[lld] [lld] Mark target section as code section when merging code sections into a data section. (PR #72030)

[Jacek Caban via llvm-commits]

https://github.com/cjacek created https://github.com/llvm/llvm-project/pull/72030

I noticed this MS link.exe behavior while experimenting with #69101. This seems reasonable, esp. in case of uninitialized data and will be useful for a test case in #69101.

>From 285156c80ecbf6a9942367b840c1496300561b13 Mon Sep 17 00:00:00 2001
From: Jacek Caban <jacek at codeweavers.com>
Date: Sat, 11 Nov 2023 15:57:48 +0100
Subject: [PATCH] [lld] Mark target section as code section when merging code
 sections into a data section.

---
 lld/COFF/Writer.cpp        |  7 +++++
 lld/test/COFF/code-merge.s | 56 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 63 insertions(+)
 create mode 100644 lld/test/COFF/code-merge.s

diff --git a/lld/COFF/Writer.cpp b/lld/COFF/Writer.cpp
index 960328d686852a3..80edda03a0a47e7 100644
--- a/lld/COFF/Writer.cpp
+++ b/lld/COFF/Writer.cpp
@@ -346,6 +346,13 @@ void OutputSection::merge(OutputSection *other) {
   contribSections.insert(contribSections.end(), other->contribSections.begin(),
                          other->contribSections.end());
   other->contribSections.clear();
+
+  // MS link.exe compatibility: when merging a code section into a data section,
+  // mark the target section as a code section.
+  if (other->header.Characteristics & IMAGE_SCN_CNT_CODE) {
+    header.Characteristics |= IMAGE_SCN_CNT_CODE;
+    header.Characteristics &= ~IMAGE_SCN_CNT_INITIALIZED_DATA | IMAGE_SCN_CNT_UNINITIALIZED_DATA;
+  }
 }
 
 // Write the section header to a given buffer.
diff --git a/lld/test/COFF/code-merge.s b/lld/test/COFF/code-merge.s
new file mode 100644
index 000000000000000..5728fbc93ff9b9d
--- /dev/null
+++ b/lld/test/COFF/code-merge.s
@@ -0,0 +1,56 @@
+# REQUIRES: x86
+
+# Test that merging code section into a data section changes its characteristics.
+
+# RUN: llvm-mc -triple x86_64-windows-msvc %s -filetype=obj -o %t.obj
+# RUN: lld-link -machine:amd64 -dll -noentry -out:%t.dll %t.obj -merge:.testx=.testd -merge:.testx2=.testbss
+# RUN: llvm-readobj --sections %t.dll | FileCheck %s
+
+# CHECK:      Sections [
+# CHECK-NEXT:   Section {
+# CHECK-NEXT:     Number: 1
+# CHECK-NEXT:     Name: .testbss (2E 74 65 73 74 62 73 73)
+# CHECK-NEXT:     VirtualSize: 0xA
+# CHECK-NEXT:     VirtualAddress: 0x1000
+# CHECK-NEXT:     RawDataSize: 512
+# CHECK-NEXT:     PointerToRawData: 0x400
+# CHECK-NEXT:     PointerToRelocations: 0x0
+# CHECK-NEXT:     PointerToLineNumbers: 0x0
+# CHECK-NEXT:     RelocationCount: 0
+# CHECK-NEXT:     LineNumberCount: 0
+# CHECK-NEXT:     Characteristics [ (0x40000020)
+# CHECK-NEXT:       IMAGE_SCN_CNT_CODE (0x20)
+# CHECK-NEXT:       IMAGE_SCN_MEM_READ (0x40000000)
+# CHECK-NEXT:     ]
+# CHECK-NEXT:   }
+# CHECK-NEXT:   Section {
+# CHECK-NEXT:     Number: 2
+# CHECK-NEXT:     Name: .testd (2E 74 65 73 74 64 00 00)
+# CHECK-NEXT:     VirtualSize: 0xA
+# CHECK-NEXT:     VirtualAddress: 0x2000
+# CHECK-NEXT:     RawDataSize: 512
+# CHECK-NEXT:     PointerToRawData: 0x600
+# CHECK-NEXT:     PointerToRelocations: 0x0
+# CHECK-NEXT:     PointerToLineNumbers: 0x0
+# CHECK-NEXT:     RelocationCount: 0
+# CHECK-NEXT:     LineNumberCount: 0
+# CHECK-NEXT:     Characteristics [ (0x40000020)
+# CHECK-NEXT:       IMAGE_SCN_CNT_CODE (0x20)
+# CHECK-NEXT:       IMAGE_SCN_MEM_READ (0x40000000)
+# CHECK-NEXT:     ]
+# CHECK-NEXT:   }
+# CHECK-NEXT: ]
+
+        .section .testx, "xr"
+        movq $1, %rax
+        retq
+
+        .section .testx2, "xr"
+        movq $2, %rax
+        retq
+
+        .section .testd, "dr"
+        .word 1
+
+        .section .testbss, "br"
+        .word 0