Fix symbolizer incorrectly extracting symbol sizes

Currently the symbolizer script assumes that all symbols belong to the same section when calculating symbol sizes, which is wrong. Handle all sections separately. Rely on readelf instead of nm.
google · Nov 29, 2022 · 054a4f0 · 054a4f0
1 parent 3113208
commit 054a4f0
Showing 1 changed file with 42 additions and 19 deletions.
diff --git a/tools/symbolizer.py b/tools/symbolizer.py
@@ -16,6 +16,9 @@
     '^(?P<time>\[ *[TC0-9\.]+\]) ?(?P<body>.*)$'
 )
 
+# A decimal number.
+DECNUM_RE = '[0-9]+'
+
 # A hexadecimal number without the leading 0x.
 HEXNUM_RE = '[0-9A-Fa-f]+'
 
@@ -72,13 +75,19 @@
     '$'
 )
 
-# Matches a single line of `nm -S` output.
-NM_RE = re.compile(
-    '^(?P<offset>' + HEXNUM_RE + ')( (?P<size>' + HEXNUM_RE + '))?' +
-    ' [a-zA-Z] (?P<symbol>[^ ]+)$'
+# Matches a single relevant line of `readelf -Ws` output.
+READELF_RE = re.compile(
+    '^[ ]*' +
+    '(?P<num>' + DECNUM_RE + '):[ ]+' +
+    '(?P<offset>' + HEXNUM_RE + ')[ ]+' +
+    '(?P<size>' + DECNUM_RE + ')[ ]+' +
+    '(?P<type>(OBJECT|FUNC|NOTYPE))[ ]+' +
+    '(?P<bind>(LOCAL|GLOBAL))[ ]+' +
+    '(?P<vis>DEFAULT)[ ]+' +
+    '(?P<section>' + DECNUM_RE + ')[ ]+' +
+    '(?P<symbol>[^ ]+)$'
 )
 
-
 class Symbolizer(object):
     def __init__(self, binary_path):
         self.proc = subprocess.Popen(
@@ -139,23 +148,37 @@ class SymbolOffsetTable(object):
     this symbol's offset.
     """
     def __init__(self, binary_path):
-        output = subprocess.check_output(['nm', '-Sn', binary_path]).decode('ascii')
-        self.offsets = defaultdict(dict)
-        prev_symbol = None
-        prev_offset, prev_size = 0, 0
+        output = subprocess.check_output(['readelf', '-Ws', binary_path]).decode('ascii')
+
+        # Extract symbols for each section.
+        sections = defaultdict(dict)
         for line in output.split('\n'):
-            match = NM_RE.match(line)
-            if match != None:
-                offset = int(match.group('offset'), 16)
-                size = 0 if not match.group('size') else int(match.group('size'), 16)
+            match = READELF_RE.match(line)
+            if match == None:
+                continue
+            symbol = match.group('symbol')
+            # Ignore init_module and cleanup_module, as those are aliases.
+            if symbol == 'init_module' or symbol == 'cleanup_module':
+                continue
+            offset = int(match.group('offset'), 16)
+            size = match.group('size')
+            section = match.group('section')
+            sections[section][offset] = (symbol, size)
+
+        # Calculate and store sizes for each symbol.
+        self.offsets = defaultdict(dict)
+        for section in sections.values():
+            prev_symbol = None
+            prev_offset = None
+            prev_size = None
+            for offset in sorted(section.keys()):
                 if prev_symbol:
-                    ksyms_size = offset - prev_offset
-                    if ksyms_size >= prev_size:
-                        prev_size = ksyms_size
+                    prev_size = offset - prev_offset
                     self.offsets[prev_symbol][prev_size] = prev_offset
-                prev_symbol = match.group('symbol')
-                prev_offset, prev_size = offset, size
-        self.offsets[prev_symbol][0] = prev_offset
+                (symbol, _) = section[offset]
+                prev_symbol, prev_offset = symbol, offset
+            (_, size) = section[prev_offset]
+            self.offsets[prev_symbol][size] = prev_offset
 
     def lookup_offset(self, symbol, size):
         offsets = self.offsets.get(symbol)