blob: 0f8ea200f54980b9f840b3df5b4206a6cd8193ca [file] [log] [blame] [edit]
# Copyright (c) 2013 The Chromium OS Authors.
#
# SPDX-License-Identifier: GPL-2.0+
#
(Please read 'How to change from MAKEALL' if you are used to that tool)
What is this?
=============
This tool handles building U-Boot to check that you have not broken it
with your patch series. It can build each individual commit and report
which boards fail on which commits, and which errors come up. It aims
to make full use of multi-processor machines.
A key feature of buildman is its output summary, which allows warnings,
errors or image size increases in a particular commit or board to be
quickly identified and the offending commit pinpointed. This can be a big
help for anyone working with >10 patches at a time.
Caveats
=======
Buildman is still in its infancy. It is already a very useful tool, but
expect to find problems and send patches.
Buildman can be stopped and restarted, in which case it will continue
where it left off. This should happen cleanly and without side-effects.
If not, it is a bug, for which a patch would be welcome.
Buildman gets so tied up in its work that it can ignore the outside world.
You may need to press Ctrl-C several times to quit it. Also it will print
out various exceptions when stopped.
Theory of Operation
===================
(please read this section in full twice or you will be perpetually confused)
Buildman is a builder. It is not make, although it runs make. It does not
produce any useful output on the terminal while building, except for
progress information (except with -v, see below). All the output (errors,
warnings and binaries if you ask for them) is stored in output
directories, which you can look at while the build is progressing, or when
it is finished.
Buildman produces a concise summary of which boards succeeded and failed.
It shows which commit introduced which board failure using a simple
red/green colour coding. Full error information can be requested, in which
case it is de-duped and displayed against the commit that introduced the
error. An example workflow is below.
Buildman stores image size information and can report changes in image size
from commit to commit. An example of this is below.
Buildman starts multiple threads, and each thread builds for one board at
a time. A thread starts at the first commit, configures the source for your
board and builds it. Then it checks out the next commit and does an
incremental build. Eventually the thread reaches the last commit and stops.
If errors or warnings are found along the way, the thread will reconfigure
after every commit, and your build will be very slow. This is because a
file that produces just a warning would not normally be rebuilt in an
incremental build.
Buildman works in an entirely separate place from your U-Boot repository.
It creates a separate working directory for each thread, and puts the
output files in the working directory, organised by commit name and board
name, in a two-level hierarchy.
Buildman is invoked in your U-Boot directory, the one with the .git
directory. It clones this repository into a copy for each thread, and the
threads do not affect the state of your git repository. Any checkouts done
by the thread affect only the working directory for that thread.
Buildman automatically selects the correct tool chain for each board. You
must supply suitable tool chains, but buildman takes care of selecting the
right one.
Buildman generally builds a branch (with the -b flag), and in this case
builds the upstream commit as well, for comparison. It cannot build
individual commits at present, unless (maybe) you point it at an empty
branch. Put all your commits in a branch, set the branch's upstream to a
valid value, and all will be well. Otherwise buildman will perform random
actions. Use -n to check what the random actions might be.
If you just want to build the current source tree, leave off the -b flag
and add -e. This will display results and errors as they happen. You can
still look at them later using -se. Note that buildman will assume that the
source has changed, and will build all specified boards in this case.
Buildman is optimised for building many commits at once, for many boards.
On multi-core machines, Buildman is fast because it uses most of the
available CPU power. When it gets to the end, or if you are building just
a few commits or boards, it will be pretty slow. As a tip, if you don't
plan to use your machine for anything else, you can use -T to increase the
number of threads beyond the default.
Buildman lets you build all boards, or a subset. Specify the subset by passing
command-line arguments that list the desired board name, architecture name,
SOC name, or anything else in the boards.cfg file. Multiple arguments are
allowed. Each argument will be interpreted as a regular expression, so
behaviour is a superset of exact or substring matching. Examples are:
* 'tegra20' All boards with a Tegra20 SoC
* 'tegra' All boards with any Tegra Soc (Tegra20, Tegra30, Tegra114...)
