From e6b8dc3cc580fabee26c84a4187c519f21bc0ae5 Mon Sep 17 00:00:00 2001 From: Pavol Rusnak Date: Mon, 13 Jun 2016 01:24:47 +0200 Subject: [PATCH] add perf option to emu.sh (-p) --- emu.sh | 28 +- src/.gitignore | 1 + vendor/flamegraph/flamegraph.pl | 1077 +++++++++++++++++++++++ vendor/flamegraph/stackcollapse-perf.pl | 257 ++++++ 4 files changed, 1357 insertions(+), 6 deletions(-) create mode 100755 vendor/flamegraph/flamegraph.pl create mode 100755 vendor/flamegraph/stackcollapse-perf.pl diff --git a/emu.sh b/emu.sh index fe2d4361fc..11faffa154 100755 --- a/emu.sh +++ b/emu.sh @@ -1,8 +1,24 @@ #!/bin/bash + +ARGS="${ARGS:--O0 -X heapsize=100000}" +MAIN="${MAIN:-main.py}" +BROWSER="${BROWSER:-chromium}" + cd `dirname $0`/src -if [ "$1" == -d ]; then - shift - gdb --args ../vendor/micropython/unix/micropython $* -O0 -X heapsize=100000 main.py -else - ../vendor/micropython/unix/micropython $* -O0 -X heapsize=100000 main.py -fi + +case "$1" in + "-d") + shift + gdb --args ../vendor/micropython/unix/micropython $ARGS $* $MAIN + ;; + "-p") + shift + ../vendor/micropython/unix/micropython $ARGS $* $MAIN & + perf record -F 100 -p $! -g -- sleep 600 + perf script > perf.trace + ../vendor/flamegraph/stackcollapse-perf.pl perf.trace | ../vendor/flamegraph/flamegraph.pl > perf.svg + $BROWSER perf.svg + ;; + *) + ../vendor/micropython/unix/micropython $ARGS $* $MAIN +esac diff --git a/src/.gitignore b/src/.gitignore index ca80b43567..4b148320e0 100644 --- a/src/.gitignore +++ b/src/.gitignore @@ -1 +1,2 @@ trezor.config +perf.* diff --git a/vendor/flamegraph/flamegraph.pl b/vendor/flamegraph/flamegraph.pl new file mode 100755 index 0000000000..5286d9f859 --- /dev/null +++ b/vendor/flamegraph/flamegraph.pl @@ -0,0 +1,1077 @@ +#!/usr/bin/perl -w +# +# flamegraph.pl flame stack grapher. +# +# This takes stack samples and renders a call graph, allowing hot functions +# and codepaths to be quickly identified. Stack samples can be generated using +# tools such as DTrace, perf, SystemTap, and Instruments. +# +# USAGE: ./flamegraph.pl [options] input.txt > graph.svg +# +# grep funcA input.txt | ./flamegraph.pl [options] > graph.svg +# +# Then open the resulting .svg in a web browser, for interactivity: mouse-over +# frames for info, click to zoom, and ctrl-F to search. +# +# Options are listed in the usage message (--help). +# +# The input is stack frames and sample counts formatted as single lines. Each +# frame in the stack is semicolon separated, with a space and count at the end +# of the line. These can be generated using DTrace with stackcollapse.pl, +# and other tools using the stackcollapse variants. +# +# An optional extra column of counts can be provided to generate a differential +# flame graph of the counts, colored red for more, and blue for less. This +# can be useful when using flame graphs for non-regression testing. +# See the header comment in the difffolded.pl program for instructions. +# +# The output graph shows relative presence of functions in stack samples. The +# ordering on the x-axis has no meaning; since the data is samples, time order +# of events is not known. The order used sorts function names alphabetically. +# +# While intended to process stack samples, this can also process stack traces. +# For example, tracing stacks for memory allocation, or resource usage. You +# can use --title to set the title to reflect the content, and --countname +# to change "samples" to "bytes" etc. +# +# There are a few different palettes, selectable using --color. By default, +# the colors are selected at random (except for differentials). Functions +# called "-" will be printed gray, which can be used for stack separators (eg, +# between user and kernel stacks). +# +# HISTORY +# +# This was inspired by Neelakanth Nadgir's excellent function_call_graph.rb +# program, which visualized