Merge pull request #120 from compnerd/die-must-die

Relax DIE verification

Author: adrian-prantl

include/llvm/DebugInfo/DWARF/DWARFVerifier.h

@@ -97,6 +97,9 @@ class DWARFVerifier {
   /// lies between to valid DIEs.
   std::map<uint64_t, std::set<uint32_t>> ReferenceToDIEOffsets;
   uint32_t NumDebugLineErrors = 0;
+  // Used to relax some checks that do not currently work portably
+  bool IsObjectFile;
+  bool IsMachOObject;
 
   raw_ostream &error() const;
   raw_ostream &warn() const;
@@ -286,8 +289,8 @@ class DWARFVerifier {
 
 public:
   DWARFVerifier(raw_ostream &S, DWARFContext &D,
-                DIDumpOptions DumpOpts = DIDumpOptions::getForSingleDIE())
-      : OS(S), DCtx(D), DumpOpts(std::move(DumpOpts)) {}
+                DIDumpOptions DumpOpts = DIDumpOptions::getForSingleDIE());
+
   /// Verify the information in any of the following sections, if available:
   /// .debug_abbrev, debug_abbrev.dwo
   ///

lib/DebugInfo/DWARF/DWARFVerifier.cpp

@@ -379,20 +379,42 @@ unsigned DWARFVerifier::verifyDieRanges(const DWARFDie &Die,
   // Build RI for this DIE and check that ranges within this DIE do not
   // overlap.
   DieRangeInfo RI(Die);
-  for (auto Range : Ranges) {
-    if (!Range.valid()) {
-      ++NumErrors;
-      error() << "Invalid address range " << Range << "\n";
-      continue;
-    }
 
-    // Verify that ranges don't intersect.
-    const auto IntersectingRange = RI.insert(Range);
-    if (IntersectingRange != RI.Ranges.end()) {
-      ++NumErrors;
-      error() << "DIE has overlapping address ranges: " << Range << " and "
-              << *IntersectingRange << "\n";
-      break;
+  // TODO support object files better
+  //
+  // Some object file formats (i.e. non-MachO) support COMDAT.  ELF in
+  // particular does so by placing each function into a section.  The DWARF data
+  // for the function at that point uses a section relative DW_FORM_addrp for
+  // the DW_AT_low_pc and a DW_FORM_data4 for the offset as the DW_AT_high_pc.
+  // In such a case, when the Die is the CU, the ranges will overlap, and we
+  // will flag valid conflicting ranges as invalid.
+  //
+  // For such targets, we should read the ranges from the CU and partition them
+  // by the section id.  The ranges within a particular section should be
+  // disjoint, although the ranges across sections may overlap.  We would map
+  // the child die to the entity that it references and the section with which
+  // it is associated.  The child would then be checked against the range
+  // information for the associated section.
+  //
+  // For now, simply elide the range verification for the CU DIEs if we are
+  // processing an object file.
+
+  if (!IsObjectFile || IsMachOObject || Die.getTag() == DW_TAG_subprogram) {
+    for (auto Range : Ranges) {
+      if (!Range.valid()) {
+        ++NumErrors;
+        error() << "Invalid address range " << Range << "\n";
+        continue;
+      }
+
+      // Verify that ranges don't intersect.
+      const auto IntersectingRange = RI.insert(Range);
+      if (IntersectingRange != RI.Ranges.end()) {
+        ++NumErrors;
+        error() << "DIE has overlapping address ranges: " << Range << " and "
+                << *IntersectingRange << "\n";
+        break;
+      }
     }
   }
 
@@ -721,6 +743,15 @@ void DWARFVerifier::verifyDebugLineRows() {
   }
 }
 
+DWARFVerifier::DWARFVerifier(raw_ostream &S, DWARFContext &D,
+                             DIDumpOptions DumpOpts)
+    : OS(S), DCtx(D), DumpOpts(std::move(DumpOpts)) {
+  if (const auto *F = DCtx.getDWARFObj().getFile()) {
+    IsObjectFile = F->isRelocatableObject();
+    IsMachOObject = F->isMachO();
+  }
+}
+
 bool DWARFVerifier::handleDebugLine() {
   NumDebugLineErrors = 0;
   OS << "Verifying .debug_line...\n";

test/tools/llvm-dwarfdump/X86/debug-verify-object.s

@@ -0,0 +1,57 @@
+# RUN: llvm-mc -triple x86_64-unknown-linux-gnu -filetype obj -o - %s | llvm-dwarfdump --verify -
+
+	.text
+
+	.section	.text.f,"ax",@progbits
+	.globl	f
+	.type	f,@function
+f:
+.Lfunc_begin0:
+	pushq	$32
+	popq	%rax
+	retq
+.Lfunc_end0:
+	.size	f, .Lfunc_end0-f
+
+	.section	.text.g,"ax",@progbits
+	.globl	g
+	.type	g,@function
+g:
+.Lfunc_begin1:
+	pushq   $64
+	popq    %rax
+	retq
+.Lfunc_end1:
+	.size	g, .Lfunc_end1-g
+
+	.section	.debug_abbrev,"",@progbits
+	.byte	1                       # Abbreviation Code
+	.byte	17                      # DW_TAG_compile_unit
+	.byte	0                       # DW_CHILDREN_no
+	.byte	17                      # DW_AT_low_pc
+	.byte	1                       # DW_FORM_addr
+	.byte	85                      # DW_AT_ranges
+	.byte	23                      # DW_FORM_sec_offset
+	.byte	0                       # EOM(1)
+	.byte	0                       # EOM(2)
+	.byte	0                       # EOM(3)
+
+	.section	.debug_info,"",@progbits
+.Lcu_begin0:
+	.long	20                      # Length of Unit
+	.short	4                       # DWARF version number
+	.long	.debug_abbrev           # Offset Into Abbrev. Section
+	.byte	8                       # Address Size (in bytes)
+	.byte	1                       # Abbrev [1] 0xb:0x1f DW_TAG_compile_unit
+	.quad	0                       # DW_AT_low_pc
+	.long	.Ldebug_ranges0         # DW_AT_ranges
+
+	.section        .debug_ranges,"",@progbits
+.Ldebug_ranges0:
+	.quad	.Lfunc_begin0
+	.quad	.Lfunc_end0
+	.quad	.Lfunc_begin1
+	.quad	.Lfunc_end1
+	.quad	0
+	.quad	0
+