* '^tegra[23]0$' All boards with either Tegra20 or Tegra30 SoC
* 'powerpc' All PowerPC boards
While the default is to OR the terms together, you can also make use of
the '&' operator to limit the selection:
* 'freescale & arm sandbox' All Freescale boards with ARM architecture,
plus sandbox
You can also use -x to specifically exclude some boards. For example:
buildmand arm -x nvidia,freescale,.*ball$
means to build all arm boards except nvidia, freescale and anything ending
with 'ball'.
It is convenient to use the -n option to see what will be built based on
the subset given.
Buildman does not store intermediate object files. It optionally copies
the binary output into a directory when a build is successful. Size
information is always recorded. It needs a fair bit of disk space to work,
typically 250MB per thread.
Setting up
==========
1. Get the U-Boot source. You probably already have it, but if not these
steps should get you started with a repo and some commits for testing.
$ cd /path/to/u-boot
$ git clone git://git.denx.de/u-boot.git .
$ git checkout -b my-branch origin/master
$ # Add some commits to the branch, reading for testing
2. Create ~/.buildman to tell buildman where to find tool chains. As an
example:
# Buildman settings file
[toolchain]
root: /
rest: /toolchains/*
eldk: /opt/eldk-4.2
arm: /opt/linaro/gcc-linaro-arm-linux-gnueabihf-4.8-2013.08_linux
aarch64: /opt/linaro/gcc-linaro-aarch64-none-elf-4.8-2013.10_linux
[toolchain-alias]
x86: i386
blackfin: bfin
sh: sh4
nds32: nds32le
openrisc: or32
This selects the available toolchain paths. Add the base directory for
each of your toolchains here. Buildman will search inside these directories
and also in any '/usr' and '/usr/bin' subdirectories.
Make sure the tags (here root: rest: and eldk:) are unique.
The toolchain-alias section indicates that the i386 toolchain should be used
to build x86 commits.
2. Check the available toolchains
Run this check to make sure that you have a toolchain for every architecture.
$ ./tools/buildman/buildman --list-tool-chains
Scanning for tool chains
- scanning path '/'
- looking in '/.'
- looking in '/bin'
- looking in '/usr/bin'
- found '/usr/bin/gcc'
Tool chain test: OK
- found '/usr/bin/c89-gcc'
Tool chain test: OK
- found '/usr/bin/c99-gcc'
Tool chain test: OK
- found '/usr/bin/x86_64-linux-gnu-gcc'
Tool chain test: OK
- scanning path '/toolchains/powerpc-linux'
- looking in '/toolchains/powerpc-linux/.'
- looking in '/toolchains/powerpc-linux/bin'
- found '/toolchains/powerpc-linux/bin/powerpc-linux-gcc'
Tool chain test: OK
- looking in '/toolchains/powerpc-linux/usr/bin'
- scanning path '/toolchains/nds32le-linux-glibc-v1f'
- looking in '/toolchains/nds32le-linux-glibc-v1f/.'
- looking in '/toolchains/nds32le-linux-glibc-v1f/bin'
- found '/toolchains/nds32le-linux-glibc-v1f/bin/nds32le-linux-gcc'
Tool chain test: OK
- looking in '/toolchains/nds32le-linux-glibc-v1f/usr/bin'
- scanning path '/toolchains/nios2'
- looking in '/toolchains/nios2/.'
- looking in '/toolchains/nios2/bin'
- found '/toolchains/nios2/bin/nios2-linux-gcc'
Tool chain test: OK
- found '/toolchains/nios2/bin/nios2-linux-uclibc-gcc'
Tool chain test: OK
- looking in '/toolchains/nios2/usr/bin'
- found '/toolchains/nios2/usr/bin/nios2-linux-gcc'
Tool chain test: OK
- found '/toolchains/nios2/usr/bin/nios2-linux-uclibc-gcc'
Tool chain test: OK
- scanning path '/toolchains/microblaze-unknown-linux-gnu'
- looking in '/toolchains/microblaze-unknown-linux-gnu/.'
- looking in '/toolchains/microblaze-unknown-linux-gnu/bin'
- found '/toolchains/microblaze-unknown-linux-gnu/bin/microblaze-unknown-linux-gnu-gcc'
Tool chain test: OK
- found '/toolchains/microblaze-unknown-linux-gnu/bin/mb-linux-gcc'
Tool chain test: OK
- looking in '/toolchains/microblaze-unknown-linux-gnu/usr/bin'
- scanning path '/toolchains/mips-linux'
- looking in '/toolchains/mips-linux/.'
