File.........: C - Misc. tutorials.txt Copyright....: (C) 2010 Yann E. MORIN License......: Creative Commons Attribution Share Alike (CC-by-sa), v2.5 Misc. tutorials / ________________/ Using crosstool-NG on FreeBSD (and other *BSD) | -----------------------------------------------+ Contributed by: Titus von Boxberg Prerequisites and instructions for using ct-ng for building a cross toolchain on FreeBSD as host. 0) Tested on FreeBSD 8.0 1) Install (at least) the following ports archivers/lzma textproc/gsed devel/gmake devel/patch shells/bash devel/bison lang/gawk devel/automake110 ftp/wget Of course, you should have /usr/local/bin in your PATH. 2) run ct-ng's configure with the following tool configuration: ./configure --with-sed=/usr/local/bin/gsed --with-make=/usr/local/bin/gmake \ --with-patch=/usr/local/bin/gpatch [...other configure parameters as you like...] 3) proceed as described in general documentation but use gmake instead of make Using crosstool-NG on MacOS-X | ------------------------------+ Contributed by: Titus von Boxberg Prerequisites and instructions for using crosstool-NG for building a cross toolchain on MacOS as host. 0) Mac OS Snow Leopard, with Developer Tools 3.2 installed, or Mac OS Leopard, with Developer Tools & newer gcc (>= 4.3) installed via macports 1) You have to use a case sensitive file system for ct-ng's build and target directories. Use a disk or disk image with a case sensitive fs that you mount somewhere. 2) Install macports (or similar easy means of installing 3rd party software), make sure that macport's bin dir is in the front (!) of your PATH. Furtheron assuming it is /opt/local/bin. 3) Install (at least) the following macports lzmautils libtool binutils gsed gawk gcc43 (only necessary for Leopard OSX 10.5) gcc_select (only necessary for OSX 10.5, or Xcode > 4) 4) Prerequisites On Leopard, make sure that the macport's gcc is called with the default commands (gcc, g++,...), via macport's gcc_select On OSX 10.7 Lion / when using Xcode >= 4 make sure that the default commands (gcc, g++, etc.) point to gcc-4.2, NOT llvm-gcc-4.2 by using macport's gcc_select feature. With MacPorts >= 1.9.2 the command is: "sudo port select --set gcc gcc42" This also requires (like written above) that macport's bin dir comes before standard directories in your PATH environment variable because the gcc symlink is installed in /opt/local/bin and the default /usr/bin/gcc is not removed by the gcc select command! Explanation: llvm-gcc-4.2 (with Xcode 4.1 it is on my machine "gcc version 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)") cannot boostrap gcc. See http://llvm.org/bugs/show_bug.cgi?id=9571 5) run ct-ng's configure with the following tool configuration (assuming you have installed the tools via macports in /opt/local): ./configure --with-sed=/opt/local/bin/gsed \ --with-libtool=/opt/local/bin/glibtool \ --with-libtoolize=/opt/local/bin/glibtoolize \ --with-objcopy=/opt/local/bin/gobjcopy \ --with-objdump=/opt/local/bin/gobjdump \ --with-readelf=/opt/local/bin/greadelf \ --with-grep=/opt/local/bin/ggrep \ [...other configure parameters as you like...] 6) proceed as described in standard documentation ----- HINTS: - Apparently, GNU make's builtin variable .LIBPATTERNS is misconfigured under MacOS: It does not include lib%.dylib. This affects build of (at least) gdb-7.1 Put 'lib%.a lib%.so lib%.dylib' as .LIBPATTERNS into your environment before executing ct-ng build. See http://www.gnu.org/software/make/manual/html_node/Libraries_002fSearch.html as an explanation. - ct-ng menuconfig will not work on Snow Leopard 10.6.3 since libncurses is broken with this release. MacOS <= 10.6.2 and >= 10.6.4 are ok. Using Mercurial to hack crosstool-NG | -------------------------------------+ NOTE: this section was applicable as long as when we were using Mercurial (Hg) as the DVCS. Now we've switched to git, this section is no longer current. We keep it as a reference, since it still contains a few useful hints. Please help rewrite this section. ;-) Contributed by: Titus von Boxberg PREREQUISITES: Configuring Mercurial: You need mercurial with the following extensions: - mq : http://mercurial.selenic.com/wiki/MqExtension - patchbomb : http://mercurial.selenic.com/wiki/PatchbombExtension Usually, these two extensions are already part of the installation package. The mq extension maintains a separate queue of your local changes that you can change at any later time. With the patchbomb extension you can email those patches directly from your local repo. Your configuration file for mercurial, e.g. ~/.hgrc should contain at least the following sections (but have a look at `man hgrc`): # --- [email] # configure sending patches