author "Yann E. MORIN" <>
Tue Aug 10 13:25:52 2010 +0200 (2010-08-10)
changeset 2075 edc7c7958e80
parent 1937 384b7f8fc780
permissions -rw-r--r--
scripts: better sanitise the patch names

When generating the fixes-patches, better sanitise the generated
file names, as it seems to raise a few issues for some people.
     1 File.........: overview.txt
     2 Content......: Overview of how crosstool-NG works.
     3 Copyrigth....: (C) 2007 Yann E. MORIN <>
     4 License......: Creative Commons Attribution Share Alike (CC-by-sa), v2.5
     6 ____________________
     7                    /
     8 Table Of Content  /
     9 _________________/
    12 Introduction
    13 History
    14 Referring to crosstool-NG
    15 Installing crosstool-NG
    16   Install method
    17   The hacker's way
    18   Preparing for packaging
    19   Shell completion
    20   Contributed code
    21 Configuring crosstool-NG
    22   Interesting config options
    23   Re-building an existing toolchain
    24   Using as a backend for a build-system
    25 Running crosstool-NG
    26   Stopping and restarting a build
    27   Testing all toolchains at once
    28   Overriding the number of // jobs
    29   Note on // jobs
    30   Tools wrapper
    31 Using the toolchain
    32   The 'populate' script
    33 Toolchain types
    34   Seemingly-native toolchains
    35 Contributing
    36   Sending a bug report
    37   Sending patches
    38 Internals
    39   Makefile front-end
    40   Kconfig parser
    41   Architecture-specific
    42   Adding a new version of a component
    43   Build scripts
    46 ________________
    47                /
    48 Introduction  /
    49 _____________/
    51 crosstool-NG aims at building toolchains. Toolchains are an essential component
    52 in a software development project. It will compile, assemble and link the code
    53 that is being developed. Some pieces of the toolchain will eventually end up
    54 in the resulting binary/ies: static libraries are but an example.
    56 So, a toolchain is a very sensitive piece of software, as any bug in one of the
    57 components, or a poorly configured component, can lead to execution problems,
    58 ranging from poor performance, to applications ending unexpectedly, to
    59 mis-behaving software (which more than often is hard to detect), to hardware
    60 damage, or even to human risks (which is more than regrettable).
    62 Toolchains are made of different piece of software, each being quite complex
    63 and requiring specially crafted options to build and work seamlessly. This
    64 is usually not that easy, even in the not-so-trivial case of native toolchains.
    65 The work reaches a higher degree of complexity when it comes to cross-
    66 compilation, where it can become quite a nightmare...
    68 Some cross-toolchains exist on the internet, and can be used for general
    69 development, but they have a number of limitations:
    70   - they can be general purpose, in that they are configured for the majority:
    71     no optimisation for your specific target,
    72   - they can be prepared for a specific target and thus are not easy to use,
    73     nor optimised for, or even supporting your target,
    74   - they often are using aging components (compiler, C library, etc...) not
    75     supporting special features of your shiny new processor;
    76 On the other side, these toolchain offer some advantages:
    77   - they are ready to use and quite easy to install and setup,
    78   - they are proven if used by a wide community.
    80 But once you want to get all the juice out of your specific hardware, you will
    81 want to build your own toolchain. This is where crosstool-NG comes into play.
    83 There are also a number of tools that build toolchains for specific needs,
    84 which are not really scalable. Examples are:
    85   - buildroot ( whose main purpose is to build root file
    86     systems, hence the name. But once you have your toolchain with buildroot,
    87     part of it is installed in the root-to-be, so if you want to build a whole
    88     new root, you either have to save the existing one as a template and
    89     restore it later, or restart again from scratch. This is not convenient,
    90   - ptxdist (, whose purpose is very
    91     similar to buildroot,
    92   - other projects ( for example), which are again used to
    93     build root file systems.
    95 crosstool-NG is really targeted at building toolchains, and only toolchains.
    96 It is then up to you to use it the way you want.
    99 ___________
   100           /
   101 History  /
   102 ________/
   104 crosstool was first 'conceived' by Dan Kegel, who offered it to the community
   105 as a set of scripts, a repository of patches, and some pre-configured, general
   106 purpose setup files to be used to configure crosstool. This is available at
   107, and the subversion repository is hosted on
   108 google at
   110 I once managed to add support for uClibc-based toolchains, but it did not make
   111 into mainline, mostly because I didn't have time to port the patch forward to
   112 the new versions, due in part to the big effort it was taking.
   114 So I decided to clean up crosstool in the state it was, re-order the things
   115 in place, add appropriate support for what I needed, that is uClibc support
   116 and a menu-driven configuration, named the new implementation crosstool-NG,
   117 (standing for crosstool Next Generation, as many other comunity projects do,
   118 and as a wink at the TV series "Star Trek: The Next Generation" ;-) ) and
   119 made it available to the community, in case it was of interest to any one.
   122 _____________________________
   123                             /
   124 Referring to crosstool-NG  /
   125 __________________________/
   128 The long name of the project is crosstool-NG:
   129   * no leading uppercase (except as first word in a sentence)
   130   * crosstool and NG separated with a hyphen (dash)
   131   * NG in uppercase
   133 Crosstool-NG can also be referred to by its short name CT-NG:
   134   * all in uppercase
   135   * CT and NG separated with a hyphen (dash)
   137 The long name is preferred over the short name, except in mail subjects, where
   138 the short name is a better fit.
