[DebugInfo] Fix reading addresses in DWARFDebugAddr.

[ikudrin]

Summary

As addresses in the address tables may have relocations, thus, the relocations should be resolved to read the correct address. That is especially important for targets that use RELA relocations because in that case addends are stored in relocation sections.

Thanks, @jhenderson for spotting this!

[jhenderson]

llvm/test/tools/llvm-dwarfdump/X86/debug_addr_rela.s

1–2

I'm not sure I entirely follow why you go through the dance you are doing here? Why not just write the YAML directly?

[dblaikie]

llvm/test/tools/llvm-dwarfdump/X86/debug_addr_rela.s

1–2

I think the asm/yaml checking isn't needed here (I guess it's an attempt to demonstrate that the values in the section are all zeros, and only by applying the relocation do we get the addend applied - but I don't think the test needs to demonstrate that the bytes are all zeros - the test should just verify that the address printed by dwarfdump includes the addend, however it got there)

So please remove the YAML testing & keep the dwarfdump testing.

[jhenderson]

llvm/test/tools/llvm-dwarfdump/X86/debug_addr_rela.s

1–2

FWIW, it _might_ be better to use YAML as input directly, rather than needing to write assembly, but I'm happy to go with whichever is easier to read.

[dblaikie]

llvm/test/tools/llvm-dwarfdump/X86/debug_addr_rela.s

1–2

Ah, sure - yeah, I'm not strongly invested in assembly versus YAML. Dealer's choice.

[ikudrin]

• Removed the YAML part from the test.

Yes, initially, I added it to demonstrate that there is no real address encoded in the raw data of .debug_addr. I feel like I heard about an architecture where addends sometimes were encoded both in relocations and in the relocation target.

[dblaikie]

In D74404#1869936, @ikudrin wrote:

• Removed the YAML part from the test.

Yes, initially, I added it to demonstrate that there is no real address encoded in the raw data of .debug_addr. I feel like I heard about an architecture where addends sometimes were encoded both in relocations and in the relocation target.

Yeah, I believe some of those exist - so it's certainly worth it to have looked into that to validate your understanding, but just doesn't make as much sense as a permanent test. (the other aspect: the dump test as-is fails before your change & passes after, so that should be sufficient)

llvm/lib/DebugInfo/DWARF/DWARFDebugAddr.cpp

return createStringError(errc::invalid_argument,
                        ".debug_addr table at offset 0x%" PRIx64
                        " contains data of size %" PRIu32
                        " which is not a multiple of addr size %" PRIu8,
                        HeaderOffset, DataSize, HeaderData.AddrSize);
   }
   Data.setAddressSize(HeaderData.AddrSize);
   uint32_t AddrCount = DataSize / HeaderData.AddrSize;
   for (uint32_t I = 0; I < AddrCount; ++I)
-    if (HeaderData.AddrSize == 4)
-      Addrs.push_back(Data.getU32(OffsetPtr));
-    else
-      Addrs.push_back(Data.getU64(OffsetPtr));
+    Addrs.push_back(Data.getRelocatedAddress(OffsetPtr));
   return Error::success();
 }
 
 void DWARFDebugAddrTable::dump(raw_ostream &OS, DIDumpOptions DumpOpts) const {
   if (DumpOpts.Verbose)
     OS << format("0x%8.8" PRIx32 ": ", HeaderOffset);
   OS << format("Addr Section: length = 0x%8.8" PRIx32
                ", version = 0x%4.4" PRIx16 ", "

llvm/test/tools/llvm-dwarfdump/X86/debug_addr_rela.s

+# RUN: llvm-mc %s -filetype obj -triple x86_64-pc-linux -o %t.o
+# RUN: obj2yaml %t.o | FileCheck %s --check-prefix=YAML
+# RUN: llvm-dwarfdump -debug-addr %t.o | FileCheck %s
+
+## This checks that we use DWARFDataExtractor::getRelocatedAddress() to read
+## addresses of an address table. The raw data in the .debug_addr section does
+## not contain the full address, thus, it is required to resolve a RELA
+## relocation to recover the real value.
+
+# YAML:      - Name:            .debug_addr
+# YAML-NEXT:   Type:
+# YAML-NEXT:   AddressAlign:
+# YAML-NEXT:   Content:         0C000000050008000000000000000000
+# YAML:      - Name:            .rela.debug_addr
+# YAML-NEXT:   Type:            SHT_RELA
+# YAML-NEXT:   Link:
+# YAML-NEXT:   AddressAlign:
+# YAML-NEXT:   EntSize:
+# YAML-NEXT:   Info:            .debug_addr
+# YAML-NEXT:   Relocations:
+# YAML-NEXT:     - Offset:          0x0000000000000008
+# YAML-NEXT:       Symbol:          .text
+# YAML-NEXT:       Type:            R_X86_64_64
+# YAML-NEXT:       Addend:          42
+
+# CHECK:      .debug_addr contents
+# CHECK-NEXT: length = 0x0000000c, version = 0x0005, addr_size = 0x08, seg_size = 0x00
+# CHECK-NEXT: Addrs: [
+# CHECK-NEXT: 0x000000000000002a
+# CHECK-NEXT: ]
+
+    .text
+    .space  0x2a
+.Lfoo:
+
+    .section .debug_addr,"",@progbits
+    .long   .LAddr0end-.LAddr0version   # Length
+.LAddr0version:
+    .short  5                           # Version
+    .byte   8                           # Address size
+    .byte   0                           # Segment selector size
+.LAddr0table:
+    .quad   .Lfoo
+.LAddr0end: