Orderfile: Use linker start of text for offset base.

The instrumentation code outputs offsets relative to the start of text
as defined by a special symbol (linker_script_start_of_text). This
currently corresponds to the first symbol appearing in text, which
also happens to be the first SymbolInfo produced by
symbol_extractor. To protect against changes to the symbol_extractor
this change uses the special symbol for the dump offset base rather
than the lowest SymbolInfo offset seen.

Bug: 893981
Change-Id: Ia3d265ae56e6a3b7362abf177c9cbe8d2eda36fb
Reviewed-on: https://chromium-review.googlesource.com/c/1325989Reviewed-by: Egor Pasko <pasko@chromium.org>
Commit-Queue: Matthew Cary <mattcary@chromium.org>
Cr-Commit-Position: refs/heads/master@{#606454}

tools/cygprofile/process_profiles.py

@@ -49,6 +49,7 @@ class SymbolOffsetProcessor(object):
   higher-level operations can be done in different orders without the caller
   managing all the state.
   """
+  START_OF_TEXT_SYMBOL = 'linker_script_start_of_text'
 
   def __init__(self, binary_filename):
     self._binary_filename = binary_filename
@@ -147,7 +148,8 @@ class SymbolOffsetProcessor(object):
         symbols.append(self.OffsetToPrimaryMap()[o - 1].name)
       else:
         not_found += 1
-    logging.warning('%d offsets do not have matching symbol', not_found)
+    if not_found:
+      logging.warning('%d offsets do not have matching symbol', not_found)
     return symbols
 
   def OffsetsPrimarySize(self, offsets):
@@ -234,7 +236,7 @@ class SymbolOffsetProcessor(object):
     for i in items:
       dump_offset = get(i)
       idx = dump_offset / 4
-      assert idx < len(dump_offset_to_symbol_info), (
+      assert dump_offset >= 0 and idx < len(dump_offset_to_symbol_info), (
           'Dump offset out of binary range')
       symbol_info = dump_offset_to_symbol_info[idx]
       if symbol_info is None:
@@ -254,12 +256,17 @@ class SymbolOffsetProcessor(object):
         section, maps it to a symbol, or None.
     """
     if self._offset_to_symbol_info is None:
-      min_offset = min(s.offset for s in self.SymbolInfos())
+      start_syms = [s for s in self.SymbolInfos()
+                    if s.name == self.START_OF_TEXT_SYMBOL]
+      assert len(start_syms) == 1, 'Can\'t find unique start of text symbol'
+      start_of_text = start_syms[0].offset
       max_offset = max(s.offset + s.size for s in self.SymbolInfos())
-      text_length_words = (max_offset - min_offset) / 4
+      text_length_words = (max_offset - start_of_text) / 4
       self._offset_to_symbol_info = [None for _ in xrange(text_length_words)]
       for s in self.SymbolInfos():
-        offset = s.offset - min_offset
+        offset = s.offset - start_of_text
+        if offset < 0:
+          continue
         for i in range(offset / 4, (offset + s.size) / 4):
           self._offset_to_symbol_info[i] = s
     return self._offset_to_symbol_info

tools/cygprofile/process_profiles_unittest.py

@@ -16,9 +16,10 @@ from test_utils import (ProfileFile,
                         TestProfileManager)
 
 class ProcessProfilesTestCase(unittest.TestCase):
+  START_SYMBOL = process_profiles.SymbolOffsetProcessor.START_OF_TEXT_SYMBOL
 
   def setUp(self):
-    self.symbol_0 = SimpleTestSymbol('0', 0, 0)
+    self.symbol_0 = SimpleTestSymbol(self.START_SYMBOL, 0, 0)
     self.symbol_1 = SimpleTestSymbol('1', 8, 16)
     self.symbol_2 = SimpleTestSymbol('2', 32, 8)
     self.symbol_3 = SimpleTestSymbol('3', 40, 12)
@@ -39,6 +40,18 @@ class ProcessProfilesTestCase(unittest.TestCase):
     offset_to_symbol_info = processor._GetDumpOffsetToSymbolInfo()
     self.assertListEqual(self.offset_to_symbol_info, offset_to_symbol_info)
 
+  def testSymbolsBeforeStart(self):
+    self.symbol_infos = [SimpleTestSymbol(s.name, s.offset + 8, s.size)
+                         for s in self.symbol_infos]
+    self.symbol_infos.append(SimpleTestSymbol('early', 0, 4))
+    processor = TestSymbolOffsetProcessor(self.symbol_infos)
+    offset_to_symbol_info = processor._GetDumpOffsetToSymbolInfo()
+    # The 'early' symbol should be omitted.
+    self.assertEqual(([None, None] + [self.symbol_infos[1]] * 4 +
+                      [None] * 2 + [self.symbol_infos[2]] * 2 +
+                      [self.symbol_infos[3]] * 3),
+                     offset_to_symbol_info)
+
   def testGetReachedOffsetsFromDump(self):
     processor = TestSymbolOffsetProcessor(self.symbol_infos)
     # 2 hits for symbol_1, 0 for symbol_2, 1 for symbol_3
@@ -60,7 +73,7 @@ class ProcessProfilesTestCase(unittest.TestCase):
 
   def testGetOrderedSymbols(self):
     processor = TestSymbolOffsetProcessor(self.symbol_infos)
-    self.assertListEqual(['1', '3', '0'],
+    self.assertListEqual(['1', '3', self.START_SYMBOL],
                          processor.GetOrderedSymbols([8, 41, 6, 0]))
 
   def testOffsetToSymbolsMap(self):