[obj2yaml] Support CREL

[MaskRay]

This decoder feature complements the yaml2obj encoder added by #91280.

[llvmbot]

@llvm/pr-subscribers-objectyaml

Author: Fangrui Song (MaskRay)

Changes

This decoder feature complements the yaml2obj encoder added by #91280.

---

Full diff: https://github.com/llvm/llvm-project/pull/98116.diff

2 Files Affected:

• (added) llvm/test/tools/obj2yaml/ELF/crel-section.yaml (+36)
• (modified) llvm/tools/obj2yaml/elf2yaml.cpp (+35-20)

diff --git a/llvm/test/tools/obj2yaml/ELF/crel-section.yaml b/llvm/test/tools/obj2yaml/ELF/crel-section.yaml
new file mode 100644
index 0000000000000..c660a397de829
--- /dev/null
+++ b/llvm/test/tools/obj2yaml/ELF/crel-section.yaml
@@ -0,0 +1,36 @@
+## Test how we dump SHT_CREL sections.
+# RUN: yaml2obj %s -o %t1
+# RUN: obj2yaml %t1 | FileCheck %s --check-prefix=YAML
+
+# YAML:      - Name:    .crel.dyn
+# YAML-NEXT:   Type:    SHT_CREL
+
+--- !ELF
+FileHeader:
+  Class: ELFCLASS64
+  Data:  ELFDATA2LSB
+  Type:  ET_DYN
+Sections:
+  - Name:    .crel.dyn
+    Type:    SHT_CREL
+    Content: 00
+## Trigger the .dynsym emission.
+DynamicSymbols: []
+
+## Test the behavior when the sh_entsize field is broken.
+## Here we use the 0xFE value instead of expected 0x18/0x10.
+
+# RUN: yaml2obj -DTYPE=SHT_RELA --docnum=2 %s -o %t2
+# RUN: not obj2yaml %t2 2>&1 | FileCheck %s --check-prefix=ERR1
+
+# ERR1: unable to decode LEB128 at offset 0x00000000: malformed uleb128, extends past end
+
+--- !ELF
+FileHeader:
+  Class: ELFCLASS64
+  Data:  ELFDATA2LSB
+  Type:  ET_DYN
+Sections:
+  - Name:    .foo
+    Type:    SHT_CREL
+    EntSize: 0xFE
diff --git a/llvm/tools/obj2yaml/elf2yaml.cpp b/llvm/tools/obj2yaml/elf2yaml.cpp
index 6b9af906736c3..9b4644bde36c0 100644
--- a/llvm/tools/obj2yaml/elf2yaml.cpp
+++ b/llvm/tools/obj2yaml/elf2yaml.cpp
@@ -596,6 +596,7 @@ ELFDumper<ELFT>::dumpSections() {
       return [this](const Elf_Shdr *S) { return dumpSymtabShndxSection(S); };
     case ELF::SHT_REL:
     case ELF::SHT_RELA:
+    case ELF::SHT_CREL:
       return [this](const Elf_Shdr *S) { return dumpRelocSection(S); };
     case ELF::SHT_RELR:
       return [this](const Elf_Shdr *S) { return dumpRelrSection(S); };
@@ -1165,27 +1166,41 @@ ELFDumper<ELFT>::dumpRelocSection(const Elf_Shdr *Shdr) {
   if (Shdr->sh_size != 0)
     S->Relocations.emplace();
 
-  if (Shdr->sh_type == ELF::SHT_REL) {
-    auto Rels = Obj.rels(*Shdr);
-    if (!Rels)
-      return Rels.takeError();
-    for (const Elf_Rel &Rel : *Rels) {
-      ELFYAML::Relocation R;
-      if (Error E = dumpRelocation(&Rel, *SymTabOrErr, R))
-        return std::move(E);
-      S->Relocations->push_back(R);
-    }
+  std::vector<Elf_Rel> Rels;
+  std::vector<Elf_Rela> Relas;
+  if (Shdr->sh_type == ELF::SHT_CREL) {
+    Expected<ArrayRef<uint8_t>> ContentOrErr = Obj.getSectionContents(*Shdr);
+    if (!ContentOrErr)
+      return ContentOrErr.takeError();
+    auto Crel = Obj.decodeCrel(*ContentOrErr);
+    if (!Crel)
+      return Crel.takeError();
+    Rels = std::move(Crel->first);
+    Relas = std::move(Crel->second);
+  } else if (Shdr->sh_type == ELF::SHT_REL) {
+    auto R = Obj.rels(*Shdr);
+    if (!R)
+      return R.takeError();
+    Rels = std::move(*R);
   } else {
-    auto Rels = Obj.relas(*Shdr);
-    if (!Rels)
-      return Rels.takeError();
-    for (const Elf_Rela &Rel : *Rels) {
-      ELFYAML::Relocation R;
-      if (Error E = dumpRelocation(&Rel, *SymTabOrErr, R))
-        return std::move(E);
-      R.Addend = Rel.r_addend;
-      S->Relocations->push_back(R);
-    }
+    auto R = Obj.relas(*Shdr);
+    if (!R)
+      return R.takeError();
+    Relas = std::move(*R);
+  }
+
+  for (const Elf_Rel &Rel : Rels) {
+    ELFYAML::Relocation R;
+    if (Error E = dumpRelocation(&Rel, *SymTabOrErr, R))
+      return std::move(E);
+    S->Relocations->push_back(R);
+  }
+  for (const Elf_Rela &Rel : Relas) {
+    ELFYAML::Relocation R;
+    if (Error E = dumpRelocation(&Rel, *SymTabOrErr, R))
+      return std::move(E);
+    R.Addend = Rel.r_addend;
+    S->Relocations->push_back(R);
   }
 
   return S.release();

[jh7370]

LGTM, with one query.

[jh7370]

Dumb question maybe, but what's the point of this particular symbol?

[MaskRay]

The unused symbol makes symbols referenced by relocations start at higher indexes.
This property is somewhat useful to test llvm-objcopy. It's probably not so useful after decodeCrel has been tested, but I think it does not hurt here...