DWARF] DWARF v5: .debug_rnglists: adding support for DW_RLE_offset_pair and DW_RLE_base_address

[wolfgangp]

Summary

Barring the usage of the .debug_addr section (DW_RLE_startx_endx etc), these 2 encodings were the only ones not supported yet., DW_RLE_offset_pair the only encoding usable in dwo files.

Note that in non-verbose printing, we don't print anything for DW_RLE_base_address. In verbose mode the entry is printed explicitly.

[aprantl]

lib/DebugInfo/DWARF/DWARFDebugRnglists.cpp

229

Just to be safe: Is this an internal consistency check or should this be an error handler instead?

[dblaikie]

lib/DebugInfo/DWARF/DWARFDebugRnglists.cpp

134

Some of our range dumping can now print which section an address refers to - should we do that for this range dumping? If so, we'd need to extract the section info here, I think?

213–233

Rather than grouping these under the same switch, then using a conditional - might it be worth having separate cases? The common verbose part could be factored into a local lambda, maybe?

[wolfgangp]

lib/DebugInfo/DWARF/DWARFDebugRnglists.cpp

229

It's a consistency check. I was debating whether the assert was necessary at all since we're in fairly tightly controlled context, but after Dave's suggestion it's a moot point.

[wolfgangp]

lib/DebugInfo/DWARF/DWARFDebugRnglists.cpp

134

Currently the section is displayed when we're dumping the specific ranges a DIE refers to (DW_AT_ranges attribute). When we're dumping the contents of the entire .debug_ranges section we don't do it. Do you think we should do it there too?

The full support for .debug_rnglists will turn the rangelists into DWARFAddressRangesVectors, so it will behave the same as the current implementation in this respect.

[wolfgangp]

Addressed review comment on separating DW_RLW_offset_pair and DW_RLE_start_length when dumping their contents.

[dblaikie]

Seems good to me

lib/DebugInfo/DWARF/DWARFDebugRnglists.cpp

while (*OffsetPtr < End) {
     case dwarf::DW_RLE_startx_endx:
       return createError("unsupported rnglists encoding DW_RLE_startx_endx at "
                          "offset 0x%" PRIx32,
                          *OffsetPtr - 1);
     case dwarf::DW_RLE_startx_length:
       return createError("unsupported rnglists encoding DW_RLE_startx_length "
                          "at offset 0x%" PRIx32,
                          *OffsetPtr - 1);
-    case dwarf::DW_RLE_offset_pair:
-      return createError("unsupported rnglists encoding DW_RLE_offset_pair at "
-                         "offset 0x%" PRIx32,
-                         *OffsetPtr - 1);
-    case dwarf::DW_RLE_base_address:
-      return createError("unsupported rnglists encoding DW_RLE_base_address at "
-                         "offset 0x%" PRIx32,
+    case dwarf::DW_RLE_offset_pair: {
+      uint32_t PreviousOffset = *OffsetPtr - 1;
+      uint64_t StartingOffset = Data.getULEB128(OffsetPtr);
+      uint64_t EndingOffset = Data.getULEB128(OffsetPtr);
+      if (End < *OffsetPtr)
+        return createError("read past end of table when reading "
+                           "DW_RLE_offset_pair encoding at offset 0x%" PRIx32,
+                           PreviousOffset);
+      CurrentRanges.push_back(
+          RangeListEntry{EntryOffset, Encoding, StartingOffset, EndingOffset});
+      break;
+    }
+    case dwarf::DW_RLE_base_address: {
+      if ((End - *OffsetPtr) < HeaderData.AddrSize)
+        return createError("insufficient space remaining in table for "
+                           "DW_RLE_base_address encoding at offset 0x%" PRIx32,
                            *OffsetPtr - 1);
+      uint64_t Base = Data.getAddress(OffsetPtr);
+      CurrentRanges.push_back(RangeListEntry{EntryOffset, Encoding, Base, 0});
+      break;
+    }
     case dwarf::DW_RLE_start_end: {
       if ((End - *OffsetPtr) < unsigned(HeaderData.AddrSize * 2))
         return createError("insufficient space remaining in table for "
                            "DW_RLE_start_end encoding "
                            "at offset 0x%" PRIx32,
                            *OffsetPtr - 1);
       uint64_t Start = Data.getAddress(OffsetPtr);
       uint64_t End = Data.getAddress(OffsetPtr);
return createError(
         "starting at offset 0x%" PRIx32,
         HeaderOffset);
   return Error::success();
 }
 