   140 When referring to a specific version of crosstool-NG, append the version number
   141 either as:
   142   * crosstool-NG X.Y.Z
   143     - the long name, a space, and the version string
   144   * crosstool-ng-X.Y.Z
   145     - the long name in lowercase, a hyphen (dash), and the version string
   146     - this is used to name the release tarballs
   147   * crosstool-ng-X.Y.Z+hg_id
   148     - the long name in lowercase, a hyphen, the version string, and the Hg id
   149       (as returned by: ct-ng version)
   150     - this is used to differentiate between releases and snapshots
   152 The frontend to crosstool-NG is the command ct-ng:
   153   * all in lowercase
   154   * ct and ng separated by a hyphen (dash)
   157 ___________________________
   158                           /
   159 Installing crosstool-NG  /
   160 ________________________/
   162 There are two ways you can use crosstool-NG:
   163  - build and install it, then get rid of the sources like you'd do for most
   164    programs,
   165  - or only build it and run from the source directory.
   167 The former should be used if you got crosstool-NG from a packaged tarball, see
   168 "Install method", below, while the latter is most useful for developpers that
   169 use a clone of the repository, and want to submit patches, see "The Hacker's
   170 way", below.
   172 Install method |
   173 ---------------+
   175 If you go for the install, then you just follow the classical, but yet easy
   176 ./configure way:
   177   ./configure --prefix=/some/place
   178   make
   179   make install
   180   export PATH="${PATH}:/some/place/bin"
   182 You can then get rid of crosstool-NG source. Next create a directory to serve
   183 as a working place, cd in there and run:
   184   ct-ng help
   186 See below for complete usage.
   188 The Hacker's way |
   189 -----------------+
   191 If you go the hacker's way, then the usage is a bit different, although very
   192 simple:
   193   ./configure --local
   194   make
   196 Now, *do not* remove crosstool-NG sources. They are needed to run crosstool-NG!
   197 Stay in the directory holding the sources, and run:
   198   ./ct-ng help
   200 See below for complete usage.
   202 Now, provided you used a clone of the repository, you can send me your changes.
   203 See the section titled CONTRIBUTING, below, for how to submit changees.
   205 Preparing for packaging |
   206 ------------------------+
   208 If you plan on packaging crosstool-NG, you surely don't want to install it
   209 in your root file system. The install procedure of crosstool-NG honors the
   210 DESTDIR variable:
   212   ./configure --prefix=/usr
   213   make
   214   make DESTDIR=/packaging/place install
   216 Shell completion |
   217 -----------------+
   219 crosstool-NG comes with a shell script fragment that defines bash-compatible
   220 completion. That shell fragment is currently not installed automatically, but
   221 this is planned.
   223 To install the shell script fragment, you have two options:
   224  - install system-wide, most probably by copying ct-ng.comp into
   225    /etc/bash_completion.d/
   226  - install for a single user, by copying ct-ng.comp into ${HOME}/ and
   227    sourcing this file from your ${HOME}/.bashrc
   229 Contributed code |
   230 -----------------+
   232 Some people contibuted code that couldn't get merged for various reasons. This
   233 code is available as lzma-compressed patches, in the contrib/ sub-directory.
   234 These patches are to be applied to the source of crosstool-NG, prior to
   235 installing, using something like the following:
   236   lzcat contrib/foobar.patch.lzma |patch -p1
   238 There is no guarantee that a particuliar contribution applies to the current
   239 version of crosstool-ng, or that it will work at all. Use contributions at
   240 your own risk.
   243 ____________________________
   244                            /
   245 Configuring crosstool-NG  /
   246 _________________________/
   248 crosstool-NG is configured with a configurator presenting a menu-stuctured set
   249 of options. These options let you specify the way you want your toolchain
   250 built, where you want it installed, what architecture and specific processor it
   251 will support, the version of the components you want to use, etc... The
   252 value for those options are then stored in a configuration file.
   254 The configurator works the same way you configure your Linux kernel. It is
   255 assumed you now how to handle this.
   257 To enter the menu, type:
   258   ct-ng menuconfig
   260 Almost every config item has a help entry. Read them carefully.
   262 String and number options can refer to environment variables. In such a case,
   263 you must use the shell syntax: ${VAR}. You shall neither single- nor double-
   264 quote the string/number options.
   266 There are three environment variables that are computed by crosstool-NG, and
   267 that you can use:
   269 CT_TARGET:
   270   It represents the target tuple you are building for. You can use it for
   271   example in the installation/prefix directory, such as:
   272     /opt/x-tools/${CT_TARGET}
   274 CT_TOP_DIR:
   275   The top directory where crosstool-NG is running. You shouldn't need it in
   276   most cases. There is one case where you may need it: if you have local
   277   patches and you store them in your running directory, you can refer to them
   278   by using CT_TOP_DIR, such as:
   279     ${CT_TOP_DIR}/patches.myproject
   281 CT_VERSION:
   282   The version of crosstool-NG you are using. Not much use for you, but it's
   283   there if you need it.
   285 Interesting config options |
   286 ---------------------------+
   289   If you already have some tarballs in a direcotry, enter it here. That will
   290   speed up the retrieving phase, where crosstool-NG would otherwise download
   291   those tarballs.
   294   This is where the toolchain will be installed in (and for now, where it
   295   will run from). Common use is to add the target tuple in the directory
   296   path, such as (see above):
   297     /opt/x-tools/${CT_TARGET}
   300   An identifier for your toolchain, will take place in the vendor part of the
   301   target tuple. It shall *not* contain spaces or dashes. Usually, keep it
   302   to a one-word string, or use underscores to separate words if you need.
   303   Avoid dots, commas, and special characters.
   306   An alias for the toolchian. It will be used as a prefix to the toolchain
   307   tools. For example, you will have ${CT_TARGET_ALIAS}-gcc
   309 Also, if you think you don't see enough versions, you can try to enable one of
   310 those:
   313   Show obsolete versions or tools. Most of the time, you don't want to base
   314   your toolchain on too old a version (of gcc, for example). But at times, it
   315   can come handy to use such an old version for regression tests. Those old
   316   versions are hidden behind CT_OBSOLETE. Those versions (or features) are so
   317   marked because maintaining support for those in crosstool-NG would be too
   318   costly, time-wise, and time is dear.
