Skip to content

[X86][LLD] Handle R_X86_64_CODE_4_GOTPC32_TLSDESC relocation type #116909

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 2 commits into from
Nov 27, 2024
Merged

[X86][LLD] Handle R_X86_64_CODE_4_GOTPC32_TLSDESC relocation type #116909

merged 2 commits into from
Nov 27, 2024

Conversation

fzou1
Copy link
Contributor

@fzou1 fzou1 commented Nov 20, 2024

For

lea name@tlsdesc(%rip), %reg

add

R_X86_64_CODE_4_GOTPC32_TLSDESC = 45

in #116908.

Linker can treat R_X86_64_CODE_4_GOTPC32_TLSDESC as R_X86_64_GOTPC32_TLSDESC or convert the instruction above to

mov $name@tpoff, %reg

if the first byte of the instruction at the relocation offset - 4 is 0xd5 (namely, encoded w/REX2 prefix) when possible.

Binutils patch: bminor/binutils-gdb@a533c8d
Binutils mailthread: https://sourceware.org/pipermail/binutils/2023-December/131463.html
ABI discussion: https://groups.google.com/g/x86-64-abi/c/ACwD-UQXVDs/m/vrgTenKyFwAJ
Blog: https://kanrobert.github.io/rfc/All-about-APX-relocation

@llvmbot
Copy link
Member

llvmbot commented Nov 20, 2024

@llvm/pr-subscribers-lld

Author: Feng Zou (fzou1)

Changes

For

lea name@tlsdesc(%rip), %reg

add

R_X86_64_CODE_4_GOTPC32_TLSDESC = 45

in #116908.

Linker can treat R_X86_64_CODE_4_GOTPC32_TLSDESC as R_X86_64_GOTPC32_TLSDESC or convert the instruction above to

mov $name@tpoff, %reg

if the first byte of the instruction at the relocation offset - 4 is 0xd5 (namely, encoded w/REX2 prefix) when possible.

Binutils patch: bminor/binutils-gdb@a533c8d
Binutils mailthread: https://sourceware.org/pipermail/binutils/2023-December/131463.html
ABI discussion: https://groups.google.com/g/x86-64-abi/c/ACwD-UQXVDs/m/vrgTenKyFwAJ
Blog: https://kanrobert.github.io/rfc/All-about-APX-relocation

Full diff: https://github.com/llvm/llvm-project/pull/116909.diff

3 Files Affected:

  • (modified) lld/ELF/Arch/X86_64.cpp (+23-10)
  • (modified) lld/test/ELF/invalid/x86-64-tlsdesc-gd.s (+6-1)
  • (modified) lld/test/ELF/x86-64-tlsdesc-gd.s (+39-18)
diff --git a/lld/ELF/Arch/X86_64.cpp b/lld/ELF/Arch/X86_64.cpp
index e9267bd4128d18..194af96ea0321b 100644
--- a/lld/ELF/Arch/X86_64.cpp
+++ b/lld/ELF/Arch/X86_64.cpp
@@ -99,8 +99,11 @@ X86_64::X86_64(Ctx &ctx) : TargetInfo(ctx) {
 
 int X86_64::getTlsGdRelaxSkip(RelType type) const {
   // TLSDESC relocations are processed separately. See relaxTlsGdToLe below.
-  return type == R_X86_64_GOTPC32_TLSDESC || type == R_X86_64_TLSDESC_CALL ? 1
-                                                                           : 2;
+  return type == R_X86_64_GOTPC32_TLSDESC ||
+                 type == R_X86_64_CODE_4_GOTPC32_TLSDESC ||
+                 type == R_X86_64_TLSDESC_CALL
+             ? 1
+             : 2;
 }
 