function entry and return trace events. As Neel +# wrote: "The output displayed is inspired by Roch's CallStackAnalyzer which +# was in turn inspired by the work on vftrace by Jan Boerhout". See: +# https://blogs.oracle.com/realneel/entry/visualizing_callstacks_via_dtrace_and +# +# Copyright 2016 Netflix, Inc. +# Copyright 2011 Joyent, Inc. All rights reserved. +# Copyright 2011 Brendan Gregg. All rights reserved. +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License (the "License"). +# You may not use this file except in compliance with the License. +# +# You can obtain a copy of the license at docs/cddl1.txt or +# http://opensource.org/licenses/CDDL-1.0. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at docs/cddl1.txt. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# 11-Oct-2014 Adrien Mahieux Added zoom. +# 21-Nov-2013 Shawn Sterling Added consistent palette file option +# 17-Mar-2013 Tim Bunce Added options and more tunables. +# 15-Dec-2011 Dave Pacheco Support for frames with whitespace. +# 10-Sep-2011 Brendan Gregg Created this. + +use strict; + +use Getopt::Long; + +# tunables +my $encoding; +my $fonttype = "Verdana"; +my $imagewidth = 1200; # max width, pixels +my $frameheight = 16; # max height is dynamic +my $fontsize = 12; # base text size +my $fontwidth = 0.59; # avg width relative to fontsize +my $minwidth = 0.1; # min function width, pixels +my $nametype = "Function:"; # what are the names in the data? +my $countname = "samples"; # what are the counts in the data? +my $colors = "hot"; # color theme +my $bgcolor1 = "#eeeeee"; # background color gradient start +my $bgcolor2 = "#eeeeb0"; # background color gradient stop +my $nameattrfile; # file holding function attributes +my $timemax; # (override the) sum of the counts +my $factor = 1; # factor to scale counts by +my $hash = 0; # color by function name +my $palette = 0; # if we use consistent palettes (default off) +my %palette_map; # palette map hash +my $pal_file = "palette.map"; # palette map file name +my $stackreverse = 0; # reverse stack order, switching merge end +my $inverted = 0; # icicle graph +my $negate = 0; # switch differential hues +my $titletext = ""; # centered heading +my $titledefault = "Flame Graph"; # overwritten by --title +my $titleinverted = "Icicle Graph"; # " " +my $searchcolor = "rgb(230,0,230)"; # color for search highlighting +my $help = 0; + +sub usage { + die < outfile.svg\n + --title # change title text + --width # width of image (default 1200) + --height # height of each frame (default 16) + --minwidth # omit smaller functions (default 0.1 pixels) + --fonttype # font type (default "Verdana") + --fontsize # font size (default 12) + --countname # count type label (default "samples") + --nametype # name type label (default "Function:") + --colors # set color palette. choices are: hot (default), mem, io, + # wakeup, chain, java, js, perl, red, green, blue, aqua, + # yellow, purple, orange + --hash # colors are keyed by function name hash + --cp # use consistent palette (palette.map) + --reverse # generate stack-reversed flame graph + --inverted # icicle graph + --negate # switch differential hues (blue<->red) + --help # this message + + eg, + $0 --title="Flame Graph: malloc()" trace.txt > graph.svg +USAGE_END +} + +GetOptions( + 'fonttype=s' => \$fonttype, + 'width=i' => \$imagewidth, + 'height=i' => \$frameheight, + 'encoding=s' => \$encoding, + 'fontsize=f' => \$fontsize, + 'fontwidth=f' => \$fontwidth, + 'minwidth=f' => \$minwidth, + 'title=s' => \$titletext, + 'nametype=s' => \$nametype, + 'countname=s' => \$countname, + 'nameattr=s' => \$nameattrfile, + 'total=s' => \$timemax, + 'factor=f' => \$factor, + 'colors=s' => \$colors, + 'hash' => \$hash, + 'cp' => \$palette, + 'reverse' => \$stackreverse, + 'inverted' => \$inverted, + 'negate' => \$negate, + 