This is the mail archive of the mailing list for the binutils project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Commit: Use reloc_at to detect partial debug_line sections

Hi Guys,

  I am checking in the patch below to fix the heuristic in
  read_debug_line_header that detect partial .debug_line sections.
  Previously it used to check for a negative length, but this is not
  guaranteed to happen.  (It depends upon the relocs used to compute the
  correct values for the debug_line header).  Instead we can now use the
  reloc_at function to detect that a reloc is needed to compute the
  correct header length, and take the appropriate steps.

  Tested with no regressions on lots of toolchains.


2015-06-05  Nick Clifton  <>

	* dwarf.c (read_debug_line_header): Use reloc_at to detect
	incomplete .debug_line headers

diff --git a/binutils/dwarf.c b/binutils/dwarf.c
index 71b0e97..c3dc814 100644
--- a/binutils/dwarf.c
+++ b/binutils/dwarf.c
@@ -2761,18 +2761,20 @@ read_debug_line_header (struct dwarf_section * section,
   if (linfo->li_length + initial_length_size > section->size)
-      /* If the length is just a bias against the initial_length_size then
-	 this means that the field has a relocation against it which has not
-	 been applied.  (Ie we are dealing with an object file, not a linked
-	 binary).  Do not complain but instead assume that the rest of the
-	 section applies to this particular header.  */
-      if (linfo->li_length == - initial_length_size)
+      /* If the length field has a relocation against it, then we should
+	 not complain if it is inaccurate (and probably negative).  This
+	 happens in object files when the .debug_line section is actually
+	 comprised of several different .debug_line.* sections, (some of
+	 which may be removed by linker garbage collection), and a relocation
+	 is used to compute the correct length once that is done.  */
+      if (reloc_at (section, (hdrptr - section->start) - offset_size))
-	  linfo->li_length = section->size - initial_length_size;
+	  linfo->li_length = (end - data) - initial_length_size;
-	  warn (_("The line info appears to be corrupt - the section is too small\n"));
+	  warn (_("The length field (0x%lx) in the debug_line header is wrong - the section is too small\n"),
+		(long) linfo->li_length);
 	  return NULL;

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]