[systemd-devel] [PATCH] [RFC] systemd-bootchart: merge bootchart.

Zbigniew Jędrzejewski-Szmek zbyszek at in.waw.pl
Wed Oct 31 11:30:21 PDT 2012


On Wed, Oct 17, 2012 at 04:01:12PM -0700, Auke Kok wrote:
> << do not merge this patch - it has not been fully tested yet >>
Hi,

> This is an initial RFC patch merging bootchart almost verbatim
> into systemd.
> 
> Bootchart is renamed to 'systemd-bootchart' and installed as
> $libdir/systemd/systemd-bootchart. The configuration file
> will reside in $sysconfdir/systemd/bootchart.conf.
Does this mean that there'll be a file in /etc/systemd/ always?
This goes against the practice of moving default configuration
to /usr.

Would be nicer to leverage conf-files.[ch] and conf-parser.[ch].

Also, if the binary is supposed to be invoked directly,
then is should go into /usr/bin, not /usr/libexec.

> The code is almost unchanged. Graphing routines and logging code
> do not use or leverage any additional systemd data that may be
> available. This work is planned after the initial merge.
Some notes inline below. Also some suggestions for rewordings
in the documentation. Nothing important really.

> The goal of this patch is to get comments on the overall integration
> style. Man pages were added to cover the various ways of starting
> bootchart:
> 
> - as replacement init process
> - as standalone program
> - not implemented yet: as early process by systemd
> - not implemented yet: as regular systemd service unit
> 
> Signed-off-by: Auke Kok <auke-jan.h.kok at intel.com>
> ---
>  Makefile.am                  |   16 +
>  configure.ac                 |    9 +
>  man/bootchart.conf.xml       |  158 ++++++
>  man/systemd-bootchart.xml    |  117 +++++
>  src/bootchart/README         |   83 ++++
>  src/bootchart/bootchart.c    |  352 +++++++++++++
>  src/bootchart/bootchart.conf |   20 +
>  src/bootchart/bootchart.h    |  117 +++++
>  src/bootchart/log.c          |  420 ++++++++++++++++
>  src/bootchart/svg.c          | 1120 ++++++++++++++++++++++++++++++++++++++++++
>  10 files changed, 2412 insertions(+)
>  create mode 100644 man/bootchart.conf.xml
>  create mode 100644 man/systemd-bootchart.xml
>  create mode 100644 src/bootchart/README
>  create mode 100644 src/bootchart/bootchart.c
>  create mode 100644 src/bootchart/bootchart.conf
>  create mode 100644 src/bootchart/bootchart.h
>  create mode 100644 src/bootchart/log.c
>  create mode 100644 src/bootchart/svg.c
> 
> diff --git a/Makefile.am b/Makefile.am
> index 8e88c42..03dc535 100644
> --- a/Makefile.am
> +++ b/Makefile.am
> @@ -2882,6 +2882,22 @@ EXTRA_DIST += \
>  	units/systemd-readahead-done.service.in
>  
>  # ------------------------------------------------------------------------------
> +if ENABLE_BOOTCHART
> +systemd_bootchart_SOURCES = \
> +	src/bootchart/bootchart.c \
> +	src/bootchart/bootchart.h \
> +	src/bootchart/log.c \
> +	src/bootchart/svg.c
> +
> +MANPAGES += \
> +	man/systemd-bootchart.1
> +	man/bootchart.conf.5
> +
> +rootlibexec_PROGRAMS += \
> +	systemd-bootchart
> +endif
> +
> +# ------------------------------------------------------------------------------
>  if ENABLE_QUOTACHECK
>  rootlibexec_PROGRAMS += \
>  	systemd-quotacheck
> diff --git a/configure.ac b/configure.ac
> index 62e83be..75fcd19 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -509,6 +509,14 @@ fi
>  AM_CONDITIONAL(ENABLE_READAHEAD, [test "$have_readahead" = "yes"])
>  
>  # ------------------------------------------------------------------------------
> +have_bootchart=no
> +AC_ARG_ENABLE(bootchart, AS_HELP_STRING([--disable-bootchart], [disable bootchart tool]))
> +if test "x$enable_bootchart" != "xno"; then
> +        have_bootchart=yes
> +fi
> +AM_CONDITIONAL(ENABLE_BOOTCHART, [test "$have_bootchart" = "yes"])
> +
> +# ------------------------------------------------------------------------------
>  have_quotacheck=no
>  AC_ARG_ENABLE(quotacheck, AS_HELP_STRING([--disable-quotacheck], [disable quotacheck tools]))
>  if test "x$enable_quotacheck" != "xno"; then
> @@ -868,6 +876,7 @@ AC_MSG_RESULT([
>          binfmt:                  ${have_binfmt}
>          vconsole:                ${have_vconsole}
>          readahead:               ${have_readahead}
> +        bootchart:               ${have_bootchart}
>          quotacheck:              ${have_quotacheck}
>          randomseed:              ${have_randomseed}
>          logind:                  ${have_logind}
> diff --git a/man/bootchart.conf.xml b/man/bootchart.conf.xml
> new file mode 100644
> index 0000000..f8fcfe0
> --- /dev/null
> +++ b/man/bootchart.conf.xml
> @@ -0,0 +1,158 @@
> +<?xml version='1.0'?> <!--*-nxml-*-->
> +<?xml-stylesheet type="text/xsl" href="http://docbook.sourceforge.net/release/xsl/current/xhtml/docbook.xsl"?>
> +<!DOCTYPE refentry PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
> +        "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd">
> +
> +<!--
> +  This file is part of systemd.
> +
> +  Copyright 2012 Intel Corporation
> +
> +  Authors:
> +    Auke Kok <auke-jan.h.kok at intel.com>
> +
> +  systemd is free software; you can redistribute it and/or modify it
> +  under the terms of the GNU Lesser General Public License as published by
> +  the Free Software Foundation; either version 2.1 of the License, or
> +  (at your option) any later version.
> +
> +  systemd is distributed in the hope that it will be useful, but
> +  WITHOUT ANY WARRANTY; without even the implied warranty of
> +  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> +  Lesser General Public License for more details.
> +
> +  You should have received a copy of the GNU Lesser General Public License
> +  along with systemd; If not, see <http://www.gnu.org/licenses/>.
> +-->
> +
> +<refentry id="bootchart.conf">
> +        <refentryinfo>
> +                <title>bootchart.conf</title>
> +                <productname>systemd</productname>
> +
> +                <authorgroup>
> +                        <author>
> +                                <contrib>Developer</contrib>
> +                                <firstname>Auke</firstname>
> +                                <surname>Kok</surname>
> +                                <email>auke-jan.h.kok at intel.com</email>
> +                        </author>
> +                </authorgroup>
> +        </refentryinfo>
> +
> +        <refmeta>
> +                <refentrytitle>bootchart.conf</refentrytitle>
> +                <manvolnum>5</manvolnum>
> +        </refmeta>
> +
> +        <refnamediv>
> +                <refname>bootchart.conf</refname>
> +                <refpurpose>Boot performance analysis graphing tool configuration file</refpurpose>
> +        </refnamediv>
> +
> +        <refsynopsisdiv>
> +                <para><filename>/etc/systemd/bootchart.conf</filename></para>
> +        </refsynopsisdiv>
> +
> +        <refsect1>
> +                <title>Description</title>
> +
> +                <para>When starting, systemd-bootchart will read the
> +                configuration file <filename>bootchart.conf</filename>.
> +                This configuration file determines logging parameters and
> +                graph output.</para>
> +
> +        </refsect1>
> +
> +        <refsect1>
> +                <title>Options</title>
> +
> +                <variablelist class='bootchart-directives'>
> +
> +                        <varlistentry>
> +                                <term><varname>samples=500</varname></term>
> +                                <listitem><para>Configure the amount of samples to
> +                                record total before bootchart exits. Each sample will
s/record total before/record before/

> +                                record at intervals defined by freq=.</para></listitem>
freq= should be <option>freq=</option>
> +                        </varlistentry>
> +
> +                        <varlistentry>
> +                                <term><varname>freq=25</varname></term>
> +                                <listitem><para>Configure the sample log frequency.
> +                                This can be a fractional number, but must be larger than
> +                                0.0. Most systems can cope with values under 25-50 without
> +                                impacting boot time severely.</para></listitem>
> +                        </varlistentry>
In general, systemd style seems to be CamelCase, mostly without abbrevations.
Frequency would be more in line.

Also, this doesn't explain in what units the number is.

> +                        <varlistentry>
> +                                <term><varname>rel=0</varname></term>
> +                                <listitem><para>Configures whether the left axis of the
> +                                output graph equals time=0.0 (CLOCK_MONOTONIC start). This
> +                                is useful for using bootchart at post-boot time to profile
> +                                an already booted system, otherwise the graph would become
> +                                extremely large. If set to a non-zero value, the horizontal
> +                                axis starts at the first recorded sample instead of time=0.0.
> +                                </para></listitem>
> +                        </varlistentry>
Relative=?

> +                        <varlistentry>
> +                                <term><varname>filter=0</varname></term>
> +                                <listitem><para>Configures whether the resulting graph
> +                                should omit tasks that did not contribute significantly
> +                                to the boot. Processes that are too short-lived (only
> +                                seen in one sample) or that do not consume any significant
> +                                CPU time (less than 0.001sec) will not be displayed in
> +                                the output graph.</para></listitem>
> +                        </varlistentry>
> +
> +                        <varlistentry>
> +                                <term><varname>output=[path]</varname></term>
> +                                <listitem><para>Configures the output folder for writing
s/folder/directory/
s/for writing the graphs/for the graphs/
> +                                the graphs. By default, bootchart writes the graphs to
> +                                <filename>/var/log</filename>.</para></listitem>
> +                        </varlistentry>
> +
> +                        <varlistentry>
> +                                <term><varname>init=[path]</varname></term>
> +                                <listitem><para>Configures bootchart to run a non-standard
> +                                binary instead of <filename>/sbin/init</filename>. This
> +                                option is only relevant if bootchart was invoked from the
> +                                kernel command line with
> +                                init=/usr/lib/systemd/systemd-bootchart.</para></listitem>
> +                        </varlistentry>
> +                        <varlistentry>
> +                                <term><varname>pss=0</varname></term>
> +                                <listitem><para>If set to 1, enables logging and graphing
> +                                of processes PSS memory consumption.</para></listitem>
> +                        </varlistentry>
> +
> +                        <varlistentry>
> +                                <term><varname>entropy=0</varname></term>
> +                                <listitem><para>If set to 1, enables logging and graphing
> +                                of the kernel random entropy pool size.</para></listitem>
> +                        </varlistentry>
> +
> +                        <varlistentry>
> +                                <term><varname>scale_x=100</varname></term>
> +                                <listitem><para>Horizontal scaling factor for all variable
> +                                graph components.</para></listitem>
> +                        </varlistentry>
> +
> +                        <varlistentry>
> +                                <term><varname>scale_y=20</varname></term>
> +                                <listitem><para>Vertical scaling factor for all variable
> +                                graph components.</para></listitem>
> +                        </varlistentry>
Init, ShowPSS, ShowEntropy, ScaleX, ScaleY?