'help' => \$help, +) or usage(); +$help && usage(); + +# internals +my $ypad1 = $fontsize * 4; # pad top, include title +my $ypad2 = $fontsize * 2 + 10; # pad bottom, include labels +my $xpad = 10; # pad lefm and right +my $framepad = 1; # vertical padding for frames +my $depthmax = 0; +my %Events; +my %nameattr; + +if ($titletext eq "") { + unless ($inverted) { + $titletext = $titledefault; + } else { + $titletext = $titleinverted; + } +} + +if ($nameattrfile) { + # The name-attribute file format is a function name followed by a tab then + # a sequence of tab separated name=value pairs. + open my $attrfh, $nameattrfile or die "Can't read $nameattrfile: $!\n"; + while (<$attrfh>) { + chomp; + my ($funcname, $attrstr) = split /\t/, $_, 2; + die "Invalid format in $nameattrfile" unless defined $attrstr; + $nameattr{$funcname} = { map { split /=/, $_, 2 } split /\t/, $attrstr }; + } +} + +# background colors: +# - yellow gradient: default (hot, java, js, perl) +# - blue gradient: mem, chain +# - gray gradient: io, wakeup, flat colors (red, green, blue, ...) +if ($colors eq "mem" or $colors eq "chain") { + $bgcolor1 = "#eeeeee"; $bgcolor2 = "#e0e0ff"; +} +if ($colors =~ /^(io|wakeup|red|green|blue|aqua|yellow|purple|orange)$/) { + $bgcolor1 = "#f8f8f8"; $bgcolor2 = "#e8e8e8"; +} + +# SVG functions +{ package SVG; + sub new { + my $class = shift; + my $self = {}; + bless ($self, $class); + return $self; + } + + sub header { + my ($self, $w, $h) = @_; + my $enc_attr = ''; + if (defined $encoding) { + $enc_attr = qq{ encoding="$encoding"}; + } + $self->{svg} .= < + + + +SVG + } + + sub include { + my ($self, $content) = @_; + $self->{svg} .= $content; + } + + sub colorAllocate { + my ($self, $r, $g, $b) = @_; + return "rgb($r,$g,$b)"; + } + + sub group_start { + my ($self, $attr) = @_; + + my @g_attr = map { + exists $attr->{$_} ? sprintf(qq/$_="%s"/, $attr->{$_}) : () + } qw(class style onmouseover onmouseout onclick); + push @g_attr, $attr->{g_extra} if $attr->{g_extra}; + $self->{svg} .= sprintf qq/\n/, join(' ', @g_attr); + + $self->{svg} .= sprintf qq/%s<\/title>/, $attr->{title} + if $attr->{title}; # should be first element within g container + + if ($attr->{href}) { + my @a_attr; + push @a_attr, sprintf qq/xlink:href="%s"/, $attr->{href} if $attr->{href}; + # default target=_top else links will open within SVG + push @a_attr, sprintf qq/target="%s"/, $attr->{target} || "_top"; + push @a_attr, $attr->{a_extra} if $attr->{a_extra}; + $self->{svg} .= sprintf qq//, join(' ', @a_attr); + } + } + + sub group_end { + my ($self, $attr) = @_; + $self->{svg} .= qq/<\/a>\n/ if $attr->{href}; + $self->{svg} .= qq/<\/g>\n/; + } + + sub filledRectangle { + my ($self, $x1, $y1, $x2, $y2, $fill, $extra) = @_; + $x1 = sprintf "%0.1f", $x1; + $x2 = sprintf "%0.1f", $x2; + my $w = sprintf "%0.1f", $x2 - $x1; + my $h = sprintf "%0.1f", $y2 - $y1; + $extra = defined $extra ? $extra : ""; + $self->{svg} .= qq/\n/; + } + + sub stringTTF { + my ($self, $color, $font, $size, $angle, $x, $y, $str, $loc, $extra) = @_; + $x = sprintf "%0.2f", $x; + $loc = defined $loc ? $loc : "left"; + $extra = defined $extra ? $extra : ""; + $self->{svg} .= qq/$str<\/text>\n/; + } + + sub svg { + my $self = shift; + return "$self->{svg}\n"; + } + 1; +} + +sub namehash { + # Generate a vector hash for the name string, weighting early over + # later characters. We want to pick the same colors for function + # names across different flame graphs. + my $name = shift; + my $vector = 0; + my $weight = 1; + my $max = 1; + my $mod = 10; + # if module name present, trunc to 1st char + $name =~ s/.(.*?)