   321   Show experimental versions or tools. Again, you might not want to base your
   322   toolchain on too recent tools (eg. gcc) for production. But if you need a
   323   feature present only in a recent version, or a new tool, you can find them
   324   hidden behind CT_EXPERIMENTAL. Those versions (or features) did not (yet)
   325   receive thorough testing in crosstool-NG, and/or are not mature enough to
   326   be blindly trusted.
   328 Re-building an existing toolchain |
   329 ----------------------------------+
   331 If you have an existing toolchain, you can re-use the options used to build it
   332 to create a new toolchain. That needs a very little bit of effort on your side
   333 but is quite easy. The options to build a toolchain are saved with the
   334 toolchain, and you can retrieve this configuration by running:
   335   ${CT_TARGET}-ct-ng.config
   337 An alternate method is to extract the configuration from a build.log file.
   338 This will be necessary if your toolchain was build with crosstool-NG prior
   339 to 1.4.0, but can be used with build.log files from any version:
   340   ct-ng extractconfig <build.log >.config
   342 Or, if your build.log file is compressed (most probably!):
   343   bzcat build.log.bz2 |ct-ng extractconfig >.config
   345 The above commands will dump the configuration to stdout, so to rebuild a
   346 toolchain with this configuration, just redirect the output to the
   347 .config file:
   348   ${CT_TARGET}-ct-ng.config >.config
   349   ct-ng oldconfig
   351 Then, you can review and change the configuration by running:
   352   ct-ng menuconfig
   354 Using as a backend for a build-system |
   355 --------------------------------------+
   357 Crosstool-NG can be used as a backend for an automated build-system. In this
   358 case, some components that are expected to run on the target (eg. the native
   359 gdb, ltrace, DUMA...) are not available in the menuconfig, and they are not
   360 build either, as it is considered the responsibility of the build-system to
   361 build its own versions of those tools.
   363 If you want to use crosstool-NG as a backend to generate your toolchains for
   364 your build-system, you have to set and export this environment variable:
   365   CT_IS_A_BACKEND=y
   367 (case is not sensitive, you can say Y).
   370 ________________________
   371                        /
   372 Running crosstool-NG  /
   373 _____________________/
   375 To build the toolchain, simply type:
   376   ct-ng build
   378 This will use the above configuration to retrieve, extract and patch the
   379 components, build, install and eventually test your newly built toolchain.
   381 You are then free to add the toolchain /bin directory in your PATH to use
   382 it at will.
   384 In any case, you can get some terse help. Just type:
   385   ct-ng help
   386 or:
   387   man 1 ct-ng
   389 Stopping and restarting a build |
   390 --------------------------------+
   392 If you want to stop the build after a step you are debugging, you can pass the
   393 variable STOP to make:
   394   ct-ng build STOP=some_step
   396 Conversely, if you want to restart a build at a specific step you are
   397 debugging, you can pass the RESTART variable to make:
   398   ct-ng build RESTART=some_step
   400 Alternatively, you can call make with the name of a step to just do that step:
   401   ct-ng libc_headers
   402 is equivalent to:
   403   ct-ng build RESTART=libc_headers STOP=libc_headers
   405 The shortcuts +step_name and step_name+ allow to respectively stop or restart
   406 at that step. Thus:
   407   ct-ng +libc_headers              and:    ct-ng libc_headers+
   408 are equivalent to:
   409   ct-ng build STOP=libc_headers    and:    ct-ng build RESTART=libc_headers
   411 To obtain the list of acceptable steps, please call:
   412   ct-ng list-steps
   414 Note that in order to restart a build, you'll have to say 'Y' to the config
   415 option CT_DEBUG_CT_SAVE_STEPS, and that the previous build effectively went
   416 that far.
   418 Building all toolchains at once |
   419 --------------------------------+
   421 You can build all samples; simply call:
   422   ct-ng build-all
   424 Overriding the number of // jobs |
   425 ---------------------------------+
   427 If you want to override the number of jobs to run in // (the -j option to
   428 make), you can either re-enter the menuconfig, or simply add it on the command
   429 line, as such:
   430   ct-ng build.4
   432 which tells crosstool-NG to override the number of // jobs to 4.
   434 You can see the actions that support overriding the number of // jobs in
   435 the help menu. Those are the ones with [.#] after them (eg. build[.#] or
   436 build-all[.#], and so on...).
   438 Note on // jobs |
   439 ----------------+
   441 The crosstool-NG script 'ct-ng' is a Makefile-script. It does *not* execute
   442 in parallel (there is not much to gain). When speaking of // jobs, we are
   443 refering to the number of // jobs when making the *components*. That is, we
   444 speak of the number of // jobs used to build gcc, glibc, and so on...
   446 Tools wrapper |
   447 --------------+
   449 Starting with gcc-4.3 come two new dependencies: GMP and MPFR. With gcc-4.4,
   450 come three new ones: PPL, CLooG/ppl and MPC. With gcc-4.5 again comes a new
   451 dependency on libelf. These are libraries that enable advanced features to
   452 gcc. Additionally, some of those libraries can be used by binutils and gdb.
   453 Unfortunately, not all systems on which crosstool-NG runs have all of those
   454 libraries. And for those that do, the versions of those libraries may be
   455 older than the version required by gcc (and binutils and gdb). To date,
   456 Debian stable (aka Lenny) is lagging behind on some, and is missing the
   457 others.
