[ObjectYAML][DWARF] Support emitting the .debug_aranges section in ELFYAML.

[Higuoxing]

Summary

This patch enables yaml2obj to emit the .debug_aranges section in ELFYAML.

Known issues:

• The current implementation of debug_aranges doesn't support emitting segment in the (segment, address, length) tuple. I will fix it in a follow-up patch.

[jhenderson]

llvm/lib/ObjectYAML/DWARFEmitter.cpp

99–100 ↗

(On Diff #267779)

Perhaps you should move this fix into a follow-up change, and wait to add DWARF64 testing to that same change. It looks good, but I think we want to avoid confusing the two issues.

108 ↗

(On Diff #267779)

Similar to above, you should probably delay segment size support for a follow-up change.

llvm/lib/ObjectYAML/ELFYAML.cpp

1657–1659 ↗

(On Diff #267779)

This seems somewhat unrelated too? Perhaps defer to another patch, rather than trying to adopt it all at once.

llvm/test/tools/yaml2obj/ELF/DWARF/debug-aranges.yaml

6

the -> a

8

Nit: -S -> --sections (since you are using a long option already).

9

Can you get away with just one check-prefix here? It looks like they're both used together always.

76

I'd make one of these 4 byte size and the other 8 (perhaps deliberately do 8 bytes for the DWARF32 case, to deliberately show the address size and DWARF32/64 bits are unrelated). Since I'm suggesting you wait for DWARF64 support to a separate change, make them both DWARF32 temporarily, and update the test case when you add DWARF64 support.

82

Not part of this change, but I think this should be changed in the future to be an explicit Format: DWARF64 field, and just leave TotalLength to be the only length field (i.e. don't have a separate Length and TotalLength fields).

92

the -> and

93

In your follow-up change to add support for big endian, you can avoid the need for two separate YAML blobs, by using yaml2obj's -D option. It works similary to FileCheck's:

# RUN: yaml2obj %s -D ENDIAN=ELFDATA2LSB -o %tle.o
...
# RUN: yaml2obj %s -D ENDIAN=ELFDATA2MSB -o %tbe.o
...

--- !ELF
FileHeader:
  Class:   ELFCLASS64
  Data:    [[ENDIAN]]
...

179

Delete "the" from "the raw section"

217

debug_arange -> debug_aranges

[Higuoxing]

Address comments

• Remove test cases that relate to endianness or DWARF64
  • These test cases will be added in a follow-up patch.
• Correct some grammar mistakes.

[Higuoxing]

Correct docnum.

[jhenderson]

llvm/test/tools/yaml2obj/ELF/DWARF/debug-aranges.yaml

10

As you only have one check prefix, you can use --check-prefix like before, if you want.

I'm slightly surprised the code naturally produces big endian output!

49

Nit: I'd get rid of the space in the dashes on this line.

Same sort of comment on other lines below.

77

Looks like you haven't addressed my earlier comment about making the tables different address sizes. I don't think the address size should have to match the target's address size, if not desired.

80–81

It's probably a good idea for at least one table to have multiple Descriptor entries.

[grimar]

llvm/test/tools/yaml2obj/ELF/DWARF/debug-aranges.yaml

107

I wonder if we need to test such things?

The Content is a key that is often used to emit some broken output.
So why not just to test that we can emit an arbitrary content, e.g.: Content: "112233"?

[Higuoxing]

llvm/test/tools/yaml2obj/ELF/DWARF/debug-aranges.yaml

107

I want to test the default values of section header, e.g., sh_info/link/... and reuse the check tag: DWARF-BE-DEFAULT. What about having 2 check tags: DWARF-BE-HEADER, DWARF-BE-CONTENT

# RUN: yaml2obj --docnum=1 %s -o %t1.o
# RUN: llvm-readobj --sections --section-data %t1.o | FileCheck --check-prefixes=DWARF-BE-HEADER,DWARF-BE-CONTENT

#      DWARF-BE-HEADER: Index: 1
# DWARF-BE-HEADER-NEXT: Name:  .debug_aranges
...
#      DWARF-BE-CONTENT: SectionData (
# DWARF-BE-CONTENT-NEXT:   ...