directly via Mercurial from = "Your Name" # How to send email: method = smtp [smtp] # SMTP configuration (only for method=smtp) host = localhost tls = true username = password = [extensions] # The following lines enable the two extensions: hgext.mq = hgext.patchbomb = # ---- Create your local repository as a clone: hg clone http://crosstool-ng.org/hg/crosstool-ng crosstool-ng Setting up the mq extension in your local copy: cd crosstool-ng hg qinit CREATING PATCHES: Recording your changes in the patch queue maintained by mq: # First, create a new patch entry in the patch queue: hg qnew -D -U -e short_patch_name1 # if you execute `hg status` here, your modifications of the working # copy should show up. # Now the following command takes your modifications from the working copy # into the patch entry hg qrefresh -D [-e] # Now your changes are recorded, and `hg status` should show a clean # working copy Repeat the above steps for all your modifications. The command `hg qseries` informs you about the content of your patch queue. CONTRIBUTING YOUR PATCHES: Once you are satisfied with your patch series, you can (you should!) contribute them back to upstream. This is easily done using the `hg email` command. `hg email` sends your new changesets to a specified list of recipients, each patch in its own email, all ordered in the way you entered them (oldest first). The command line flag --outgoing selects all changesets that are in your local but not yet in the upstream repository. Here, these are exactly the ones you entered into your local patch queue in the section above, so --outgoing is what you want. Each email gets the subject set to: "[PATCH x of n] " where 'x' is the serial number in the email series, and 'n' is the total number of patches in the series. The body of the email is the complete patch, plus a handful of metadata, that helps properly apply the patch, keeping the log message, attribution and date, tracking file changes (move, delete, modes...) `hg email` also threads all outgoing patch emails below an introductory message. You should use the introductory message (command line flag --intro) to describe the scope and motivation for the whole patch series. The subject for the introductory message gets set to: "[PATCH 0 of n] " and you get the chance to set the . Here is a sample `hg email` complete command line: Note: replace " (at) " with "@" hg email --outgoing --intro \ --to '"Yann E. MORIN" ' \ --cc 'crossgcc (at) sourceware.org' # It then opens an editor and lets you enter the subject # and the body for the introductory message. Use `hg email` with the additional command line switch -n to first have a look at the email(s) without actually sending them. MAINTAINING YOUR PATCHES: When the patches are refined by discussing them on the mailing list, you may want to finalize and resend them. The mq extension has the idiosyncrasy of imposing a stack onto the queue: You can always reedit/refresh only the patch on top of stack. The queue consists of applied and unapplied patches (if you reached here via the above steps, all of your patches are applied), where the 'stack' consists of the applied patches, and 'top of stack' is the latest applied patch. The following output of `hg qseries` is now used as an example: 0 A short_patch_name1 1 A short_patch_name2 2 A short_patch_name3 3 A short_patch_name4 You are now able to edit patch 'short_patch_name4' (which is top of stack): # and execute again hg qrefresh -D [-e] If you want to edit e.g. patch short_patch_name2, you have to modify mq's stack so this patch gets top of stack. For this purpose see `hg help qgoto`, `hg help qpop`, and `hg help qpush`. hg qgoto short_patch_name2 # The patch queue should now look like hg qseries 0 A short_patch_name1 1 A short_patch_name2 2 U short_patch_name3 3 U short_patch_name4 # so patch # 1 (short_patch_name2) is top of stack. # and execute again hg qrefresh -D [-e] # the following command reapplies the now unapplied two patches: hg qpush -a # you can also use `hg qgoto short_patch_name4` to get there again. RESENDING YOUR REEDITED PATCHES: By mailing list policy, please resend your complete patch series. --> Go back to section "CONTRIBUTING YOUR PATCHES" and resubmit the full set. SYNCING WITH UPSTREAM AGAIN: You can sync your repo with upstream at any time by executing # first unapply all your patches: hg qpop -a # next fetch new changesets from upstream hg pull # then update your working copy hg up # optionally remove already upstream integrated patches (see below) hg qdelete # and reapply your patches if any non upstream-integrated left (but see below) hg qpush -a Eventually, your patches get included into the upstream repository which you initially