[obj2yaml] - Teach obj2yaml to dump SHT_NOBITS sections when dumping program headers.

[grimar]

Summary

SHT_NOBITS are a bit special because occupy no physical space.
This patch adds support for them.

[grimar]

Rebased on top of D77652.

[grimar]

I'll add one more test case.

[grimar]

I think I've found an issue in yaml2obj revealed by the test case I'd like to add here.
I'll investigate it in my tomorrow and update this patch after.

[MaskRay]

I feel that the dependency can be reversed, i.e. D77652 will depend on D77805.

Using file offsets for SHT_NOBITS is the only way. For an empty SHT_PROGBITS, we need to discuss several malformed cases. It is less clear what we should do.

[grimar]

In D77805#1972126, @grimar wrote:

I think I've found an issue in yaml2obj revealed by the test case I'd like to add here.
I'll investigate it in my tomorrow and update this patch after.

It needs D78005 to be landed to proced.

[jhenderson]

llvm/test/tools/obj2yaml/program-headers.yaml

482

in a different

493

Could you add some comments to these sections saying why the Address and Size values were chosen. It's not particularly clear to me how they're important to the test.

llvm/tools/obj2yaml/elf2yaml.cpp

307

"SHT_NOBITS sections usually occupy no physical space in a file. Such sections belong to a segment when they reside in the segment's virtual address space."

I added the "usually" because I think a SHT_NOBITS section sandwiched between two non NOBITS sections might need to be allocated file space to avoid things pointing to the wrong place.

[grimar]

• Addressed review comments.

Ready for review:

I feel that the dependency can be reversed, i.e. D77652 will depend on D77805.

Done.

In D77805#1972126, @grimar wrote:
I think I've found an issue in yaml2obj revealed by the test case I'd like to add here.
I'll investigate it in my tomorrow and update this patch after.

It needs D78005 to be landed to proceed.

Let's go without an additional test case for now, the issue I was talking about is
related to trailing multiple SHT_NOBITS sections (fixed in D78005).
We can add it later.

[jhenderson]

LGTM with grammar and typo fixes.