# RUN: yaml2obj --docnum=2 %s -o %t2.o
# RUN: llvm-readobj --sections --section-data %t2.o | FileCheck --check-prefixes=DWARF-BE-HEADER,ARBITRARY-CONTENT

#      ARBITRARY-CONTENT: SectionData (
# ARBITRARY-CONTENT-NEXT:   ...

[grimar]

llvm/test/tools/yaml2obj/ELF/DWARF/debug-aranges.yaml

107

Having 2 tags sounds fine to me.

[Higuoxing]

Addressed comments:

• Let the section in test case (b) be generated from an arbitraty "Content".
• Let the "AddrSize" in first entry be 0x04.
• Add one more (Address, Length) pair in the second entry.

[Higuoxing]

Correct one comment in test case.

[jhenderson]

LGTM, with one nit.

Don't forget to mark comments as "Done" before posting your latest update. The "Done" marks will be submitted when you post the diff, and it's then easy to see any comments that might have been missed/not addressed etc.

llvm/test/tools/yaml2obj/ELF/DWARF/debug-aranges.yaml

12–25

Nit: there's an extra two spaces of indentation here that aren't really needed as far as I can tell, so you can remove them.

[Higuoxing]

Remove unneeded spaces.

Thanks for reviewing!

llvm/lib/ObjectYAML/DWARFYAML.cpp

return DebugStrings.empty() && AbbrevDecls.empty() && ARanges.empty() &&
          DebugRanges.empty() && !PubNames && !PubTypes && !GNUPubNames &&
          !GNUPubTypes && CompileUnits.empty() && DebugLines.empty();
 }
 
 SetVector<StringRef> DWARFYAML::Data::getUsedSectionNames() const {
   SetVector<StringRef> SecNames;
   if (!DebugStrings.empty())
     SecNames.insert("debug_str");
+  if (!ARanges.empty())
+    SecNames.insert("debug_aranges");
   return SecNames;
 }
 
 namespace yaml {
 
 void MappingTraits<DWARFYAML::Data>::mapping(IO &IO, DWARFYAML::Data &DWARF) {
   auto oldContext = IO.getContext();
   IO.setContext(&DWARF);

llvm/lib/ObjectYAML/ELFEmitter.cpp

 }
 
 template <class ELFT>
 uint64_t emitDWARF(typename ELFT::Shdr &SHeader, StringRef Name,
                    const DWARFYAML::Data &DWARF, raw_ostream &OS) {
   uint64_t BeginOffset = OS.tell();
   if (Name == ".debug_str")
     DWARFYAML::EmitDebugStr(OS, DWARF);
+  else if (Name == ".debug_aranges")
+    DWARFYAML::EmitDebugAranges(OS, DWARF);
   else
     llvm_unreachable("unexpected emitDWARF() call");
 
   return OS.tell() - BeginOffset;
 }
 
 template <class ELFT>
 void ELFState<ELFT>::initDWARFSectionHeader(Elf_Shdr &SHeader, StringRef Name,