- looking in '/toolchains/mips-linux/bin'
- found '/toolchains/mips-linux/bin/mips-linux-gcc'
Tool chain test: OK
- looking in '/toolchains/mips-linux/usr/bin'
- scanning path '/toolchains/old'
- looking in '/toolchains/old/.'
- looking in '/toolchains/old/bin'
- looking in '/toolchains/old/usr/bin'
- scanning path '/toolchains/i386-linux'
- looking in '/toolchains/i386-linux/.'
- looking in '/toolchains/i386-linux/bin'
- found '/toolchains/i386-linux/bin/i386-linux-gcc'
Tool chain test: OK
- looking in '/toolchains/i386-linux/usr/bin'
- scanning path '/toolchains/bfin-uclinux'
- looking in '/toolchains/bfin-uclinux/.'
- looking in '/toolchains/bfin-uclinux/bin'
- found '/toolchains/bfin-uclinux/bin/bfin-uclinux-gcc'
Tool chain test: OK
- looking in '/toolchains/bfin-uclinux/usr/bin'
- scanning path '/toolchains/sparc-elf'
- looking in '/toolchains/sparc-elf/.'
- looking in '/toolchains/sparc-elf/bin'
- found '/toolchains/sparc-elf/bin/sparc-elf-gcc'
Tool chain test: OK
- looking in '/toolchains/sparc-elf/usr/bin'
- scanning path '/toolchains/arm-2010q1'
- looking in '/toolchains/arm-2010q1/.'
- looking in '/toolchains/arm-2010q1/bin'
- found '/toolchains/arm-2010q1/bin/arm-none-linux-gnueabi-gcc'
Tool chain test: OK
- looking in '/toolchains/arm-2010q1/usr/bin'
- scanning path '/toolchains/from'
- looking in '/toolchains/from/.'
- looking in '/toolchains/from/bin'
- looking in '/toolchains/from/usr/bin'
- scanning path '/toolchains/sh4-gentoo-linux-gnu'
- looking in '/toolchains/sh4-gentoo-linux-gnu/.'
- looking in '/toolchains/sh4-gentoo-linux-gnu/bin'
- found '/toolchains/sh4-gentoo-linux-gnu/bin/sh4-gentoo-linux-gnu-gcc'
Tool chain test: OK
- looking in '/toolchains/sh4-gentoo-linux-gnu/usr/bin'
- scanning path '/toolchains/avr32-linux'
- looking in '/toolchains/avr32-linux/.'
- looking in '/toolchains/avr32-linux/bin'
- found '/toolchains/avr32-linux/bin/avr32-gcc'
Tool chain test: OK
- looking in '/toolchains/avr32-linux/usr/bin'
- scanning path '/toolchains/m68k-linux'
- looking in '/toolchains/m68k-linux/.'
- looking in '/toolchains/m68k-linux/bin'
- found '/toolchains/m68k-linux/bin/m68k-linux-gcc'
Tool chain test: OK
- looking in '/toolchains/m68k-linux/usr/bin'
List of available toolchains (17):
arm : /toolchains/arm-2010q1/bin/arm-none-linux-gnueabi-gcc
avr32 : /toolchains/avr32-linux/bin/avr32-gcc
bfin : /toolchains/bfin-uclinux/bin/bfin-uclinux-gcc
c89 : /usr/bin/c89-gcc
c99 : /usr/bin/c99-gcc
i386 : /toolchains/i386-linux/bin/i386-linux-gcc
m68k : /toolchains/m68k-linux/bin/m68k-linux-gcc
mb : /toolchains/microblaze-unknown-linux-gnu/bin/mb-linux-gcc
microblaze: /toolchains/microblaze-unknown-linux-gnu/bin/microblaze-unknown-linux-gnu-gcc
mips : /toolchains/mips-linux/bin/mips-linux-gcc
nds32le : /toolchains/nds32le-linux-glibc-v1f/bin/nds32le-linux-gcc
nios2 : /toolchains/nios2/bin/nios2-linux-gcc
powerpc : /toolchains/powerpc-linux/bin/powerpc-linux-gcc
sandbox : /usr/bin/gcc
sh4 : /toolchains/sh4-gentoo-linux-gnu/bin/sh4-gentoo-linux-gnu-gcc
sparc : /toolchains/sparc-elf/bin/sparc-elf-gcc
x86_64 : /usr/bin/x86_64-linux-gnu-gcc
You can see that everything is covered, even some strange ones that won't
be used (c88 and c99). This is a feature.