`//; + foreach my $c (split //, $name) { + my $i = (ord $c) % $mod; + $vector += ($i / ($mod++ - 1)) * $weight; + $max += 1 * $weight; + $weight *= 0.70; + last if $mod > 12; + } + return (1 - $vector / $max) +} + +sub color { + my ($type, $hash, $name) = @_; + my ($v1, $v2, $v3); + + if ($hash) { + $v1 = namehash($name); + $v2 = $v3 = namehash(scalar reverse $name); + } else { + $v1 = rand(1); + $v2 = rand(1); + $v3 = rand(1); + } + + # theme palettes + if (defined $type and $type eq "hot") { + my $r = 205 + int(50 * $v3); + my $g = 0 + int(230 * $v1); + my $b = 0 + int(55 * $v2); + return "rgb($r,$g,$b)"; + } + if (defined $type and $type eq "mem") { + my $r = 0; + my $g = 190 + int(50 * $v2); + my $b = 0 + int(210 * $v1); + return "rgb($r,$g,$b)"; + } + if (defined $type and $type eq "io") { + my $r = 80 + int(60 * $v1); + my $g = $r; + my $b = 190 + int(55 * $v2); + return "rgb($r,$g,$b)"; + } + + # multi palettes + if (defined $type and $type eq "java") { + if ($name =~ m:(/|\.):) { # Java (match "/" in path) + $type = "green"; + $type = "aqua" if $name =~ m/_\[i\]/; #inline + } elsif ($name =~ /::/) { # C++ + $type = "yellow"; + } elsif ($name =~ m:_\[k\]:) { # kernel + $type = "orange" + } else { # system + $type = "red"; + } + # fall-through to color palettes + } + if (defined $type and $type eq "perl") { + if ($name =~ /::/) { # C++ + $type = "yellow"; + } elsif ($name =~ m:Perl: or $name =~ m:\.pl:) { # Perl + $type = "green"; + } elsif ($name =~ m:_\[k\]:) { # kernel + $type = "orange" + } else { # system + $type = "red"; + } + # fall-through to color palettes + } + if (defined $type and $type eq "js") { + if ($name =~ /::/) { # C++ + $type = "yellow"; + } elsif ($name =~ m:/:) { # JavaScript (match "/" in path) + $type = "green" + } elsif ($name =~ m/:/) { # JavaScript (match ":" in builtin) + $type = "aqua" + } elsif ($name =~ m/^ $/) { # Missing symbol + $type = "green" + } elsif ($name =~ m:_\[k\]:) { # kernel + $type = "orange" + } else { # system + $type = "red"; + } + # fall-through to color palettes + } + if (defined $type and $type eq "wakeup") { + $type = "aqua"; + # fall-through to color palettes + } + if (defined $type and $type eq "chain") { + if ($name =~ m:_\[w\]:) { # waker + $type = "aqua" + } else { # off-CPU + $type = "blue"; + } + # fall-through to color palettes + } + + # color palettes + if (defined $type and $type eq "red") { + my $r = 200 + int(55 * $v1); + my $x = 50 + int(80 * $v1); + return "rgb($r,$x,$x)"; + } + if (defined $type and $type eq "green") { + my $g = 200 + int(55 * $v1); + my $x = 50 + int(60 * $v1); + return "rgb($x,$g,$x)"; + } + if (defined $type and $type eq "blue") { + my $b = 205 + int(50 * $v1); + my $x = 80 + int(60 * $v1); + return "rgb($x,$x,$b)"; + } + if (defined $type and $type eq "yellow") { + my $x = 175 + int(55 * $v1); + my $b = 50 + int(20 * $v1); + return "rgb($x,$x,$b)"; + } + if (defined $type and $type eq "purple") { + my $x = 190 + int(65 * $v1); + my $g = 80 + int(60 * $v1); + return "rgb($x,$g,$x)"; + } + if (defined $type and $type eq "aqua") { + my $r = 50 + int(60 * $v1); + my $g = 165 + int(55 * $v1); + my $b = 165 + int(55 * $v1); + return "rgb($r,$g,$b)"; + } + if (defined $type and $type eq "orange") { + my $r = 190 + int(65 * $v1); + my $g = 90 + int(65 * $v1); + return "rgb($r,$g,0)"; + } + + return "rgb(0,0,0)"; +} + +sub color_scale { + my ($value, $max) = @_; + my ($r, $g, $b) = (255, 255, 255); + $value = -$value if $negate; + if ($value > 0) { + $g = $b = int(210 * ($max - $value) / $max); + } elsif ($value < 0) { + $r = $g = int(210 * ($max + $value) / $max); + } + return "rgb($r,$g,$b)"; +} + +sub color_map { + my ($colors, $func) = @_; + if (exists $palette_map{$func}) { + return $palette_map{$func}; + } else { + $palette_map{$func} = color($colors, $hash, $func); + return $palette_map{$func}; + } +} + +sub write_palette { + open(FILE, ">$pal_file"); + foreach my $key (sort keys %palette_map) { + print FILE $key."->".$palette_map{$key}."