llvm/test/tools/yaml2obj/ELF/DWARF/debug-aranges.yaml

+## Test that yaml2obj emits .debug_aranges section.
+
+## a) Generate the .debug_aranges section from the "DWARF" entry.
+
+## Generate and verify a big endian object file.
+
+# RUN: yaml2obj --docnum=1 -DENDIAN=ELFDATA2MSB %s -o %t1.o
+# RUN: llvm-readobj --sections --section-data %t1.o | \
+# RUN:   FileCheck %s -DADDRALIGN=1 -DSIZE=96 --check-prefixes=DWARF-BE-HEADER,DWARF-BE-CONTENT
+
+#       DWARF-BE-HEADER: Index: 1
+#  DWARF-BE-HEADER-NEXT: Name: .debug_aranges (1)
+#  DWARF-BE-HEADER-NEXT: Type: SHT_PROGBITS (0x1)
+#  DWARF-BE-HEADER-NEXT: Flags [ (0x0)
+#  DWARF-BE-HEADER-NEXT: ]
+#  DWARF-BE-HEADER-NEXT: Address: 0x0
+#  DWARF-BE-HEADER-NEXT: Offset: 0x40
+#  DWARF-BE-HEADER-NEXT: Size: [[SIZE]]
+#  DWARF-BE-HEADER-NEXT: Link: 0
+#  DWARF-BE-HEADER-NEXT: Info: 0
+#  DWARF-BE-HEADER-NEXT: AddressAlignment: [[ADDRALIGN]]
+#  DWARF-BE-HEADER-NEXT: EntrySize: 0
+#      DWARF-BE-CONTENT: SectionData (
+# DWARF-BE-CONTENT-NEXT:   0000: 0000002C 00020000 00000400 00000000
+##                               |        |   |        | |  |
+##                               |        |   |        | |  +------- Padding zeros (4-byte)
+##                               |        |   |        | +- SegSize (1-byte) 0x00
+##                               |        |   |        +- AddrSize (1-byte) 0x04
+##                               |        |   +-------- CuOffset (4-byte) 0x00
+##                               |        +--- Version (2-byte) 0x02
+##                               +------- InitialLength (4-byte) 0x2c
+##
+# DWARF-BE-CONTENT-NEXT:   0010: 00001234 00000020 00000000 00000000
+##                               |        |        |
+##                               |        |        +---------------- Terminating Entry
+##                               |        +------- Length (4-byte) 0x20
+##                               +------- Address (4-byte) 0x1234
+##
+# DWARF-BE-CONTENT-NEXT:   0020: 0000002C 00020000 00650800 00000000
+##                               |        |   |        | |  |
+##                               |        |   |        | |  +------- Padding zeros (4-byte)
+##                               |        |   |        | +- SegSize (1-byte) 0x00
+##                               |        |   |        +- AddrSize (1-byte) 0x08
+##                               |        |   +-------- CuOffset (4-byte) 0x00
+##                               |        +--- Version (2-byte) 0x02
+##                               +------- InitialLength (4-byte) 0x2c
+##
+# DWARF-BE-CONTENT-NEXT:   0030: 00000000 00005678 00000000 00000020
+##                               |                 |
+##                               |                 +---------------- Length (8-byte) 0x20
+##                               +---------------- Address (8-byte) 0x5678
+##
+# DWARF-BE-CONTENT-NEXT:   0040: 00000000 56780000 00000000 00000010
+##                               |                 |
+##                               |                 +---------------- Length (8-byte) 0x10
+##                               +---------------- Address (8-byte) 0x5678
+##
+# DWARF-BE-CONTENT-NEXT:   0050: 00000000 00000000 00000000 00000000
+##                               |
+##                               +---------------------------------- Terminating entry
+##
+# DWARF-BE-CONTENT-NEXT: )
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    [[ENDIAN]]
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+DWARF:
+  debug_aranges:
+    - Length:
+        TotalLength: 0x2c
+      Version:  2
+      CuOffset: 0
+      AddrSize: 0x04
+      SegSize:  0x00
+      Descriptors:
+        - Address: 0x00001234
+          Length:  0x20
+    - Length:
+        TotalLength: 0x2c
+      Version:  2
+      CuOffset: 0x65
+      AddrSize: 0x08
+      SegSize:  0x00
+      Descriptors:
+        - Address: 0x0000000000005678
+          Length:  0x20
+        - Address: 0x0000000056780000
+          Length:  0x10
+
+## b) Generate the .debug_aranges section from raw section content.
+
+# RUN: yaml2obj --docnum=2 %s -o %t2.o
+# RUN: llvm-readobj --sections --section-data %t2.o | \
+# RUN:   FileCheck %s -DADDRALIGN=0 -DSIZE=3 --check-prefixes=DWARF-BE-HEADER,ARBITRARY-CONTENT
+
+#      ARBITRARY-CONTENT: SectionData (
+# ARBITRARY-CONTENT-NEXT:   0000: 112233
+# ARBITRARY-CONTENT-NEXT: )
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2MSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+Sections:
+  - Name:    .debug_aranges
+    Type:    SHT_PROGBITS
+    Content: "112233"
+
+## c) Generate the .debug_aranges section when the "Size" is specified.
+
+# RUN: yaml2obj --docnum=3 %s -o %t3.o
+# RUN: llvm-readobj --hex-dump=.debug_aranges %t3.o | FileCheck %s --check-prefix=SIZE