How to run it
=============
First do a dry run using the -n flag: (replace <branch> with a real, local
branch with a valid upstream)
$ ./tools/buildman/buildman -b <branch> -n
If it can't detect the upstream branch, try checking out the branch, and
doing something like 'git branch --set-upstream <branch> upstream/master'
or something similar.
As an example:
Dry run, so not doing much. But I would do this:
Building 18 commits for 1059 boards (4 threads, 1 job per thread)
Build directory: ../lcd9b
5bb3505 Merge branch 'master' of git://git.denx.de/u-boot-arm
c18f1b4 tegra: Use const for pinmux_config_pingroup/table()
2f043ae tegra: Add display support to funcmux
e349900 tegra: fdt: Add pwm binding and node
424a5f0 tegra: fdt: Add LCD definitions for Tegra
0636ccf tegra: Add support for PWM
a994fe7 tegra: Add SOC support for display/lcd
fcd7350 tegra: Add LCD driver
4d46e9d tegra: Add LCD support to Nvidia boards
991bd48 arm: Add control over cachability of memory regions
54e8019 lcd: Add CONFIG_LCD_ALIGNMENT to select frame buffer alignment
d92aff7 lcd: Add support for flushing LCD fb from dcache after update
dbd0677 tegra: Align LCD frame buffer to section boundary
0cff9b8 tegra: Support control of cache settings for LCD
9c56900 tegra: fdt: Add LCD definitions for Seaboard
5cc29db lcd: Add CONFIG_CONSOLE_SCROLL_LINES option to speed console
cac5a23 tegra: Enable display/lcd support on Seaboard
49ff541 wip
Total boards to build for each commit: 1059
This shows that it will build all 1059 boards, using 4 threads (because
we have a 4-core CPU). Each thread will run with -j1, meaning that each
make job will use a single CPU. The list of commits to be built helps you
confirm that things look about right. Notice that buildman has chosen a
'base' directory for you, immediately above your source tree.
Buildman works entirely inside the base directory, here ../lcd9b,
creating a working directory for each thread, and creating output
directories for each commit and board.
Suggested Workflow
==================
To run the build for real, take off the -n:
$ ./tools/buildman/buildman -b <branch>
Buildman will set up some working directories, and get started. After a
minute or so it will settle down to a steady pace, with a display like this:
Building 18 commits for 1059 boards (4 threads, 1 job per thread)
528 36 124 /19062 1:13:30 : SIMPC8313_SP
This means that it is building 19062 board/commit combinations. So far it
has managed to successfully build 528. Another 36 have built with warnings,
and 124 more didn't build at all. Buildman expects to complete the process
in an hour and 15 minutes. Use this time to buy a faster computer.
To find out how the build went, ask for a summary with -s. You can do this
either before the build completes (presumably in another terminal) or
afterwards. Let's work through an example of how this is used:
$ ./tools/buildman/buildman -b lcd9b -s
...
01: Merge branch 'master' of git://git.denx.de/u-boot-arm
powerpc: + galaxy5200_LOWBOOT
02: tegra: Use const for pinmux_config_pingroup/table()
03: tegra: Add display support to funcmux
04: tegra: fdt: Add pwm binding and node
05: tegra: fdt: Add LCD definitions for Tegra
06: tegra: Add support for PWM
07: tegra: Add SOC support for display/lcd
08: tegra: Add LCD driver
09: tegra: Add LCD support to Nvidia boards
10: arm: Add control over cachability of memory regions
11: lcd: Add CONFIG_LCD_ALIGNMENT to select frame buffer alignment
12: lcd: Add support for flushing LCD fb from dcache after update
arm: + lubbock
13: tegra: Align LCD frame buffer to section boundary
14: tegra: Support control of cache settings for LCD
15: tegra: fdt: Add LCD definitions for Seaboard
16: lcd: Add CONFIG_CONSOLE_SCROLL_LINES option to speed console
17: tegra: Enable display/lcd support on Seaboard
18: wip
This shows which commits have succeeded and which have failed. In this case
the build is still in progress so many boards are not built yet (use -u to
see which ones). But still we can see a few failures. The galaxy5200_LOWBOOT
never builds correctly. This could be a problem with our toolchain, or it
could be a bug in the upstream. The good news is that we probably don't need
to blame our commits. The bad news is it isn't tested on that board.