 // Opcodes for the different X86_64 jmp instructions.
@@ -390,6 +393,7 @@ RelExpr X86_64::getRelExpr(RelType type, const Symbol &s,
   case R_X86_64_GOT64:
     return R_GOTPLT;
   case R_X86_64_GOTPC32_TLSDESC:
+  case R_X86_64_CODE_4_GOTPC32_TLSDESC:
     return R_TLSDESC_PC;
   case R_X86_64_GOTPCREL:
   case R_X86_64_GOTPCRELX:
@@ -486,18 +490,24 @@ void X86_64::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
     // The original code used a pc relative relocation and so we have to
     // compensate for the -4 in had in the addend.
     write32le(loc + 8, val + 4);
-  } else if (rel.type == R_X86_64_GOTPC32_TLSDESC) {
+  } else if (rel.type == R_X86_64_GOTPC32_TLSDESC ||
+             rel.type == R_X86_64_CODE_4_GOTPC32_TLSDESC) {
     // Convert leaq x@tlsdesc(%rip), %REG to movq $x@tpoff, %REG.
     if ((loc[-3] & 0xfb) != 0x48 || loc[-2] != 0x8d ||
         (loc[-1] & 0xc7) != 0x05) {
       Err(ctx) << getErrorLoc(ctx, loc - 3)
-               << "R_X86_64_GOTPC32_TLSDESC must be used "
-                  "in leaq x@tlsdesc(%rip), %REG";
+               << "R_X86_64_GOTPC32_TLSDESC/R_X86_64_CODE_4_GOTPC32_TLSDESC "
+                  "must be used in leaq x@tlsdesc(%rip), %REG";
       return;
     }
-    loc[-3] = 0x48 | ((loc[-3] >> 2) & 1);
+    if (rel.type == R_X86_64_GOTPC32_TLSDESC) {
+      loc[-3] = 0x48 | ((loc[-3] >> 2) & 1);
+    } else {
+      loc[-3] = (loc[-3] & ~0x44) | ((loc[-3] & 0x44) >> 2);
+    }
     loc[-2] = 0xc7;
     loc[-1] = 0xc0 | ((loc[-1] >> 3) & 7);
+
     write32le(loc, val + 4);
   } else {
     // Convert call *x@tlsdesc(%REG) to xchg ax, ax.
@@ -527,14 +537,16 @@ void X86_64::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
     // Both code sequences are PC relatives, but since we are moving the
     // constant forward by 8 bytes we have to subtract the value by 8.
     write32le(loc + 8, val - 8);
-  } else if (rel.type == R_X86_64_GOTPC32_TLSDESC) {
+  } else if (rel.type == R_X86_64_GOTPC32_TLSDESC ||
+             rel.type == R_X86_64_CODE_4_GOTPC32_TLSDESC) {
     // Convert leaq x@tlsdesc(%rip), %REG to movq x@gottpoff(%rip), %REG.
-    assert(rel.type == R_X86_64_GOTPC32_TLSDESC);
+    assert(rel.type == R_X86_64_GOTPC32_TLSDESC ||
+           rel.type == R_X86_64_CODE_4_GOTPC32_TLSDESC);
     if ((loc[-3] & 0xfb) != 0x48 || loc[-2] != 0x8d ||
         (loc[-1] & 0xc7) != 0x05) {
       Err(ctx) << getErrorLoc(ctx, loc - 3)
-               << "R_X86_64_GOTPC32_TLSDESC must be used "
-                  "in leaq x@tlsdesc(%rip), %REG";
+               << "R_X86_64_GOTPC32_TLSDESC/R_X86_64_CODE_4_GOTPC32_TLSDESC "
+                  "must be used in leaq x@tlsdesc(%rip), %REG";
       return;
     }
     loc[-2] = 0x8b;
@@ -830,6 +842,7 @@ void X86_64::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const {
     }
     break;
   case R_X86_64_GOTPC32_TLSDESC:
+  case R_X86_64_CODE_4_GOTPC32_TLSDESC:
   case R_X86_64_TLSDESC_CALL:
   case R_X86_64_TLSGD:
     if (rel.expr == R_RELAX_TLS_GD_TO_LE) {
diff --git a/lld/test/ELF/invalid/x86-64-tlsdesc-gd.s b/lld/test/ELF/invalid/x86-64-tlsdesc-gd.s
index a04087bb096a57..625248da144312 100644
--- a/lld/test/ELF/invalid/x86-64-tlsdesc-gd.s
+++ b/lld/test/ELF/invalid/x86-64-tlsdesc-gd.s
@@ -8,8 +8,13 @@
 ## GD to IE relaxation.
 # RUN: not ld.lld %t.o %t1.so -o /dev/null 2>&1 | FileCheck -DINPUT=%t.o %s
 
