1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/tools/rb/fix-linux-stack.pl Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,259 @@
1.4 +#!/usr/bin/perl
1.5 +# vim:sw=4:ts=4:et:
1.6 +# This Source Code Form is subject to the terms of the Mozilla Public
1.7 +# License, v. 2.0. If a copy of the MPL was not distributed with this
1.8 +# file, You can obtain one at http://mozilla.org/MPL/2.0/.
1.9 +
1.10 +# $Id: fix-linux-stack.pl,v 1.16 2008/05/05 21:51:11 dbaron%dbaron.org Exp $
1.11 +#
1.12 +# This script uses addr2line (part of binutils) to process the output of
1.13 +# nsTraceRefcnt's Linux stack walking code. This is useful for two
1.14 +# things:
1.15 +# (1) Getting line number information out of
1.16 +# |nsTraceRefcnt::WalkTheStack|'s output in debug builds.
1.17 +# (2) Getting function names out of |nsTraceRefcnt::WalkTheStack|'s
1.18 +# output on optimized builds (where it mostly prints UNKNOWN
1.19 +# because only a handful of symbols are exported from component
1.20 +# libraries).
1.21 +#
1.22 +# Use the script by piping output containing stacks (such as raw stacks
1.23 +# or make-tree.pl balance trees) through this script.
1.24 +
1.25 +use strict;
1.26 +use IPC::Open2;
1.27 +use File::Basename;
1.28 +
1.29 +# XXX Hard-coded to gdb defaults (works on Fedora).
1.30 +my $global_debug_dir = '/usr/lib/debug';
1.31 +
1.32 +# We record several things for each file encountered.
1.33 +#
1.34 +# - {pipe_read}, {pipe_write}: these constitute a bidirectional pipe to an
1.35 +# addr2line process that gives symbol information for a file.
1.36 +#
1.37 +# - {cache}: this table holds the results of lookups that we've done
1.38 +# previously for (pre-adjustment) addresses, which lets us avoid redundant
1.39 +# calls to addr2line.
1.40 +#
1.41 +# - {address_adjustment}: addr2line wants offsets relative to the base address
1.42 +# for shared libraries, but it wants addresses including the base address
1.43 +# offset for executables. This holds the appropriate address adjustment to
1.44 +# add to an offset within file. See bug 230336.
1.45 +#
1.46 +my %file_infos;
1.47 +
1.48 +sub set_address_adjustment($$) {
1.49 + my ($file, $file_info) = @_;
1.50 +
1.51 + # find out if it's an executable (as opposed to a shared library)
1.52 + my $elftype;
1.53 + open(ELFHDR, '-|', 'readelf', '-h', $file);
1.54 + while (<ELFHDR>) {
1.55 + if (/^\s*Type:\s+(\S+)/) {
1.56 + $elftype = $1;
1.57 + last;
1.58 + }
1.59 + }
1.60 + close(ELFHDR);
1.61 +
1.62 + # If it's an executable, make adjustment the base address.
1.63 + # Otherwise, leave it zero.
1.64 + my $adjustment = 0;
1.65 + if ($elftype eq 'EXEC') {
1.66 + open(ELFSECS, '-|', 'readelf', '-S', $file);
1.67 + while (<ELFSECS>) {
1.68 + if (/^\s*\[\s*\d+\]\s+\.text\s+\w+\s+(\w+)\s+(\w+)\s+/) {
1.69 + # Subtract the .text section's offset within the
1.70 + # file from its base address.
1.71 + $adjustment = hex($1) - hex($2);
1.72 + last;
1.73 + }
1.74 + }
1.75 + close(ELFSECS);
1.76 + }
1.77 +
1.78 + $file_info->{address_adjustment} = $adjustment;
1.79 +}
1.80 +
1.81 +# Files sometimes contain a link to a separate object file that contains
1.82 +# the debug sections of the binary, removed so that a smaller file can
1.83 +# be shipped, but kept separately so that it can be obtained by those
1.84 +# who want it.
1.85 +# See http://sources.redhat.com/gdb/current/onlinedocs/gdb_16.html#SEC154
1.86 +# for documentation of debugging information in separate files.