Commit 12 broke lubbock. That's what the '+ lubbock' means. The failure
is never fixed by a later commit, or you would see lubbock again, in green,
without the +.
To see the actual error:
$ ./tools/buildman/buildman -b <branch> -se lubbock
...
12: lcd: Add support for flushing LCD fb from dcache after update
arm: + lubbock
+common/libcommon.o: In function `lcd_sync':
+/u-boot/lcd9b/.bm-work/00/common/lcd.c:120: undefined reference to `flush_dcache_range'
+arm-none-linux-gnueabi-ld: BFD (Sourcery G++ Lite 2010q1-202) 2.19.51.20090709 assertion fail /scratch/julian/2010q1-release-linux-lite/obj/binutils-src-2010q1-202-arm-none-linux-gnueabi-i686-pc-linux-gnu/bfd/elf32-arm.c:12572
+make: *** [/u-boot/lcd9b/.bm-work/00/build/u-boot] Error 139
13: tegra: Align LCD frame buffer to section boundary
14: tegra: Support control of cache settings for LCD
15: tegra: fdt: Add LCD definitions for Seaboard
16: lcd: Add CONFIG_CONSOLE_SCROLL_LINES option to speed console
-/u-boot/lcd9b/.bm-work/00/common/lcd.c:120: undefined reference to `flush_dcache_range'
+/u-boot/lcd9b/.bm-work/00/common/lcd.c:125: undefined reference to `flush_dcache_range'
17: tegra: Enable display/lcd support on Seaboard
18: wip
So the problem is in lcd.c, due to missing cache operations. This information
should be enough to work out what that commit is doing to break these
boards. (In this case pxa did not have cache operations defined).
If you see error lines marked with - that means that the errors were fixed
by that commit. Sometimes commits can be in the wrong order, so that a
breakage is introduced for a few commits and fixed by later commits. This
shows up clearly with buildman. You can then reorder the commits and try
again.
At commit 16, the error moves - you can see that the old error at line 120
is fixed, but there is a new one at line 126. This is probably only because
we added some code and moved the broken line further down the file.
If many boards have the same error, then -e will display the error only
once. This makes the output as concise as possible. To see which boards have
each error, use -l.
Buildman tries to distinguish warnings from errors, and shows warning lines
separately with a 'w' prefix.
The full build output in this case is available in:
../lcd9b/12_of_18_gd92aff7_lcd--Add-support-for/lubbock/
done: Indicates the build was done, and holds the return code from make.
This is 0 for a good build, typically 2 for a failure.
err: Output from stderr, if any. Errors and warnings appear here.
log: Output from stdout. Normally there isn't any since buildman runs
in silent mode for now.
toolchain: Shows information about the toolchain used for the build.
sizes: Shows image size information.
It is possible to get the build output there also. Use the -k option for
this. In that case you will also see some output files, like:
System.map toolchain u-boot u-boot.bin u-boot.map autoconf.mk
(also SPL versions u-boot-spl and u-boot-spl.bin if available)
Checking Image Sizes
====================
A key requirement for U-Boot is that you keep code/data size to a minimum.
Where a new feature increases this noticeably it should normally be put
behind a CONFIG flag so that boards can leave it off and keep the image
size more or less the same with each new release.
To check the impact of your commits on image size, use -S. For example:
$ ./tools/buildman/buildman -b us-x86 -sS
Summary of 10 commits for 1066 boards (4 threads, 1 job per thread)
01: MAKEALL: add support for per architecture toolchains
02: x86: Add function to get top of usable ram
x86: (for 1/3 boards) text -272.0 rodata +41.0
03: x86: Add basic cache operations
04: x86: Permit bootstage and timer data to be used prior to relocation
x86: (for 1/3 boards) data +16.0
05: x86: Add an __end symbol to signal the end of the U-Boot binary
x86: (for 1/3 boards) text +76.0
06: x86: Rearrange the output input to remove BSS
x86: (for 1/3 boards) bss -2140.0
07: x86: Support relocation of FDT on start-up
x86: + coreboot-x86
08: x86: Add error checking to x86 relocation code
09: x86: Adjust link device tree include file
10: x86: Enable CONFIG_OF_CONTROL on coreboot
You can see that image size only changed on x86, which is good because this
series is not supposed to change any other board. From commit 7 onwards the
build fails so we don't get code size numbers. The numbers are fractional
because they are an average of all boards for that architecture. The
intention is to allow you to quickly find image size problems introduced by
your commits.