> +
> +                </variablelist>
> +        </refsect1>
> +
> +        <refsect1>
> +                  <title>See Also</title>
> +                  <para>
> +                          <citerefentry><refentrytitle>systemd-bootchart</refentrytitle><manvolnum>1</manvolnum></citerefentry>
> +                  </para>
> +        </refsect1>
> +
> +</refentry>
> diff --git a/man/systemd-bootchart.xml b/man/systemd-bootchart.xml
> new file mode 100644
> index 0000000..4d53824
> --- /dev/null
> +++ b/man/systemd-bootchart.xml
> @@ -0,0 +1,117 @@
> +<?xml version='1.0'?> <!--*-nxml-*-->
> +<?xml-stylesheet type="text/xsl" href="http://docbook.sourceforge.net/release/xsl/current/xhtml/docbook.xsl"?>
> +<!DOCTYPE refentry PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
> +        "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd">
> +
> +<!--
> +  This file is part of systemd.
> +
> +  Copyright 2012 Intel Corporation
> +
> +  Authors:
> +    Auke Kok <auke-jan.h.kok at intel.com>
> +
> +  systemd is free software; you can redistribute it and/or modify it
> +  under the terms of the GNU Lesser General Public License as published by
> +  the Free Software Foundation; either version 2.1 of the License, or
> +  (at your option) any later version.
> +
> +  systemd is distributed in the hope that it will be useful, but
> +  WITHOUT ANY WARRANTY; without even the implied warranty of
> +  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> +  Lesser General Public License for more details.
> +
> +  You should have received a copy of the GNU Lesser General Public License
> +  along with systemd; If not, see <http://www.gnu.org/licenses/>.
> +-->
> +
> +<refentry id="systemd-bootchart">
> +        <refentryinfo>
> +                <title>systemd-bootchart</title>
> +                <productname>systemd</productname>
> +
> +                <authorgroup>
> +                        <author>
> +                                <contrib>Developer</contrib>
> +                                <firstname>Auke</firstname>
> +                                <surname>Kok</surname>
> +                                <email>auke-jan.h.kok at intel.com</email>
> +                        </author>
> +                </authorgroup>
> +        </refentryinfo>
> +
> +        <refmeta>
> +                <refentrytitle>systemd-bootchart</refentrytitle>
> +                <manvolnum>1</manvolnum>
> +        </refmeta>
> +
> +        <refnamediv>
> +                <refname>systemd-bootchart</refname>
> +                <refpurpose>Boot performance analysis graphing tool</refpurpose>
> +        </refnamediv>
> +
> +        <refsect1>
> +                <title>Description</title>
> +
> +                <para>Systemd-bootchart is an boot time analysis tool. It represents
s/an/a/

> +                various aspects of the system as graph elements. These graph
> +                elements allow the user to determine resource usage, efficiency
> +                and performance issues.</para>
> +
> +        </refsect1>
> +
> +        <refsect1>
> +                <title>Invocation</title>
> +
> +                <para>systemd-bootchart can be invoked in several different ways:</para>
> +
> +                <variablelist class='bootchart-invocation'>
> +
> +                        <varlistentry>
> +                                <title>Kernel invocation</title>
> +                                <listitem><para>The kernel can invoke systemd-bootchart
> +                                instead of the init process. In itself, systemd-bootchart
> +                                will invoke <filename>/sbin/init</filename> if invoked in
> +                                this matter.</para></listitem>
> +                        </varlistentry>
<command>systemd-bootchart</command> can be specified as the init binary invoked by the kernel
with the kernel <option>init=</option> option.
<command>systemd-bootchart</command> will then in
turn invoke <filename>/sbin/init</filename> or the binary specified with
<option>init=</option> option in
<citerefentry><refentrytitle>bootchart.conf</refentrytitle>
<manvolnum>5</manvolnum></citerefentry>.

> +
> +                        <varlistentry>
> +                                <title>Started as a standalone program</title>
> +                                <listitem><para>One can execute systemd-bootchart as
> +                                normal application from the commandline. In this mode
> +                                it is highly recommended to pass the "-r" flag in order
> +                                to not graph the time elapsed since boot and before
> +                                systemd-bootchart was started, as it may result in
> +                                extremely large graphs.
> +                                </para></listitem>
> +                        </varlistentry>
> +                </variablelist>
> +        </refsect1>
> +
> +        <refsect1>
> +                <title>Options</title>
> +
> +                <para>Please read systemd-bootchart --help or the bootchart.conf manual
<programlisting>systemd-bootchart --help</programlisting>
<citerefentry>...

> +                page for information about the various options that influence how
> +                systemd-bootchart operates.</para>
> +        </refsect1>
> +
> +        <refsect1>
> +                <title>Output</title>
> +
> +                <para>Systemd-bootchart generates SVG graphs. In order to render these
> +                on a graphica display any SVG capable viewer can be used. It should be
> +                noted that the SVG render engines in most browsers (including Chrome
> +                and Firefox) are many times faster than dedicated graphical applications
> +                like Gimp and Inkscape.  Just point your browser at "file:///var/log"!
> +                </para>
Would be nice to explicitly say that the graphs go to /var/log/.

> +        </refsect1>
> +
> +        <refsect1>
> +                  <title>See Also</title>
> +                  <para>
> +                          <citerefentry><refentrytitle>bootchart.conf</refentrytitle><manvolnum>5</manvolnum></citerefentry>
> +                  </para>
> +        </refsect1>
> +
> +</refentry>
> diff --git a/src/bootchart/README b/src/bootchart/README
> new file mode 100644
> index 0000000..a209caf
> --- /dev/null
> +++ b/src/bootchart/README
> @@ -0,0 +1,83 @@
> +
> +Bootchart - a 'startup' graphing tool
> +
> +--
> +
> +Bootchart is a tool, usually run at system startup, that collects and graphs
> +the CPU and disk load of the system as it works. The output of bootchart is
> +an SVG graph. Normally, bootchart is invoked as `bootchartd` by the kernel
systemd-bootchart

> +by passing "init=/sbin/bootchartd" to the kernel. Bootchart will then fork
> +init off to resume normal system startup, while monitoring and logging
> +startup information in the background.
> +
> +After collecting a certain amount of data (usually 15-30 seconds) the logging
> +stops and a graph is generated from the logged information. This graph
> +contains vital clues to which resources are being used, in which order, and
> +where possible problems exist in the startup sequence of the system.
> +
> +Of course, bootchart can also be used at any moment in time to collect and
> +graph some data for an amount of time. Bootchart does not even require root
> +privileges to do so, and will happily run as a normal user. Bootchart graphs
> +are by default written time-stamped in /var/log.
OK, so this text can go into systemd-bootchart manpage.

> +This version of bootchart was implemented from scratch and inspired by former
> +incantations of bootchart:
> +
> +- The original bash/shell code implemented bootchart. This version logged all
> +data into a compressed tarball for later processing, and did not create a graph
> +on it's own.
> +
> +- The C-code implementation found in Ubuntu. This version replaced above shell
> +code version with a faster and efficient data logger, but still did not graph
> +code itself.
> +
> +- the original Java-based bootchart, the original graphing program that created
> +a bootchart graph from logged data.
> +
> +- the pybootchartgui.py program, which created a graph based on the data logged
> +by either standalone data logger.
> +
> +The version you are looking at combines these 2 parts into a single program,
> +which makes running it and creating graphs a bit more efficient and simple.
> +You can now run a single program at startup instead of 2. There are no timing
> +problems (the graphing stage will never run if the logging stage didn't
> +finish). The logged data isn't being written to disc first, then read again.
> +Also, the data kept in memory is reduced to the absolute minimum needed to
> +keep memory use low.
HISTORY section of the manpage?

> +
> +--
> +
> +Requirements: glibc. Your kernel must have procfs support and several
> +proc output options enabled:
> +  CONFIG_PROC_FS
> +  CONFIG_SCHEDSTATS
> +  CONFIG_SCHED_DEBUG
> +at a minimum. bootchartd itself does not require any graphics library
> +to generate the SVG output file.
This part into REQUIREMENTS in /README.

> +
> +--
> +
> +Configuration: please see bootchartd --help, as well as /etc/bootchartd.conf
> +and/or /usr/share/doc/bootchart/bootchartd.conf.example for a list of
> +configurable options.
This part is outdated.

> +
> +--
> +
> +Many thanks to those who contributed ideas and code:
> + - Ziga Mahkovec - Original bootchart author
> + - Anders Norgaard - PyBootchartgui
> + - Michael Meeks - bootchart2
> + - Scott James Remnant - Ubuntu C-based logger
> + - Arjan van der Ven - for the idea to merge bootgraph.pl functionality
> +
> +--
> +
> +For bugs, please contact the author or current maintainer:
> +Auke Kok <auke-jan.h.kok at intel.com>
> +
> +--
> +
> +Download bootchart releases here: http://foo-projects.org/~sofar/bootchart/
> +Source code is hosted here: git://github.com/sofar/bootchart
> diff --git a/src/bootchart/bootchart.c b/src/bootchart/bootchart.c
> new file mode 100644
> index 0000000..2eeb37a
> --- /dev/null
> +++ b/src/bootchart/bootchart.c
> @@ -0,0 +1,352 @@
> +/*
> + * bootchart.c
> + *
> + * Copyright (C) 2009-2012 Intel Coproration
> + *
> + * Authors:
> + *   Auke Kok <auke-jan.h.kok at intel.com>
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; version 2
> + * of the License.
> + */
> +
> +
> +#include <sys/time.h>
> +#include <sys/types.h>
> +#include <sys/resource.h>
> +#include <stdio.h>
> +#include <signal.h>
> +#include <stdlib.h>
> +#include <string.h>
> +#include <unistd.h>
> +#include <time.h>
> +#include <getopt.h>
> +#include <limits.h>
> +#include <errno.h>
> +
> +
> +#include "bootchart.h"
> +
> +double graph_start;
> +double log_start;
> +double sampletime[MAXSAMPLES];
> +struct ps_struct *ps_first;
> +struct block_stat_struct blockstat[MAXSAMPLES];
> +int entropy_avail[MAXSAMPLES];
> +struct cpu_stat_struct cpustat[MAXCPUS];
> +int pscount;
> +int cpus;
> +double interval;
> +FILE *of;
> +int overrun = 0;
> +static int exiting = 0;
> +
> +/* graph defaults */
> +int entropy = 0;
> +int initcall = 1;
> +int relative;
> +int filter = 1;
> +int pss = 0;
> +int samples;
> +int len = 500; /* we record len+1 (1 start sample) */
> +double hz = 25.0;   /* 20 seconds log time */
> +double scale_x = 100.0; /* 100px = 1sec */
> +double scale_y = 20.0;  /* 16px = 1 process bar */
> +
> +char init_path[PATH_MAX] = "/sbin/init";
> +char output_path[PATH_MAX] = "/var/log";
> +
> +static struct rlimit rlim;
> +
> +static void signal_handler(int sig)
> +{
> +	if (sig++)
> +		sig--;
> +	exiting = 1;
> +}
> +
> +
> +int main(int argc, char *argv[])
> +{
> +	struct sigaction sig;
> +	struct ps_struct *ps;
> +	char output_file[PATH_MAX];
> +	char datestr[200];
> +	time_t t;
> +	FILE *f;
FILE _cleanup_fclose_ *f = NULL;

> +	int gind;
> +	int i;
> +
> +	memset(&t, 0, sizeof(time_t));
> +
> +	rlim.rlim_cur = 4096;
> +	rlim.rlim_max = 4096;
> +	(void) setrlimit(RLIMIT_NOFILE, &rlim);
> +
> +	f = fopen("/etc/systemd/bootchart.conf", "r");
"re"

> +	if (f) {
> +		char buf[256];
> +		char *key;
> +		char *val;
> +
> +		while (fgets(buf, 80, f) != NULL) {
> +			char *c;
> +
> +			c = strchr(buf, '\n');
> +			if (c) *c = 0; /* remove trailing \n */
> +
> +			if (buf[0] == '#')
> +				continue; /* comment line */
> +
> +			key = strtok(buf, "=");
> +			if (!key)
> +				continue;
> +			val = strtok(NULL, "=");
> +			if (!val)
> +				continue;
> +
> +			// todo: filter leading/trailing whitespace
> +
> +			if (!strcmp(key, "samples"))
!strcmp -> streq

> +				len = atoi(val);
> +			if (!strcmp(key, "freq"))
> +				hz = atof(val);
> +			if (!strcmp(key, "rel"))
> +				relative = atoi(val);
> +			if (!strcmp(key, "filter"))
> +				filter = atoi(val);
> +			if (!strcmp(key, "pss"))
> +				pss = atoi(val);
> +			if (!strcmp(key, "output"))
> +				strncpy(output_path, val, PATH_MAX - 1);
Is this copy at all necessary? Right now it gives an (potentially)
unterminated string. But argv should not go away, so output_path = val
should be enough.

