Re: [PATCH v2] scripts/ksize: Add kernel build size report

[josh@joshtriplett.org]

On Wed, Nov 19, 2014 at 09:01:19AM -0800, Darren Hart wrote:
> ksize generates hierarchical build size reports from vmlinux, *.o, and
> built-in.o files.
> 
> ksize is useful in preparing minimal configurations and comparing
> similar configurations across kernel versions.
> 
> Signed-off-by: Darren Hart <dvhart@linux.intel.com>
> Reviewed-by: Josh Triplett <josh@joshtriplett.org>

I've applied this patch to tiny/ksize in the tinification tree
(https://git.kernel.org/cgit/linux/kernel/git/josh/linux.git/), and I
plan to submit it upstream during the next merge window.

If someone would prefer to take this patch upstream through a different
tree, please let me know.

- Josh Triplett

>  scripts/ksize | 340 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 340 insertions(+)
>  create mode 100755 scripts/ksize
> 
> diff --git a/scripts/ksize b/scripts/ksize
> new file mode 100755
> index 0000000..0a52121
> --- /dev/null
> +++ b/scripts/ksize
> @@ -0,0 +1,340 @@
> +#!/usr/bin/env python
> +#
> +# Copyright (c) 2011-2014 Intel Corporation.
> +# All rights reserved.
> +#
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License version 2 as
> +# published by the Free Software Foundation.
> +#
> +# This program 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
> +# General Public License for more details.
> +#
> +# Author: Darren Hart <dvhart@linux.intel.com>
> +#
> +
> +"""
> +Display details of the kernel build size.
> +
> +The generated report is comprised of many sub-reports, starting with vmlinux,
> +and descending into each component built-in.o.
> +
> +The first line of each report block is the table header, including the report
> +title and the column labels. Next is the report totals for the top level file
> +(vmlinux or built-in.o). This is followed by itemized sizes for any component
> +*.o object files and all built-in.o files from one directory down (the
> +built-in.o components are labeled with their parent directory to avoid
> +displaying "built-in.o" on nearly every line). The final lines display the sum
> +of all the itemized components and delta between the total and the sum.
> +
> +An example report from an x86_64 allnoconfig build follows in part:
> +
> +Linux Kernel (vmlinux)                  total |       text       data        bss
> +--------------------------------------------------------------------------------
> +vmlinux                               2201904 |     864548     121612    1215744
> +--------------------------------------------------------------------------------
> +arch/x86                               282709 |     171021      65448      46240
> +kernel                                 249960 |     234355       7201       8404
> +mm                                     190369 |     154171      14154      22044
> +fs                                     163867 |     160820       1351       1696
> +drivers                                 44429 |      41353       2052       1024
> +lib                                     37143 |      37053         85          5
> +init                                    21535 |       5189      16285         61
> +security                                 3674 |       3658          8          8
> +net                                       122 |        122          0          0
> +--------------------------------------------------------------------------------
> +sum                                    993808 |     807742     106584      79482
> +delta                                 1208096 |      56806      15028    1136262
> +
> +...
> +
> +drivers                                 total |       text       data        bss
> +--------------------------------------------------------------------------------
> +drivers/built-in.o                      44429 |      41353       2052       1024
> +--------------------------------------------------------------------------------
> +drivers/base                            32427 |      31267       1060        100
> +drivers/char                             9980 |       8412        656        912
> +drivers/rtc                              1155 |       1155          0          0
> +drivers/clocksource                       674 |        406        256         12
> +drivers/video                              62 |         46         16          0
> +--------------------------------------------------------------------------------
> +sum                                     44298 |      41286       1988       1024
> +delta                                     131 |         67         64          0
> +
> +The report may optionally display an additional level of drivers/* reports:
