Merge pull request #251 from jan-konczak-cs-put/master

enhancing stackcollapse-perf.pl "--inline" option: caching addr2line and getting rid of '??' in graphs

stackcollapse-perf.pl

@@ -88,8 +88,11 @@ sub remember_stack {
 
 my $show_inline = 0;
 my $show_context = 0;
+
+my $srcline_in_input = 0; # if there are extra lines with source location (perf script -F+srcline)
 GetOptions('inline' => \$show_inline,
            'context' => \$show_context,
+           'srcline' => \$srcline_in_input,
            'pid' => \$include_pid,
            'kernel' => \$annotate_kernel,
            'jit' => \$annotate_jit,
@@ -106,6 +109,7 @@ sub remember_stack {
 	--kernel	# annotate kernel functions with a _[k]
 	--jit		# annotate jit functions with a _[j]
 	--context	# adds source context to --inline
+	--srcline	# parses output of 'perf script -F+srcline' and adds source context
 	--addrs		# include raw addresses where symbols can't be found
 	--event-filter=EVENT	# event name filter\n
 [1] perf script must emit both PID and TIDs for these to work; eg, Linux < 4.1:
@@ -119,16 +123,49 @@ sub remember_stack {
 	$annotate_kernel = $annotate_jit = 1;
 }
 
+my %inlineCache;
+
+my %nmCache;
+
+sub inlineCacheAdd {
+	my ($pc, $mod, $result) = @_;
+   if (defined($inlineCache{$pc})) {
+      $inlineCache{$pc}{$mod} = $result;
+   } else {
+      $inlineCache{$pc} = {$mod => $result};
+   }
+}
+
 # for the --inline option
 sub inline {
-	my ($pc, $mod) = @_;
+	my ($pc, $rawfunc, $mod) = @_;
+
+	return $inlineCache{$pc}{$mod} if defined($inlineCache{$pc}{$mod});
 
 	# capture addr2line output
 	my $a2l_output = `addr2line -a $pc -e $mod -i -f -s -C`;
 
 	# remove first line
 	$a2l_output =~ s/^(.*\n){1}//;
 
+	if ($a2l_output =~ /\?\?\n\?\?:0/) {
+		# if addr2line fails and rawfunc is func+offset, then fall back to it
+		if ($rawfunc =~ /^(.+)\+0x([0-9a-f]+)$/) {
+			my $func = $1;
+			my $addr = hex $2;
+
+			$nmCache{$mod}=`nm $mod` unless defined $nmCache{$mod};
+
+			if ($nmCache{$mod} =~ /^([0-9a-f]+) . \Q$func\E$/m) {
+			   my $base = hex $1;
+				my $newPc = sprintf "0x%x", $base+$addr;
+				my $result = inline($newPc, '', $mod);
+				inlineCacheAdd($pc, $mod, $result);
+				return $result;
+			}
+		}
+	}
+
 	my @fullfunc;
 	my $one_item = "";
 	for (split /^/, $a2l_output) {
@@ -149,7 +186,11 @@ sub inline {
 		}
 	}
 
-	return join(";", @fullfunc);
+	my $result = join ";" , @fullfunc;
+
+	inlineCacheAdd($pc, $mod, $result);
+
+	return $result;
 }
 
 my @stack;
@@ -256,18 +297,25 @@ sub inline {
 
 		my ($pc, $rawfunc, $mod) = ($1, $2, $3);
 
+		if ($show_inline == 1 && $mod !~ m/(perf-\d+.map|kernel\.|\[[^\]]+\])/) {
+			my $inlineRes = inline($pc, $rawfunc, $mod);
+			# - empty result this happens e.g., when $mod does not exist or is a path to a compressed kernel module
+			#   if this happens, the user will see error message from addr2line written to stderr
+			# - if addr2line results in "??" , then it's much more sane to fall back than produce a '??' in graph
+			if($inlineRes ne "" and $inlineRes ne "??" and $inlineRes ne "??:??:0" ) {
+				unshift @stack, $inlineRes;
+				next;
+			}
+		}
+
 		# Linux 4.8 included symbol offsets in perf script output by default, eg:
 		# 7fffb84c9afc cpu_startup_entry+0x800047c022ec ([kernel.kallsyms])
 		# strip these off:
 		$rawfunc =~ s/\+0x[\da-f]+$//;
 
-		if ($show_inline == 1 && $mod !~ m/(perf-\d+.map|kernel\.|\[[^\]]+\])/) {
-			unshift @stack, inline($pc, $mod);
-			next;
-		}
-
 		next if $rawfunc =~ /^\(/;		# skip process names
 
+		my $is_unknown=0;
 		my @inline;
 		for (split /\->/, $rawfunc) {
 			my $func = $_;
@@ -278,6 +326,7 @@ sub inline {
 					$func =~ s/.*\///;
 				} else {
 					$func = "unknown";
+					$is_unknown=1;
 				}
 
 				if ($include_addrs) {
@@ -331,6 +380,42 @@ sub inline {
 			} elsif ($annotate_jit == 1 && $mod =~ m:/tmp/perf-\d+\.map:) {
 				$func .= "_[j]";	# jitted
 			}
+
+			#
+			# Source lines
+			#
+			#
+			# Sample outputs:
+			#   | a.out 35081 252436.005167:     667783 cycles:
+			#   |                   408ebb some_method_name+0x8b (/full/path/to/a.out)
+			#   |   uniform_int_dist.h:300
+			#   |                   4069f5 main+0x935 (/full/path/to/a.out)
+			#   |   file.cpp:137
+			#   |             7f6d2148eb25 __libc_start_main+0xd5 (/lib64/libc-2.33.so)
+			#   |   libc-2.33.so[27b25]
+			#
+			#   | a.out 35081 252435.738165:     306459 cycles:
+			#   |             7f6d213c2750 [unknown] (/usr/lib64/libkmod.so.2.3.6)
+			#   |   libkmod.so.2.3.6[6750]
+			#
+			#   | a.out 35081 252435.738373:     315813 cycles:
+			#   |             7f6d215ca51b __strlen_avx2+0x4b (/lib64/libc-2.33.so)
+			#   |   libc-2.33.so[16351b]
+			#   |             7ffc71ee9580 [unknown] ([unknown])			
+			#   |
+			#
+			#   | a.out 35081 252435.718940:     247984 cycles:
+			#   |         ffffffff814f9302 up_write+0x32 ([kernel.kallsyms])
+			#   |   [kernel.kallsyms][ffffffff814f9302]
+			if($srcline_in_input and not $is_unknown){
+				$_ = <>;
+				chomp;
+				s/\[.*?\]//g;
+				s/^\s*//g;
+				s/\s*$//g;
+				$func.=':'.$_ unless $_ eq "";
+			}
+
 			push @inline, $func;
 		}