> +			if (!strcmp(key, "init"))
> +				strncpy(init_path, val, PATH_MAX - 1);
Likewise.

> +			if (!strcmp(key, "scale_x"))
> +				scale_x = atof(val);
> +			if (!strcmp(key, "scale_y"))
> +				scale_y = atof(val);
> +			if (!strcmp(key, "entropy"))
> +				entropy = atoi(val);
> +		}
> +		fclose(f);
> +	}
> +
> +	while (1) {
> +		static struct option opts[] = {
> +			{"rel", 0, NULL, 'r'},
> +			{"freq", 1, NULL, 'f'},
> +			{"samples", 1, NULL, 'n'},
> +			{"pss", 0, NULL, 'p'},
> +			{"output", 1, NULL, 'o'},
> +			{"init", 1, NULL, 'i'},
> +			{"filter", 0, NULL, 'F'},
> +			{"help", 0, NULL, 'h'},
> +			{"scale-x", 1, NULL, 'x'},
> +			{"scale-y", 1, NULL, 'y'},
> +			{"entropy", 0, NULL, 'e'},
> +			{NULL, 0, NULL, 0}
> +		};
> +
> +		gind = 0;
> +
> +		i = getopt_long(argc, argv, "erpf:n:o:i:Fhx:y:", opts, &gind);
> +		if (i == -1)
> +			break;
> +		switch (i) {
> +		case 'r':
> +			relative = 1;
> +			break;
> +		case 'f':
> +			hz = atof(optarg);
> +			break;
> +		case 'F':
> +			filter = 0;
> +			break;
> +		case 'n':
> +			len = atoi(optarg);
> +			break;
> +		case 'o':
> +			strncpy(output_path, optarg, PATH_MAX - 1);
Likewise, no need for strncpy.

> +			break;
> +		case 'i':
> +			strncpy(init_path, optarg, PATH_MAX - 1);
Likewise, no need for strncpy.

> +			break;
> +		case 'p':
> +			pss = 1;
> +			break;
> +		case 'x':
> +			scale_x = atof(optarg);
> +			break;
> +		case 'y':
> +			scale_y = atof(optarg);
> +			break;
> +		case 'e':
> +			entropy = 1;
> +			break;
> +		case 'h':
> +			fprintf(stderr, "Usage: %s [OPTIONS]\n", argv[0]);
> +			fprintf(stderr, " --rel,     -r            Record time relative to recording\n");
> +			fprintf(stderr, " --freq,    -f N          Sample frequency [%f]\n", hz);
> +			fprintf(stderr, " --samples, -n N          Stop sampling at [%d] samples\n", len);
> +			fprintf(stderr, " --scale-x, -x N          Scale the graph horizontally [%f] \n", scale_x);
> +			fprintf(stderr, " --scale-y, -y N          Scale the graph vertically [%f] \n", scale_y);
> +			fprintf(stderr, " --pss,     -p            Enable PSS graph (CPU intensive)\n");
> +			fprintf(stderr, " --entropy, -e            Enable the entropy_avail graph\n");
> +			fprintf(stderr, " --output,  -o [PATH]     Path to output files [%s]\n", output_path);
> +			fprintf(stderr, " --init,    -i [PATH]     Path to init executable [%s]\n", init_path);
> +			fprintf(stderr, " --filter,  -F            Disable filtering of processes from the graph\n");
> +			fprintf(stderr, "                          that are of less importance or short-lived\n");
> +			fprintf(stderr, " --help,    -h            Display this message\n");
> +			fprintf(stderr, "See the installed README and bootchartd.conf.example for more information.\n");
GNU style says that --help output goes to stdout, when invoked
as --help or -h.

Just for readability, why not spell this as
fprintf(...,
	" --rel,     -r            Record time relative to recording\n"
	" --freq,    -f N          Sample frequency [%f]\n"
	...,
	hz
);

> +			exit (EXIT_SUCCESS);
> +			break;
> +		default:
> +			break;
> +		}
> +	}
> +
> +	if (len > MAXSAMPLES) {
> +		fprintf(stderr, "Error: samples exceeds maximum\n");
> +		exit(EXIT_FAILURE);
> +	}
> +
> +	if (hz <= 0.0) {
> +		fprintf(stderr, "Error: Frequency needs to be > 0\n");
> +		exit(EXIT_FAILURE);
> +	}
> +
> +	/*
> +	 * If the kernel executed us through init=/sbin/bootchartd, then
> +	 * fork:
> +	 * - parent execs executable specified via init_path[] (/sbin/init by default) as pid=1
> +	 * - child logs data
> +	 */
> +	if (getpid() == 1) {
> +		if (fork()) {
> +			/* parent */
> +			execl(init_path, init_path, NULL);
> +		}
> +	}
No error handling?

> +
> +	/* start with empty ps LL */
> +	ps_first = malloc(sizeof(struct ps_struct));
malloc+memset -> calloc

> +	if (!ps_first) {
> +		perror("malloc(ps_struct)");
> +		exit(EXIT_FAILURE);
> +	}
> +	memset(ps_first, 0, sizeof(struct ps_struct));

> +
> +	/* handle TERM/INT nicely */
> +	memset(&sig, 0, sizeof(struct sigaction));
> +	sig.sa_handler = signal_handler;
> +	sigaction(SIGHUP, &sig, NULL);
> +
> +	interval = (1.0 / hz) * 1000000000.0;
> +
> +	log_uptime();
> +
> +	/* main program loop */
> +	while (!exiting) {
> +		int res;
> +		double sample_stop;
> +		struct timespec req;
> +		time_t newint_s;
> +		long newint_ns;
> +		double elapsed;
> +		double timeleft;
> +
> +		sampletime[samples] = gettime_ns();
> +
> +		/* wait for /proc to become available, discarding samples */
> +		if (!(graph_start > 0.0))
Can't this be reversed and simplified?
> +			log_uptime();
> +		else
> +			log_sample(samples);
> +
> +		sample_stop = gettime_ns();
> +
> +		elapsed = (sample_stop - sampletime[samples]) * 1000000000.0;
> +		timeleft = interval - elapsed;
> +
> +		newint_s = (time_t)(timeleft / 1000000000.0);
> +		newint_ns = (long)(timeleft - (newint_s * 1000000000.0));
> +
> +		/*
> +		 * check if we have not consumed our entire timeslice. If we
> +		 * do, don't sleep and take a new sample right away.
> +		 * we'll lose all the missed samples and overrun our total
> +		 * time
> +		 */
> +		if ((newint_ns > 0) || (newint_s > 0)) {
> +			req.tv_sec = newint_s;
> +			req.tv_nsec = newint_ns;
> +
> +			res = nanosleep(&req, NULL);
> +			if (res) {
> +				if (errno == EINTR) {
> +					/* caught signal, probably HUP! */
> +					break;
> +				}
> +				perror("nanosleep()");
> +				exit (EXIT_FAILURE);
 > +			}
> +		} else {
> +			overrun++;
> +			/* calculate how many samples we lost and scrap them */
> +			len = len + ((int)(newint_ns / interval));
> +		}
> +
> +		samples++;
> +
> +		if (samples > len)
> +			break;
> +
> +	}
> +
> +	/* do some cleanup, close fd's */
> +	ps = ps_first;
> +	while (ps->next_ps) {
> +		ps = ps->next_ps;
> +		if (ps->schedstat)
> +			close(ps->schedstat);
The condition doesn't seem to be right, valid fd can be 0.

> +		if (ps->sched)
> +			close(ps->sched);
Here too.

> +		if (ps->smaps)
> +			fclose(ps->smaps);
> +	}
> +	closedir(proc);
This seems slightly risky, since proc is opened in a completely
different place and could be NULL.

> +
> +	t = time(NULL);
> +	strftime(datestr, sizeof(datestr), "%Y%m%d-%H%M", localtime(&t));
> +	snprintf(output_file, PATH_MAX, "%s/bootchart-%s.svg", output_path, datestr);
Hm, I understand that having a separate file is the most convinient,
but what about throwing it into the journal? That way it will be
automatically rotated. Maybe add an option to output to the journal.

> +
> +	of = fopen(output_file, "w");
> +	if (!of) {
> +		perror("open output_file");
> +		exit (EXIT_FAILURE);
> +	}
> +
> +	svg_do();
> +
> +	fprintf(stderr, "bootchartd: Wrote %s\n", output_file);
> +	fclose(of);
> +
> +	/* nitpic cleanups */
> +	ps = ps_first;
> +	while (ps->next_ps) {
> +		struct ps_struct *old = ps;
> +		ps = ps->next_ps;
> +		free(old->sample);
> +		free(old);
> +	}
> +	free(ps->sample);
> +	free(ps);
> +
> +	/* don't complain when overrun once, happens most commonly on 1st sample */
> +	if (overrun > 1)
> +		fprintf(stderr, "bootchartd: Warning: sample time overrun %i times\n", overrun);
> +
> +	return 0;
> +}
> diff --git a/src/bootchart/bootchart.conf b/src/bootchart/bootchart.conf
> new file mode 100644
> index 0000000..06c1b47
> --- /dev/null
> +++ b/src/bootchart/bootchart.conf
> @@ -0,0 +1,20 @@
> +
> +#  This file is part of systemd.
> +#
> +#  systemd is free software; you can redistribute it and/or modify it
> +#  under the terms of the GNU Lesser General Public License as published by
> +#  the Free Software Foundation; either version 2.1 of the License, or
> +#  (at your option) any later version.
> +#
> +# See systemd-bootchart.conf(5) for details
> +
> +#samples=500
> +#freq=25
> +#rel=0
> +#filter=1
> +#output=<folder name, defaults to /var/log>
> +#init=/path/to/init-binary
> +#pss=0
> +#entropy=0
> +#scale_x=100
> +#scale_y=20
> diff --git a/src/bootchart/bootchart.h b/src/bootchart/bootchart.h
> new file mode 100644
> index 0000000..0d8bed1
> --- /dev/null
> +++ b/src/bootchart/bootchart.h
> @@ -0,0 +1,117 @@
> +/*
> + * bootchart.h
> + *
> + * Copyright (C) 2009-2012 Intel Coproration
> + *
> + * Authors:
> + *   Auke Kok <auke-jan.h.kok at intel.com>
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; version 2
> + * of the License.
> + */
> +
> +#include <dirent.h>
> +
> +#define MAXCPUS        16
> +#define MAXPIDS     65535
> +#define MAXSAMPLES   8192
> +
> +
> +struct block_stat_struct {
> +	/* /proc/vmstat pgpgin & pgpgout */
> +	int bi;
> +	int bo;
> +};
> +
> +struct cpu_stat_sample_struct {
> +	/* /proc/schedstat fields 10 & 11 (after name) */
> +	double runtime;
> +	double waittime;
> +};
> +
> +struct cpu_stat_struct {
> +	/* per cpu array */
> +	struct cpu_stat_sample_struct sample[MAXSAMPLES];
> +};
> +
> +/* per process, per sample data we will log */
> +struct ps_sched_struct {
> +	/* /proc/<n>/schedstat fields 1 & 2 */
> +	double runtime;
> +	double waittime;
> +	int pss;
> +};
> +
> +/* process info */
> +struct ps_struct {
> +	struct ps_struct *next_ps;    /* SLL pointer */
> +	struct ps_struct *parent;     /* ppid ref */
> +	struct ps_struct *children;   /* children */
> +	struct ps_struct *next;       /* siblings */
> +
> +	/* must match - otherwise it's a new process with same PID */
> +	char name[16];
> +	int pid;
> +	int ppid;
> +
> +	/* cache fd's */
> +	int sched;
> +	int schedstat;
> +	FILE *smaps;
> +
> +	/* index to first/last seen timestamps */
> +	int first;
> +	int last;
> +
> +	/* records actual start time, may be way before bootchart runs */
> +	double starttime;
> +
> +	/* record human readable total cpu time */
> +	double total;
> +
> +	/* largest PSS size found */
> +	int pss_max;
> +
> +	/* for drawing connection lines later */
> +	double pos_x;
> +	double pos_y;
> +
> +	struct ps_sched_struct *sample;
> +};
> +
> +extern int entropy_avail[];
> +
> +extern double graph_start;
> +extern double log_start;
> +extern double sampletime[];
> +extern struct ps_struct *ps_first;
> +extern struct block_stat_struct blockstat[];
> +extern struct cpu_stat_struct cpustat[];
> +extern int pscount;
> +extern int relative;
> +extern int filter;
> +extern int pss;
> +extern int entropy;
> +extern int initcall;
> +extern int samples;
> +extern int cpus;
> +extern int len;
> +extern double hz;
> +extern double scale_x;
> +extern double scale_y;
> +extern int overrun;
> +extern double interval;
> +
> +extern char output_path[PATH_MAX];
> +extern char init_path[PATH_MAX];
> +
> +extern FILE *of;
> +extern DIR *proc;
> +
> +extern double gettime_ns(void);
> +extern void log_uptime(void);
> +extern void log_sample(int sample);
> +
> +extern void svg_do(void);
> diff --git a/src/bootchart/log.c b/src/bootchart/log.c
> new file mode 100644
> index 0000000..89c7b35
> --- /dev/null
> +++ b/src/bootchart/log.c
> @@ -0,0 +1,420 @@
> +/*
> + * log.c
> + *
> + * Copyright (C) 2009-2012 Intel Coproration
> + *
> + * Authors:
> + *   Auke Kok <auke-jan.h.kok at intel.com>
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; version 2
> + * of the License.
> + */
> +
> +#define _GNU_SOURCE 1
Always defined by the build system.