\n"; + } + close(FILE); +} + +sub read_palette { + if (-e $pal_file) { + open(FILE, $pal_file) or die "can't open file $pal_file: $!"; + while ( my $line = ) { + chomp($line); + (my $key, my $value) = split("->",$line); + $palette_map{$key}=$value; + } + close(FILE) + } +} + +my %Node; # Hash of merged frame data +my %Tmp; + +# flow() merges two stacks, storing the merged frames and value data in %Node. +sub flow { + my ($last, $this, $v, $d) = @_; + + my $len_a = @$last - 1; + my $len_b = @$this - 1; + + my $i = 0; + my $len_same; + for (; $i <= $len_a; $i++) { + last if $i > $len_b; + last if $last->[$i] ne $this->[$i]; + } + $len_same = $i; + + for ($i = $len_a; $i >= $len_same; $i--) { + my $k = "$last->[$i];$i"; + # a unique ID is constructed from "func;depth;etime"; + # func-depth isn't unique, it may be repeated later. + $Node{"$k;$v"}->{stime} = delete $Tmp{$k}->{stime}; + if (defined $Tmp{$k}->{delta}) { + $Node{"$k;$v"}->{delta} = delete $Tmp{$k}->{delta}; + } + delete $Tmp{$k}; + } + + for ($i = $len_same; $i <= $len_b; $i++) { + my $k = "$this->[$i];$i"; + $Tmp{$k}->{stime} = $v; + if (defined $d) { + $Tmp{$k}->{delta} += $i == $len_b ? $d : 0; + } + } + + return $this; +} + +# parse input +my @Data; +my $last = []; +my $time = 0; +my $delta = undef; +my $ignored = 0; +my $line; +my $maxdelta = 1; + +# reverse if needed +foreach (<>) { + chomp; + $line = $_; + if ($stackreverse) { + # there may be an extra samples column for differentials + # XXX todo: redo these REs as one. It's repeated below. + my($stack, $samples) = (/^(.*)\s+?(\d+(?:\.\d*)?)$/); + my $samples2 = undef; + if ($stack =~ /^(.*)\s+?(\d+(?:\.\d*)?)$/) { + $samples2 = $samples; + ($stack, $samples) = $stack =~ (/^(.*)\s+?(\d+(?:\.\d*)?)$/); + unshift @Data, join(";", reverse split(";", $stack)) . " $samples $samples2"; + } else { + unshift @Data, join(";", reverse split(";", $stack)) . " $samples"; + } + } else { + unshift @Data, $line; + } +} + +# process and merge frames +foreach (sort @Data) { + chomp; + # process: folded_stack count + # eg: func_a;func_b;func_c 31 + my ($stack, $samples) = (/^(.*)\s+?(\d+(?:\.\d*)?)$/); + unless (defined $samples and defined $stack) { + ++$ignored; + next; + } + + # there may be an extra samples column for differentials: + my $samples2 = undef; + if ($stack =~ /^(.*)\s+?(\d+(?:\.\d*)?)$/) { + $samples2 = $samples; + ($stack, $samples) = $stack =~ (/^(.*)\s+?(\d+(?:\.\d*)?)$/); + } + $delta = undef; + if (defined $samples2) { + $delta = $samples2 - $samples; + $maxdelta = abs($delta) if abs($delta) > $maxdelta; + } + + # clean up SVG breaking characters: + $stack =~ tr/<>/()/; + + # for chain graphs, annotate waker frames with "_[w]", for later + # coloring. This is a hack, but has a precedent ("_[k]" from perf). + if ($colors eq "chain") { + my @parts = split ";-;", $stack; + my @newparts = (); + $stack = shift @parts; + $stack .= ";-;"; + foreach my $part (@parts) { + $part =~ s/;/_[w];/g; + $part .= "_[w]"; + push @newparts, $part; + } + $stack .= join ";-;", @parts; + } + + # merge frames and populate %Node: + $last = flow($last, [ '', split ";", $stack ], $time, $delta); + + if (defined $samples2) { + $time += $samples2; + } else { + $time += $samples; + } +} +flow($last, [], $time, $delta); + +warn "Ignored $ignored lines with invalid format\n" if $ignored; +unless ($time) { + warn "ERROR: No stack counts found\n"; + my $im = SVG->new(); + # emit an error message SVG, for tools automating flamegraph use + my $imageheight = $fontsize * 5; + $im->header($imagewidth, $imageheight); + $im->stringTTF($im->colorAllocate(0, 0, 0), $fonttype, $fontsize + 2, + 0.0, int($imagewidth / 2), $fontsize * 2, + "ERROR: No valid input provided to flamegraph.pl.", "middle"); + print $im->svg; + exit 2; +} +if ($timemax and $timemax < $time) { + warn "Specified --total $timemax is less than actual total $time, so ignored\n" + if $timemax/$time > 0.02; # only warn is significant (e.g., not rounding etc) + undef $timemax; +} +$timemax ||= $time; + +my $widthpertime = ($imagewidth - 2 * $xpad) / $timemax; +my $minwidth_time = $minwidth / $widthpertime; + +# prune blocks that are too narrow and determine max depth +while (my ($id, $node) = each %Node) { + my ($func, $depth, $etime) = split ";", $id; + my $stime = $node->{stime}; + die "missing start for $id" if not defined $stime; + + if (($etime-$stime) < $minwidth_time) { + delete $Node{$id}; + next; + } + $depthmax = $depth if $depth > $depthmax; +} + +# draw canvas, and embed interactive JavaScript program +my $imageheight = ($depthmax * $frameheight) + $ypad1 + $ypad2; +my $im = SVG->new(); +$im->header($imagewidth, $imageheight); +my $inc = < + + + + + + + +INC +$im->include($inc); +$im->filledRectangle(0, 0, $imagewidth, $imageheight, 'url(#background)'); +my ($white, $black, $vvdgrey, $vdgrey) = ( + $im->colorAllocate(255, 255, 255), + $im->colorAllocate(0, 0, 0), + $im->colorAllocate(40, 40, 40), + $im->colorAllocate(160, 160, 160), + ); +$im->stringTTF($black, $fonttype, $fontsize + 5, 0.0, int($imagewidth / 2), $fontsize * 2, $titletext, "middle"); +$im->stringTTF($black, $fonttype, $fontsize, 0.0, $xpad, $imageheight - ($ypad2 / 2), " ", "", 'id="details"'); +$im->stringTTF($black, $fonttype, $fontsize, 0.0, $xpad, $fontsize * 2, + "Reset Zoom", "", 'id="unzoom" onclick="unzoom()" style="opacity:0.0;cursor:pointer"'); +$im->stringTTF($black, $fonttype, $fontsize, 0.0, $imagewidth - $xpad - 100, + $fontsize * 2, "Search", "", 'id="search" onmouseover="searchover()" onmouseout="searchout()" onclick="search_prompt()" style="opacity:0.1;cursor:pointer"'); +$im->stringTTF($black, $fonttype, $fontsize, 0.0, $imagewidth - $xpad - 100, $imageheight - ($ypad2 / 2), " ", "", 'id="matched"'); + +if ($palette) { + read_palette(); +} + +# draw frames +while (my ($id, $node) = each %Node) { + my ($func, $depth, $etime) = split ";", $id; + my $stime = $node->{stime}; + my $delta = $node->{delta}; + + $etime = $timemax if $func eq "" and $depth == 0; + + my $x1 = $xpad + $stime * $widthpertime; + my $x2 = $xpad + $etime * $widthpertime; + my ($y1, $y2); + unless ($inverted) { + $y1 = $imageheight - $ypad2 - ($depth + 1) * $frameheight + $framepad; + $y2 = $imageheight - $ypad2 - $depth * $frameheight; + } else { + $y1 = $ypad1 + $depth * $frameheight; + $y2 = $ypad1 + ($depth + 1) * $frameheight - $framepad; + } + + my $samples = sprintf "%.0f", ($etime - $stime) * $factor; + (my $samples_txt = $samples) # add commas per perlfaq5 + =~ s/(^[-+]?\d+?(?=(?>(?:\d{3})+)(?!\d))|\G\d{3}(?=\d))/$1,/g; + + my $info; + if ($func eq "" and $depth == 0) { + $info = "all ($samples_txt $countname, 100%)"; + } else { + my $pct = sprintf "%.2f", ((100 * $samples) / ($timemax * $factor)); + my $escaped_func = $func; + $escaped_func =~ s/&/&/g; + $escaped_func =~ s//>/g; + $escaped_func =~ s/"/"/g; + $escaped_func =~ s/_\[[kwi]\]$//; # strip any annotation + unless (defined $delta) { + $info = "$escaped_func ($samples_txt $countname, $pct%)"; + } else { + my $d = $negate ? -$delta : $delta; + my $deltapct = sprintf "%.2f", ((100 * $d) / ($timemax * $factor)); + $deltapct = $d > 0 ? "+$deltapct" : $deltapct; + $info = "$escaped_func ($samples_txt $countname, $pct%; $deltapct%)"; + } + } + + my $nameattr = { %{ $nameattr{$func}||{} } }; # shallow clone + $nameattr->{class} ||= "func_g"; + $nameattr->{onmouseover} ||= "s(this)"; + $nameattr->{onmouseout} ||= "c()"; + $nameattr->{onclick} ||= "zoom(this)"; + $nameattr->{title} ||= $info; + $im->group_start($nameattr); + + my $color; + if ($func eq "-") { + $color = $vdgrey; + } elsif (defined $delta) { + $color = color_scale($delta, $maxdelta); + } elsif ($palette) { + $color = color_map($colors, $func); + } else { + $color = color($colors, $hash, $func); + } + $im->filledRectangle($x1, $y1, $x2, $y2, $color, 'rx="2" ry="2"'); + + my $chars = int( ($x2 - $x1) / ($fontsize * $fontwidth)); + my $text = ""; + if ($chars >= 3) { # room for one char plus two dots + $func =~ s/_\[[kwi]\]$//; # strip any annotation + $text = substr $func, 0, $chars; + substr($text, -2, 2) = ".." if $chars < length $func; + $text =~ s/&/&/g; + $text =~ s//>/g; + } + $im->stringTTF($black, $fonttype, $fontsize, 0.0, $x1 + 3, 3 + ($y1 + $y2) / 2, $text, ""); + + $im->group_end($nameattr); +} + +print $im->svg; + +if ($palette) { + write_palette(); +} + +# vim: ts=8 sts=8 sw=8 noexpandtab diff --git a/vendor/flamegraph/stackcollapse-perf.pl b/vendor/flamegraph/stackcollapse-perf.pl new file mode 100755 index 0000000000..dfd6da47bc --- /dev/null +++ b/vendor/flamegraph/stackcollapse-perf.pl @@ -0,0 +1,257 @@ +#!/usr/bin/perl -w +# +# stackcolllapse-perf.pl collapse perf samples into single lines. +# +# Parses a list of multiline stacks generated by "perf script", and +# outputs a semicolon separated stack followed by a space and a count. +# If memory addresses (+0xd) are present, they are stripped, and resulting +# identical stacks are colased with their counts summed. +# +# USAGE: ./stackcollapse-perf.pl [options] infile > outfile +# +# Run "./stackcollapse-perf.pl -h" to list options. +# +# Example input: +# +# swapper 0 [000] 158665.570607: cpu-clock: +# ffffffff8103ce3b native_safe_halt ([kernel.kallsyms]) +# ffffffff8101c6a3 default_idle ([kernel.kallsyms]) +# ffffffff81013236 cpu_idle ([kernel.kallsyms]) +# ffffffff815bf03e rest_init ([kernel.kallsyms]) +# ffffffff81aebbfe start_kernel ([kernel.kallsyms].init.text) +# [...] +# +# Example output: +# +# swapper;start_kernel;rest_init;cpu_idle;default_idle;native_safe_halt 1 +# +# Input may be created and processed using: +# +# perf record -a -g -F 997 sleep 60 +# perf script | ./stackcollapse-perf.pl > out.stacks-folded +# +# The output of "perf script" should include stack traces. If these are missing +# for you, try manually selecting the perf script output; eg: +# +# perf script -f comm,pid,tid,cpu,time,event,ip,sym,dso,trace | ... +# +# This is also required for the --pid or --tid options, so that the output has +# both the PID and TID. +# +# Copyright 2012 Joyent, Inc. All rights reserved. +# Copyright 2012 Brendan Gregg. All rights reserved. +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License (the "License"). +# You may not use this file except in compliance with the License. +# +# You can obtain a copy of the license at docs/cddl1.txt or +# http://opensource.org/licenses/CDDL-1.0. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at docs/cddl1.txt. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# 02-Mar-2012 Brendan Gregg Created this. +# 02-Jul-2014 " " Added process name to stacks. + +use strict; +use Getopt::Long; + +my %collapsed; + +sub remember_stack { + my ($stack, $count) = @_; + $collapsed{$stack} += $count; +} +my $annotate_kernel = 0; # put an annotation on kernel function +my $include_pname = 1; # include process names in stacks +my $include_pid = 0; # include process ID with process name +my $include_tid = 0; # include process & thread ID with process name +my $tidy_java = 1; # condense Java signatures +my $tidy_generic = 1; # clean up function names a little +my $target_pname; # target process name from perf invocation + +my $show_inline = 0; +my $show_context = 0; +GetOptions('inline' => \$show_inline, + 'context' => \$show_context, + 'pid' => \$include_pid, + 'kernel' => \$annotate_kernel, + 'tid' => \$include_tid) +or die < outfile\n + --pid # include PID with process names [1] + --tid # include TID and PID with process names [1] + --inline # un-inline using addr2line + --kernel # annotate kernel functions with a _[k] + --context # include source