   459 This is why crosstool-NG builds its own set of libraries as part of the
   460 toolchain.
   462 The companion libraries can be built either as static libraries, or as shared
   463 libraries. The default is to build static libraries, and is the safe way.
   464 If you decide to use static companion libraries, then you can stop reading
   465 this section.
   467 But if you prefer to have shared libraries, then read on...
   469 Building shared companion libraries poses no problem at build time, as
   470 crosstool-NG correctly points gcc (and binutils and gdb) to the correct
   471 place where our own version of the libraries are installed. But it poses
   472 a problem when gcc et al. are run: the place where the libraries are is most
   473 probably not known to the host dynamic linker. Still worse, if the host system
   474 has its own versions, then would load the wrong libraries!
   476 So we have to force the dynamic linker to load the correct version. We do this
   477 by using the LD_LIBRARY_PATH variable, that informs the dynamic linker where
   478 to look for shared libraries prior to searching its standard places. But we
   479 can't impose that burden on all the system (because it'd be a nightmare to
   480 configure, and because two toolchains on the same system may use different
   481 versions of the libraries); so we have to do it on a per-toolchain basis.
   483 So we rename all binaries of the toolchain (by adding a dot '.' as their first
   484 character), and add a small program, the so-called "tools wrapper", that
   485 correctly sets LD_LIBRARY_PATH prior to running the real tool.
   487 First, the wrapper was written as a POSIX-compliant shell script. That shell
   488 script is very simple, if not trivial, and works great. The only drawback is
   489 that it does not work on host systems that lack a shell, for example the
   490 MingW32 environment. To solve the issue, the wrapper has been re-written in C,
   491 and compiled at build time. This C wrapper is much more complex than the shell
   492 script, and although it sems to be working, it's been only lightly tested.
   493 Some of the expected short-comings with this C wrapper are;
   494  - multi-byte file names may not be handled correctly
   495  - it's really big for what it does
   497 So, the default wrapper installed with your toolchain is the shell script.
   498 If you know that your system is missing a shell, then you shall use the C
   499 wrapper (and report back whether it works, or does not work, for you).
   501 A final word on the subject: do not build shared libraries. Build them
   502 static, and you'll be safe.
   505 _______________________
   506                       /
   507 Using the toolchain  /
   508 ____________________/
   510 Using the toolchain is as simple as adding the toolchain's bin directory in
   511 your PATH, such as:
   512   export PATH="${PATH}:/your/toolchain/path/bin"
   514 and then using the target tuple to tell the build systems to use your
   515 toolchain:
   516   ./configure --target=your-target-tuple
   517 or
   518   make CC=your-target-tuple-gcc
   519 or
   520   make CROSS_COMPILE=your-target-tuple-
   521 and so on...
   523 It is strongly advised not to use the toolchain sys-root directory as an
   524 install directory for your programs/packages. If you do so, you will not be
   525 able to use your toolchain for another project. It is even strongly advised
   526 that your toolchain is chmod-ed to read-only once successfully build, so that
   527 you don't go polluting your toolchain with your programs/packages' files.
   529 Thus, when you build a program/package, install it in a separate directory,
   530 eg. /your/root. This directory is the /image/ of what would be in the root file
   531 system of your target, and will contain all that your programs/packages have
   532 installed.
   534 The 'populate' script |
   535 ----------------------+
   537 When your root directory is ready, it is still missing some important bits: the
   538 toolchain's libraries. To populate your root directory with those libs, just
   539 run:
   540   your-target-tuple-populate -s /your/root -d /your/root-populated
   542 This will copy /your/root into /your/root-populated, and put the needed and only
   543 the needed libraries there. Thus you don't polute /your/root with any cruft that
   544 would no longer be needed should you have to remove stuff. /your/root always
   545 contains only those things you install in it.
   547 You can then use /your/root-populated to build up your file system image, a
   548 tarball, or to NFS-mount it from your target, or whatever you need.
   550 The populate script accepts the following options:
   552  -s src_dir
   553     Use 'src_dir' as the un-populated root directory.
   555  -d dst_dir
   556     Put the populated root directory in 'dst_dir'.
   558  -l lib1 [...]
   559     Always add specified libraries.
   561  -L file
   562     Always add libraries listed in 'file'.
   564  -f
   565     Remove 'dst_dir' if it previously existed; continue even if any library
   566     specified with -l or -L is missing.
   568  -v
   569     Be verbose, and tell what's going on (you can see exactly where libs are
   570     coming from).
   572  -h
   573     Print the help.
   575 See 'your-target-tuple-populate -h' for more information on the options.
   577 Here is how populate works:
   579   1) performs some sanity checks:
   580      - src_dir and dst_dir are specified
   581      - src_dir exists
   582      - unless forced, dst_dir does not exist
   583      - src_dir != dst_dir
   585   2) copy src_dir to dst_dir
   587   3) add forced libraries to dst_dir
   588      - build the list from -l and -L options
   589      - get forced libraries from the sysroot (see below for heuristics)
   590        - abort on the first missing library, unless -f is specified
   592   4) add all missing libraries to dst_dir
   593      - scan dst_dir for every ELF files that are 'executable' or
   594        'shared object'
   595      - list the "NEEDED Shared library" fields
   596        - check if the library is already in dst_dir/lib or dst_dir/usr/lib
   597        - if not, get the library from the sysroot
   598          - if it's in sysroot/lib, copy it to dst_dir/lib
   599          - if it's in sysroot/usr/lib, copy it to dst_dir/usr/lib
   600          - in both cases, use the SONAME of the library to create the file
   601            in dst_dir
   602          - if it was not found in the sysroot, this is an error.
