Support debug_ranges lookup when matching address for inline functions.

Summary:
Range lists are contained in a separate object file section called .debug_ranges. A range list is
indicated by a DW_AT_ranges attribute whose value is represented as an offset from the
beginning of the .debug_ranges section to the beginning of the range list.
Each entry in a range list is either:
a range list entry,
a base address selection entry, or
an end of list entry.

Reviewed By: luciang

Differential Revision: D19658819

fbshipit-source-id: ed70915465f0ac6bb8ab6f7efe415a3d8c5b04ee

folly/experimental/symbolizer/Dwarf.cpp

@@ -344,9 +344,14 @@ Dwarf::Dwarf(const ElfFile* elf)
       debugAbbrev_(getSection(".debug_abbrev")),
       debugLine_(getSection(".debug_line")),
       debugStr_(getSection(".debug_str")),
-      debugAranges_(getSection(".debug_aranges")) {
-  // NOTE: debugAranges_ is for fast address range lookup.
-  // If missing .debug_info can be used - but it's much slower (linear scan).
+      debugAranges_(getSection(".debug_aranges")),
+      debugRanges_(getSection(".debug_ranges")) {
+  // Optional sections:
+  //  - debugAranges_: for fast address range lookup.
+  //     If missing .debug_info can be used - but it's much slower (linear
+  //     scan).
+  //  - debugRanges_: contains non-contiguous address ranges of debugging
+  //    information entries. Used for inline function address lookup.
   if (debugInfo_.empty() || debugAbbrev_.empty() || debugLine_.empty() ||
       debugStr_.empty()) {
     elf_ = nullptr;
@@ -457,6 +462,7 @@ bool Dwarf::findLocation(
     LocationInfo& locationInfo,
     folly::Range<SymbolizedFrame*> inlineFrames) const {
   detail::Die die = getDieAtOffset(cu, cu.firstDie);
+  // Partial compilation unit (DW_TAG_partial_unit) is not supported.
   FOLLY_SAFE_CHECK(
       die.abbr.tag == DW_TAG_compile_unit, "expecting compile unit entry");
 
@@ -734,6 +740,40 @@ folly::Optional<T> Dwarf::getAttribute(
   return result;
 }
 
+bool Dwarf::isAddrInRangeList(uint64_t address, size_t offset, uint8_t addrSize)
+    const {
+  FOLLY_SAFE_CHECK(addrSize == 4 || addrSize == 8, "wrong address size");
+  if (debugRanges_.empty()) {
+    return false;
+  }
+
+  const bool is64BitAddr = addrSize == 8;
+  folly::StringPiece sp = debugRanges_;
+  sp.advance(offset);
+  const uint64_t maxAddr = is64BitAddr ? std::numeric_limits<uint64_t>::max()
+                                       : std::numeric_limits<uint32_t>::max();
+  uint64_t baseAddr = 0;
+  while (!sp.empty()) {
+    uint64_t begin = readOffset(sp, is64BitAddr);
+    uint64_t end = readOffset(sp, is64BitAddr);
+    // The range list entry is a base address selection entry.
+    if (begin == maxAddr) {
+      baseAddr = end;
+      continue;
+    }
+    // The range list entry is an end of list entry.
+    if (begin == 0 && end == 0) {
+      break;
+    }
+    // Check if the given address falls in the range list entry.
+    if (address >= begin + baseAddr && address < end + baseAddr) {
+      return true;
+    }
+  };
+
+  return false;
+}
+
 void Dwarf::findSubProgramDieForAddress(
     const detail::CompilationUnit& cu,
     const detail::Die& die,
@@ -744,10 +784,12 @@ void Dwarf::findSubProgramDieForAddress(
       folly::Optional<uint64_t> lowPc;
       folly::Optional<uint64_t> highPc;
       folly::Optional<bool> isHighPcAddr;
+      folly::Optional<uint64_t> rangeOffset;
       forEachAttribute(cu, childDie, [&](const detail::Attribute& attr) {
         switch (attr.spec.name) {
-          // Here DW_AT_ranges is not supported since it requires looking up
-          // in a different section (.debug_ranges).
+          case DW_AT_ranges:
+            rangeOffset = boost::get<uint64_t>(attr.attrValue);
+            break;
           case DW_AT_low_pc:
             lowPc = boost::get<uint64_t>(attr.attrValue);
             break;
@@ -761,12 +803,10 @@ void Dwarf::findSubProgramDieForAddress(
         // Iterate through all attributes until find all above.
         return true;
       });
-      if (!lowPc || !highPc || !isHighPcAddr) {
-        // Missing required fields. Keep searching other children.
-        return true;
-      }
-      if (address >= *lowPc &&
-          address < (*isHighPcAddr ? *highPc : *lowPc + *highPc)) {
+      if ((lowPc && highPc && isHighPcAddr && address >= *lowPc &&
+           address < (*isHighPcAddr ? *highPc : *lowPc + *highPc)) ||
+          (rangeOffset &&
+           isAddrInRangeList(address, *rangeOffset, cu.addrSize))) {
         subprogram = childDie;
         return false;
       }
@@ -798,10 +838,12 @@ void Dwarf::findInlinedSubroutineDieForAddress(
     folly::Optional<uint64_t> origin;
     folly::Optional<uint64_t> originRefType;
     folly::Optional<uint64_t> callLine;
+    folly::Optional<uint64_t> rangeOffset;
     forEachAttribute(cu, childDie, [&](const detail::Attribute& attr) {
       switch (attr.spec.name) {
-        // Here DW_AT_ranges is not supported since it requires looking up
-        // in a different section (.debug_ranges).
+        case DW_AT_ranges:
+          rangeOffset = boost::get<uint64_t>(attr.attrValue);
+          break;
         case DW_AT_low_pc:
           lowPc = boost::get<uint64_t>(attr.attrValue);
           break;
@@ -822,13 +864,14 @@ void Dwarf::findInlinedSubroutineDieForAddress(
       // Iterate through all until find all above attributes.
       return true;
     });
-    if (!lowPc || !highPc || !isHighPcAddr || !origin || !originRefType ||
-        !callLine) {
+    if (!origin || !originRefType || !callLine) {
       // Missing required fields. Keep searching other children.
       return true;
     }
-    if (address < *lowPc ||
-        address >= (*isHighPcAddr ? *highPc : *lowPc + *highPc)) {
+    if ((!lowPc || !highPc || !isHighPcAddr || address < *lowPc ||
+         address >= (*isHighPcAddr ? *highPc : *lowPc + *highPc)) &&
+        (!rangeOffset ||
+         !isAddrInRangeList(address, rangeOffset.value(), cu.addrSize))) {
       // Address doesn't match. Keep searching other children.
       return true;
     }

folly/experimental/symbolizer/Dwarf.h

@@ -188,6 +188,12 @@ class Dwarf {
       const detail::CompilationUnit& cu,
       const detail::Die& die,
       uint64_t attrName) const;
+  /**
+   * Check if the given address is in the range list at the given offset in
+   * .debug_ranges.
+   */
+  bool isAddrInRangeList(uint64_t address, size_t offset, uint8_t addrSize)
+      const;
 
   const ElfFile* elf_;
   const folly::StringPiece debugInfo_; // .debug_info
@@ -195,6 +201,7 @@ class Dwarf {
   const folly::StringPiece debugLine_; // .debug_line
   const folly::StringPiece debugStr_; // .debug_str
   const folly::StringPiece debugAranges_; // .debug_aranges
+  const folly::StringPiece debugRanges_; // .debug_ranges
 };
 
 class Dwarf::Section {