-# CHECK: error: [[INPUT]]:(.text+0x0): R_X86_64_GOTPC32_TLSDESC must be used in leaq x@tlsdesc(%rip), %REG
+# CHECK: error: [[INPUT]]:(.text+0x0): R_X86_64_GOTPC32_TLSDESC/R_X86_64_CODE_4_GOTPC32_TLSDESC must be used in leaq x@tlsdesc(%rip), %REG
+# CHECK-NEXT: error: [[INPUT]]:(.text+0xd): R_X86_64_GOTPC32_TLSDESC/R_X86_64_CODE_4_GOTPC32_TLSDESC must be used in leaq x@tlsdesc(%rip), %REG
 
 leaq a@tlsdesc(%rbx), %rdx
 call *a@tlscall(%rdx)
 movl %fs:(%rax), %eax
+
+leaq a@tlsdesc(%r16), %r20
+call *a@tlscall(%r20)
+movl %fs:(%rax), %eax
\ No newline at end of file
diff --git a/lld/test/ELF/x86-64-tlsdesc-gd.s b/lld/test/ELF/x86-64-tlsdesc-gd.s
index 64e9016925bb31..433c2b2723463b 100644
--- a/lld/test/ELF/x86-64-tlsdesc-gd.s
+++ b/lld/test/ELF/x86-64-tlsdesc-gd.s
@@ -19,42 +19,48 @@
 # RUN: llvm-objdump --no-print-imm-hex -d --no-show-raw-insn %t | FileCheck --check-prefix=IE %s
 
 # GD-RELA:      .rela.dyn {
-# GD-RELA-NEXT:   0x23D0 R_X86_64_TLSDESC - 0xB
-# GD-RELA-NEXT:   0x23B0 R_X86_64_TLSDESC a 0x0
-# GD-RELA-NEXT:   0x23C0 R_X86_64_TLSDESC c 0x0
+# GD-RELA-NEXT:   0x23E0 R_X86_64_TLSDESC - 0xB
+# GD-RELA-NEXT:   0x23C0 R_X86_64_TLSDESC a 0x0
+# GD-RELA-NEXT:   0x23D0 R_X86_64_TLSDESC c 0x0
 # GD-RELA-NEXT: }
 # GD-RELA:      Hex dump of section '.got':
-# GD-RELA-NEXT: 0x000023b0 00000000 00000000 00000000 00000000
 # GD-RELA-NEXT: 0x000023c0 00000000 00000000 00000000 00000000
 # GD-RELA-NEXT: 0x000023d0 00000000 00000000 00000000 00000000
+# GD-RELA-NEXT: 0x000023e0 00000000 00000000 00000000 00000000
 
 # GD-REL:       .rel.dyn {
-# GD-REL-NEXT:    0x23B8 R_X86_64_TLSDESC -
-# GD-REL-NEXT:    0x2398 R_X86_64_TLSDESC a
-# GD-REL-NEXT:    0x23A8 R_X86_64_TLSDESC c
+# GD-REL-NEXT:    0x23C8 R_X86_64_TLSDESC -
+# GD-REL-NEXT:    0x23A8 R_X86_64_TLSDESC a
+# GD-REL-NEXT:    0x23B8 R_X86_64_TLSDESC c
 # GD-REL-NEXT:  }
 # GD-REL:       Hex dump of section '.got':
-# GD-REL-NEXT:  0x00002398 00000000 00000000 00000000 00000000
 # GD-REL-NEXT:  0x000023a8 00000000 00000000 00000000 00000000
-# GD-REL-NEXT:  0x000023b8 00000000 00000000 0b000000 00000000
+# GD-REL-NEXT:  0x000023b8 00000000 00000000 00000000 00000000
+# GD-REL-NEXT:  0x000023c8 00000000 00000000 0b000000 00000000
 
-## &.rela.dyn[a]-pc = 0x23B0-0x12e7 = 4297
-# GD:            leaq 4297(%rip), %rax
+## &.rela.dyn[a]-pc = 0x23C0-0x12e7 = 4313
+# GD:            leaq 4313(%rip), %rax
 # GD-NEXT: 12e7: callq *(%rax)
 # GD-NEXT:       movl %fs:(%rax), %eax
 
-## &.rela.dyn[b]-pc = 0x23D0-0x12f3 = 4317
-# GD-NEXT:       leaq 4317(%rip), %rcx
+## &.rela.dyn[b]-pc = 0x23E0-0x12f3 = 4333
+# GD-NEXT:       leaq 4333(%rip), %rcx
 # GD-NEXT: 12f3: movq %rcx, %rax
 # GD-NEXT:       callq *(%rax)
 # GD-NEXT:       movl %fs:(%rax), %eax
 