+
+#       SIZE: Hex dump of section '.debug_aranges':
+#  SIZE-NEXT: 0x00000000 00000000 00000000 00000000 00000000 ................
+# SIZE-EMPTY:
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2MSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+Sections:
+  - Name: .debug_aranges
+    Type: SHT_PROGBITS
+    Size: 0x10
+
+## d) Test that yaml2obj emits an error message when both the "Size" and the
+## "debug_aranges" entry are specified at the same time.
+
+# RUN: not yaml2obj --docnum=4 %s 2>&1 | FileCheck %s --check-prefix=ERROR
+
+# ERROR: yaml2obj: error: cannot specify section '.debug_aranges' contents in the 'DWARF' entry and the 'Content' or 'Size' in the 'Sections' entry at the same time
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2MSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+Sections:
+  - Name: .debug_aranges
+    Type: SHT_PROGBITS
+    Size: 0x10
+DWARF:
+  debug_aranges:
+    - Length:
+        TotalLength: 0x2c
+      Version:  2
+      CuOffset: 0
+      AddrSize: 0x08
+      SegSize:  0x00
+      Descriptors:
+        - Address: 0x0000000000001234
+          Length:  0x20
+
+## e) Test that yaml2obj emits an error message when both the "Content" and the
+## "debug_aranges" entry are specified at the same time.
+
+# RUN: not yaml2obj --docnum=5 %s 2>&1 | FileCheck %s --check-prefix=ERROR
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2MSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+Sections:
+  - Name:    .debug_aranges
+    Type:    SHT_PROGBITS
+    Content: "00"
+DWARF:
+  debug_aranges:
+    - Length:
+        TotalLength: 0x2c
+      Version:  2
+      CuOffset: 0
+      AddrSize: 0x08
+      SegSize:  0x00
+      Descriptors:
+        - Address: 0x0000000000001234
+          Length:  0x20
+
+## f) Test that all the properties can be overridden by the section header when
+## the "debug_aranges" entry doesn't exist.
+
+# RUN: yaml2obj --docnum=6 %s -o %t6.o
+# RUN: llvm-readelf --sections %t6.o | FileCheck %s --check-prefix=OVERRIDDEN
+
+#      OVERRIDDEN: [Nr] Name           Type   Address          Off    Size   ES Flg Lk Inf Al
+#      OVERRIDDEN: [ 1] .debug_aranges STRTAB 0000000000002020 000050 000030 01   A  2   1  2
+# OVERRIDDEN-NEXT: [ 2] .sec           STRTAB 0000000000000000 000080 000000 00      0   0  0
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2MSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+Sections:
+  - Name:         .debug_aranges
+    Type:         SHT_STRTAB         # SHT_PROGBITS by default.
+    Flags:        [SHF_ALLOC]        # 0 by default.
+    Link:         .sec               # 0 by default.
+    EntSize:      1                  # 0 by default.
+    Info:         1                  # 0 by default.
+    AddressAlign: 2                  # 0 by default.
+    Address:      0x0000000000002020 # 0x00 by default.
+    Offset:       0x00000050         # 0x40 for the first section.
+    Size:         0x30               # Set the "Size" so that we can reuse the check tag "OVERRIDDEN".
+  - Name:         .sec               # Linked by .debug_aranges.
+    Type:         SHT_STRTAB
+
+## g) Test that all the properties can be overridden by the section header when
+## the "debug_aranges" entry exists.
+
+# RUN: yaml2obj --docnum=7 %s -o %t7.o
+# RUN: llvm-readelf --sections %t7.o | FileCheck %s --check-prefix=OVERRIDDEN
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2MSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+Sections:
+  - Name:         .debug_aranges
+    Type:         SHT_STRTAB         # SHT_PROGBITS by default.
+    Flags:        [SHF_ALLOC]        # 0 by default.
+    Link:         .sec               # 0 by default.
+    EntSize:      1                  # 0 by default.
+    Info:         1                  # 0 by default.
+    AddressAlign: 2                  # 1 by default.
+    Address:      0x0000000000002020 # 0x00 by default.
+    Offset:       0x00000050         # 0x40 for the first section.
+  - Name:         .sec               # Linked by .debug_aranges.
+    Type:         SHT_STRTAB
+DWARF:
+  debug_aranges:
+    - Length:
+        TotalLength: 0x2c
+      Version:  2
+      CuOffset: 0
+      AddrSize: 0x08
+      SegSize:  0x00
+      Descriptors:
+        - Address: 0x0000000000001234
+          Length:  0x20
+
+## h) Test that yaml2obj will not generate the .debug_aranges section when the "DWARF" entry
+## exists but the "debug_aranges" entry doesn't exist in the "DWARF" entry or the "Sections" entry.
+
+# RUN: yaml2obj --docnum=8 %s -o %t8.o
+# RUN: llvm-readelf --sections %t8.o | FileCheck /dev/null --implicit-check-not=.debug_aranges
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2MSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+DWARF: