[lld] [LLD] Match delayLoad thunk with MSVC (PR #149521)
Evgenii Kudriashov via llvm-commits
llvm-commits at lists.llvm.org
Fri Jul 18 07:37:50 PDT 2025
https://github.com/e-kud created https://github.com/llvm/llvm-project/pull/149521
Previously we saved registers in the shadow space of callee before calling __delayLoadHelper2. Now we save arguments in the shadow space of the caller and allocate shadow space for the callee.
Fixes #51941
>From d83312cdec8e339683a23ba4e8c37d3cba041194 Mon Sep 17 00:00:00 2001
From: Evgenii Kudriashov <evgenii.kudriashov at intel.com>
Date: Fri, 18 Jul 2025 07:25:16 -0700
Subject: [PATCH] [LLD] Match delayLoad thunk with MSVC
Previously we saved registers in the shadow space of callee before
calling __delayLoadHelper2. Now we save arguments in the shadow space of
the caller and allocate shadow space for the callee.
Fixes #51941
Co-authored-by: Benjamin Santerre <benjamin.santerre at gmail.com>
---
lld/COFF/DLL.cpp | 48 ++++++++++++++--------------
lld/test/COFF/delayimports.test | 2 +-
lld/test/COFF/delayimporttables.yaml | 4 +--
3 files changed, 27 insertions(+), 27 deletions(-)
diff --git a/lld/COFF/DLL.cpp b/lld/COFF/DLL.cpp
index c327da28ce138..d98126fea7234 100644
--- a/lld/COFF/DLL.cpp
+++ b/lld/COFF/DLL.cpp
@@ -244,28 +244,28 @@ static const uint8_t thunkX64[] = {
};
static const uint8_t tailMergeX64[] = {
- 0x51, // push rcx
- 0x52, // push rdx
- 0x41, 0x50, // push r8
- 0x41, 0x51, // push r9
- 0x48, 0x83, 0xEC, 0x48, // sub rsp, 48h
- 0x66, 0x0F, 0x7F, 0x04, 0x24, // movdqa xmmword ptr [rsp], xmm0
- 0x66, 0x0F, 0x7F, 0x4C, 0x24, 0x10, // movdqa xmmword ptr [rsp+10h], xmm1
- 0x66, 0x0F, 0x7F, 0x54, 0x24, 0x20, // movdqa xmmword ptr [rsp+20h], xmm2
- 0x66, 0x0F, 0x7F, 0x5C, 0x24, 0x30, // movdqa xmmword ptr [rsp+30h], xmm3
- 0x48, 0x8B, 0xD0, // mov rdx, rax
- 0x48, 0x8D, 0x0D, 0, 0, 0, 0, // lea rcx, [___DELAY_IMPORT_...]
- 0xE8, 0, 0, 0, 0, // call __delayLoadHelper2
- 0x66, 0x0F, 0x6F, 0x04, 0x24, // movdqa xmm0, xmmword ptr [rsp]
- 0x66, 0x0F, 0x6F, 0x4C, 0x24, 0x10, // movdqa xmm1, xmmword ptr [rsp+10h]
- 0x66, 0x0F, 0x6F, 0x54, 0x24, 0x20, // movdqa xmm2, xmmword ptr [rsp+20h]
- 0x66, 0x0F, 0x6F, 0x5C, 0x24, 0x30, // movdqa xmm3, xmmword ptr [rsp+30h]
- 0x48, 0x83, 0xC4, 0x48, // add rsp, 48h
- 0x41, 0x59, // pop r9
- 0x41, 0x58, // pop r8
- 0x5A, // pop rdx
- 0x59, // pop rcx
- 0xFF, 0xE0, // jmp rax
+ 0x48, 0x89, 0x4C, 0x24, 0x08, // mov qword ptr [rsp+8], rcx
+ 0x48, 0x89, 0x54, 0x24, 0x10, // mov qword ptr [rsp+10h], rdx
+ 0x4C, 0x89, 0x44, 0x24, 0x18, // mov qword ptr [rsp+18h], r8
+ 0x4C, 0x89, 0x4C, 0x24, 0x20, // mov qword ptr [rsp+20h], r9
+ 0x48, 0x83, 0xEC, 0x68, // sub rsp, 68h
+ 0x66, 0x0F, 0x7F, 0x44, 0x24, 0x20, // movdqa xmmword ptr [rsp+20h], xmm0
+ 0x66, 0x0F, 0x7F, 0x4C, 0x24, 0x30, // movdqa xmmword ptr [rsp+30h], xmm1
+ 0x66, 0x0F, 0x7F, 0x54, 0x24, 0x40, // movdqa xmmword ptr [rsp+40h], xmm2
+ 0x66, 0x0F, 0x7F, 0x5C, 0x24, 0x50, // movdqa xmmword ptr [rsp+50h], xmm3
+ 0x48, 0x8B, 0xD0, // mov rdx, rax
+ 0x48, 0x8D, 0x0D, 0, 0, 0, 0, // lea rcx, [___DELAY_IMPORT_...]
+ 0xE8, 0, 0, 0, 0, // call __delayLoadHelper2
+ 0x66, 0x0F, 0x6F, 0x44, 0x24, 0x20, // movdqa xmm0, xmmword ptr [rsp+20h]
+ 0x66, 0x0F, 0x6F, 0x4C, 0x24, 0x30, // movdqa xmm1, xmmword ptr [rsp+30h]
+ 0x66, 0x0F, 0x6F, 0x54, 0x24, 0x40, // movdqa xmm2, xmmword ptr [rsp+40h]
+ 0x66, 0x0F, 0x6F, 0x5C, 0x24, 0x50, // movdqa xmm3, xmmword ptr [rsp+50h]
+ 0x48, 0x8B, 0x4C, 0x24, 0x70, // mov rcx, qword ptr [rsp+70h]
+ 0x48, 0x8B, 0x54, 0x24, 0x78, // mov rdx, qword ptr [rsp+78h]
+ 0x4C, 0x8B, 0x84, 0x24, 0x80, 00, 00, 00, // mov r8, qword ptr [rsp+80h]
+ 0x4C, 0x8B, 0x8C, 0x24, 0x88, 00, 00, 00, // mov r9, qword ptr [rsp+88h]
+ 0x48, 0x83, 0xC4, 0x68, // add rsp, 68h
+ 0xFF, 0xE0, // jmp rax
};
static const uint8_t tailMergeUnwindInfoX64[] = {
@@ -378,8 +378,8 @@ class TailMergeChunkX64 : public NonSectionCodeChunk {
void writeTo(uint8_t *buf) const override {
memcpy(buf, tailMergeX64, sizeof(tailMergeX64));
- write32le(buf + 39, desc->getRVA() - rva - 43);
- write32le(buf + 44, helper->getRVA() - rva - 48);
+ write32le(buf + 54, desc->getRVA() - rva - 58);
+ write32le(buf + 59, helper->getRVA() - rva - 63);
}
Chunk *desc = nullptr;
diff --git a/lld/test/COFF/delayimports.test b/lld/test/COFF/delayimports.test
index f410eef35fd1d..1521155d8764f 100644
--- a/lld/test/COFF/delayimports.test
+++ b/lld/test/COFF/delayimports.test
@@ -44,7 +44,7 @@ BASEREL-NEXT: }
UNWIND: UnwindInformation [
UNWIND-NEXT: RuntimeFunction {
UNWIND-NEXT: StartAddress: (0x14000108A)
-UNWIND-NEXT: EndAddress: (0x1400010DD)
+UNWIND-NEXT: EndAddress: (0x140001101)
UNWIND-NEXT: UnwindInfoAddress: (0x140002000)
UNWIND-NEXT: UnwindInfo {
UNWIND-NEXT: Version: 1
diff --git a/lld/test/COFF/delayimporttables.yaml b/lld/test/COFF/delayimporttables.yaml
index cf54c0a7140a1..a4e0cf0cc992c 100644
--- a/lld/test/COFF/delayimporttables.yaml
+++ b/lld/test/COFF/delayimporttables.yaml
@@ -37,11 +37,11 @@
# CHECK-NEXT: UnloadDelayImportTable: 0x0
# CHECK-NEXT: Import {
# CHECK-NEXT: Symbol: left (0)
-# CHECK-NEXT: Address: 0x1400010B8
+# CHECK-NEXT: Address: 0x1400010DC
# CHECK-NEXT: }
# CHECK-NEXT: Import {
# CHECK-NEXT: Symbol: right (0)
-# CHECK-NEXT: Address: 0x1400010C4
+# CHECK-NEXT: Address: 0x1400010E8
# CHECK-NEXT: }
# CHECK-NEXT: }
More information about the llvm-commits
mailing list