-## &.rela.dyn[c]-pc = 0x23C0-0x1302 = 4286
-# GD-NEXT:       leaq 4286(%rip), %r15
+## &.rela.dyn[c]-pc = 0x23D0-0x1302 = 4302
+# GD-NEXT:       leaq 4302(%rip), %r15
 # GD-NEXT: 1302: movq %r15, %rax
 # GD-NEXT:       callq *(%rax)
 # GD-NEXT:       movl %fs:(%rax), %eax
 
+## &.rela.dyn[c]-pc = 0x23D0-0x1312 = 4286
+# GD-NEXT:       leaq 4286(%rip), %r16
+# GD-NEXT: 1312: movq %r16, %rax
+# GD-NEXT:       callq *(%rax)
+# GD-NEXT:       movl %fs:(%rax), %eax
+
 # NOREL: no relocations
 
 ## tpoff(a) = st_value(a) - tls_size = -8
@@ -71,9 +77,14 @@
 # LE-NEXT: movq %r15, %rax
 # LE-NEXT: nop
 # LE-NEXT: movl %fs:(%rax), %eax
+## tpoff(c) = st_value(c) - tls_size = -4
+# LE:      movq $-4, %r16
+# LE-NEXT: movq %r16, %rax
+# LE-NEXT: nop
+# LE-NEXT: movl %fs:(%rax), %eax
 
 # IE-REL:      .rela.dyn {
-# IE-REL-NEXT:   0x202378 R_X86_64_TPOFF64 c 0x0
+# IE-REL-NEXT:   0x202388 R_X86_64_TPOFF64 c 0x0
 # IE-REL-NEXT: }
 
 ## a is relaxed to use LE.
@@ -84,11 +95,16 @@
 # IE-NEXT:         movq %rcx, %rax
 # IE-NEXT:         nop
 # IE-NEXT:         movl %fs:(%rax), %eax
-## &.rela.dyn[c]-pc = 0x202378 - 0x2012aa = 4302
-# IE-NEXT:         movq 4302(%rip), %r15
+## &.rela.dyn[c]-pc = 0x202388 - 0x2012aa = 4318
+# IE-NEXT:         movq 4318(%rip), %r15
 # IE-NEXT: 2012aa: movq %r15, %rax
 # IE-NEXT:         nop
 # IE-NEXT:         movl %fs:(%rax), %eax
+## &.rela.dyn[c]-pc = 0x202388 - 0x2012ba = 4302
+# IE-NEXT:         movq 4302(%rip), %r16
+# IE-NEXT: 2012ba: movq %r16, %rax
+# IE-NEXT:         nop
+# IE-NEXT:         movl %fs:(%rax), %eax
 
 leaq a@tlsdesc(%rip), %rax
 call *a@tlscall(%rax)
@@ -106,6 +122,11 @@ movq %r15, %rax
 call *c@tlscall(%rax)
 movl %fs:(%rax), %eax
 
+leaq c@tlsdesc(%rip), %r16
+movq %r16, %rax
+call *c@tlscall(%rax)
+movl %fs:(%rax), %eax
+
 .section .tbss
 .globl a
 .zero 8

@llvmbot
Copy link
Member

llvmbot commented Nov 20, 2024

@llvm/pr-subscribers-lld-elf

Author: Feng Zou (fzou1)

Changes

For

lea name@tlsdesc(%rip), %reg

add

R_X86_64_CODE_4_GOTPC32_TLSDESC = 45

in #116908.

Linker can treat R_X86_64_CODE_4_GOTPC32_TLSDESC as R_X86_64_GOTPC32_TLSDESC or convert the instruction above to

mov $name@tpoff, %reg

if the first byte of the instruction at the relocation offset - 4 is 0xd5 (namely, encoded w/REX2 prefix) when possible.

Binutils patch: bminor/binutils-gdb@a533c8d
Binutils mailthread: https://sourceware.org/pipermail/binutils/2023-December/131463.html
ABI discussion: https://groups.google.com/g/x86-64-abi/c/ACwD-UQXVDs/m/vrgTenKyFwAJ
Blog: https://kanrobert.github.io/rfc/All-about-APX-relocation