context from addr2line\n +[1] perf script must emit both PID and TIDs for these to work; eg: + perf script -f comm,pid,tid,cpu,time,event,ip,sym,dso,trace +USAGE_END + +# for the --inline option +sub inline { + my ($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}//; + + my @fullfunc; + my $one_item = ""; + for (split /^/, $a2l_output) { + chomp $_; + + # remove discriminator info if exists + $_ =~ s/ \(discriminator \S+\)//; + + if ($one_item eq "") { + $one_item = $_; + } else { + if ($show_context == 1) { + unshift @fullfunc, $one_item . ":$_"; + } else { + unshift @fullfunc, $one_item; + } + $one_item = ""; + } + } + + return join(";", @fullfunc); +} + +my @stack; +my $pname; + +# +# Main loop +# +while (defined($_ = <>)) { + + # find the name of the process launched by perf, by stepping backwards + # over the args to find the first non-option (no dash): + if (/^# cmdline/) { + my @args = split ' ', $_; + foreach my $arg (reverse @args) { + if ($arg !~ /^-/) { + $target_pname = $arg; + $target_pname =~ s:.*/::; # strip pathname + last; + } + } + } + + # skip remaining comments + next if m/^#/; + chomp; + + # end of stack. save cached data. + if (m/^$/) { + if ($include_pname) { + if (defined $pname) { + unshift @stack, $pname; + } else { + unshift @stack, ""; + } + } + remember_stack(join(";", @stack), 1) if @stack; + undef @stack; + undef $pname; + next; + } + + # event record start + if (/^(\S+\s*?\S*?)\s+(\d+)\s/) { + # default "perf script" output has TID but not PID + # eg, "java 25607 4794564.109216: cycles:" + # eg, "java 12688 [002] 6544038.708352: cpu-clock:" + # eg, "V8 WorkerThread 25607 4794564.109216: cycles:" + # other combinations possible + if ($include_tid) { + $pname = "$1-?/$2"; + } elsif ($include_pid) { + $pname = "$1-?"; + } else { + $pname = $1; + } + $pname =~ tr/ /_/; + } elsif (/^(\S+\s*?\S*?)\s+(\d+)\/(\d+)/) { + # eg, "java 24636/25607 [000] 4794564.109216: cycles:" + # eg, "java 12688/12764 6544038.708352: cpu-clock:" + # eg, "V8 WorkerThread 24636/25607 [000] 94564.109216: cycles:" + # other combinations possible + if ($include_tid) { + $pname = "$1-$2/$3"; + } elsif ($include_pid) { + $pname = "$1-$2"; + } else { + $pname = $1; + } + $pname =~ tr/ /_/; + + # stack line + } elsif (/^\s*(\w+)\s*(.+) \((\S*)\)/) { + my ($pc, $rawfunc, $mod) = ($1, $2, $3); + $rawfunc.="_[k]" if ($annotate_kernel == 1 && $mod =~ m/kernel\./); + if ($show_inline == 1 && $mod !~ m/(perf-\d+.map|kernel\.|\[[^\]]+\])/) { + unshift @stack, inline($pc, $mod); + next; + } + + next if $rawfunc =~ /^\(/; # skip process names + + my @inline; + for (split /\->/, $rawfunc) { + my $func = $_; + + if ($tidy_generic) { + $func =~ s/;/:/g; + $func =~ tr/<>//d; + if ($func !~ m/\.\(.*\)\./) { + # This doesn't look like a Go method name (such as + # "net/http.(*Client).Do"), so everything after the first open + # paren is just noise. + $func =~ s/\(.*//; + } + # now tidy this horrible thing: + # 13a80b608e0a RegExp:[&<>\"\'] (/tmp/perf-7539.map) + $func =~ tr/"\'//d; + # fall through to $tidy_java + } + + if ($tidy_java and $pname eq "java") { + # along with $tidy_generic, converts the following: + # Lorg/mozilla/javascript/ContextFactory;.call(Lorg/mozilla/javascript/ContextAction;)Ljava/lang/Object; + # Lorg/mozilla/javascript/ContextFactory;.call(Lorg/mozilla/javascript/C + # Lorg/mozilla/javascript/MemberBox;.(Ljava/lang/reflect/Method;)V + # into: + # org/mozilla/javascript/ContextFactory:.call + # org/mozilla/javascript/ContextFactory:.call + # org/mozilla/javascript/MemberBox:.init + $func =~ s/^L// if $func =~ m:/:; + } + + $func .= "_[i]" if scalar(@inline) > 0; #inlined + push @inline, $func; + } + + unshift @stack, @inline; + } else { + warn "Unrecognized line: $_"; + } +} + +foreach my $k (sort { $a cmp $b } keys %collapsed) { + print "$k $collapsed{$k}\n"; +}