Note that the 'text' region and 'rodata' are split out. You should add the
two together to get the total read-only size (reported as the first column
in the output from binutil's 'size' utility).
A useful option is --step which lets you skip some commits. For example
--step 2 will show the image sizes for only every 2nd commit (so it will
compare the image sizes of the 1st, 3rd, 5th... commits). You can also use
--step 0 which will compare only the first and last commits. This is useful
for an overview of how your entire series affects code size.
You can also use -d to see a detailed size breakdown for each board. This
list is sorted in order from largest growth to largest reduction.
It is possible to go a little further with the -B option (--bloat). This
shows where U-Boot has bloated, breaking the size change down to the function
level. Example output is below:
$ ./tools/buildman/buildman -b us-mem4 -sSdB
...
19: Roll crc32 into hash infrastructure
arm: (for 10/10 boards) all -143.4 bss +1.2 data -4.8 rodata -48.2 text -91.6
paz00 : all +23 bss -4 rodata -29 text +56
u-boot: add: 1/0, grow: 3/-2 bytes: 168/-104 (64)
function old new delta
hash_command 80 160 +80
crc32_wd_buf - 56 +56
ext4fs_read_file 540 568 +28
insert_var_value_sub 688 692 +4
run_list_real 1996 1992 -4
do_mem_crc 168 68 -100
trimslice : all -9 bss +16 rodata -29 text +4
u-boot: add: 1/0, grow: 1/-3 bytes: 136/-124 (12)
function old new delta
hash_command 80 160 +80
crc32_wd_buf - 56 +56
ext4fs_iterate_dir 672 668 -4
ext4fs_read_file 568 548 -20
do_mem_crc 168 68 -100
whistler : all -9 bss +16 rodata -29 text +4
u-boot: add: 1/0, grow: 1/-3 bytes: 136/-124 (12)
function old new delta
hash_command 80 160 +80
crc32_wd_buf - 56 +56
ext4fs_iterate_dir 672 668 -4
ext4fs_read_file 568 548 -20
do_mem_crc 168 68 -100
seaboard : all -9 bss -28 rodata -29 text +48
u-boot: add: 1/0, grow: 3/-2 bytes: 160/-104 (56)
function old new delta
hash_command 80 160 +80
crc32_wd_buf - 56 +56
ext4fs_read_file 548 568 +20
run_list_real 1996 2000 +4
do_nandboot 760 756 -4
do_mem_crc 168 68 -100
colibri_t20_iris: all -9 rodata -29 text +20
u-boot: add: 1/0, grow: 2/-3 bytes: 140/-112 (28)
function old new delta
hash_command 80 160 +80
crc32_wd_buf - 56 +56
read_abs_bbt 204 208 +4
do_nandboot 760 756 -4
ext4fs_read_file 576 568 -8
do_mem_crc 168 68 -100
ventana : all -37 bss -12 rodata -29 text +4
u-boot: add: 1/0, grow: 1/-3 bytes: 136/-124 (12)
function old new delta
hash_command 80 160 +80
crc32_wd_buf - 56 +56
ext4fs_iterate_dir 672 668 -4
ext4fs_read_file 568 548 -20
do_mem_crc 168 68 -100
harmony : all -37 bss -16 rodata -29 text +8
u-boot: add: 1/0, grow: 2/-3 bytes: 140/-124 (16)
function old new delta
hash_command 80 160 +80
crc32_wd_buf - 56 +56
nand_write_oob_syndrome 428 432 +4
ext4fs_iterate_dir 672 668 -4
ext4fs_read_file 568 548 -20
do_mem_crc 168 68 -100
medcom-wide : all -417 bss +28 data -16 rodata -93 text -336
u-boot: add: 1/-1, grow: 1/-2 bytes: 88/-376 (-288)
function old new delta
crc32_wd_buf - 56 +56
do_fat_read_at 2872 2904 +32
hash_algo 16 - -16
do_mem_crc 168 68 -100
hash_command 420 160 -260
tec : all -449 bss -4 data -16 rodata -93 text -336
u-boot: add: 1/-1, grow: 1/-2 bytes: 88/-376 (-288)
function old new delta
crc32_wd_buf - 56 +56
do_fat_read_at 2872 2904 +32
hash_algo 16 - -16
do_mem_crc 168 68 -100
hash_command 420 160 -260
plutux : all -481 bss +16 data -16 rodata -93 text -388
u-boot: add: 1/-1, grow: 1/-3 bytes: 68/-408 (-340)
function old new delta
crc32_wd_buf - 56 +56
do_load_serial_bin 1688 1700 +12
hash_algo 16 - -16
do_fat_read_at 2904 2872 -32
do_mem_crc 168 68 -100
hash_command 420 160 -260
powerpc: (for 5/5 boards) all +37.4 data -3.2 rodata -41.8 text +82.4
MPC8610HPCD : all +55 rodata -29 text +84
u-boot: add: 1/0, grow: 0/-1 bytes: 176/-96 (80)
function old new delta
hash_command - 176 +176
do_mem_crc 184 88 -96
MPC8641HPCN : all +55 rodata -29 text +84
u-boot: add: 1/0, grow: 0/-1 bytes: 176/-96 (80)
function old new delta
hash_command - 176 +176
do_mem_crc 184 88 -96
MPC8641HPCN_36BIT: all +55 rodata -29 text +84
u-boot: add: 1/0, grow: 0/-1 bytes: 176/-96 (80)
function old new delta
hash_command - 176 +176
do_mem_crc 184 88 -96
sbc8641d : all +55 rodata -29 text +84
u-boot: add: 1/0, grow: 0/-1 bytes: 176/-96 (80)
function old new delta
hash_command - 176 +176
do_mem_crc 184 88 -96
xpedite517x : all -33 data -16 rodata -93 text +76
u-boot: add: 1/-1, grow: 0/-1 bytes: 176/-112 (64)
function old new delta
hash_command - 176 +176
hash_algo 16 - -16
do_mem_crc 184 88 -96
...
This shows that commit 19 has increased text size for arm (although only one
board was built) and by 96 bytes for powerpc. This increase was offset in both
cases by reductions in rodata and data/bss.
Shown below the summary lines are the sizes for each board. Below each board
are the sizes for each function. This information starts with:
add - number of functions added / removed
grow - number of functions which grew / shrunk
bytes - number of bytes of code added to / removed from all functions,
plus the total byte change in brackets
The change seems to be that hash_command() has increased by more than the
do_mem_crc() function has decreased. The function sizes typically add up to
roughly the text area size, but note that every read-only section except
rodata is included in 'text', so the function total does not exactly
correspond.
It is common when refactoring code for the rodata to decrease as the text size
increases, and vice versa.
Providing 'make' flags
======================
U-Boot's build system supports a few flags (such as BUILD_TAG) which affect
the build product. These flags can be specified in the buildman settings
file. They can also be useful when building U-Boot against other open source
software.
[make-flags]
at91-boards=ENABLE_AT91_TEST=1
snapper9260=${at91-boards} BUILD_TAG=442
snapper9g45=${at91-boards} BUILD_TAG=443
This will use 'make ENABLE_AT91_TEST=1 BUILD_TAG=442' for snapper9260
and 'make ENABLE_AT91_TEST=1 BUILD_TAG=443' for snapper9g45. A special
variable ${target} is available to access the target name (snapper9260 and
snapper9g20 in this case). Variables are resolved recursively. Note that
variables can only contain the characters A-Z, a-z, 0-9, hyphen (-) and
underscore (_).
It is expected that any variables added are dealt with in U-Boot's
config.mk file and documented in the README.
Quick Sanity Check
==================
If you have made changes and want to do a quick sanity check of the
currently checked-out source, run buildman without the -b flag. This will
build the selected boards and display build status as it runs (i.e. -v is
enabled automatically). Use -e to see errors/warnings as well.
Other options
=============
Buildman has various other command line options. Try --help to see them.