Full diff: https://github.com/llvm/llvm-project/pull/116909.diff

3 Files Affected:

  • (modified) lld/ELF/Arch/X86_64.cpp (+23-10)
  • (modified) lld/test/ELF/invalid/x86-64-tlsdesc-gd.s (+6-1)
  • (modified) lld/test/ELF/x86-64-tlsdesc-gd.s (+39-18)
diff --git a/lld/ELF/Arch/X86_64.cpp b/lld/ELF/Arch/X86_64.cpp
index e9267bd4128d18..194af96ea0321b 100644
--- a/lld/ELF/Arch/X86_64.cpp
+++ b/lld/ELF/Arch/X86_64.cpp
@@ -99,8 +99,11 @@ X86_64::X86_64(Ctx &ctx) : TargetInfo(ctx) {
 
 int X86_64::getTlsGdRelaxSkip(RelType type) const {
   // TLSDESC relocations are processed separately. See relaxTlsGdToLe below.
-  return type == R_X86_64_GOTPC32_TLSDESC || type == R_X86_64_TLSDESC_CALL ? 1
-                                                                           : 2;
+  return type == R_X86_64_GOTPC32_TLSDESC ||
+                 type == R_X86_64_CODE_4_GOTPC32_TLSDESC ||
+                 type == R_X86_64_TLSDESC_CALL
+             ? 1
+             : 2;
 }
 
 // Opcodes for the different X86_64 jmp instructions.
@@ -390,6 +393,7 @@ RelExpr X86_64::getRelExpr(RelType type, const Symbol &s,
   case R_X86_64_GOT64:
     return R_GOTPLT;
   case R_X86_64_GOTPC32_TLSDESC:
+  case R_X86_64_CODE_4_GOTPC32_TLSDESC:
     return R_TLSDESC_PC;
   case R_X86_64_GOTPCREL:
   case R_X86_64_GOTPCRELX:
@@ -486,18 +490,24 @@ void X86_64::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
     // The original code used a pc relative relocation and so we have to
     // compensate for the -4 in had in the addend.
     write32le(loc + 8, val + 4);
-  } else if (rel.type == R_X86_64_GOTPC32_TLSDESC) {
+  } else if (rel.type == R_X86_64_GOTPC32_TLSDESC ||
+             rel.type == R_X86_64_CODE_4_GOTPC32_TLSDESC) {
     // Convert leaq x@tlsdesc(%rip), %REG to movq $x@tpoff, %REG.
     if ((loc[-3] & 0xfb) != 0x48 || loc[-2] != 0x8d ||
         (loc[-1] & 0xc7) != 0x05) {
       Err(ctx) << getErrorLoc(ctx, loc - 3)
-               << "R_X86_64_GOTPC32_TLSDESC must be used "
-                  "in leaq x@tlsdesc(%rip), %REG";
+               << "R_X86_64_GOTPC32_TLSDESC/R_X86_64_CODE_4_GOTPC32_TLSDESC "
+                  "must be used in leaq x@tlsdesc(%rip), %REG";
       return;
     }
-    loc[-3] = 0x48 | ((loc[-3] >> 2) & 1);
+    if (rel.type == R_X86_64_GOTPC32_TLSDESC) {
+      loc[-3] = 0x48 | ((loc[-3] >> 2) & 1);
+    } else {
+      loc[-3] = (loc[-3] & ~0x44) | ((loc[-3] & 0x44) >> 2);
+    }
     loc[-2] = 0xc7;
     loc[-1] = 0xc0 | ((loc[-1] >> 3) & 7);
+
     write32le(loc, val + 4);
   } else {
     // Convert call *x@tlsdesc(%REG) to xchg ax, ax.
@@ -527,14 +537,16 @@ void X86_64::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
     // Both code sequences are PC relatives, but since we are moving the
     // constant forward by 8 bytes we have to subtract the value by 8.
     write32le(loc + 8, val - 8);
-  } else if (rel.type == R_X86_64_GOTPC32_TLSDESC) {
+  } else if (rel.type == R_X86_64_GOTPC32_TLSDESC ||
+             rel.type == R_X86_64_CODE_4_GOTPC32_TLSDESC) {
     // Convert leaq x@tlsdesc(%rip), %REG to movq x@gottpoff(%rip), %REG.
-    assert(rel.type == R_X86_64_GOTPC32_TLSDESC);
+    assert(rel.type == R_X86_64_GOTPC32_TLSDESC ||
+           rel.type == R_X86_64_CODE_4_GOTPC32_TLSDESC);
     if ((loc[-3] & 0xfb) != 0x48 || loc[-2] != 0x8d ||
         (loc[-1] & 0xc7) != 0x05) {
       Err(ctx) << getErrorLoc(ctx, loc - 3)
-               << "R_X86_64_GOTPC32_TLSDESC must be used "
-                  "in leaq x@tlsdesc(%rip), %REG";
+               << "R_X86_64_GOTPC32_TLSDESC/R_X86_64_CODE_4_GOTPC32_TLSDESC "
+                  "must be used in leaq x@tlsdesc(%rip), %REG";
       return;
     }
     loc[-2] = 0x8b;
@@ -830,6 +842,7 @@ void X86_64::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const {
     }
     break;
   case R_X86_64_GOTPC32_TLSDESC:
+  case R_X86_64_CODE_4_GOTPC32_TLSDESC:
   case R_X86_64_TLSDESC_CALL:
   case R_X86_64_TLSGD:
     if (rel.expr == R_RELAX_TLS_GD_TO_LE) {
diff --git a/lld/test/ELF/invalid/x86-64-tlsdesc-gd.s b/lld/test/ELF/invalid/x86-64-tlsdesc-gd.s
index a04087bb096a57..625248da144312 100644
--- a/lld/test/ELF/invalid/x86-64-tlsdesc-gd.s
+++ b/lld/test/ELF/invalid/x86-64-tlsdesc-gd.s
@@ -8,8 +8,13 @@
 ## GD to IE relaxation.
 # RUN: not ld.lld %t.o %t1.so -o /dev/null 2>&1 | FileCheck -DINPUT=%t.o %s
 
-# CHECK: error: [[INPUT]]:(.text+0x0): R_X86_64_GOTPC32_TLSDESC must be used in leaq x@tlsdesc(%rip), %REG
+# CHECK: error: [[INPUT]]:(.text+0x0): R_X86_64_GOTPC32_TLSDESC/R_X86_64_CODE_4_GOTPC32_TLSDESC must be used in leaq x@tlsdesc(%rip), %REG
+# CHECK-NEXT: error: [[INPUT]]:(.text+0xd): R_X86_64_GOTPC32_TLSDESC/R_X86_64_CODE_4_GOTPC32_TLSDESC must be used in leaq x@tlsdesc(%rip), %REG
 
 leaq a@tlsdesc(%rbx), %rdx
 call *a@tlscall(%rdx)
 movl %fs:(%rax), %eax
+
+leaq a@tlsdesc(%r16), %r20
+call *a@tlscall(%r20)
+movl %fs:(%rax), %eax
\ No newline at end of file
diff --git a/lld/test/ELF/x86-64-tlsdesc-gd.s b/lld/test/ELF/x86-64-tlsdesc-gd.s
index 64e9016925bb31..433c2b2723463b 100644
--- a/lld/test/ELF/x86-64-tlsdesc-gd.s
+++ b/lld/test/ELF/x86-64-tlsdesc-gd.s
@@ -19,42 +19,48 @@
 # RUN: llvm-objdump --no-print-imm-hex -d --no-show-raw-insn %t | FileCheck --check-prefix=IE %s
 
 # GD-RELA:      .rela.dyn {
-# GD-RELA-NEXT:   0x23D0 R_X86_64_TLSDESC - 0xB
-# GD-RELA-NEXT:   0x23B0 R_X86_64_TLSDESC a 0x0
-# GD-RELA-NEXT:   0x23C0 R_X86_64_TLSDESC c 0x0
+# GD-RELA-NEXT:   0x23E0 R_X86_64_TLSDESC - 0xB
+# GD-RELA-NEXT:   0x23C0 R_X86_64_TLSDESC a 0x0
+# GD-RELA-NEXT:   0x23D0 R_X86_64_TLSDESC c 0x0
 # GD-RELA-NEXT: }
 # GD-RELA:      Hex dump of section '.got':
-# GD-RELA-NEXT: 0x000023b0 00000000 00000000 00000000 00000000
 # GD-RELA-NEXT: 0x000023c0 00000000 00000000 00000000 00000000
 # GD-RELA-NEXT: 0x000023d0 00000000 00000000 00000000 00000000
+# GD-RELA-NEXT: 0x000023e0 00000000 00000000 00000000 00000000
 
 # GD-REL:       .rel.dyn {
-# GD-REL-NEXT:    0x23B8 R_X86_64_TLSDESC -
-# GD-REL-NEXT:    0x2398 R_X86_64_TLSDESC a
-# GD-REL-NEXT:    0x23A8 R_X86_64_TLSDESC c
+# GD-REL-NEXT:    0x23C8 R_X86_64_TLSDESC -
+# GD-REL-NEXT:    0x23A8 R_X86_64_TLSDESC a
+# GD-REL-NEXT:    0x23B8 R_X86_64_TLSDESC c
 # GD-REL-NEXT:  }
 # GD-REL:       Hex dump of section '.got':
-# GD-REL-NEXT:  0x00002398 00000000 00000000 00000000 00000000
 # GD-REL-NEXT:  0x000023a8 00000000 00000000 00000000 00000000
-# GD-REL-NEXT:  0x000023b8 00000000 00000000 0b000000 00000000
+# GD-REL-NEXT:  0x000023b8 00000000 00000000 00000000 00000000
+# GD-REL-NEXT:  0x000023c8 00000000 00000000 0b000000 00000000
 
-## &.rela.dyn[a]-pc = 0x23B0-0x12e7 = 4297
-# GD:            leaq 4297(%rip), %rax
+## &.rela.dyn[a]-pc = 0x23C0-0x12e7 = 4313
+# GD:            leaq 4313(%rip), %rax
 # GD-NEXT: 12e7: callq *(%rax)
 # GD-NEXT:       movl %fs:(%rax), %eax
 
-## &.rela.dyn[b]-pc = 0x23D0-0x12f3 = 4317
-# GD-NEXT:       leaq 4317(%rip), %rcx
+## &.rela.dyn[b]-pc = 0x23E0-0x12f3 = 4333
+# GD-NEXT:       leaq 4333(%rip), %rcx
 # GD-NEXT: 12f3: movq %rcx, %rax
 # GD-NEXT:       callq *(%rax)
 # GD-NEXT:       movl %fs:(%rax), %eax
 
-## &.rela.dyn[c]-pc = 0x23C0-0x1302 = 4286
-# GD-NEXT:       leaq 4286(%rip), %r15
+## &.rela.dyn[c]-pc = 0x23D0-0x1302 = 4302
+# GD-NEXT:       leaq 4302(%rip), %r15
 # GD-NEXT: 1302: movq %r15, %rax
 # GD-NEXT:       callq *(%rax)
 # GD-NEXT:       movl %fs:(%rax), %eax
 
+## &.rela.dyn[c]-pc = 0x23D0-0x1312 = 4286
+# GD-NEXT:       leaq 4286(%rip), %r16
+# GD-NEXT: 1312: movq %r16, %rax
+# GD-NEXT:       callq *(%rax)
+# GD-NEXT:       movl %fs:(%rax), %eax
+
 # NOREL: no relocations
 
 ## tpoff(a) = st_value(a) - tls_size = -8
@@ -71,9 +77,14 @@
 # LE-NEXT: movq %r15, %rax
 # LE-NEXT: nop
 # LE-NEXT: movl %fs:(%rax), %eax
+## tpoff(c) = st_value(c) - tls_size = -4
+# LE:      movq $-4, %r16
+# LE-NEXT: movq %r16, %rax
+# LE-NEXT: nop
+# LE-NEXT: movl %fs:(%rax), %eax
 
 # IE-REL:      .rela.dyn {
-# IE-REL-NEXT:   0x202378 R_X86_64_TPOFF64 c 0x0
+# IE-REL-NEXT:   0x202388 R_X86_64_TPOFF64 c 0x0
 # IE-REL-NEXT: }
 
 ## a is relaxed to use LE.
@@ -84,11 +95,16 @@
 # IE-NEXT:         movq %rcx, %rax
 # IE-NEXT:         nop
 # IE-NEXT:         movl %fs:(%rax), %eax
-## &.rela.dyn[c]-pc = 0x202378 - 0x2012aa = 4302
-# IE-NEXT:         movq 4302(%rip), %r15
+## &.rela.dyn[c]-pc = 0x202388 - 0x2012aa = 4318
+# IE-NEXT:         movq 4318(%rip), %r15
 # IE-NEXT: 2012aa: movq %r15, %rax
 # IE-NEXT:         nop
 # IE-NEXT:         movl %fs:(%rax), %eax
+## &.rela.dyn[c]-pc = 0x202388 - 0x2012ba = 4302
+# IE-NEXT:         movq 4302(%rip), %r16
+# IE-NEXT: 2012ba: movq %r16, %rax
+# IE-NEXT:         nop
+# IE-NEXT:         movl %fs:(%rax), %eax
 
 leaq a@tlsdesc(%rip), %rax
 call *a@tlscall(%rax)
@@ -106,6 +122,11 @@ movq %r15, %rax
 call *c@tlscall(%rax)
 movl %fs:(%rax), %eax
 