 static void dumpRangeEntry(raw_ostream &OS,
                            DWARFDebugRnglists::RangeListEntry Entry,
                            uint8_t AddrSize, uint8_t MaxEncodingStringLength,
-                           DIDumpOptions DumpOpts) {
+                           uint64_t &CurrentBase, DIDumpOptions DumpOpts) {
   if (DumpOpts.Verbose) {
     // Print the section offset in verbose mode.
     OS << format("0x%8.8" PRIx32 ":", Entry.Offset);
     auto EncodingString = dwarf::RangeListEncodingString(Entry.EntryKind);
     // Unsupported encodings should have been reported during parsing.
     assert(!EncodingString.empty() && "Unknown range entry encoding");
     OS << format(" [%s%*c", EncodingString.data(),
                  MaxEncodingStringLength - EncodingString.size() + 1, ']');
     if (Entry.EntryKind != dwarf::DW_RLE_end_of_list)
       OS << ": ";
   }
 
   switch (Entry.EntryKind) {
   case dwarf::DW_RLE_end_of_list:
     OS << (DumpOpts.Verbose ? "" : "<End of list>");
     break;
+  case dwarf::DW_RLE_base_address:
+    // In non-verbose mode we do not print anything for this entry.
+    CurrentBase = Entry.Value0;
+    if (!DumpOpts.Verbose)
+      return;
+    OS << format(" 0x%*.*" PRIx64, AddrSize * 2, AddrSize * 2, Entry.Value0);
+    break;
   case dwarf::DW_RLE_start_length:
+  case dwarf::DW_RLE_offset_pair: {
     if (DumpOpts.Verbose) {
       // Make the address range display its contents in raw form rather than
       // as an interval (i.e. without brackets).
       DumpOpts.DisplayRawContents = true;
       DWARFAddressRange(Entry.Value0, Entry.Value1)
           .dump(OS, AddrSize, DumpOpts);
       OS << " => ";
     }
     DumpOpts.DisplayRawContents = false;
-    DWARFAddressRange(Entry.Value0, Entry.Value0 + Entry.Value1)
-        .dump(OS, AddrSize, DumpOpts);
+    uint64_t Start = Entry.Value0;
+    uint64_t End = Entry.Value1;
+    if (Entry.EntryKind == dwarf::DW_RLE_start_length)
+      End += Entry.Value0;
+    else {
+      assert(Entry.EntryKind == dwarf::DW_RLE_offset_pair &&
+             "Expected EntryKind to be DW_RLE_offset_pair");
+      Start += CurrentBase;
+      End += CurrentBase;
+    }
+    DWARFAddressRange(Start, End).dump(OS, AddrSize, DumpOpts);
     break;
+  }
   case dwarf::DW_RLE_start_end:
     DWARFAddressRange(Entry.Value0, Entry.Value1).dump(OS, AddrSize, DumpOpts);
     break;
   default:
     llvm_unreachable("Unsupported range list encoding");
   }
   OS << "\n";
 }
for (const auto &Off : Offsets) {
       if (DumpOpts.Verbose)
         OS << format(" => 0x%8.8" PRIx32,
                      Off + HeaderOffset + sizeof(HeaderData));
     }
     OS << "\n]\n";
   }
   OS << "Ranges:\n";
 
+  uint64_t CurrentBase = 0;
   for (const auto &List : Ranges)
     for (const auto &Entry : List)
       dumpRangeEntry(OS, Entry, HeaderData.AddrSize, MaxEncodingStringLength,
-                     DumpOpts);
+                     CurrentBase, DumpOpts);
 }
 
 uint32_t DWARFDebugRnglists::length() const {
   if (HeaderData.Length == 0)
     return 0;
   // TODO: DWARF64 support.
   return HeaderData.Length + sizeof(uint32_t);
 }

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

 # TERSE-NEXT:   Range List Header: length = 0x00000008, version = 0x0005, addr_size = 0x08, seg_size = 0x00, offset_entry_count = 0x00000000
 
 # VERBOSE-NEXT: 0x{{[0-9a-f]*}}:
 # VERBOSE-SAME: Range List Header: length = 0x00000008, version = 0x0005, addr_size = 0x08, seg_size = 0x00, offset_entry_count = 0x00000000
 