> +
> +    drivers/base                        total |       text       data        bss
> +    ----------------------------------------------------------------------------
> +    drivers/base/built-in.o             32427 |      31267       1060        100
> +    ----------------------------------------------------------------------------
> +    drivers/base/*.o                    32253 |      31121       1032        100
> +    ----------------------------------------------------------------------------
> +    sum                                 32253 |      31121       1032        100
> +    delta                                 174 |        146         28          0
> +
> +    ...
> +"""
> +
> +import sys
> +import getopt
> +import os
> +import struct
> +import termios
> +import fcntl
> +import glob
> +from subprocess import *
> +
> +
> +def usage():
> +    print 'Usage: ksize [OPTION]...'
> +    print '  -d,                 display an additional level of drivers detail'
> +    print '  -h, --help          display this help and exit'
> +    print ''
> +    print 'Run ksize from the top-level Linux kernel build directory. Always'
> +    print 'perform a "make clean" before running a build to be measured by'
> +    print 'ksize to avoid contaminating the report with unassociated build'
> +    print 'artifacts'
> +
> +
> +def term_width():
> +    """
> +    Determine the width of the terminal for formatting the report
> +
> +    Prefer the COLUMNS environment variable, and fall back to termios.
> +    The width will be limited to the range of [70, 100], and will default to 80
> +    if none can be determined, or if redirecting to a file.
> +    """
> +    minw = 70
> +    maxw = 100
> +
> +    if os.environ.has_key('COLUMNS'):
> +        return max(minw, min(int(os.environ['COLUMNS']), maxw))
> +
> +    try:
> +        (rows,cols) = struct.unpack('hh', fcntl.ioctl(1, termios.TIOCGWINSZ, "CCRR"))
> +        return max(minw, min(cols, maxw))
> +    except IOError:
> +        # Probably redirecting to a file
> +        pass
> +    except struct.error, err:
> +        print "Error:", err
> +        sys.exit(1)
> +
> +    return 80
> +
> +
> +def fmt_title(title, maxw):
> +    """
> +    Format the title to fit within a maximum width
> +
> +    The title will be shifted left and prefixed with a '<' character as
> +    necessary to fit within the maximum width.
> +
> +    Args:
> +      title (str): Title to be formatted
> +      maxw (int): Maximum width
> +    """
> +    if len(title) <= maxw:
> +        return title
> +    else:
> +        return "<%s" % (title[-(maxw - 1):])
> +
> +
> +class Sizes(object):
> +    """
> +    Storage class for 'size -t' output
> +    """
> +    def __init__(self, title="", files=None):
> +        """
> +        Create a new Sizes container and populate it with the sum of the sizes
> +        from the files list.
> +
> +        Args:
> +          title (str, optional): Title to display via the show method
> +          files (list of str, optional): Files to pass to 'size -t'
> +        """
> +        self.title = title
> +        self.text = self.data = self.bss = self.total = 0
> +
> +        if files:
> +            p = Popen("size -t " + " ".join(files),
> +                      shell=True, stdout=PIPE, stderr=PIPE)
> +            output = p.communicate()[0].splitlines()
> +
> +            if len(output) > 2:
> +                sizes = output[-1].split()[0:4]
> +                self.text = int(sizes[0])
> +                self.data = int(sizes[1])
> +                self.bss = int(sizes[2])
> +                self.total = int(sizes[3])
> +
> +    def __add__(self, that):
> +        if not (isinstance(self, Sizes) and isinstance(that, Sizes)):
> +            raise TypeError
> +        sum = Sizes()
> +        sum.text = self.text + that.text
> +        sum.data = self.data + that.data
> +        sum.bss = self.bss + that.bss
> +        sum.total = self.total + that.total
> +        return sum
> +
> +    def __sub__(self, that):
> +        if not (isinstance(self, Sizes) and isinstance(that, Sizes)):
> +            raise TypeError
> +        diff = Sizes()
> +        diff.text = self.text - that.text
> +        diff.data = self.data - that.data
> +        diff.bss = self.bss - that.bss
> +        diff.total = self.total - that.total
> +        return diff
> +
> +    def show(self, cols, indent="", alt_title=None):
> +        """
> +        Print a row in a report for sizes represented by this object
> +
> +        Args:
> +          cols (int): Width of the report in characters
> +          indent (str, optional): Literal indentation string, all spaces
> +          alt_title (str, optional): An alternate title to display
> +        """
> +        max_title = cols - 46 - len(indent)
> +        if alt_title is not None:
> +            title = fmt_title(alt_title, max_title)
> +        else:
> +            title = fmt_title(self.title, max_title)
> +        print "%s%-*s %10d | %10d %10d %10d" % (
> +                indent, max_title, title, self.total,
> +                self.text, self.data, self.bss)
> +
> +
> +class Report(object):
> +    """
> +    Container of sizes and sub reports
> +    """
> +    @staticmethod
> +    def create(title, filename, subglobs):
> +        """
> +        Named constructor to create hierarchies of Report objects
> +
> +        Args:
> +          title (str): Title of the report
> +          filename (str): Top level build object filename
> +          subglobs (list of str): Shell globs matching the components of the top
> +            level filename
> +        """
> +        r = Report(title, [filename])
> +
> +        # Create the .o object file report for this level
> +        path = os.path.dirname(filename)
> +        files = [p for p in glob.iglob(path + "/*.o") if not p.endswith("/built-in.o")]
> +        r.parts.append(Report(path + "/*.o", files))
> +
> +        # Create the sub-reports based on each built-in.o
> +        for g in subglobs:
> +            for f in glob.glob(g):
> +                path = os.path.dirname(f)
> +                r.parts.append(Report.create(path, f, [path + "/*/built-in.o"]))
> +
> +        # Display in descending total size order
> +        r.parts.sort(reverse=True)
> +
> +        # Calculate the sum and deltas from each component report
> +        for b in r.parts:
> +            r.totals += b.sizes
> +        r.totals.title = "sum"
> +
> +        r.deltas = r.sizes - r.totals
> +        r.deltas.title = "delta"
> +
> +        return r
> +
> +    def __init__(self, title, files):
> +        """
> +        Create a new singular Report object, only called by Report.create()
> +
> +        Args:
> +          title (str, optional): Title to display via the show method
> +          files (list of str, optional): Files to construct Sizes
> +        """
> +        self.files = files
> +        self.title = title
> +        self.parts = list()
> +        self.sizes = Sizes(title, files)
> +        self.totals = Sizes("sum")
> +        self.deltas = Sizes("delta")
> +
> +    def show(self, cols, indent=""):
> +        """
> +        Print the Report as a table with Sizes as rows
> +
> +        Args:
> +          cols (int): Width of the report in characters
> +          indent (str, optional): Literal indentation string, all spaces
> +        """
> +        max_title = cols - 46 - len(indent)
> +        title = fmt_title(self.title, max_title)
> +        rule = str.ljust(indent, cols, '-')
> +
> +        # Print report table header
> +        print "%s%-*s %10s | %10s %10s %10s" % (
> +                indent, max_title, title, "total", "text", "data", "bss")
> +
> +        # Print top level report filename instead of title (usually path)
> +        print rule
> +        self.sizes.show(cols, indent, self.files[0])
> +        print rule
> +
> +        # Print component sizes (*.o and */built-in.o)
> +        for p in self.parts:
> +            if p.sizes.total > 0:
> +                p.sizes.show(cols, indent)
> +        print rule
> +
> +        # Print the sum of the components, and the delta with the total
> +        self.totals.show(cols, indent)
> +        self.deltas.show(cols, indent)
> +        print "\n"
> +
> +    def __cmp__(self, that):
> +        if not isinstance(that, Report):
> +            raise TypeError
> +        return cmp(self.sizes.total, that.sizes.total)
> +
> +
> +def main(argv):
> +    try:
> +        opts, args = getopt.getopt(argv[1:], "dh", ["help"])
> +    except getopt.GetoptError, err:
> +        print '%s' % str(err)
> +        usage()
> +        return 2
> +
> +    driver_detail = False
> +    for o, a in opts:
> +        if o == '-d':
> +            driver_detail = True
> +        elif o in ('-h', '--help'):
> +            usage()
> +            return 0
> +        else:
> +            assert False, "unhandled option"
> +
> +    cols = term_width()
> +
> +    globs = ["arch/*/built-in.o", "*/built-in.o"]
> +    vmlinux = Report.create("Linux Kernel (vmlinux)", "vmlinux", globs)
> +
> +    vmlinux.show(cols)
> +    for b in vmlinux.parts:
> +        if b.totals.total > 0 and len(b.parts) > 1:
> +            b.show(cols)
> +        if b.title == "drivers" and driver_detail:
> +            for d in b.parts:
> +                if d.totals.total > 0 and len(d.parts) > 1:
> +                    d.show(cols, "    ")
> +
> +
> +if __name__ == "__main__":
> +    sys.exit(main(sys.argv))
> -- 
> 2.1.1
>