   605 ___________________
   606                   /
   607 Toolchain types  /
   608 ________________/
   610 There are four kinds of toolchains you could encounter.
   612 First off, you must understand the following: when it comes to compilers there
   613 are up to four machines involved:
   614   1) the machine configuring the toolchain components: the config machine
   615   2) the machine building the toolchain components:    the build machine
   616   3) the machine running the toolchain:                the host machine
   617   4) the machine the toolchain is generating code for: the target machine
   619 We can most of the time assume that the config machine and the build machine
   620 are the same. Most of the time, this will be true. The only time it isn't
   621 is if you're using distributed compilation (such as distcc). Let's forget
   622 this for the sake of simplicity.
   624 So we're left with three machines:
   625  - build
   626  - host
   627  - target
   629 Any toolchain will involve those three machines. You can be as pretty sure of
   630 this as "2 and 2 are 4". Here is how they come into play:
   632 1) build == host == target
   633     This is a plain native toolchain, targetting the exact same machine as the
   634     one it is built on, and running again on this exact same machine. You have
   635     to build such a toolchain when you want to use an updated component, such
   636     as a newer gcc for example.
   637     crosstool-NG calls it "native".
   639 2) build == host != target
   640     This is a classic cross-toolchain, which is expected to be run on the same
   641     machine it is compiled on, and generate code to run on a second machine,
   642     the target.
   643     crosstool-NG calls it "cross".
   645 3) build != host == target
   646     Such a toolchain is also a native toolchain, as it targets the same machine
   647     as it runs on. But it is build on another machine. You want such a
   648     toolchain when porting to a new architecture, or if the build machine is
   649     much faster than the host machine.
   650     crosstool-NG calls it "cross-native".
   652 4) build != host != target
   653     This one is called a canadian-toolchain (*), and is tricky. The three
   654     machines in play are different. You might want such a toolchain if you
   655     have a fast build machine, but the users will use it on another machine,
   656     and will produce code to run on a third machine.
   657     crosstool-NG calls it "canadian".
   659 crosstool-NG can build all these kinds of toolchains (or is aiming at it,
   660 anyway!)
   662 (*) The term Canadian Cross came about because at the time that these issues
   663     were all being hashed out, Canada had three national political parties.
   667 ________________
   668                /
   669 Contributing  /
   670 _____________/
   672 Sending a bug report |
   673 ---------------------+
   675 If you need to send a bug report, please send a mail with subject
   676 prefixed with "[CT_NG]" with to following destinations:
   677     TO: yann.morin.1998 (at)
   678     CC: crossgcc (at)
   680 Sending patches |
   681 ----------------+
   683 If you want to enhance crosstool-NG, there's a to-do list in the TODO file.
   685 Patches should come with the appropriate SoB line. A SoB line is typically
   686 something like:
   687    Signed-off-by: John DOE <>
   689 The SoB line is clearly described in Documentation/SubmittingPatches , section
   690 12, of your favourite Linux kernel source tree.
   693 How to Use Mercurial |
   694 ---------------------+
   696 For larger or more frequent contributions, mercurial should be used.
   700 Configuring Mercurial:
   701   You need mercurial with the following extensions:
   702    - mq        :
   703    - patchbomb :
   704   Usually, these two extensions are already part of the installation package.
   705   The mq extension maintains a separate queue of your local changes
   706   that you can change at any later time.
   707   With the patchbomb extension you can email those patches directly
   708   from your local repo.
   710   Your configuration file for mercurial, e.g. ~/.hgrc should contain
   711   at least the following sections (but have a look at `man hgrc`):
   712   # ---
   713   [email]
   714   # configure sending patches directly via Mercurial
   715   from = "Your Name" <your@email.address>
   716   # How to send email:
   717   method = smtp
   719   [smtp]
   720   # SMTP configuration (only for method=smtp)
   721   host = localhost
   722   tls = true
   723   username =
   724   password =
   726   [extensions]
   727   # The following lines enable the two extensions:
   728 =
   729   hgext.patchbomb =
   730   # ----
   732 Create your local repository as a clone:
   733   hg clone crosstool-ng
   735 Setting up the mq extension in your local copy:
   736   cd crosstool-ng
   737   hg qinit
   742 Recording your changes in the patch queue maintained by mq:
   743   # First, create a new patch entry in the patch queue:
   744   hg qnew -D -U -e short_patch_name1
   745   <edit patch description as commit message (see below for an example)>
   747   <now edit the ct-ng sources and check them>
   749   # if you execute `hg status` here, your modifications of the working
   750   # copy should show up.
   752   # Now the following command takes your modifications from the working copy
   753   # into the patch entry
   754   hg qrefresh -D [-e]
   755   <reedit patch description [-e] if desired>
   757   # Now your changes are recorded, and `hg status` should show a clean
   758   # working copy
   760 Repeat the above steps for all your modifications.
   761 The command `hg qseries` informs you about the content of your patch queue.
   766 Once you are satisfied with your patch series, you can (you should!)
   767 contribute them back to upstream.
   768 This is easily done using the `hg email` command.
   770 `hg email` sends your new changesets to a specified list of recipients,
   771 each patch in its own email, all ordered in the way you entered them (oldest
   772 first). The command line flag --outgoing selects all changesets that are in
   773 your local but not yet in the upstream repository. Here, these are exactly
   774 the ones you entered into your local patch queue in the section above, so
   775 --outgoing is what you want.
   777 Each email gets the subject set to:  "[PATCH x of n] <series summary>"
   778 where 'x' is the serial number in the email series, and 'n' is the total number
   779 of patches in the series. The body of the email is the complete patch, plus
   780 a handful of metadata, that helps properly apply the patch, keeping the log
   781 message, attribution and date, tracking file changes (move, delete, modes...)