 # BOTH-NOT:     Offsets:
 # BOTH:         Ranges:
 # BOTH-NOT:     [
+
+# TERSE-NEXT:   Range List Header: length = 0x0000000e, version = 0x0005, addr_size = 0x08, seg_size = 0x00, offset_entry_count = 0x00000000
+
+# VERBOSE-NEXT: 0x{{[0-9a-f]*}}:
+# VERBOSE-SAME: Range List Header: length = 0x0000000e, version = 0x0005, addr_size = 0x08, seg_size = 0x00, offset_entry_count = 0x00000000
+
+# BOTH-NEXT:    Ranges:
+# TERSE-NEXT:   [0x0000000000000800, 0x0000000000001000)
+# TERSE-NEXT:   <End of list>
+
+# VERBOSE-NEXT: 0x000000b1: [DW_RLE_offset_pair]:  0x0000000000000800, 0x0000000000001000 =>
+# VERBOSE-SAME:                                   [0x0000000000000800, 0x0000000000001000)
+# VERBOSE-NEXT: 0x000000b6: [DW_RLE_end_of_list]
+
+# TERSE-NEXT:   Range List Header: length = 0x00000017, version = 0x0005, addr_size = 0x08, seg_size = 0x00, offset_entry_count = 0x00000000
+
+# VERBOSE-NEXT: 0x{{[0-9a-f]*}}:
+# VERBOSE-SAME: Range List Header: length = 0x00000017, version = 0x0005, addr_size = 0x08, seg_size = 0x00, offset_entry_count = 0x00000000
+
+# BOTH-NEXT:    Ranges:
+# TERSE-NEXT:   [0x0000000000001800, 0x0000000000002000)
+# TERSE-NEXT:   <End of list>
+
+# VERBOSE-NEXT: 0x000000c3: [DW_RLE_base_address]:  0x0000000000001000
+# VERBOSE-NEXT: 0x000000cc: [DW_RLE_offset_pair ]:  0x0000000000000800, 0x0000000000001000 =>
+# VERBOSE-SAME:                                    [0x0000000000001800, 0x0000000000002000)
+# VERBOSE-NEXT: 0x000000d1: [DW_RLE_end_of_list ]
+
 # BOTH-NOT:     Range List Header:
 
 # ERR-NOT:  error:
 # ERR:      error: unsupported rnglists encoding DW_RLE_base_addressx at offset 0x82
 # ERR-NEXT: error: unsupported rnglists encoding DW_RLE_startx_endx at offset 0x91
 # ERR-NEXT: error: unsupported rnglists encoding DW_RLE_startx_length at offset 0xa1
-# ERR-NEXT: error: unsupported rnglists encoding DW_RLE_offset_pair at offset 0xb1
-# ERR-NEXT: error: unsupported rnglists encoding DW_RLE_base_address at offset 0xc
 # ERR-NOT:  error:
 
 .section .debug_rnglists,"",@progbits
 
 # First table (tests DW_RLE_end_of_list, start_end, and start_length encodings)
 .long 55 # Table length
 .short 5 # Version
 .byte 8  # Address size

 
 # First range list
 .byte 3          # DW_RLE_startx_length
 .byte 2            # Start address (index in .debug_addr)
 .byte 42           # Length
 .byte 0          # DW_RLE_end_of_list
 
 # Seventh table (testing DW_RLE_offset_pair)
-.long 12 # Table length
+.long 14 # Table length
 .short 5 # Version
 .byte 8  # Address size
 .byte 0  # Segment selector size
 .long 0  # Offset entry count
 
 # First range list
 .byte 4          # DW_RLE_offset_pair
-.byte 3            # Start offset (index in .debug_addr)
-.byte 19           # End offset (index in .debug_addr)
+.byte 0x80, 0x10   # Start offset
+.byte 0x80, 0x20   # End offset (index in .debug_addr)
 .byte 0          # DW_RLE_end_of_list
 
-# Eigth table (testing DW_RLE_base_address)
-.long 18 # Table length
+# Eigth table (testing DW_RLE_base_address and its impact
+# on DW_RLE_offset_pair)
+.long 23 # Table length
 .short 5 # Version
 .byte 8  # Address size
 .byte 0  # Segment selector size
 .long 0  # Offset entry count
 
 # First range list
 .byte 5          # DW_RLE_base_address
 .quad 0x1000       # Base address
+.byte 4          # DW_RLE_offset_pair
+.byte 0x80, 0x10 # Start offset
+.byte 0x80, 0x20 # End offset
 .byte 0          # DW_RLE_end_of_list