> +#include <unistd.h>
> +#include <stdlib.h>
> +#include <limits.h>
> +#include <sys/types.h>
> +#include <sys/stat.h>
> +#include <stdio.h>
> +#include <string.h>
> +#include <dirent.h>
> +#include <fcntl.h>
> +#include <time.h>
> +
> +
> +#include "bootchart.h"
> +
> +/*
> + * Alloc a static 4k buffer for stdio - primarily used to increase
> + * PSS buffering from the default 1k stdin buffer to reduce
> + * read() overhead.
> + */
> +static char smaps_buf[4096];
> +DIR *proc;
> +
> +
> +double gettime_ns(void)
> +{
> +	struct timespec now;
> +
> +	clock_gettime(CLOCK_MONOTONIC, &now);
> +
> +	return (now.tv_sec + (now.tv_nsec / 1000000000.0));
> +}
> +
> +
> +void log_uptime(void)
> +{
> +	FILE *f;
> +	char str[32];
> +	double uptime;
There's read_one_line_file() to read a line.

> +
> +	f = fopen("/proc/uptime", "r");
> +	if (!f)
> +		return;
> +	if (!fscanf(f, "%s %*s", str)) {
> +		fclose(f);
> +		return;
> +	}
> +	fclose(f);
> +	uptime = strtod(str, NULL);
> +
> +	log_start = gettime_ns();
> +
> +	/* start graph at kernel boot time */
> +	if (relative)
> +		graph_start = log_start;
> +	else
> +		graph_start = log_start - uptime;
> +}
> +
> +
> +static char *bufgetline(char *buf)
> +{
> +	char *c;
> +
> +	if (!buf)
> +		return NULL;
> +
> +	c = strchr(buf, '\n');
> +	if (c)
> +		c++;
> +	return c;
> +}
> +
> +
> +void log_sample(int sample)
> +{
> +	static int vmstat;
> +	static int schedstat;
> +	FILE *st;
> +	char buf[4095];
> +	char key[256];
> +	char val[256];
> +	char rt[256];
> +	char wt[256];
> +	char *m;
> +	int c;
> +	int p;
> +	int mod;
> +	static int e_fd;
> +	ssize_t s;
> +	ssize_t n;
> +	struct dirent *ent;
> +
> +	if (!vmstat) {
> +		/* block stuff */
> +		vmstat = open("/proc/vmstat", O_RDONLY);
> +		if (vmstat == -1) {
> +			perror("open /proc/vmstat");
> +			exit (EXIT_FAILURE);
> +		}
> +	}
> +
> +	n = pread(vmstat, buf, sizeof(buf) - 1, 0);
> +	if (n <= 0) {
> +		close(vmstat);
> +		return;
> +	}
> +	buf[n] = '\0';
> +
> +	m = buf;
> +	while (m) {
> +		if (sscanf(m, "%s %s", key, val) < 2)
> +			goto vmstat_next;
> +		if (!strcmp(key, "pgpgin"))
> +			blockstat[sample].bi = atoi(val);
> +		if (!strcmp(key, "pgpgout")) {
> +			blockstat[sample].bo = atoi(val);
> +			break;
> +		}
> +vmstat_next:
> +		m = bufgetline(m);
> +		if (!m)
> +			break;
> +	}
> +
> +	if (!schedstat) {
> +		/* overall CPU utilization */
> +		schedstat = open("/proc/schedstat", O_RDONLY);
> +		if (schedstat == -1) {
> +			perror("open /proc/schedstat");
> +			exit (EXIT_FAILURE);
> +		}
> +	}
> +
> +	n = pread(schedstat, buf, sizeof(buf) - 1, 0);
> +	if (n <= 0) {
> +		close(schedstat);
> +		return;
> +	}
> +	buf[n] = '\0';
> +
> +	m = buf;
> +	while (m) {
> +		if (sscanf(m, "%s %*s %*s %*s %*s %*s %*s %s %s", key, rt, wt) < 3)
> +			goto schedstat_next;
> +
> +		if (strstr(key, "cpu")) {
> +			c = atoi((const char*)(key+3));
> +			if (c > MAXCPUS)
> +				/* Oops, we only have room for MAXCPUS data */
> +				break;
> +			cpustat[c].sample[sample].runtime = atoll(rt);
> +			cpustat[c].sample[sample].waittime = atoll(wt);
> +
> +			if (c == cpus)
> +				cpus = c + 1;
> +		}
> +schedstat_next:
> +		m = bufgetline(m);
> +		if (!m)
> +			break;
> +	}
> +
> +	if (entropy) {
> +		if (!e_fd) {
> +			e_fd = open("/proc/sys/kernel/random/entropy_avail", O_RDONLY);
> +		}
> +
> +		if (e_fd) {
> +			n = pread(e_fd, buf, sizeof(buf) - 1, 0);
> +			if (n > 0)
> +				entropy_avail[sample] = atoi(buf);
> +		}
> +	}
> +
> +	/* all the per-process stuff goes here */
> +	if (!proc) {
> +		/* find all processes */
> +		proc = opendir("/proc");
> +		if (!proc)
> +			return;
> +	} else {
> +		rewinddir(proc);
> +	}
> +
> +	while ((ent = readdir(proc)) != NULL) {
> +		char filename[PATH_MAX];
> +		int pid;
> +		struct ps_struct *ps;
> +
> +		if ((ent->d_name[0] < '0') || (ent->d_name[0] > '9'))
> +			continue;
> +
> +		pid = atoi(ent->d_name);
> +
> +		if (pid >= MAXPIDS)
> +			continue;
> +
> +		ps = ps_first;
> +		while (ps->next_ps) {
> +			ps = ps->next_ps;
> +			if (ps->pid == pid)
> +				break;
> +		}
> +
> +		/* end of our LL? then append a new record */
> +		if (ps->pid != pid) {
> +			char t[32];
> +			struct ps_struct *parent;
> +
> +			ps->next_ps = malloc(sizeof(struct ps_struct));
malloc + memset -> calloc
> +			if (!ps->next_ps) {
> +				perror("malloc(ps_struct)");
> +				exit (EXIT_FAILURE);
> +			}
> +			memset(ps->next_ps, 0, sizeof(struct ps_struct));
> +			ps = ps->next_ps;
> +			ps->pid = pid;
> +
> +			ps->sample = malloc(sizeof(struct ps_sched_struct) * (len + 1));
> +			if (!ps->sample) {
> +				perror("malloc(ps_struct)");
> +				exit (EXIT_FAILURE);
> +			}
> +			memset(ps->sample, 0, sizeof(struct ps_sched_struct) * (len + 1));
> +
> +			pscount++;
> +
> +			/* mark our first sample */
> +			ps->first = sample;
> +
> +			/* get name, start time */
> +			if (!ps->sched) {
> +				sprintf(filename, "/proc/%d/sched", pid);
> +				ps->sched = open(filename, O_RDONLY);
> +				if (ps->sched == -1)
> +					continue;
> +			}
> +
> +			s = pread(ps->sched, buf, sizeof(buf) - 1, 0);
> +			if (s <= 0) {
> +				close(ps->sched);
> +				continue;
> +			}
> +
> +			if (!sscanf(buf, "%s %*s %*s", key))
> +				continue;
> +
> +			strncpy(ps->name, key, 16);
> +			/* discard line 2 */
> +			m = bufgetline(buf);
> +			if (!m)
> +				continue;
> +
> +			m = bufgetline(m);
> +			if (!m)
> +				continue;
> +
> +			if (!sscanf(m, "%*s %*s %s", t))
> +				continue;
> +
> +			ps->starttime = strtod(t, NULL) / 1000.0;
> +
> +			/* ppid */
> +			sprintf(filename, "/proc/%d/stat", pid);
> +			st = fopen(filename, "r");
> +			if (!st)
> +				continue;
> +			if (!fscanf(st, "%*s %*s %*s %i", &p)) {
> +				fclose(st);
> +				continue;
> +			}
> +			fclose(st);
> +			ps->ppid = p;
> +
> +			/*
> +			 * setup child pointers
> +			 *
> +			 * these are used to paint the tree coherently later
> +			 * each parent has a LL of children, and a LL of siblings
> +			 */
> +			if (pid == 1)
> +				continue; /* nothing to do for init atm */
> +
> +			/* kthreadd has ppid=0, which breaks our tree ordering */
> +			if (ps->ppid == 0)
> +				ps->ppid = 1;
> +
> +			parent = ps_first;
> +			while ((parent->next_ps && parent->pid != ps->ppid))
> +				parent = parent->next_ps;
> +
> +			if ((!parent) || (parent->pid != ps->ppid)) {
> +				/* orphan */
> +				ps->ppid = 1;
> +				parent = ps_first->next_ps;
> +			}
> +
> +			ps->parent = parent;
> +
> +			if (!parent->children) {
> +				/* it's the first child */
> +				parent->children = ps;
> +			} else {
> +				/* walk all children and append */
> +				struct ps_struct *children;
> +				children = parent->children;
> +				while (children->next)
> +					children = children->next;
> +				children->next = ps;
> +			}
> +		}
> +
> +		/* else -> found pid, append data in ps */
> +
> +		/* below here is all continuous logging parts - we get here on every
> +		 * iteration */
> +
> +		/* rt, wt */
> +		if (!ps->schedstat) {
> +			sprintf(filename, "/proc/%d/schedstat", pid);
oom check?

> +			ps->schedstat = open(filename, O_RDONLY);
> +			if (ps->schedstat == -1)
> +				continue;
> +		}
> +
> +		if (pread(ps->schedstat, buf, sizeof(buf) - 1, 0) <= 0) {
> +			/* clean up our file descriptors - assume that the process exited */
> +			close(ps->schedstat);
> +			if (ps->sched)
> +				close(ps->sched);
> +			//if (ps->smaps)
> +			//	fclose(ps->smaps);
?

> +			continue;
> +		}
> +		if (!sscanf(buf, "%s %s %*s", rt, wt))
> +			continue;
> +
> +		ps->last = sample;
> +		ps->sample[sample].runtime = atoll(rt);
> +		ps->sample[sample].waittime = atoll(wt);
> +
> +		ps->total = (ps->sample[ps->last].runtime
> +				 - ps->sample[ps->first].runtime)
> +				 / 1000000000.0;
> +
> +		if (!pss)
> +			goto catch_rename;
> +		/* Pss */
> +		if (!ps->smaps) {
> +			sprintf(filename, "/proc/%d/smaps", pid);
oom_check?

> +			ps->smaps = fopen(filename, "r");
> +			setvbuf(ps->smaps, smaps_buf, _IOFBF, sizeof(smaps_buf));
> +			if (!ps->smaps)
> +				continue;
> +		} else {
> +			rewind(ps->smaps);
> +		}
> +
> +		while (1) {
> +			int pss_kb;
> +
> +			/* skip one line, this contains the object mapped */
> +			if (fgets(buf, sizeof(buf), ps->smaps) == NULL)
> +				break;
> +			/* then there's a 28 char 14 line block */
> +			if (fread(buf, 1, 28 * 14, ps->smaps) != 28 * 14)
> +				break;
> +
> +			pss_kb = atoi(&buf[61]);
> +			ps->sample[sample].pss += pss_kb;
> +		}
> +
> +		if (ps->sample[sample].pss > ps->pss_max)
> +			ps->pss_max = ps->sample[sample].pss;
> +
> +catch_rename:
> +		/* catch process rename, try to randomize time */
> +		mod = (hz < 4.0) ? 4.0 : (hz / 4.0);
> +		if (((samples - ps->first) + pid) % (int)(mod) == 0) {
> +
> +			/* re-fetch name */
> +			/* get name, start time */
> +			if (!ps->sched) {
> +				sprintf(filename, "/proc/%d/sched", pid);
oom check?
> +				ps->sched = open(filename, O_RDONLY);
> +				if (ps->sched == -1)
> +					continue;
> +			}
> +			if (pread(ps->sched, buf, sizeof(buf) - 1, 0) <= 0) {
> +				/* clean up file descriptors */
> +				close(ps->sched);
> +				if (ps->schedstat)
> +					close(ps->schedstat);
> +				//if (ps->smaps)
> +				//	fclose(ps->smaps);
> +				continue;
> +			}
> +
> +			if (!sscanf(buf, "%s %*s %*s", key))
> +				continue;
> +
> +			strncpy(ps->name, key, 16);
> +		}
> +	}
> +}
> diff --git a/src/bootchart/svg.c b/src/bootchart/svg.c
> new file mode 100644
> index 0000000..68ec539
> --- /dev/null
> +++ b/src/bootchart/svg.c
> @@ -0,0 +1,1120 @@
> +/*
> + * svg.c
> + *
> + * Copyright (C) 2009-2012 Intel Coproration
> + *
> + * Authors:
> + *   Auke Kok <auke-jan.h.kok at intel.com>
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; version 2
> + * of the License.
> + */
> +
> +#include <stdio.h>
> +#include <stdarg.h>
> +#include <stdlib.h>
> +#include <string.h>
> +#include <time.h>
> +#include <limits.h>
> +#include <unistd.h>
> +#include <sys/utsname.h>
> +
> +#include "bootchart.h"
> +
> +
> +#define time_to_graph(t) ((t) * scale_x)
> +#define ps_to_graph(n) ((n) * scale_y)
> +#define kb_to_graph(m) ((m) * scale_y * 0.0001)
> +#define to_color(n) (192.0 - ((n) * 192.0))
> +
> +#define max(x, y) (((x) > (y)) ? (x) : (y))
> +#define min(x, y) (((x) < (y)) ? (x) : (y))
> +
> +static char str[8092];
> +
> +#define svg(a...) do { snprintf(str, 8092, ## a); fputs(str, of); fflush(of); } while (0)
Why not normal fprintf?

> +
> +static const char *colorwheel[12] = {
> +	"rgb(255,32,32)",  // red
> +	"rgb(32,192,192)", // cyan
> +	"rgb(255,128,32)", // orange
> +	"rgb(128,32,192)", // blue-violet
> +	"rgb(255,255,32)", // yellow
> +	"rgb(192,32,128)", // red-violet
> +	"rgb(32,255,32)",  // green
> +	"rgb(255,64,32)",  // red-orange
> +	"rgb(32,32,255)",  // blue
> +	"rgb(255,192,32)", // yellow-orange
> +	"rgb(192,32,192)", // violet
> +	"rgb(32,192,32)"   // yellow-green
> +};
> +
> +static double idletime = -1.0;
> +static int pfiltered = 0;
> +static int pcount = 0;
> +static int kcount = 0;
> +static float psize = 0;
> +static float ksize = 0;
> +static float esize = 0;
> +
> +
> +static void svg_header(void)
> +{
> +	float w;
> +	float h;
> +
> +	/* min width is about 1600px due to the label */
> +	w = 150.0 + 10.0 + time_to_graph(sampletime[samples-1] - graph_start);
> +	w = ((w < 1600.0) ? 1600.0 : w);
> +
> +	/* height is variable based on pss, psize, ksize */
> +	h = 400.0 + (scale_y * 30.0) /* base graphs and title */
> +	    + (pss ? (100.0 * scale_y) + (scale_y * 7.0) : 0.0) /* pss estimate */
> +	    + psize + ksize + esize;
> +
> +	svg("<?xml version=\"1.0\" standalone=\"no\"?>\n");
> +	svg("<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\" ");
> +	svg("\"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n");
> +
> +	//svg("<g transform=\"translate(10,%d)\">\n", 1000 + 150 + (pcount * 20));
> +	svg("<svg width=\"%.0fpx\" height=\"%.0fpx\" version=\"1.1\" ",
> +	    w, h);
> +	svg("xmlns=\"http://www.w3.org/2000/svg\">\n\n");
> +
> +	/* write some basic info as a comment, including some help */
> +	svg("<!-- This file is a bootchart SVG file. It is best rendered in a browser -->\n");
> +	svg("<!-- such as Chrome/Chromium, firefox. Other applications that render    -->\n");
> +	svg("<!-- these files properly but much more slow are ImageMagick, gimp,      -->\n");
> +	svg("<!-- inkscape, etc.. To display the files on your system, just point     -->\n");
> +	svg("<!-- your browser to file:///var/log/ and click. This bootchart was      -->\n\n");
> +
> +	svg("<!-- generated by bootchart version %s, running with options:  -->\n", VERSION);
> +	svg("<!-- hz=\"%f\" n=\"%d\" -->\n", hz, len);
> +	svg("<!-- x=\"%f\" y=\"%f\" -->\n", scale_x, scale_y);
> +	svg("<!-- rel=\"%d\" f=\"%d\" -->\n", relative, filter);
> +	svg("<!-- p=\"%d\" e=\"%d\" -->\n", pss, entropy);
> +	svg("<!-- o=\"%s\" i=\"%s\" -->\n\n", output_path, init_path);
> +
> +	/* style sheet */
> +	svg("<defs>\n  <style type=\"text/css\">\n    <![CDATA[\n");
> +
> +	svg("      rect       { stroke-width: 1; }\n");
> +	svg("      rect.cpu   { fill: rgb(64,64,240); stroke-width: 0; fill-opacity: 0.7; }\n");
> +	svg("      rect.wait  { fill: rgb(240,240,0); stroke-width: 0; fill-opacity: 0.7; }\n");
> +	svg("      rect.bi    { fill: rgb(240,128,128); stroke-width: 0; fill-opacity: 0.7; }\n");
> +	svg("      rect.bo    { fill: rgb(192,64,64); stroke-width: 0; fill-opacity: 0.7; }\n");
> +	svg("      rect.ps    { fill: rgb(192,192,192); stroke: rgb(128,128,128); fill-opacity: 0.7; }\n");
> +	svg("      rect.krnl  { fill: rgb(240,240,0); stroke: rgb(128,128,128); fill-opacity: 07; }\n");
> +	svg("      rect.box   { fill: rgb(240,240,240); stroke: rgb(192,192,192); }\n");
> +	svg("      rect.clrw  { stroke-width: 0; fill-opacity: 0.7;}\n");
> +	svg("      line       { stroke: rgb(64,64,64); stroke-width: 1; }\n");
> +	svg("//    line.sec1  { }\n");
> +	svg("      line.sec5  { stroke-width: 2; }\n");
> +	svg("      line.sec01 { stroke: rgb(224,224,224); stroke-width: 1; }\n");
> +	svg("      line.dot   { stroke-dasharray: 2 4; }\n");
> +	svg("      line.idle  { stroke: rgb(64,64,64); stroke-dasharray: 10 6; stroke-opacity: 0.7; }\n");
> +
> +	svg("      .run       { font-size: 8; font-style: italic; }\n");
> +	svg("      text       { font-family: Verdana, Helvetica; font-size: 10; }\n");
> +	svg("      text.sec   { font-size: 8; }\n");
> +	svg("      text.t1    { font-size: 24; }\n");
> +	svg("      text.t2    { font-size: 12; }\n");
> +	svg("      text.idle  { font-size: 18; }\n");
> +
> +	svg("    ]]>\n   </style>\n</defs>\n\n");
> +
> +}
> +
> +
> +static void svg_title(void)
> +{
> +	char cmdline[256] = "";
> +	char filename[PATH_MAX];
> +	char buf[256];
> +	char rootbdev[16] = "Unknown";
> +	char model[256] = "Unknown";
> +	char date[256] = "Unknown";
> +	char cpu[256] = "Unknown";
> +	char build[256] = "Unknown";
> +	char *c;
> +	FILE *f;
> +	time_t t;
> +	struct utsname uts;
> +
> +	/* grab /proc/cmdline */
> +	f = fopen("/proc/cmdline", "r");
> +	if (f) {
> +		if (!fgets(cmdline, 255, f))
> +			sprintf(cmdline, "Unknown");
> +		fclose(f);
> +	}
> +
> +	/* extract root fs so we can find disk model name in sysfs */
> +	c = strstr(cmdline, "root=/dev/");
> +	if (c) {
> +		strncpy(rootbdev, &c[10], 3);
> +		rootbdev[3] = '\0';
> +	}
> +	sprintf(filename, "/sys/block/%s/device/model", rootbdev);
> +	f = fopen(filename, "r");
> +	if (f) {
> +		if (!fgets(model, 255, f))
> +			fprintf(stderr, "Error reading disk model for %s\n", rootbdev);
> +		fclose(f);
> +	}
> +
> +	/* various utsname parameters */
> +	if (uname(&uts))
> +		fprintf(stderr, "Error getting uname info\n");
> +
> +	/* date */
> +	t = time(NULL);
> +	strftime(date, sizeof(date), "%a, %d %b %Y %H:%M:%S %z", localtime(&t));
> +
> +	/* CPU type */
> +	f = fopen("/proc/cpuinfo", "r");
> +	if (f) {
> +		while (fgets(buf, 255, f)) {
> +			if (strstr(buf, "model name")) {
> +				strncpy(cpu, &buf[13], 255);
> +				break;
> +			}
> +		}
> +		fclose(f);
> +	}
> +
> +	/* Build - 1st line from /etc/system-release */
> +	f = fopen("/etc/system-release", "r");
> +	if (f) {
> +		if (fgets(buf, 255, f))
> +			strncpy(build, buf, 255);
> +		fclose(f);
> +	}
> +
> +	svg("<text class=\"t1\" x=\"0\" y=\"30\">Bootchart for %s - %s</text>\n",
> +	    uts.nodename, date);
> +	svg("<text class=\"t2\" x=\"20\" y=\"50\">System: %s %s %s %s</text>\n",
> +	    uts.sysname, uts.release, uts.version, uts.machine);
> +	svg("<text class=\"t2\" x=\"20\" y=\"65\">CPU: %s</text>\n",
> +	    cpu);
> +	svg("<text class=\"t2\" x=\"20\" y=\"80\">Disk: %s</text>\n",
> +	    model);
> +	svg("<text class=\"t2\" x=\"20\" y=\"95\">Boot options: %s</text>\n",
> +	    cmdline);
> +	svg("<text class=\"t2\" x=\"20\" y=\"110\">Build: %s</text>\n",
> +	    build);
> +	svg("<text class=\"t2\" x=\"20\" y=\"125\">Log start time: %.03fs</text>\n", log_start);
> +	svg("<text class=\"t2\" x=\"20\" y=\"140\">Idle time: ");
> +
> +	if (idletime >= 0.0)
> +		svg("%.03fs", idletime);
> +	else
> +		svg("Not detected");
> +	svg("</text>\n");
> +	svg("<text class=\"sec\" x=\"20\" y=\"155\">Graph data: %.03f samples/sec, recorded %i total, dropped %i samples, %i processes, %i filtered</text>\n",
> +	    hz, len, overrun, pscount, pfiltered);
> +}
> +
> +
> +static void svg_graph_box(int height)
> +{
> +	double d = 0.0;
> +	int i = 0;
> +
> +	/* outside box, fill */
> +	svg("<rect class=\"box\" x=\"%.03f\" y=\"0\" width=\"%.03f\" height=\"%.03f\" />\n",
> +	    time_to_graph(0.0),
> +	    time_to_graph(sampletime[samples-1] - graph_start),
> +	    ps_to_graph(height));
> +
> +	for (d = graph_start; d <= sampletime[samples-1];
> +	     d += (scale_x < 2.0 ? 60.0 : scale_x < 10.0 ? 1.0 : 0.1)) {
> +		/* lines for each second */
> +		if (i % 50 == 0)
> +			svg("  <line class=\"sec5\" x1=\"%.03f\" y1=\"0\" x2=\"%.03f\" y2=\"%.03f\" />\n",
> +			    time_to_graph(d - graph_start),
> +			    time_to_graph(d - graph_start),
> +			    ps_to_graph(height));
> +		else if (i % 10 == 0)
> +			svg("  <line class=\"sec1\" x1=\"%.03f\" y1=\"0\" x2=\"%.03f\" y2=\"%.03f\" />\n",
> +			    time_to_graph(d - graph_start),
> +			    time_to_graph(d - graph_start),
> +			    ps_to_graph(height));
> +		else
> +			svg("  <line class=\"sec01\" x1=\"%.03f\" y1=\"0\" x2=\"%.03f\" y2=\"%.03f\" />\n",
> +			    time_to_graph(d - graph_start),
> +			    time_to_graph(d - graph_start),
> +			    ps_to_graph(height));
> +
> +		/* time label */
> +		if (i % 10 == 0)
> +			svg("  <text class=\"sec\" x=\"%.03f\" y=\"%.03f\" >%.01fs</text>\n",
> +			    time_to_graph(d - graph_start),
> +			    -5.0,
> +			    d - graph_start);
> +
> +		i++;
> +	}
> +}
> +
> +
> +static void svg_pss_graph(void)
> +{
> +	struct ps_struct *ps;
> +	int i;
> +
> +	svg("\n\n<!-- Pss memory size graph -->\n");
> +
> +	svg("\n  <text class=\"t2\" x=\"5\" y=\"-15\">Memory allocation - Pss</text>\n");
> +
> +	/* vsize 1000 == 1000mb */
> +	svg_graph_box(100);
> +	/* draw some hlines for usable memory sizes */
> +	for (i = 100000; i < 1000000; i += 100000) {
> +		svg("  <line class=\"sec01\" x1=\"%.03f\" y1=\"%.0f\" x2=\"%.03f\" y2=\"%.0f\"/>\n",
> +			time_to_graph(.0),
> +			kb_to_graph(i),
> +			time_to_graph(sampletime[samples-1] - graph_start),
> +			kb_to_graph(i));
> +		svg("  <text class=\"sec\" x=\"%.03f\" y=\"%.0f\">%dM</text>\n",
> +		    time_to_graph(sampletime[samples-1] - graph_start) + 5,
> +		    kb_to_graph(i), (1000000 - i) / 1000);
> +	}
> +	svg("\n");
> +
> +	/* now plot the graph itself */
> +	for (i = 1; i < samples ; i++) {
> +		int bottom;
> +		int top;
> +
> +		bottom = 0;
> +		top = 0;
> +
> +		/* put all the small pss blocks into the bottom */
> +		ps = ps_first;
> +		while (ps->next_ps) {
> +			ps = ps->next_ps;
> +			if (!ps)
> +				continue;
> +			if (ps->sample[i].pss <= (100 * scale_y))
> +				top += ps->sample[i].pss;
> +		};
> +		svg("    <rect class=\"clrw\" style=\"fill: %s\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +		    "rgb(64,64,64)",
> +		    time_to_graph(sampletime[i - 1] - graph_start),
> +		    kb_to_graph(1000000.0 - top),
> +		    time_to_graph(sampletime[i] - sampletime[i - 1]),
> +		    kb_to_graph(top - bottom));
> +
> +		bottom = top;
> +
> +		/* now plot the ones that are of significant size */
> +		ps = ps_first;
> +		while (ps->next_ps) {
> +			ps = ps->next_ps;
> +			if (!ps)
> +				continue;
> +			/* don't draw anything smaller than 2mb */
> +			if (ps->sample[i].pss > (100 * scale_y)) {
> +				top = bottom + ps->sample[i].pss;
> +				svg("    <rect class=\"clrw\" style=\"fill: %s\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +				    colorwheel[ps->pid % 12],
> +				    time_to_graph(sampletime[i - 1] - graph_start),
> +				    kb_to_graph(1000000.0 - top),
> +				    time_to_graph(sampletime[i] - sampletime[i - 1]),
> +				    kb_to_graph(top - bottom));
> +				bottom = top;
> +			}
> +		}
> +	}
> +
> +	/* overlay all the text labels */
> +	for (i = 1; i < samples ; i++) {
> +		int bottom;
> +		int top;
> +
> +		bottom = 0;
> +		top = 0;
> +