When doing builds, Buildman's return code will reflect the overall result:
0 (success) No errors or warnings found
128 Errors found
129 Warnings found
How to change from MAKEALL
==========================
Buildman includes most of the features of MAKEALL and is generally faster
and easier to use. In particular it builds entire branches: if a particular
commit introduces an error in a particular board, buildman can easily show
you this, even if a later commit fixes that error.
The reasons to deprecate MAKEALL are:
- We don't want to maintain two build systems
- Buildman is typically faster
- Buildman has a lot more features
But still, many people will be sad to lose MAKEALL. If you are used to
MAKEALL, here are a few pointers.
First you need to set up your tool chains - see the 'Setting up' section
for details. Once you have your required toolchain(s) detected then you are
ready to go.
To build the current source tree, run buildman without a -b flag:
./tools/buildman/buildman <list of things to build>
This will build the current source tree for the given boards and display
the results and errors.
However buildman usually works on entire branches, and for that you must
specify a board flag:
./tools/buildman/buildman -b <branch_name> <list of things to build>
followed by (afterwards, or perhaps concurrently in another terminal):
./tools/buildman/buildman -b <branch_name> -s <list of things to build>
to see the results of the build. Rather than showing you all the output,
buildman just shows a summary, with red indicating that a commit introduced
an error and green indicating that a commit fixed an error. Use the -e
flag to see the full errors and -l to see which boards caused which errors.
If you really want to see build results as they happen, use -v when doing a
build (and -e to see the errors/warnings too).
You don't need to stick around on that branch while buildman is running. It
checks out its own copy of the source code, so you can change branches,
add commits, etc. without affecting the build in progress.
The <list of things to build> can include board names, architectures or the
like. There are no flags to disambiguate since ambiguities are rare. Using
the examples from MAKEALL:
Examples:
- build all Power Architecture boards:
MAKEALL -a powerpc
MAKEALL --arch powerpc
MAKEALL powerpc
** buildman -b <branch> powerpc
- build all PowerPC boards manufactured by vendor "esd":
MAKEALL -a powerpc -v esd
** buildman -b <branch> esd
- build all PowerPC boards manufactured either by "keymile" or "siemens":
MAKEALL -a powerpc -v keymile -v siemens
** buildman -b <branch> keymile siemens
- build all Freescale boards with MPC83xx CPUs, plus all 4xx boards:
MAKEALL -c mpc83xx -v freescale 4xx
** buildman -b <branch> mpc83xx freescale 4xx
Buildman automatically tries to use all the CPUs in your machine. If you
are building a lot of boards it will use one thread for every CPU core
it detects in your machine. This is like MAKEALL's BUILD_NBUILDS option.
You can use the -T flag to change the number of threads. If you are only
building a few boards, buildman will automatically run make with the -j
flag to increase the number of concurrent make tasks. It isn't normally
that helpful to fiddle with this option, but if you use the BUILD_NCPUS
option in MAKEALL then -j is the equivalent in buildman.
Buildman puts its output in ../<branch_name> by default but you can change
this with the -o option. Buildman normally does out-of-tree builds: use -i
to disable that if you really want to. But be careful that once you have
used -i you pollute buildman's copies of the source tree, and you will need
to remove the build directory (normally ../<branch_name>) to run buildman
in normal mode (without -i).
Buildman doesn't keep the output result normally, but use the -k option to
do this.
Please read 'Theory of Operation' a few times as it will make a lot of
things clearer.
Some options you might like are:
-B shows which functions are growing/shrinking in which commit - great
for finding code bloat.
-S shows image sizes for each commit (just an overall summary)
-u shows boards that you haven't built yet
--step 0 will build just the upstream commit and the last commit of your
branch. This is often a quick sanity check that your branch doesn't
break anything. But note this does not check bisectability!
TODO
====
This has mostly be written in my spare time as a response to my difficulties
in testing large series of patches. Apart from tidying up there is quite a
bit of scope for improvement. Things like better error diffs and easier
access to log files. Also it would be nice if buildman could 'hunt' for
problems, perhaps by building a few boards for each arch, or checking
commits for changed files and building only boards which use those files.
Credits
=======
Thanks to Grant Grundler <grundler@chromium.org> for his ideas for improving
the build speed by building all commits for a board instead of the other
way around.
Simon Glass
sjg@chromium.org
Halloween 2012
Updated 12-12-12
Updated 23-02-13