   783 `hg email` also threads all outgoing patch emails below an introductory
   784 message. You should use the introductory message (command line flag --intro)
   785 to describe the scope and motivation for the whole patch series. The subject
   786 for the introductory message gets set to:  "[PATCH 0 of n] <series summary>"
   787 and you get the chance to set the <series summary>.
   789 Here is a sample `hg email` complete command line:
   790 Note: replace " (at) " with "@"
   792   hg email --outgoing --intro   \
   793            --to '"Yann E. MORIN" <yann.morin.1998 (at)>'    \
   794            --cc 'crossgcc (at)'
   796   # It then opens an editor and lets you enter the subject
   797   # and the body for the introductory message.
   799 Use `hg email` with the additional command line switch -n to
   800 first have a look at the email(s) without actually sending them.
   805 When the patches are refined by discussing them on the mailing list,
   806 you may want to finalize and resend them.
   808 The mq extension has the idiosyncrasy of imposing a stack onto the queue:
   809 You can always reedit/refresh only the patch on top of stack.
   810 The queue consists of applied and unapplied patches
   811 (if you reached here via the above steps, all of your patches are applied),
   812 where the 'stack' consists of the applied patches, and 'top of stack'
   813 is the latest applied patch.
   815 The following output of `hg qseries` is now used as an example:
   816   0 A short_patch_name1
   817   1 A short_patch_name2
   818   2 A short_patch_name3
   819   3 A short_patch_name4
   821 You are now able to edit patch 'short_patch_name4' (which is top of stack):
   822   <Edit the sources>
   823   # and execute again
   824   hg qrefresh -D [-e]
   825   <and optionally [-e] reedit the commit message>
   827 If you want to edit e.g. patch short_patch_name2, you have to modify
   828 mq's stack so this patch gets top of stack.
   829 For this purpose see `hg help qgoto`, `hg help qpop`, and `hg help qpush`.
   831   hg qgoto short_patch_name2
   832   # The patch queue should now look like
   833   hg qseries
   834     0 A short_patch_name1
   835     1 A short_patch_name2
   836     2 U short_patch_name3
   837     3 U short_patch_name4
   838   # so patch # 1 (short_patch_name2) is top of stack.
   839   <now reedit the sources for short_patch_name2>
   840   # and execute again
   841   hg qrefresh -D [-e]
   842   <and optionally [-e] reedit the commit message>
   843   # the following command reapplies the now unapplied two patches:
   844   hg qpush -a
   845   # you can also use `hg qgoto short_patch_name4` to get there again.
   850 By mailing list policy, please resend your complete patch series.
   851 --> Go back to section "CONTRIBUTING YOUR PATCHES" and resubmit the full set.
   856 You can sync your repo with upstream at any time by executing
   857   # first unapply all your patches:
   858   hg qpop -a
   859   # next fetch new changesets from upstream
   860   hg pull
   861   # then update your working copy
   862   hg up
   863   # optionally remove already upstream integrated patches (see below)
   864   hg qdelete <short_name_of_already_applied_patch>
   865   # and reapply your patches if any non upstream-integrated left (but see below)
   866   hg qpush -a
   868 Eventually, your patches get included into the upstream repository
   869 which you initially cloned.
   870 In this case, before executing the hg qpush -a from above
   871 you should manually "hg qdelete" the patches that are already integrated upstream.
   874 HOW TO FORMAT COMMIT MESSAGES (aka patch desciptions):
   876 Commit messages should look like (without leading pipes):
   877  |component: short, one-line description
   878  |
   879  |optional longer description
   880  |on multiple lines if needed
   882 Here is an example commit message (see revision a53a5e1d61db):
   883  |comp-libs/cloog: fix building
   884  |
   885  |For CLooG/PPL 0.15.3, the directory name was simply cloog-ppl.
   886  |For any later versions, the directory name does have the version, such as
   887  |cloog-ppl-0.15.4.
   889 _____________
   890             /
   891 Internals  /
   892 __________/
   894 Internally, crosstool-NG is script-based. To ease usage, the frontend is
   895 Makefile-based.
   897 Makefile front-end |
   898 -------------------+
   900 The entry point to crosstool-NG is the Makefile script "ct-ng". Calling this
   901 script with an action will act exactly as if the Makefile was in the current
   902 working directory and make was called with the action as rule. Thus:
   903   ct-ng menuconfig
   905 is equivalent to having the Makefile in CWD, and calling:
   906   make menuconfig
   908 Having ct-ng as it is avoids copying the Makefile everywhere, and acts as a
   909 traditional command.
   911 ct-ng loads sub- Makefiles from the library directory $(CT_LIB_DIR), as set up
   912 at configuration time with ./configure.
   914 ct-ng also searches for config files, sub-tools, samples, scripts and patches in
   915 that library directory.
   917 Because of a stupid make behavior/bug I was unable to track down, implicit make
   918 rules are disabled: installing with --local would triger those rules, and mconf
   919 was unbuildable.
   921 Kconfig parser |
   922 ---------------+
   924 The kconfig language is a hacked version, vampirised from the Linux kernel
   925 (, and (heavily) adapted to my needs.
   927 The list of the most notable changes (at least the ones I remember) follows:
   928 - the CONFIG_ prefix has been replaced with CT_
   929 - a leading | in prompts is skipped, and subsequent leading spaces are not
   930   trimmed; otherwise leading spaces are silently trimmed
   931 - removed the warning about undefined environment variable
   933 The kconfig parsers (conf and mconf) are not installed pre-built, but as
   934 source files. Thus you can have the directory where crosstool-NG is installed,
   935 exported (via NFS or whatever) and have clients with different architectures
   936 use the same crosstool-NG installation, and most notably, the same set of
   937 patches.
   939 Architecture-specific |
   940 ----------------------+
   942 Note: this chapter is not really well written, and might thus be a little bit
   943 complex to understand. To get a better grasp of what an architecture is, the
   944 reader is kindly encouraged to look at the "arch/" sub-directory, and to the
   945 existing architectures to see how things are laid out.
   947 An architecture is defined by:
   949  - a human-readable name, in lower case letters, with numbers as appropriate.
   950    The underscore is allowed; space and special characters are not.
   951      Eg.: arm, x86_64
   952  - a file in "config/arch/", named after the architecture's name, and suffixed
   953    with ".in".
   954      Eg.: config/arch/
   955  - a file in "scripts/build/arch/", named after the architecture's name, and
   956    suffixed with ".sh".
   957      Eg.: scripts/build/arch/
   959 The architecture's ".in" file API:
   960  > the config option "ARCH_%arch%" (where %arch% is to be replaced with the
   961    actual architecture name).
   962    That config option must have *neither* a type, *nor* a prompt! Also, it can
   963    *not* depend on any other config option (EXPERIMENTAL is managed as above).
   964      Eg.:
   965        config ARCH_arm
   966    + mandatory:
   967        defines a (terse) help entry for this architecture:
   968        Eg.:
   969          config ARCH_arm
   970            help
   971              The ARM architecture.
   972    + optional:
   973        selects adequate associated config options.
   974        Note: 64-bit architectures *shall* select ARCH_64
   975        Eg.:
   976          config ARCH_arm
   977            select ARCH_SUPPORTS_BOTH_ENDIAN
   978            select ARCH_DEFAULT_LE
   979            help
   980              The ARM architecture.
   981        Eg.:
   982          config ARCH_x86_64
   983             select ARCH_64
   984             help
   985               The x86_64 architecture.
   987  > other target-specific options, at your discretion. Note however that to
   988    avoid name-clashing, such options shall be prefixed with "ARCH_%arch%",
   989    where %arch% is again replaced by the actual architecture name.
   990    (Note: due to historical reasons, and lack of time to clean up the code,
   991     I may have left some config options that do not completely conform to
   992     this, as the architecture name was written all upper case. However, the
   993     prefix is unique among architectures, and does not cause harm).
   995 The architecture's ".sh" file API:
   996  > the function "CT_DoArchTupleValues"
   997    + parameters: none
   998    + environment:
   999      - all variables from the ".config" file,
  1000      - the two variables "target_endian_eb" and "target_endian_el" which are
  1001        the endianness suffixes
  1002    + return value: 0 upon success, !0 upon failure
  1003    + provides:
  1004      - mandatory
  1005      - the environment variable CT_TARGET_ARCH
  1006      - contains:
  1007        the architecture part of the target tuple.
  1008        Eg.: "armeb" for big endian ARM
  1009             "i386" for an i386
  1010    + provides:
  1011      - optional
  1012      - the environment variable CT_TARGET_SYS
  1013      - contains:
  1014        the sytem part of the target tuple.
  1015        Eg.: "gnu" for glibc on most architectures
  1016             "gnueabi" for glibc on an ARM EABI
  1017      - defaults to:
  1018        - for glibc-based toolchain: "gnu"
  1019        - for uClibc-based toolchain: "uclibc"
  1020    + provides:
  1021      - optional
  1022      - the environment variables to configure the cross-gcc (defaults)
  1023        - CT_ARCH_WITH_ARCH    : the gcc ./configure switch to select architecture level         ( "--with-arch=${CT_ARCH_ARCH}"   )
  1024        - CT_ARCH_WITH_ABI     : the gcc ./configure switch to select ABI level                  ( "--with-abi=${CT_ARCH_ABI}"     )
  1025        - CT_ARCH_WITH_CPU     : the gcc ./configure switch to select CPU instruction set        ( "--with-cpu=${CT_ARCH_CPU}"     )
  1026        - CT_ARCH_WITH_TUNE    : the gcc ./configure switch to select scheduling                 ( "--with-tune=${CT_ARCH_TUNE}"   )
  1027        - CT_ARCH_WITH_FPU     : the gcc ./configure switch to select FPU type                   ( "--with-fpu=${CT_ARCH_FPU}"     )
  1028        - CT_ARCH_WITH_FLOAT   : the gcc ./configure switch to select floating point arithmetics ( "--with-float=soft" or /empty/  )
  1029    + provides:
  1030      - optional
  1031      - the environment variables to pass to the cross-gcc to build target binaries (defaults)
  1032        - CT_ARCH_ARCH_CFLAG   : the gcc switch to select architecture level                     ( "-march=${CT_ARCH_ARCH}"            )
  1033        - CT_ARCH_ABI_CFLAG    : the gcc switch to select ABI level                              ( "-mabi=${CT_ARCH_ABI}"              )
  1034        - CT_ARCH_CPU_CFLAG    : the gcc switch to select CPU instruction set                    ( "-mcpu=${CT_ARCH_CPU}"              )
  1035        - CT_ARCH_TUNE_CFLAG   : the gcc switch to select scheduling                             ( "-mtune=${CT_ARCH_TUNE}"            )
  1036        - CT_ARCH_FPU_CFLAG    : the gcc switch to select FPU type                               ( "-mfpu=${CT_ARCH_FPU}"              )
  1037        - CT_ARCH_FLOAT_CFLAG  : the gcc switch to choose floating point arithmetics             ( "-msoft-float" or /empty/           )
  1038        - CT_ARCH_ENDIAN_CFLAG : the gcc switch to choose big or little endian                   ( "-mbig-endian" or "-mlittle-endian" )
  1039      - default to:
  1040        see above.