> +		/* put all the small pss blocks into the bottom */
> +		ps = ps_first;
> +		while (ps->next_ps) {
> +			ps = ps->next_ps;
> +			if (!ps)
> +				continue;
> +			if (ps->sample[i].pss <= (100 * scale_y))
> +				top += ps->sample[i].pss;
> +		};
> +
> +		bottom = top;
> +
> +		/* now plot the ones that are of significant size */
> +		ps = ps_first;
> +		while (ps->next_ps) {
> +			ps = ps->next_ps;
> +			if (!ps)
> +				continue;
> +			/* don't draw anything smaller than 2mb */
> +			if (ps->sample[i].pss > (100 * scale_y)) {
> +				top = bottom + ps->sample[i].pss;
> +				/* draw a label with the process / PID */
> +				if ((i == 1) || (ps->sample[i - 1].pss <= (100 * scale_y)))
> +					svg("  <text x=\"%.03f\" y=\"%.03f\">%s [%i]</text>\n",
> +					    time_to_graph(sampletime[i] - graph_start),
> +					    kb_to_graph(1000000.0 - bottom - ((top -  bottom) / 2)),
> +					    ps->name,
> +					    ps->pid);
> +				bottom = top;
> +			}
> +		}
> +	}
> +
> +	/* debug output - full data dump */
> +	svg("\n\n<!-- PSS map - csv format -->\n");
> +	ps = ps_first;
> +	while (ps->next_ps) {
> +		ps = ps->next_ps;
> +		if (!ps)
> +			continue;
> +		svg("<!-- %s [%d] pss=", ps->name, ps->pid);
> +		for (i = 0; i < samples ; i++) {
> +			svg("%d," , ps->sample[i].pss);
> +		}
> +		svg(" -->\n");
> +	}
> +
> +}
> +
> +static void svg_io_bi_bar(void)
> +{
> +	double max = 0.0;
> +	double range;
> +	int max_here = 0;
> +	int i;
> +
> +	svg("<!-- IO utilization graph - In -->\n");
> +
> +	svg("<text class=\"t2\" x=\"5\" y=\"-15\">IO utilization - read</text>\n");
> +
> +	/*
> +	 * calculate rounding range
> +	 *
> +	 * We need to round IO data since IO block data is not updated on
> +	 * each poll. Applying a smoothing function loses some burst data,
> +	 * so keep the smoothing range short.
> +	 */
> +	range = 0.25 / (1.0 / hz);
> +	if (range < 2.0)
> +		range = 2.0; /* no smoothing */
> +
> +	/* surrounding box */
> +	svg_graph_box(5);
> +
> +	/* find the max IO first */
> +	for (i = 1; i < samples; i++) {
> +		int start;
> +		int stop;
> +		double tot;
> +
> +		start = max(i - ((range / 2) - 1), 0);
> +		stop = min(i + (range / 2), samples - 1);
> +
> +		tot = (double)(blockstat[stop].bi - blockstat[start].bi)
> +		      / (stop - start);
> +		if (tot > max) {
> +			max = tot;
> +			max_here = i;
> +		}
> +		tot = (double)(blockstat[stop].bo - blockstat[start].bo)
> +		      / (stop - start);
> +		if (tot > max)
> +			max = tot;
> +	}
> +
> +	/* plot bi */
> +	for (i = 1; i < samples; i++) {
> +		int start;
> +		int stop;
> +		double tot;
> +		double pbi;
> +
> +		start = max(i - ((range / 2) - 1), 0);
> +		stop = min(i + (range / 2), samples);
> +
> +		tot = (double)(blockstat[stop].bi - blockstat[start].bi)
> +		      / (stop - start);
> +		pbi = tot / max;
> +
> +		if (pbi > 0.001)
> +			svg("<rect class=\"bi\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +			    time_to_graph(sampletime[i - 1] - graph_start),
> +			    (scale_y * 5) - (pbi * (scale_y * 5)),
> +			    time_to_graph(sampletime[i] - sampletime[i - 1]),
> +			    pbi * (scale_y * 5));
> +
> +		/* labels around highest value */
> +		if (i == max_here) {
> +			svg("  <text class=\"sec\" x=\"%.03f\" y=\"%.03f\">%0.2fmb/sec</text>\n",
> +			    time_to_graph(sampletime[i] - graph_start) + 5,
> +			    ((scale_y * 5) - (pbi * (scale_y * 5))) + 15,
> +			    max / 1024.0 / (interval / 1000000000.0));
> +		}
> +	}
> +}
> +
> +static void svg_io_bo_bar(void)
> +{
> +	double max = 0.0;
> +	double range;
> +	int max_here = 0;
> +	int i;
> +
> +	svg("<!-- IO utilization graph - out -->\n");
> +
> +	svg("<text class=\"t2\" x=\"5\" y=\"-15\">IO utilization - write</text>\n");
> +
> +	/*
> +	 * calculate rounding range
> +	 *
> +	 * We need to round IO data since IO block data is not updated on
> +	 * each poll. Applying a smoothing function loses some burst data,
> +	 * so keep the smoothing range short.
> +	 */
> +	range = 0.25 / (1.0 / hz);
> +	if (range < 2.0)
> +		range = 2.0; /* no smoothing */
> +
> +	/* surrounding box */
> +	svg_graph_box(5);
> +
> +	/* find the max IO first */
> +	for (i = 1; i < samples; i++) {
> +		int start;
> +		int stop;
> +		double tot;
> +
> +		start = max(i - ((range / 2) - 1), 0);
> +		stop = min(i + (range / 2), samples - 1);
> +
> +		tot = (double)(blockstat[stop].bi - blockstat[start].bi)
> +		      / (stop - start);
> +		if (tot > max)
> +			max = tot;
> +		tot = (double)(blockstat[stop].bo - blockstat[start].bo)
> +		      / (stop - start);
> +		if (tot > max) {
> +			max = tot;
> +			max_here = i;
> +		}
> +	}
> +
> +	/* plot bo */
> +	for (i = 1; i < samples; i++) {
> +		int start;
> +		int stop;
> +		double tot;
> +		double pbo;
> +
> +		start = max(i - ((range / 2) - 1), 0);
> +		stop = min(i + (range / 2), samples);
> +
> +		tot = (double)(blockstat[stop].bo - blockstat[start].bo)
> +		      / (stop - start);
> +		pbo = tot / max;
> +
> +		if (pbo > 0.001)
> +			svg("<rect class=\"bo\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +			    time_to_graph(sampletime[i - 1] - graph_start),
> +			    (scale_y * 5) - (pbo * (scale_y * 5)),
> +			    time_to_graph(sampletime[i] - sampletime[i - 1]),
> +			    pbo * (scale_y * 5));
> +
> +		/* labels around highest bo value */
> +		if (i == max_here) {
> +			svg("  <text class=\"sec\" x=\"%.03f\" y=\"%.03f\">%0.2fmb/sec</text>\n",
> +			    time_to_graph(sampletime[i] - graph_start) + 5,
> +			    ((scale_y * 5) - (pbo * (scale_y * 5))),
> +			    max / 1024.0 / (interval / 1000000000.0));
> +		}
> +	}
> +}
> +
> +
> +static void svg_cpu_bar(void)
> +{
> +	int i;
> +
> +	svg("<!-- CPU utilization graph -->\n");
> +
> +	svg("<text class=\"t2\" x=\"5\" y=\"-15\">CPU utilization</text>\n");
> +	/* surrounding box */
> +	svg_graph_box(5);
> +
> +	/* bars for each sample, proportional to the CPU util. */
> +	for (i = 1; i < samples; i++) {
> +		int c;
> +		double trt;
> +		double ptrt;
> +
> +		ptrt = trt = 0.0;
> +
> +		for (c = 0; c < cpus; c++)
> +			trt += cpustat[c].sample[i].runtime - cpustat[c].sample[i - 1].runtime;
> +
> +		trt = trt / 1000000000.0;
> +
> +		trt = trt / (double)cpus;
> +
> +		if (trt > 0.0)
> +			ptrt = trt / (sampletime[i] - sampletime[i - 1]);
> +
> +		if (ptrt > 1.0)
> +			ptrt = 1.0;
> +
> +		if (ptrt > 0.001) {
> +			svg("<rect class=\"cpu\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +			    time_to_graph(sampletime[i - 1] - graph_start),
> +			    (scale_y * 5) - (ptrt * (scale_y * 5)),
> +			    time_to_graph(sampletime[i] - sampletime[i - 1]),
> +			    ptrt * (scale_y * 5));
> +		}
> +	}
> +}
> +
> +static void svg_wait_bar(void)
> +{
> +	int i;
> +
> +	svg("<!-- Wait time aggregation box -->\n");
> +
> +	svg("<text class=\"t2\" x=\"5\" y=\"-15\">CPU wait</text>\n");
> +
> +	/* surrounding box */
> +	svg_graph_box(5);
> +
> +	/* bars for each sample, proportional to the CPU util. */
> +	for (i = 1; i < samples; i++) {
> +		int c;
> +		double twt;
> +		double ptwt;
> +
> +		ptwt = twt = 0.0;
> +
> +		for (c = 0; c < cpus; c++)
> +			twt += cpustat[c].sample[i].waittime - cpustat[c].sample[i - 1].waittime;
> +
> +		twt = twt / 1000000000.0;
> +
> +		twt = twt / (double)cpus;
> +
> +		if (twt > 0.0)
> +			ptwt = twt / (sampletime[i] - sampletime[i - 1]);
> +
> +		if (ptwt > 1.0)
> +			ptwt = 1.0;
> +
> +		if (ptwt > 0.001) {
> +			svg("<rect class=\"wait\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +			    time_to_graph(sampletime[i - 1] - graph_start),
> +			    ((scale_y * 5) - (ptwt * (scale_y * 5))),
> +			    time_to_graph(sampletime[i] - sampletime[i - 1]),
> +			    ptwt * (scale_y * 5));
> +		}
> +	}
> +}
> +
> +
> +static void svg_entropy_bar(void)
> +{
> +	int i;
> +
> +	svg("<!-- entropy pool graph -->\n");
> +
> +	svg("<text class=\"t2\" x=\"5\" y=\"-15\">Entropy pool size</text>\n");
> +	/* surrounding box */
> +	svg_graph_box(5);
> +
> +	/* bars for each sample, scale 0-4096 */
> +	for (i = 1; i < samples; i++) {
> +		/* svg("<!-- entropy %.03f %i -->\n", sampletime[i], entropy_avail[i]); */
> +		svg("<rect class=\"cpu\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +		    time_to_graph(sampletime[i - 1] - graph_start),
> +		    ((scale_y * 5) - ((entropy_avail[i] / 4096.) * (scale_y * 5))),
> +		    time_to_graph(sampletime[i] - sampletime[i - 1]),
> +		    (entropy_avail[i] / 4096.) * (scale_y * 5));
> +	}
> +}
> +
> +
> +static struct ps_struct *get_next_ps(struct ps_struct *ps)
> +{
> +	/*
> +	 * walk the list of processes and return the next one to be
> +	 * painted
> +	 */
> +	if (ps == ps_first)
> +		return ps->next_ps;
> +
> +	/* go deep */
> +	if (ps->children)
> +		return ps->children;
> +
> +	/* find siblings */
> +	if (ps->next)
> +		return ps->next;
> +
> +	/* go back for parent siblings */
> +	while (1) {
> +		if (ps->parent)
> +			if (ps->parent->next)
> +				return ps->parent->next;
> +		ps = ps->parent;
> +		if (!ps)
> +			return ps;
> +	}
> +
> +	return NULL;
> +}
> +
> +
> +static int ps_filter(struct ps_struct *ps)
> +{
> +	if (!filter)
> +		return 0;
> +
> +	/* can't draw data when there is only 1 sample (need start + stop) */
> +	if (ps->first == ps->last)
> +		return -1;
> +
> +	/* don't filter kthreadd */
> +	if (ps->pid == 2)
> +		return 0;
> +
> +	/* drop stuff that doesn't use any real CPU time */
> +	if (ps->total <= 0.001)
> +		return -1;
> +
> +	return 0;
> +}
> +
> +
> +static void svg_do_initcall(int count_only)
> +{
> +	FILE *f;
> +	double t;
> +	char func[256];
> +	int ret;
> +	int usecs;
> +
> +	/* can't plot initcall when disabled or in relative mode */
> +	if (!initcall || relative) {
> +		kcount = 0;
> +		return;
> +	}
> +
> +	if (!count_only) {
> +		svg("<!-- initcall -->\n");
> +
> +		svg("<text class=\"t2\" x=\"5\" y=\"-15\">Kernel init threads</text>\n");
> +		/* surrounding box */
> +		svg_graph_box(kcount);
> +	}
> +
> +	kcount = 0;
> +
> +	/*
> +	 * Initcall graphing - parses dmesg buffer and displays kernel threads
> +	 * This somewhat uses the same methods and scaling to show processes
> +	 * but looks a lot simpler. It's overlaid entirely onto the PS graph
> +	 * when appropriate.
> +	 */
> +
> +	f = popen("dmesg", "r");
> +	if (!f)
> +		return;
> +
> +	while (!feof(f)) {
> +		int c;
> +		int z = 0;
> +		char l[256];
> +
> +		if (fgets(l, sizeof(l) - 1, f) == NULL)
> +			continue;
> +
> +		c = sscanf(l, "[%lf] initcall %s %*s %d %*s %d %*s",
> +			   &t, func, &ret, &usecs);
> +		if (c != 4) {
> +			/* also parse initcalls done by module loading */
> +			c = sscanf(l, "[%lf] initcall %s %*s %*s %d %*s %d %*s",
> +				   &t, func, &ret, &usecs);
> +			if (c != 4)
> +				continue;
> +		}
> +
> +		/* chop the +0xXX/0xXX stuff */
> +		while(func[z] != '+')
> +			z++;
> +		func[z] = 0;
> +
> +		if (count_only) {
> +			/* filter out irrelevant stuff */
> +			if (usecs >= 1000)
> +				kcount++;
> +			continue;
> +		}
> +
> +		svg("<!-- thread=\"%s\" time=\"%.3f\" elapsed=\"%d\" result=\"%d\" -->\n",
> +		    func, t, usecs, ret);
> +
> +		if (usecs < 1000)
> +			continue;
> +
> +		/* rect */
> +		svg("  <rect class=\"krnl\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +		    time_to_graph(t - (usecs / 1000000.0)),
> +		    ps_to_graph(kcount),
> +		    time_to_graph(usecs / 1000000.0),
> +		    ps_to_graph(1));
> +
> +		/* label */
> +		svg("  <text x=\"%.03f\" y=\"%.03f\">%s <tspan class=\"run\">%.03fs</tspan></text>\n",
> +		    time_to_graph(t - (usecs / 1000000.0)) + 5,
> +		    ps_to_graph(kcount) + 15,
> +		    func,
> +		    usecs / 1000000.0);
> +
> +		kcount++;
> +	}
> +
> +	fclose(f);
> +}
> +
> +
> +static void svg_ps_bars(void)
> +{
> +	struct ps_struct *ps;
> +	int i = 0;
> +	int j = 0;
> +	int w;
> +	int pid;
> +
> +	svg("<!-- Process graph -->\n");
> +
> +	svg("<text class=\"t2\" x=\"5\" y=\"-15\">Processes</text>\n");
> +
> +	/* surrounding box */
> +	svg_graph_box(pcount);
> +
> +	/* pass 2 - ps boxes */
> +	ps = ps_first;
> +	while ((ps = get_next_ps(ps))) {
> +		double starttime;
> +		int t;
> +
> +		if (!ps)
> +			continue;
> +
> +		/* leave some trace of what we actually filtered etc. */
> +		svg("<!-- %s [%i] ppid=%i runtime=%.03fs -->\n", ps->name, ps->pid,
> +		    ps->ppid, ps->total);
> +
> +		/* it would be nice if we could use exec_start from /proc/pid/sched,
> +		 * but it's unreliable and gives bogus numbers */
> +		starttime = sampletime[ps->first];
> +
> +		if (!ps_filter(ps)) {
> +			/* remember where _to_ our children need to draw a line */
> +			ps->pos_x = time_to_graph(starttime - graph_start);
> +			ps->pos_y = ps_to_graph(j+1); /* bottom left corner */
> +		} else {
> +			/* hook children to our parent coords instead */
> +			ps->pos_x = ps->parent->pos_x;
> +			ps->pos_y = ps->parent->pos_y;
> +
> +			/* if this is the last child, we might still need to draw a connecting line */
> +			if ((!ps->next) && (ps->parent))
> +				svg("  <line class=\"dot\" x1=\"%.03f\" y1=\"%.03f\" x2=\"%.03f\" y2=\"%.03f\" />\n",
> +				    ps->parent->pos_x,
> +				    ps_to_graph(j-1) + 10.0, /* whee, use the last value here */
> +				    ps->parent->pos_x,
> +				    ps->parent->pos_y);
> +			continue;
> +		}
> +
> +		svg("  <rect class=\"ps\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +		    time_to_graph(starttime - graph_start),
> +		    ps_to_graph(j),
> +		    time_to_graph(sampletime[ps->last] - starttime),
> +		    ps_to_graph(1));
> +
> +		/* paint cpu load over these */
> +		for (t = ps->first + 1; t < ps->last; t++) {
> +			double rt, prt;
> +			double wt, wrt;
> +
> +			/* calculate over interval */
> +			rt = ps->sample[t].runtime - ps->sample[t-1].runtime;
> +			wt = ps->sample[t].waittime - ps->sample[t-1].waittime;
> +
> +			prt = (rt / 1000000000) / (sampletime[t] - sampletime[t-1]);
> +			wrt = (wt / 1000000000) / (sampletime[t] - sampletime[t-1]);
> +
> +			/* this can happen if timekeeping isn't accurate enough */
> +			if (prt > 1.0)
> +				prt = 1.0;
> +			if (wrt > 1.0)
> +				wrt = 1.0;
> +
> +			if ((prt < 0.1) && (wrt < 0.1)) /* =~ 26 (color threshold) */
> +				continue;
> +
> +			svg("    <rect class=\"wait\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +			    time_to_graph(sampletime[t - 1] - graph_start),
> +			    ps_to_graph(j),
> +			    time_to_graph(sampletime[t] - sampletime[t - 1]),
> +			    ps_to_graph(wrt));
> +
> +			/* draw cpu over wait - TODO figure out how/why run + wait > interval */
> +			svg("    <rect class=\"cpu\" x=\"%.03f\" y=\"%.03f\" width=\"%.03f\" height=\"%.03f\" />\n",
> +			    time_to_graph(sampletime[t - 1] - graph_start),
> +			    ps_to_graph(j + (1.0 - prt)),
> +			    time_to_graph(sampletime[t] - sampletime[t - 1]),
> +			    ps_to_graph(prt));
> +		}
> +
> +		/* determine where to display the process name */
> +		if (sampletime[ps->last] - sampletime[ps->first] < 1.5)
> +			/* too small to fit label inside the box */
> +			w = ps->last;
> +		else
> +			w = ps->first;
> +
> +		/* text label of process name */
> +		svg("  <text x=\"%.03f\" y=\"%.03f\">%s [%i] <tspan class=\"run\">%.03fs</tspan></text>\n",
> +		    time_to_graph(sampletime[w] - graph_start) + 5.0,
> +		    ps_to_graph(j) + 14.0,
> +		    ps->name,
> +		    ps->pid,
> +		    (ps->sample[ps->last].runtime - ps->sample[ps->first].runtime) / 1000000000.0);
> +		/* paint lines to the parent process */
> +		if (ps->parent) {
> +			/* horizontal part */
> +			svg("  <line class=\"dot\" x1=\"%.03f\" y1=\"%.03f\" x2=\"%.03f\" y2=\"%.03f\" />\n",
> +			    time_to_graph(starttime - graph_start),
> +			    ps_to_graph(j) + 10.0,
> +			    ps->parent->pos_x,
> +			    ps_to_graph(j) + 10.0);
> +
> +			/* one vertical line connecting all the horizontal ones up */
> +			if (!ps->next)
> +				svg("  <line class=\"dot\" x1=\"%.03f\" y1=\"%.03f\" x2=\"%.03f\" y2=\"%.03f\" />\n",
> +				    ps->parent->pos_x,
> +				    ps_to_graph(j) + 10.0,
> +				    ps->parent->pos_x,
> +				    ps->parent->pos_y);
> +		}
> +
> +		j++; /* count boxes */
> +
> +		svg("\n");
> +	}
> +
> +	/* last pass - determine when idle */
> +	pid = getpid();
> +	/* make sure we start counting from the point where we actually have
> +	 * data: assume that bootchart's first sample is when data started
> +	 */
> +	ps = ps_first;
> +	while (ps->next_ps) {
> +		ps = ps->next_ps;
> +		if (ps->pid == pid)
> +			break;
> +	}
> +
> +	for (i = ps->first; i < samples - (hz / 2); i++) {
> +		double crt;
> +		double brt;
> +		int c;
> +
> +		/* subtract bootchart cpu utilization from total */
> +		crt = 0.0;
> +		for (c = 0; c < cpus; c++)
> +			crt += cpustat[c].sample[i + ((int)hz / 2)].runtime - cpustat[c].sample[i].runtime;
> +		brt = ps->sample[i + ((int)hz / 2)].runtime - ps->sample[i].runtime;
> +
> +		/*
> +		 * our definition of "idle":
> +		 *
> +		 * if for (hz / 2) we've used less CPU than (interval / 2) ...
> +		 * defaults to 4.0%, which experimentally, is where atom idles
> +		 */
> +		if ((crt - brt) < (interval / 2.0)) {
> +			idletime = sampletime[i] - graph_start;
> +			svg("\n<!-- idle detected at %.03f seconds -->\n",
> +			    idletime);
> +			svg("<line class=\"idle\" x1=\"%.03f\" y1=\"%.03f\" x2=\"%.03f\" y2=\"%.03f\" />\n",
> +			    time_to_graph(idletime),
> +			    -scale_y,
> +			    time_to_graph(idletime),
> +			    ps_to_graph(pcount) + scale_y);
> +			svg("<text class=\"idle\" x=\"%.03f\" y=\"%.03f\">%.01fs</text>\n",
> +			    time_to_graph(idletime) + 5.0,
> +			    ps_to_graph(pcount) + scale_y,
> +			    idletime);
> +			break;
> +		}
> +	}
> +}
> +
> +
> +static void svg_top_ten_cpu(void)
> +{
> +	struct ps_struct *top[10];
> +	struct ps_struct emptyps;
> +	struct ps_struct *ps;
> +	int n, m;
> +
> +	memset(&emptyps, 0, sizeof(struct ps_struct));
> +	for (n=0; n < 10; n++)
> +		top[n] = &emptyps;
> +
> +	/* walk all ps's and setup ptrs */
> +	ps = ps_first;
> +	while ((ps = get_next_ps(ps))) {
> +		for (n = 0; n < 10; n++) {
> +			if (ps->total <= top[n]->total)
> +				continue;
> +			/* cascade insert */
> +			for (m = 9; m > n; m--)
> +				top[m] = top[m-1];
> +			top[n] = ps;
> +			break;
> +		}
> +	}
> +
> +	svg("<text class=\"t2\" x=\"20\" y=\"0\">Top CPU consumers:</text>\n");
> +	for (n = 0; n < 10; n++)
> +		svg("<text class=\"t3\" x=\"20\" y=\"%d\">%3.03fs - %s[%d]</text>\n",
> +		    20 + (n * 13),
> +		    top[n]->total,
> +		    top[n]->name,
> +		    top[n]->pid);
> +}
> +
> +
> +static void svg_top_ten_pss(void)
> +{
> +	struct ps_struct *top[10];
> +	struct ps_struct emptyps;
> +	struct ps_struct *ps;
> +	int n, m;
> +
> +	memset(&emptyps, 0, sizeof(struct ps_struct));
> +	for (n=0; n < 10; n++)
> +		top[n] = &emptyps;
> +
> +	/* walk all ps's and setup ptrs */
> +	ps = ps_first;
> +	while ((ps = get_next_ps(ps))) {
> +		for (n = 0; n < 10; n++) {
> +			if (ps->pss_max <= top[n]->pss_max)
> +				continue;
> +			/* cascade insert */
> +			for (m = 9; m > n; m--)
> +				top[m] = top[m-1];
> +			top[n] = ps;
> +			break;
> +		}
> +	}
> +
> +	svg("<text class=\"t2\" x=\"20\" y=\"0\">Top PSS consumers:</text>\n");
> +	for (n = 0; n < 10; n++)
> +		svg("<text class=\"t3\" x=\"20\" y=\"%d\">%dK - %s[%d]</text>\n",
> +		    20 + (n * 13),
> +		    top[n]->pss_max,
> +		    top[n]->name,
> +		    top[n]->pid);
> +}
> +
> +
> +void svg_do(void)
> +{
> +	struct ps_struct *ps;
> +
> +	memset(&str, 0, sizeof(str));
> +
> +	ps = ps_first;
> +
> +	/* count initcall thread count first */
> +	svg_do_initcall(1);
> +	ksize = (kcount ? ps_to_graph(kcount) + (scale_y * 2) : 0);
> +
> +	/* then count processes */
> +	while ((ps = get_next_ps(ps))) {
> +		if (!ps_filter(ps))
> +			pcount++;
> +		else
> +			pfiltered++;
> +	}
> +	psize = ps_to_graph(pcount) + (scale_y * 2);
> +
> +	esize = (entropy ? scale_y * 7 : 0);
> +
> +	/* after this, we can draw the header with proper sizing */
> +	svg_header();
> +
> +	svg("<g transform=\"translate(10,400)\">\n");
> +	svg_io_bi_bar();
> +	svg("</g>\n\n");
> +
> +	svg("<g transform=\"translate(10,%.03f)\">\n", 400.0 + (scale_y * 7.0));
> +	svg_io_bo_bar();
> +	svg("</g>\n\n");
> +
> +	svg("<g transform=\"translate(10,%.03f)\">\n", 400.0 + (scale_y * 14.0));
> +	svg_cpu_bar();
> +	svg("</g>\n\n");
> +
> +	svg("<g transform=\"translate(10,%.03f)\">\n", 400.0 + (scale_y * 21.0));
> +	svg_wait_bar();
> +	svg("</g>\n\n");
> +
> +	if (kcount) {
> +		svg("<g transform=\"translate(10,%.03f)\">\n", 400.0 + (scale_y * 28.0));
> +		svg_do_initcall(0);
> +		svg("</g>\n\n");
> +	}
> +
> +	svg("<g transform=\"translate(10,%.03f)\">\n", 400.0 + (scale_y * 28.0) + ksize);
> +	svg_ps_bars();
> +	svg("</g>\n\n");
> +
> +	svg("<g transform=\"translate(10,  0)\">\n");
> +	svg_title();
> +	svg("</g>\n\n");
> +
> +	svg("<g transform=\"translate(10,200)\">\n");
> +	svg_top_ten_cpu();
> +	svg("</g>\n\n");
> +
> +	if (entropy) {
> +		svg("<g transform=\"translate(10,%.03f)\">\n", 400.0 + (scale_y * 28.0) + ksize + psize);
> +		svg_entropy_bar();
> +		svg("</g>\n\n");
> +	}
> +
> +	if (pss) {
> +		svg("<g transform=\"translate(10,%.03f)\">\n", 400.0 + (scale_y * 28.0) + ksize + psize + esize);
> +		svg_pss_graph();
> +		svg("</g>\n\n");
> +
> +		svg("<g transform=\"translate(410,200)\">\n");
> +		svg_top_ten_pss();
> +		svg("</g>\n\n");
> +	}
> +
> +	/* svg footer */
> +	svg("\n</svg>\n");
> +}
> -- 

HTH,
Zbyszek



More information about the systemd-devel mailing list