1.87 +# On Fedora distributions, these files can be obtained by installing
1.88 +# *-debuginfo RPM packages.
1.89 +sub separate_debug_file_for($) {
1.90 + my ($file) = @_;
1.91 + # We can read the .gnu_debuglink section using either of:
1.92 + # objdump -s --section=.gnu_debuglink $file
1.93 + # readelf -x .gnu_debuglink $file
1.94 + # Since readelf prints things backwards on little-endian platforms
1.95 + # for some versions only (backwards on Fedora Core 6, forwards on
1.96 + # Fedora 7), use objdump.
1.97 +
1.98 + # See if there's a .gnu_debuglink section
1.99 + my $have_debuglink = 0;
1.100 + open(ELFSECS, '-|', 'readelf', '-S', $file);
1.101 + while (<ELFSECS>) {
1.102 + if (/^\s*\[\s*\d+\]\s+\.gnu_debuglink\s+\w+\s+(\w+)\s+(\w+)\s+/) {
1.103 + $have_debuglink = 1;
1.104 + last;
1.105 + }
1.106 + }
1.107 + close(ELFSECS);
1.108 + return '' unless ($have_debuglink);
1.109 +
1.110 + # Determine the endianness of the shared library.
1.111 + my $endian = '';
1.112 + open(ELFHDR, '-|', 'readelf', '-h', $file);
1.113 + while (<ELFHDR>) {
1.114 + if (/^\s*Data:\s+.*(little|big) endian.*$/) {
1.115 + $endian = $1;
1.116 + last;
1.117 + }
1.118 + }
1.119 + close(ELFHDR);
1.120 + if ($endian ne 'little' && $endian ne 'big') {
1.121 + print STDERR "Warning: could not determine endianness of $file.\n";
1.122 + return '';
1.123 + }
1.124 +
1.125 +
1.126 + # Read the debuglink section as an array of words, in hexidecimal.
1.127 + open(DEBUGLINK, '-|', 'objdump', '-s', '--section=.gnu_debuglink', $file);
1.128 + my @words;
1.129 + while (<DEBUGLINK>) {
1.130 + if ($_ =~ /^ [0-9a-f]* ([0-9a-f ]{8}) ([0-9a-f ]{8}) ([0-9a-f ]{8}) ([0-9a-f ]{8}).*/) {
1.131 + push @words, $1, $2, $3, $4;
1.132 + }
1.133 + }
1.134 + close(DEBUGLINK);
1.135 +
1.136 + while (@words[$#words] eq ' ') {
1.137 + pop @words;
1.138 + }
1.139 +
1.140 + if ($#words < 1) {
1.141 + print STDERR "Warning: .gnu_debuglink section in $file too short.\n";
1.142 + return '';
1.143 + }
1.144 +
1.145 + my @chars;
1.146 + while ($#words >= 0) {
1.147 + my $w = shift @words;
1.148 + if ($w =~ /^([0-9a-f]{2})([0-9a-f]{2})([0-9a-f]{2})([0-9a-f]{2})$/) {
1.149 + push @chars, $1, $2, $3, $4;
1.150 + } else {
1.151 + print STDERR "Warning: malformed objdump output for $file.\n";
1.152 + return '';
1.153 + }
1.154 + }
1.155 +
1.156 + my @hash_bytes = map(hex, @chars[$#chars - 3 .. $#chars]);
1.157 + $#chars -= 4;
1.158 +
1.159 + my $hash;
1.160 + if ($endian eq 'little') {
1.161 + $hash = ($hash_bytes[3] << 24) | ($hash_bytes[2] << 16) | ($hash_bytes[1] << 8) | $hash_bytes[0];
1.162 + } else {
1.163 + $hash = ($hash_bytes[0] << 24) | ($hash_bytes[1] << 16) | ($hash_bytes[2] << 8) | $hash_bytes[3];
1.164 + }
1.165 +
1.166 + # The string ends with a null-terminator and then 0 to three bytes
1.167 + # of padding to fill the current 32-bit unit. (This padding is
1.168 + # usually null bytes, but I've seen null-null-H, on Ubuntu x86_64.)
