[llvm-readobj][RISCV] Support dumping PT_RISCV_ATTRIBUTES

[StephenFan]

Summary

This patch drops the prefix PT_RISCV_ when dumping PT_RISCV_ATTRIBUTES.

GNU readelf dumps it as RISCV_ATTRIBUT. Because GNU readelf uses
something like %-14.14s so only the first 14 bytes are printed.

[jhenderson]

LGTM. Does llvm-objdump need something equivalent? (I don't remember whether we interpret platform-specific program headers in llvm-objdump)

llvm/test/tools/llvm-readobj/ELF/program-headers.test

611

Nit: delete extra blank line.

[MaskRay]

Supporting dump PT_RISCV_ATTRIBUTES

Support dumping

But it is weird that riscv gnu toolchain's readelf dumps PT_RISCV_ATTRIBUTES as RISCV_ATTRIBUT.

riscv gnu toolchain's readelf => GNU readelf

When describing binutils-gdb behavior, prefer the official binutils-gdb to a downstream repository like riscv-gnu-toolchain.

But it is weird that riscv gnu toolchain's readelf dumps PT_RISCV_ATTRIBUTES as RISCV_ATTRIBUT.

Because it uses something like %-14.14s so only the first 14 bytes are printed. Many long program header types like PT_AARCH64_MEMTAG_MTE are affected as well.

For target specific program headers, we generally drop the PT_xxxx_prefix. This patch also adopts this approach.

Actually I am unsure we should drop the machine prefix. GNU readelf uses the prefix and it seems clearer to have the prefix.

llvm/tools/llvm-readobj/ELFDumper.cpp

1436

if (Seg.consume_front("PT_RISCV_"))
  return Seg.str();

[StephenFan]

1. Remove blank line.
2. Replace drop_front with consume_front

[StephenFan]

In D128493#3607134, @MaskRay wrote:

Supporting dump PT_RISCV_ATTRIBUTES

Support dumping

But it is weird that riscv gnu toolchain's readelf dumps PT_RISCV_ATTRIBUTES as RISCV_ATTRIBUT.

riscv gnu toolchain's readelf => GNU readelf

When describing binutils-gdb behavior, prefer the official binutils-gdb to a downstream repository like riscv-gnu-toolchain.

But it is weird that riscv gnu toolchain's readelf dumps PT_RISCV_ATTRIBUTES as RISCV_ATTRIBUT.

Because it uses something like %-14.14s so only the first 14 bytes are printed. Many long program header types like PT_AARCH64_MEMTAG_MTE are affected as well.

For target specific program headers, we generally drop the PT_xxxx_prefix. This patch also adopts this approach.

Actually I am unsure we should drop the machine prefix. GNU readelf uses the prefix and it seems clearer to have the prefix.

Thanks for such a detailed explanation!

[MaskRay]

ATTRIBUTES may be fine for now since the GNU readelf output isn't ideal RISCV_ATTRIBUT.

llvm/include/llvm/BinaryFormat/ELF.h

enum {
   PT_ARM_EXIDX = 0x70000001,
   PT_ARM_UNWIND = 0x70000001,
 
   // MIPS program header types.
   PT_MIPS_REGINFO = 0x70000000,  // Register usage information.
   PT_MIPS_RTPROC = 0x70000001,   // Runtime procedure table.
   PT_MIPS_OPTIONS = 0x70000002,  // Options segment.
   PT_MIPS_ABIFLAGS = 0x70000003, // Abiflags segment.
+
+  // RISCV program header types.
+  PT_RISCV_ATTRIBUTES = 0x70000003,
 };
 
 // Segment flag bits.
 enum : unsigned {
   PF_X = 1,                // Execute
   PF_W = 2,                // Write
   PF_R = 4,                // Read
   PF_MASKOS = 0x0ff00000,  // Bits for operating system-specific semantics.

llvm/test/tools/llvm-readobj/ELF/program-headers.test

- Type:     0x70000001 ## PT_ARM_EXIDX, PT_MIPS_RTPROC
     VAddr:    0x1000
     FirstSec: .foo.begin
     LastSec:  .foo.end
 ## Case 20: the PT_MIPS_OPTIONS segment.
   - Type:     0x70000002 ## PT_MIPS_OPTIONS
     VAddr:    0x1000
     FirstSec: .foo.begin
     LastSec:  .foo.end
-## Case 21: the PT_MIPS_ABIFLAGS segment.
-  - Type:     0x70000003 ## PT_MIPS_ABIFLAGS
+## Case 21: the PT_MIPS_ABIFLAGS/PT_RISCV_ATTRIBUTES segment.
+  - Type:     0x70000003 ## PT_MIPS_ABIFLAGS/PT_RISCV_ATTRIBUTES
     VAddr:    0x1000
     FirstSec: .foo.begin
     LastSec:  .foo.end
 ## Case 22: the PT_HIPROC segment.
   - Type:     0x7fffffff ## PT_HIPROC
     VAddr:    0x1000
     FirstSec: .foo.begin
     LastSec: .foo.end

 # MIPS-LLVM:        Type: PT_MIPS_REGINFO (0x70000000)
 # MIPS-LLVM:      ProgramHeader {
 # MIPS-LLVM-NEXT:   Type: PT_MIPS_RTPROC (0x70000001)
 # MIPS-LLVM:      ProgramHeader {
 # MIPS-LLVM-NEXT:   Type: PT_MIPS_OPTIONS (0x70000002)
 # MIPS-LLVM:      ProgramHeader {
 # MIPS-LLVM-NEXT:   Type: PT_MIPS_ABIFLAGS (0x70000003)
 
+## CHECK how we dump RISCV specific program headers.
+# RUN: yaml2obj --docnum=1 -DBITS=64 -DMACHINE=EM_RISCV %s -o %triscv.elf
+# RUN: llvm-readelf --program-headers %triscv.elf | FileCheck %s --check-prefix=RISCV-GNU
+# RUN: llvm-readobj --program-headers %triscv.elf | FileCheck %s --check-prefix=RISCV-LLVM
+
+# RISCV-GNU:       ATTRIBUTES  0x000548 0x0000000000001000 0x0000000000001000 0x000003 0x000003 0x1
+# RISCV-LLVM:      ProgramHeader {
+# RISCV-LLVM:        Type: PT_RISCV_ATTRIBUTES (0x70000003)
+
 ## Check that llvm-readelf reports a warning when a program interpreter
 ## name is non-null-terminated or when PT_INTERP has an offset that
 ## goes past the end of the file.
 # RUN: yaml2obj --docnum=2 %s -o %t.err
 
 ## Show the size of the output produced. It is used in the YAML below.
 # RUN: wc -c < %t.err | FileCheck %s --check-prefix=SIZE
 # SIZE: 616

llvm/tools/llvm-readobj/ELFDumper.cpp

static StringRef segmentTypeToString(unsigned Arch, unsigned Type) {
   case ELF::EM_MIPS_RS3_LE:
     switch (Type) {
       LLVM_READOBJ_ENUM_CASE(ELF, PT_MIPS_REGINFO);
       LLVM_READOBJ_ENUM_CASE(ELF, PT_MIPS_RTPROC);
       LLVM_READOBJ_ENUM_CASE(ELF, PT_MIPS_OPTIONS);
       LLVM_READOBJ_ENUM_CASE(ELF, PT_MIPS_ABIFLAGS);
     }
     break;
+  case ELF::EM_RISCV:
+    switch (Type) { LLVM_READOBJ_ENUM_CASE(ELF, PT_RISCV_ATTRIBUTES); }
   }
 
   switch (Type) {
     LLVM_READOBJ_ENUM_CASE(ELF, PT_NULL);
     LLVM_READOBJ_ENUM_CASE(ELF, PT_LOAD);
     LLVM_READOBJ_ENUM_CASE(ELF, PT_DYNAMIC);
     LLVM_READOBJ_ENUM_CASE(ELF, PT_INTERP);
     LLVM_READOBJ_ENUM_CASE(ELF, PT_NOTE);
static std::string getGNUPtType(unsigned Arch, unsigned Type) {
   // E.g. "PT_ARM_EXIDX" -> "EXIDX".
   if (Seg.startswith("PT_ARM_"))
     return Seg.drop_front(7).str();
 
   // E.g. "PT_MIPS_REGINFO" -> "REGINFO".
   if (Seg.startswith("PT_MIPS_"))
     return Seg.drop_front(8).str();
 
+  // E.g. "PT_RISCV_ATTRIBUTES"
+  if (Seg.consume_front("PT_RISCV_"))
+    return Seg.str();
+
   // E.g. "PT_LOAD" -> "LOAD".
   assert(Seg.startswith("PT_"));
   return Seg.drop_front(3).str();
 }
 
 const EnumEntry<unsigned> ElfSegmentFlags[] = {
   LLVM_READOBJ_ENUM_ENT(ELF, PF_X),
   LLVM_READOBJ_ENUM_ENT(ELF, PF_W),