.got[0] - 0x127f = 0x2378 - 0x127f = 4345
## &.got[1] - 0x1286 = 0x2350 - 0x1286 = 4298 ## &.got[1] - 0x1286 = 0x2380 - 0x1286 = 4346
## &.got[2] - 0x128e = 0x2348 - 0x128e = 4282 ## &.got[2] - 0x128e = 0x2378 - 0x128e = 4330
## &.got[3] - 0x1296 = 0x2350 - 0x1296 = 4282 ## &.got[3] - 0x1296 = 0x2380 - 0x1296 = 4330
## &.got[0] - 0x12a0 = 0x2376 - 0x12a0 = 4310
## &.got[1] - 0x12aa = 0x237e - 0x12aa = 4308
## &.got[0] - 0x12b4 = 0x2376 - 0x12b4 = 4290
## &.got[1] - 0x12be = 0x237e - 0x12be = 4288
## &.got[0] - 0x12c8 = 0x2376 - 0x12c8 = 4270
# CHECK: 1278: addq 4297(%rip), %rax # CHECK: 1278: addq 4345(%rip), %rax
# CHECK-NEXT: 127f: addq 4298(%rip), %rax # CHECK-NEXT: 127f: addq 4346(%rip), %rax
# CHECK-NEXT: 1286: addq 4282(%rip), %r16 # CHECK-NEXT: 1286: addq 4330(%rip), %r16
# CHECK-NEXT: 128e: addq 4282(%rip), %r16 # CHECK-NEXT: 128e: addq 4330(%rip), %r16
# CHECK-NEXT: 1296: addq %r8, 4310(%rip), %r16
# CHECK-NEXT: 12a0: addq 4308(%rip), %rax, %r12
# CHECK-NEXT: 12aa: {nf} addq %r8, 4290(%rip), %r16
# CHECK-NEXT: 12b4: {nf} addq 4288(%rip), %rax, %r12
# CHECK-NEXT: 12be: {nf} addq 4270(%rip), %r12
addq foo@GOTTPOFF(%rip), %rax addq foo@GOTTPOFF(%rip), %rax
addq bar@GOTTPOFF(%rip), %rax addq bar@GOTTPOFF(%rip), %rax
# EGPR
addq foo@GOTTPOFF(%rip), %r16 addq foo@GOTTPOFF(%rip), %r16
addq bar@GOTTPOFF(%rip), %r16 addq bar@GOTTPOFF(%rip), %r16
# NDD
addq %r8, foo@GOTTPOFF(%rip), %r16
addq bar@GOTTPOFF(%rip), %rax, %r12
# NDD + NF
{nf} addq %r8, foo@GOTTPOFF(%rip), %r16
{nf} addq bar@GOTTPOFF(%rip), %rax, %r12
# NF
{nf} addq foo@GOTTPOFF(%rip), %r12
.section .tbss,"awT",@nobits .section .tbss,"awT",@nobits
foo: foo: