[yaml2obj] - Add a way to set sh_entsize for relocation sections.

[grimar]

Summary

We are missing ability to override the sh_entsize field for
SHT_REL[A] sections. It would be useful for writing test cases.

[jhenderson]

LGTM, with suggestions.

Idle thought: I wonder if it's worth introducing some kind of command-line macro option to yaml2obj? E.g. -DMACHINE=EM_X86_64 would replace Machine: MACHINE (or some similar syntax). There are a growing number of cases where we have two identical YAML blobs except for some minor difference.

llvm/test/tools/yaml2obj/ELF/reloc-sec-entry-size.yaml

2

set -> sets

33

sh_entsizes -> sh_entsize

('sh_entsize' is an adjective, so doesn't need pluralizing)

40

Check we can set sh_entsize fields to arbitrary values.

58

Same as above

65

Same as above

[grimar]

In D73621#1848823, @jhenderson wrote:

Idle thought: I wonder if it's worth introducing some kind of command-line macro option to yaml2obj? E.g. -DMACHINE=EM_X86_64 would replace Machine: MACHINE (or some similar syntax). There are a growing number of cases where we have two identical YAML blobs except for some minor difference.

Sounds good. I'd give it a try.

[MaskRay]

In D73621#1848823, @jhenderson wrote:

LGTM, with suggestions.

Idle thought: I wonder if it's worth introducing some kind of command-line macro option to yaml2obj? E.g. -DMACHINE=EM_X86_64 would replace Machine: MACHINE (or some similar syntax). There are a growing number of cases where we have two identical YAML blobs except for some minor difference.

LG

[grimar]

In D73621#1848843, @grimar wrote:

In D73621#1848823, @jhenderson wrote:

Idle thought: I wonder if it's worth introducing some kind of command-line macro option to yaml2obj? E.g. -DMACHINE=EM_X86_64 would replace Machine: MACHINE (or some similar syntax). There are a growing number of cases where we have two identical YAML blobs except for some minor difference.

Sounds good. I'd give it a try.

One more idea: what if we make FileHeader to be optional?
e.g. a valid YAML can be something like this:

--- !ELF
Sections:
.....
Symbols:
....

When there is no FileHeader, yaml2obj could require --platform=X option (or alike) to be specified.
It can be similar to linker's -m (emulation) option and can accept the same strings for X:

  std::pair<ELFKind, uint16_t> ret =
      StringSwitch<std::pair<ELFKind, uint16_t>>(s)
          .Cases("aarch64elf", "aarch64linux", "aarch64_elf64_le_vec",
                 {ELF64LEKind, EM_AARCH64})
          .Cases("armelf", "armelf_linux_eabi", {ELF32LEKind, EM_ARM})
...
          .Cases("elf_amd64", "elf_x86_64", {ELF64LEKind, EM_X86_64})
          .Case("elf_i386", {ELF32LEKind, EM_386})
          .Case("elf_iamcu", {ELF32LEKind, EM_IAMCU})
          .Default({ELFNoneKind, EM_NONE});

Perhaps we also could add --elf-type=X to switch the output type: ET_DYN/ET_REL/ET_EXEC
(and have something by default).

A benefit is that we often do not need or use fields of the file header in our tests.
Making it optional can reduce the YAML descriptions.

[jhenderson]

In D73621#1851587, @grimar wrote:

In D73621#1848843, @grimar wrote:

In D73621#1848823, @jhenderson wrote:

Idle thought: I wonder if it's worth introducing some kind of command-line macro option to yaml2obj? E.g. -DMACHINE=EM_X86_64 would replace Machine: MACHINE (or some similar syntax). There are a growing number of cases where we have two identical YAML blobs except for some minor difference.

Sounds good. I'd give it a try.

One more idea: what if we make FileHeader to be optional?
e.g. a valid YAML can be something like this:

--- !ELF
Sections:
.....
Symbols:
....

When there is no FileHeader, yaml2obj could require --platform=X option (or alike) to be specified.
It can be similar to linker's -m (emulation) option and can accept the same strings for X:

  std::pair<ELFKind, uint16_t> ret =
      StringSwitch<std::pair<ELFKind, uint16_t>>(s)
          .Cases("aarch64elf", "aarch64linux", "aarch64_elf64_le_vec",
                 {ELF64LEKind, EM_AARCH64})
          .Cases("armelf", "armelf_linux_eabi", {ELF32LEKind, EM_ARM})
...
          .Cases("elf_amd64", "elf_x86_64", {ELF64LEKind, EM_X86_64})
          .Case("elf_i386", {ELF32LEKind, EM_386})
          .Case("elf_iamcu", {ELF32LEKind, EM_IAMCU})
          .Default({ELFNoneKind, EM_NONE});

Perhaps we also could add --elf-type=X to switch the output type: ET_DYN/ET_REL/ET_EXEC
(and have something by default).

A benefit is that we often do not need or use fields of the file header in our tests.
Making it optional can reduce the YAML descriptions.

My concern with this approach is that we can keep on adding switches, with very little gain. Taken to an extreme, you could specify sections, symbols, relocations etc via command-line switches and then not have any YAML at all. I don't think that really helps readability. A generic default file header might be okay though, but what would the parameters be? X86_64? ARM? It probably doesn't matter too much, for the cases its used in, but it is perhaps something we should be aware of.

llvm/lib/ObjectYAML/ELFEmitter.cpp

 void ELFState<ELFT>::writeSectionContent(
     Elf_Shdr &SHeader, const ELFYAML::RelocationSection &Section,
     ContiguousBlobAccumulator &CBA) {
   assert((Section.Type == llvm::ELF::SHT_REL ||
           Section.Type == llvm::ELF::SHT_RELA) &&
          "Section type is not SHT_REL nor SHT_RELA");
 
   bool IsRela = Section.Type == llvm::ELF::SHT_RELA;
+  if (Section.EntSize)
+    SHeader.sh_entsize = *Section.EntSize;
+  else
     SHeader.sh_entsize = IsRela ? sizeof(Elf_Rela) : sizeof(Elf_Rel);
-  SHeader.sh_size = SHeader.sh_entsize * Section.Relocations.size();
+  SHeader.sh_size = (IsRela ? sizeof(Elf_Rela) : sizeof(Elf_Rel)) *
+                    Section.Relocations.size();
 
   // For relocation section set link to .symtab by default.
   unsigned Link = 0;
   if (Section.Link.empty() && SN2I.lookup(".symtab", Link))
     SHeader.sh_link = Link;
 
   if (!Section.RelocatableSec.empty())
     SHeader.sh_info = toSectionIndex(Section.RelocatableSec, Section.Name);

llvm/test/tools/yaml2obj/ELF/reloc-sec-entry-size.yaml

+## Test how yaml2obj sets values for sh_entsize fields of relocation sections.
+
+# RUN: yaml2obj --docnum=1 %s -o %t64
+# RUN: llvm-readelf --sections %t64 | FileCheck %s --check-prefix=ELF64
+
+# RUN: yaml2obj --docnum=2 %s -o %t32
+# RUN: llvm-readelf --sections %t32 | FileCheck %s --check-prefix=ELF32
+
+# ELF64: Name          Type Address          Off    Size   ES
+# ELF64: .rela.default RELA 0000000000000000 000040 000000 18
+# ELF64: .rel.default  REL  0000000000000000 000040 000000 10
+# ELF64: .relr.default RELR 0000000000000000 000040 000000 08
+# ELF64: .rela.custom  RELA 0000000000000000 000040 000000 ff
+# ELF64: .rel.custom   REL  0000000000000000 000040 000000 ff
+# ELF64: .relr.custom  RELR 0000000000000000 000040 000000 ff
+
+# ELF32: Name          Type  Address  Off    Size   ES
+# ELF32: .rela.default RELA  00000000 000034 000000 0c
+# ELF32: .rel.default  REL   00000000 000034 000000 08
+# ELF32: .relr.default RELR  00000000 000034 000000 04
+# ELF32: .rela.custom  RELA  00000000 000034 000000 ff
+# ELF32: .rel.custom   REL   00000000 000034 000000 ff
+# ELF32: .relr.custom  RELR  00000000 000034 000000 ff
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2LSB
+  Type:    ET_REL
+  Machine: EM_X86_64
+Sections:
+## Check default sh_entsize field values.
+  - Name: .rela.default
+    Type: SHT_RELA
+  - Name: .rel.default
+    Type: SHT_REL
+  - Name: .relr.default
+    Type: SHT_RELR
+## Check we can set sh_entsize fields to arbitrary values.
+  - Name:    .rela.custom
+    Type:    SHT_RELA
+    EntSize: 0xFF
+  - Name:    .rel.custom
+    Type:    SHT_REL
+    EntSize: 0xFF
+  - Name:    .relr.custom
+    Type:    SHT_RELR
+    EntSize: 0xFF
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS32
+  Data:    ELFDATA2LSB
+  Type:    ET_REL
+  Machine: EM_386
+Sections:
+## Check default sh_entsizes field values.
+  - Name: .rela.default
+    Type: SHT_RELA
+  - Name: .rel.default
+    Type: SHT_REL
+  - Name: .relr.default
+    Type: SHT_RELR
+## Check we can set sh_entsize fields to arbitrary values.
+  - Name:    .rela.custom
+    Type:    SHT_RELA
+    EntSize: 0xFF
+  - Name:    .rel.custom
+    Type:    SHT_REL
+    EntSize: 0xFF
+  - Name:    .relr.custom
+    Type:    SHT_RELR
+    EntSize: 0xFF