1.169 + my $terminator = 1;
1.170 + while ($chars[$terminator] ne '00') {
1.171 + if ($terminator == $#chars) {
1.172 + print STDERR "Warning: missing null terminator in " .
1.173 + ".gnu_debuglink section of $file.\n";
1.174 + return '';
1.175 + }
1.176 + ++$terminator;
1.177 + }
1.178 + if ($#chars - $terminator > 3) {
1.179 + print STDERR "Warning: Excess padding in .gnu_debuglink section " .
1.180 + "of $file.\n";
1.181 + return '';
1.182 + }
1.183 + $#chars = $terminator - 1;
1.184 +
1.185 + my $basename = join('', map { chr(hex($_)) } @chars);
1.186 +
1.187 + # Now $basename and $hash represent the information in the
1.188 + # .gnu_debuglink section.
1.189 + #printf STDERR "%x: %s\n", $hash, $basename;
1.190 +
1.191 + my @possible_results = (
1.192 + dirname($file) . $basename,
1.193 + dirname($file) . '.debug/' . $basename,
1.194 + $global_debug_dir . dirname($file) . '/' . $basename
1.195 + );
1.196 + foreach my $result (@possible_results) {
1.197 + if (-f $result) {
1.198 + # XXX We should check the hash.
1.199 + return $result;
1.200 + }
1.201 + }
1.202 +
1.203 + return '';
1.204 +}
1.205 +
1.206 +sub get_file_info($) {
1.207 + my ($file) = @_;
1.208 + my $file_info = $file_infos{$file};
1.209 + unless (defined $file_info) {
1.210 + my $debug_file = separate_debug_file_for($file);
1.211 + $debug_file = $file if ($debug_file eq '');
1.212 +
1.213 + my $pid = open2($file_info->{pipe_read}, $file_info->{pipe_write},
1.214 + '/usr/bin/addr2line', '-C', '-f', '-e', $debug_file);
1.215 +
1.216 + set_address_adjustment($file, $file_info);
1.217 +
1.218 + $file_infos{$file} = $file_info;
1.219 + }
1.220 + return $file_info;
1.221 +}
1.222 +
1.223 +# Ignore SIGPIPE as a workaround for addr2line crashes in some situations.
1.224 +$SIG{PIPE} = 'IGNORE';
1.225 +
1.226 +select STDOUT; $| = 1; # make STDOUT unbuffered
1.227 +while (<>) {
1.228 + my $line = $_;
1.229 + if ($line =~ /^([ \|0-9-]*)(.*) ?\[([^ ]*) \+(0x[0-9A-F]{1,8})\](.*)$/) {
1.230 + my $before = $1; # allow preservation of balance trees
1.231 + my $badsymbol = $2;
1.232 + my $file = $3;
1.233 + my $address = hex($4);
1.234 + my $after = $5; # allow preservation of counts
1.235 +
1.236 + if (-f $file) {
1.237 + my $file_info = get_file_info($file);
1.238 + my $result = $file_info->{cache}->{$address};
1.239 + if (not defined $result) {
1.240 + my $address2 = $address + $file_info->{address_adjustment};
1.241 + my $out = $file_info->{pipe_write};
1.242 + my $in = $file_info->{pipe_read};
1.243 + printf {$out} "0x%X\n", $address2;
1.244 + chomp(my $symbol = <$in>);
1.245 + chomp(my $fileandline = <$in>);
1.246 + if (!$symbol || $symbol eq '??') { $symbol = $badsymbol; }
1.247 + if (!$fileandline || $fileandline eq '??:0') {
1.248 + $fileandline = $file;
1.249 + }
1.250 + $result = "$symbol ($fileandline)";
1.251 + $file_info->{cache}->{$address} = $result;
1.252 + }
1.253 + print "$before$result$after\n";
1.254 + } else {
1.255 + print STDERR "Warning: File \"$file\" does not exist.\n";
1.256 + print $line;
1.257 + }
1.258 +
1.259 + } else {
1.260 + print $line;
1.261 + }
1.262 +}