llvm/test/tools/obj2yaml/program-headers.yaml

493

arbitraty -> arbitrary

499

the size which -> a size that

504

arbitraty -> arbitrary

517

arbitraty -> arbitrary
different from -> different to

llvm/test/tools/obj2yaml/program-headers.yaml

Sections:
   - Name:    .non-alloc.2
     Type:    SHT_NULL
     Flags:   [ ]
     Size:    0x10
   - Name:    .alloc.2
     Type:    SHT_PROGBITS
     Flags:   [ SHF_ALLOC ]
     Size:    0x1
+
+## Check how we dump segments which contain SHT_NOBITS sections.
+# RUN: yaml2obj --docnum=5 %s -o %t5
+# RUN: obj2yaml %t5 | FileCheck %s --check-prefix=NOBITS
+
+# NOBITS:      ProgramHeaders:
+# NOBITS-NEXT:   - Type:  PT_LOAD
+# NOBITS-NEXT:     Flags: [ PF_W, PF_R ]
+# NOBITS-NEXT:     Sections:
+# NOBITS-NEXT:       - Section: .bss
+# NOBITS-NEXT:   - Type:  PT_LOAD
+# NOBITS-NEXT:     Flags: [ PF_W, PF_R ]
+# NOBITS-NEXT:     Sections:
+# NOBITS-NEXT:       - Section: .data.1
+# NOBITS-NEXT:       - Section: .bss
+# NOBITS-NEXT:   - Type:  PT_LOAD
+# NOBITS-NEXT:     Flags: [ PF_W, PF_R ]
+# NOBITS-NEXT:     Sections:
+# NOBITS-NEXT:       - Section: .data.1
+# NOBITS-NEXT:       - Section: .bss
+# NOBITS-NEXT:       - Section: .data.2
+# NOBITS-NEXT:   - Type:  PT_LOAD
+# NOBITS-NEXT:     Flags: [ PF_W, PF_R ]
+# NOBITS-NEXT:     Sections:
+# NOBITS-NEXT:       - Section: .bss
+# NOBITS-NEXT:       - Section: .data.2
+# NOBITS-NEXT:   - Type:  PT_LOAD
+# NOBITS-NEXT:     Flags: [ PF_W, PF_R ]
+# NOBITS-NEXT:     Sections:
+# NOBITS-NEXT:       - Section: .foo.bss
+# NOBITS-NEXT:       - Section: .bar.bss
+# NOBITS-NEXT:     VAddr: 0x0000000200000000
+# NOBITS-NEXT: Sections:
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2LSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+ProgramHeaders:
+## Case 1: the segment contains a single SHT_NOBITS section.
+  - Type:  PT_LOAD
+    Flags: [ PF_W, PF_R ]
+    Sections:
+      - Section: .bss
+## Case 2: the SHT_NOBITS section is the last section in the segment.
+  - Type:  PT_LOAD
+    Flags: [ PF_W, PF_R ]
+    Sections:
+      - Section: .data.1
+      - Section: .bss
+## Case 3: the SHT_NOBITS section is in the middle of the segment.
+  - Type:  PT_LOAD
+    Flags: [ PF_W, PF_R ]
+    Sections:
+      - Section: .data.1
+      - Section: .bss
+      - Section: .data.2
+## Case 4: the SHT_NOBITS section is the first section in the segment.
+  - Type:  PT_LOAD
+    Flags: [ PF_W, PF_R ]
+    Sections:
+      - Section: .bss
+      - Section: .data.2
+## Case 5: another two SHT_NOBITS sections in a different segment.
+  - Type:  PT_LOAD
+    Flags: [ PF_W, PF_R ]
+    Sections:
+      - Section: .foo.bss
+      - Section: .bar.bss
+    VAddr: 0x200000000
+Sections:
+  - Name:    .data.1
+    Type:    SHT_PROGBITS
+    Flags:   [ SHF_WRITE, SHF_ALLOC ]
+## Use an arbitraty address and size.
+    Address: 0x1000
+    Size:    0x1
+  - Name:  .bss
+    Type:  SHT_NOBITS
+    Flags: [ SHF_WRITE, SHF_ALLOC ]
+## Use the size which is larger than the file size.
+    Size:  0x00000000FFFFFFFF
+  - Name:  .data.2
+    Type:  SHT_PROGBITS
+    Flags: [ SHF_WRITE, SHF_ALLOC ]
+## Use an arbitraty size.
+    Size:  0x1
+  - Name:    .foo.bss
+    Type:    SHT_NOBITS
+    Flags:   [ SHF_WRITE, SHF_ALLOC ]
+## Set an arbitrary address and size so that this section can be used
+## to start a different non-overlapping segment.
+## I.e. its address is larger than addresses of previous sections.
+    Size:    0x10
+    Address: 0x200000000
+  - Name:    .bar.bss
+    Type:    SHT_NOBITS
+    Flags:   [ SHF_WRITE, SHF_ALLOC ]
+## Use an arbitraty size that is different from the size of
+## the previous section.
+    Size:    0x20

llvm/tools/obj2yaml/elf2yaml.cpp

 }
 
 template <class ELFT>
 static bool isInSegment(const ELFYAML::Section &Sec,
                         const typename ELFT::Shdr &SHdr,
                         const typename ELFT::Phdr &Phdr) {
   if (Sec.Type == ELF::SHT_NULL)
     return false;
-  return SHdr.sh_offset >= Phdr.p_offset &&
-         (SHdr.sh_offset + SHdr.sh_size <= Phdr.p_offset + Phdr.p_filesz);
+
+  // A section is within a segment when its location in a file is within the
+  // [p_offset, p_offset + p_filesz] region.
+  bool FileOffsetsMatch =
+      SHdr.sh_offset >= Phdr.p_offset &&
+
+  if (FileOffsetsMatch)
+    return true;
+
+  // SHT_NOBITS sections usually occupy no physical space in a file. Such
+  // sections belong to a segment when they reside in the segment's virtual
+  // address space.
+  if (Sec.Type != ELF::SHT_NOBITS)
+    return false;
+
+  return SHdr.sh_addr >= Phdr.p_vaddr &&
+         SHdr.sh_addr <= Phdr.p_vaddr + Phdr.p_memsz;
 }
 
 template <class ELFT>
 Expected<std::vector<ELFYAML::ProgramHeader>>
 ELFDumper<ELFT>::dumpProgramHeaders(
     ArrayRef<std::unique_ptr<ELFYAML::Chunk>> Chunks) {
   std::vector<ELFYAML::ProgramHeader> Ret;
   Expected<typename ELFT::PhdrRange> PhdrsOrErr = Obj.program_headers();