+leaq c@tlsdesc(%rip), %r16
+movq %r16, %rax
+call *c@tlscall(%rax)
+movl %fs:(%rax), %eax
+
 .section .tbss
 .globl a
 .zero 8

@fzou1
[X86][LLD] Handle R_X86_64_CODE_4_GOTPC32_TLSDESC relocation type
2685e0e 
For

  lea name@tlsdesc(%rip), %reg

add

  R_X86_64_CODE_4_GOTPC32_TLSDESC = 45

in llvm#116908.

Linker can treat R_X86_64_CODE_4_GOTPC32_TLSDESC as R_X86_64_GOTPC32_TLSDESC or
convert the instruction above to

  mov $name@tpoff, %reg

if the first byte of the instruction at the relocation offset - 4 is 0xd5
(namely, encoded w/REX2 prefix) when possible.

Binutils patch: bminor/binutils-gdb@a533c8d
Binutils mailthread: https://sourceware.org/pipermail/binutils/2023-December/131463.html
ABI discussion: https://groups.google.com/g/x86-64-abi/c/ACwD-UQXVDs/m/vrgTenKyFwAJ
Blog: https://kanrobert.github.io/rfc/All-about-APX-relocation
@fzou1 fzou1 force-pushed the lld-code-4-tlsdesc branch from 796fa80 to 2685e0e Compare November 22, 2024 03:24
@fzou1
Copy link
Contributor Author

fzou1 commented Nov 22, 2024

@MaskRay / @KanRobert , could you please review this PR? Thanks.

KanRobert
KanRobert reviewed Nov 25, 2024
View reviewed changes
lld/ELF/Arch/X86_64.cpp
@@ -486,18 +490,26 @@ void X86_64::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
// The original code used a pc relative relocation and so we have to
// compensate for the -4 in had in the addend.
write32le(loc + 8, val + 4);
} else if (rel.type == R_X86_64_GOTPC32_TLSDESC) {
} else if (rel.type == R_X86_64_GOTPC32_TLSDESC ||
rel.type == R_X86_64_CODE_4_GOTPC32_TLSDESC) {
// Convert leaq x@tlsdesc(%rip), %REG to movq $x@tpoff, %REG.
if ((loc[-3] & 0xfb) != 0x48 || loc[-2] != 0x8d ||
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seems you're missing some tests. The (loc[-3] & 0xfb) != 0x48 is used to check the X and B bit for REX prefix, it would fail for REX2 prefix.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Never mind. I got something wrong. It works for REX2 too if we don't support

{rex2} leaq x@tlsdesc(%rip), %r8

for this relocation.

Currently, compiler won't emit such code.

KanRobert
KanRobert approved these changes Nov 25, 2024
View reviewed changes
Copy link
Contributor

@KanRobert KanRobert left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@fzou1
Copy link
Contributor Author

fzou1 commented Nov 26, 2024

@MaskRay , friendly ping.

@fzou1
Copy link
Contributor Author

fzou1 commented Nov 27, 2024

I'll commit first. Any further comments are welcome.

@fzou1 fzou1 merged commit fad5ed6 into llvm:main Nov 27, 2024
8 checks passed
@fzou1 fzou1 deleted the lld-code-4-tlsdesc branch November 27, 2024 01:31
@MaskRay
Copy link
Member

MaskRay commented Nov 28, 2024

I'll commit first. Any further comments are welcome.

This patch looks good. I delegated the review to Kan.

Note, as a new contributor, it's important to avoid rushing through reviews.
A friendly ping and a quick land the next day can be perceived as rushing.

@fzou1
Copy link
Contributor Author

fzou1 commented Nov 28, 2024

I'll commit first. Any further comments are welcome.

This patch looks good. I delegated the review to Kan.

Note, as a new contributor, it's important to avoid rushing through reviews. A friendly ping and a quick land the next day can be perceived as rushing.

Sorry for that. Noted. Thanks.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@KanRobert KanRobert KanRobert approved these changes

@MaskRay MaskRay Awaiting requested review from MaskRay

@phoebewang phoebewang Awaiting requested review from phoebewang

Assignees
No one assigned
Labels
lld:ELF lld
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@fzou1 @llvmbot @MaskRay @KanRobert