  1041    + provides:
  1042      - optional
  1043      - the environement variables to configure the core and final compiler, specific to this architecture:
  1044        - CT_ARCH_CC_CORE_EXTRA_CONFIG   : additional, architecture specific core gcc ./configure flags
  1045        - CT_ARCH_CC_EXTRA_CONFIG        : additional, architecture specific final gcc ./configure flags
  1046      - default to:
  1047        - all empty
  1048    + provides:
  1049      - optional
  1050      - the architecture-specific CFLAGS and LDFLAGS:
  1051        - CT_ARCH_TARGET_CLFAGS
  1052        - CT_ARCH_TARGET_LDFLAGS
  1053      - default to:
  1054        - all empty
  1056 You can have a look at "config/arch/" and "scripts/build/arch/" for
  1057 a quite complete example of what an actual architecture description looks like.
  1059 Kernel specific |
  1060 ----------------+
  1062 A kernel is defined by:
  1064  - a human-readable name, in lower case letters, with numbers as appropriate.
  1065    The underscore is allowed; space and special characters are not (although
  1066    they are internally replaced with underscores.
  1067      Eg.: linux, bare-metal
  1068  - a file in "config/kernel/", named after the kernel name, and suffixed with
  1069    ".in".
  1070      Eg.: config/kernel/, config/kernel/
  1071  - a file in "scripts/build/kernel/", named after the kernel name, and suffixed
  1072    with ".sh".
  1073      Eg.: scripts/build/kernel/, scripts/build/kernel/
  1075 The kernel's ".in" file must contain:
  1076  > an optional lines containing exactly "# EXPERIMENTAL", starting on the
  1077    first column, and without any following space or other character.
  1078    If this line is present, then this kernel is considered EXPERIMENTAL,
  1079    and correct dependency on EXPERIMENTAL will be set.
  1081  > the config option "KERNEL_%kernel_name%" (where %kernel_name% is to be
  1082    replaced with the actual kernel name, with all special characters and
  1083    spaces replaced by underscores).
  1084    That config option must have *neither* a type, *nor* a prompt! Also, it can
  1085    *not* depends on EXPERIMENTAL.
  1086      Eg.: KERNEL_linux, KERNEL_bare_metal
  1087    + mandatory:
  1088        defines a (terse) help entry for this kernel.
  1089        Eg.:
  1090          config KERNEL_bare_metal
  1091            help
  1092              Build a compiler for use without any kernel.
  1093    + optional:
  1094        selects adequate associated config options.
  1095        Eg.:
  1096          config KERNEL_bare_metal
  1097            select BARE_METAL
  1098            help
  1099              Build a compiler for use without any kernel.
  1101  > other kernel specific options, at your discretion. Note however that, to
  1102    avoid name-clashing, such options should be prefixed with
  1103    "KERNEL_%kernel_name%", where %kernel_name% is again tp be replaced with
  1104    the actual kernel name.
  1105    (Note: due to historical reasons, and lack of time to clean up the code,
  1106     I may have left some config options that do not completely conform to
  1107     this, as the kernel name was written all upper case. However, the prefix
  1108     is unique among kernels, and does not cause harm).
  1110 The kernel's ".sh" file API:
  1111  > is a bash script fragment
  1113  > defines the function CT_DoKernelTupleValues
  1114    + see the architecture's CT_DoArchTupleValues, except for:
  1115    + set the environment variable CT_TARGET_KERNEL, the kernel part of the
  1116      target tuple
  1117    + return value: ignored
  1119  > defines the function "do_kernel_get":
  1120    + parameters: none
  1121    + environment:
  1122       - all variables from the ".config" file.
  1123    + return value: 0 for success, !0 for failure.
  1124    + behavior: download the kernel's sources, and store the tarball into
  1125      "${CT_TARBALLS_DIR}". To this end, a functions is available, that
  1126      abstracts downloading tarballs:
  1127      - CT_DoGet <tarball_base_name> <URL1 [URL...]>
  1128        Eg.: CT_DoGet linux-
  1129      Note: retrieving sources from svn, cvs, git and the likes is not supported
  1130      by CT_DoGet. You'll have to do this by hand, as it is done for eglibc in
  1131      "scripts/build/libc/"
  1133  > defines the function "do_kernel_extract":
  1134    + parameters: none
  1135    + environment:
  1136       - all variables from the ".config" file,
  1137    + return value: 0 for success, !0 for failure.
  1138    + behavior: extract the kernel's tarball into "${CT_SRC_DIR}", and apply
  1139      required patches. To this end, a function is available, that abstracts
  1140      extracting tarballs:
  1141      - CT_ExtractAndPatch <tarball_base_name>
  1142        Eg.: CT_ExtractAndPatch linux-
  1144  > defines the function "do_kernel_headers":
  1145    + parameters: none
  1146    + environment:
  1147       - all variables from the ".config" file,
  1148    + return value: 0 for success, !0 for failure.
  1149    + behavior: install the kernel headers (if any) in "${CT_SYSROOT_DIR}/usr/include"
  1151  > defines any kernel-specific helper functions
  1152    These functions, if any, must be prefixed with "do_kernel_%CT_KERNEL%_",
  1153    where '%CT_KERNEL%' is to be replaced with the actual kernel name, to avoid
  1154    any name-clashing.
  1156 You can have a look at "config/kernel/" and "scripts/build/kernel/"
  1157 as an example of what a complex kernel description looks like.
  1159 Adding a new version of a component |
  1160 ------------------------------------+
  1162 When a new component, such as the Linux kernel, gcc or any other is released,
  1163 adding the new version to crosstool-NG is quite easy. There is a script that
  1164 will do all that for you:
  1165   scripts/
  1167 Run it with no option to get some help.
  1169 Build scripts |
  1170 --------------+
  1172 To Be Written later...