Home Original page

a patch for coff_find_nearest_line()

Jaka Močnik jaka@xlab.si
Wed Aug 13 14:48:00 GMT 2008 
hello!

there is a bug in coff_find_nearest_line() (binutils 2.18), that
sometimes returns no line number information when this info in fact
exists ... it can be reproduced by using addr2line tool on a coff file
(any coff will do, I believe ...) to look up the same address twice in a
row.

the bug is due to faulty priming of the look-up loop, which works like
this:

when an appropriate line number info for address A is found at index I
in the section's lineno array, this information is returned (via the
line_ptr out parameter to coff_find_nearest_line()), and the lookup loop
breaks with the loop counter having the value I+1. this value I+1 is
then remembered as the starting point for further lookups of addresses
>= A.

now consider the following lineno array entries:

...
lineno[i]: offset X
lineno[i+1]: offset X+10
...

looking up line number for address X+2, we end up with the info from
lineno[i]. furthermore, i+1 is remembered as the starting index for
priming further lookups.

now try to lookup line number for address X+2 once again (or any address
A, X+2 <= A < X+10). since the address is larger or equal than the
previously looked up one, the lookup loop starts at index i+1.
naturally, no info for any address < X+10 is found, and
coff_find_nearest_line() returns wrong information.

the attached simple patch (against binutils 2.18) remedies the problem.

regards,
  jaKa

-- 
email: jaka@xlab.si
w3:    http://www.gmajna.net/svojat/jaka/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: binutils-2.18-coff-find-nearest-line-priming.diff
Type: text/x-patch
Size: 373 bytes
Desc: not available
URL: <https://sourceware.org/pipermail/binutils/attachments/20080813/5e8afb07/attachment.bin>
More information about the Binutils mailing list