cloned. In this case, before executing the hg qpush -a from above you should manually "hg qdelete" the patches that are already integrated upstream. HOW TO FORMAT COMMIT MESSAGES (aka patch descriptions): Commit messages should look like (without leading pipes): |component: short, one-line description | |optional longer description |on multiple lines if needed | |Signed-off-by: as documented in section 7 of ct-ng's documentation Here is an example commit message (see revision 8bb5151c5b01): kernel/linux: fix type in version strings I missed refreshing the patch before pushing. :-( Signed-off-by: "Yann E. MORIN" Using crosstool-NG on Windows | ------------------------------+ Contributed by: Ray Donnelly Prerequisites and instructions for using crosstool-NG for building a cross toolchain on Windows (Cygwin) as build and, optionally Windows (hereafter) MinGW-w64 as host. 0. Use Cygwin64 if you can. DLL base-address problems are lessened that way and if you bought a 64-bit CPU, you may as well use it. 1. You must enable Case Sensitivity in the Windows Kernel (this is only really necessary for Linux targets, but at present, crosstool-ng refuses to operate on case insensitive filesystems). The registry key for this is: HKLM\SYSTEM\CurrentControlSet\Control\Session Manager\kernel\obcaseinsensitive Read more at: https://cygwin.com/cygwin-ug-net/using-specialnames.html 2. Using setup{,-x86_64}.exe, install the default packages and also the following ones: (tested versions in brackets, please test newer versions and report successes via pull requests changing this list and failures to: https://github.com/crosstool-ng/crosstool-ng/issues autoconf (13-1), make (4.1-1), gcc-g++ (4.9.3-1), gperf (3.0.4-2), bison (3.0.4-1), flex (2.5.39-1), texinfo (6.0-1), wget (1.16.3-1), patch (2.7.4-1), libtool (2.4.6-2), automake (9-1), diffutils (3.3-3), libncurses-devel (6.0-1.20151017), help2man (1.44.1-1) mingw64-i686-gcc-g++* (4.9.2-2), mingw64-x86_64-gcc-g++* (4.9.2-2) Leave "Select required packages (RECOMMENDED)" ticked. Notes: 2.1 The packages marked with * are only needed if your host is MinGW-w64. 2.2 Unfortunately, wget pulls in an awful lot of dependencies, including Python 2.7, Ruby, glib and Tcl. 3. Although nativestrict symlinks seem like the best idea, extracting glibc fails when they are enabled, so just don't set anything here. If your host is MinGW-w64 then these 'Cygwin-special' symlinks won't work, but you can dereference them by using tar options --dereference and --hard-dereference when making a final tarball. I plan to investigate and fix or at least work around the extraction problem. Read more at: https://cygwin.com/cygwin-ug-net/using-cygwinenv.html 4. collect2.exe will attempt to run ld which is a shell script that runs either ld.exe or gold.exe so you need to make sure that a working shell is in your path. Eventually I will replace this with a native program for MinGW-w64 host. Using crosstool-NG to build Xtensa toolchains | ----------------------------------------------+ Contributed by: Max Filippov Xtensa cores are highly configurable: endianness, instruction set, register set of a core is chosen at processor configuration time. New registers and instructions may be added by designers, making each core configuration unique. Toolchain components cannot know about features of each individual core and need to be configured in order to be compatible with particular architecture variant. This configuration includes: - definitions of instruction formats, names and properties for assembler, disassembler and debugger; - definitions of register names and properties for assembler, disassembler and debugger; - selection of predefined features, such as endianness, presence of certain processor options or instructions for compiler, debugger C library and OS kernels; - macros with instruction sequences for saving and restoring special, user or coprocessor registers for OS kernels. This configuration is provided in form of source files, that must replace corresponding files in binutils, gcc, gdb or newlib source trees or be added to OS kernel source tree. This set of files is usually distributed as archive known as Xtensa configuration overlay. Tensilica provides such an overlay as part of the processor download, however, it needs to be reformatted to match the specific format required by the crosstool-NG. For a script to convert the overlay file, and additional information, please see http://wiki.linux-xtensa.org/index.php/Toolchain_Overlay_File The current version of crosstool-NG requires that the overlay file name has the format xtensa_.tar, where CORE_NAME can be any user selected name. To make crosstool-NG use overlay file located at /xtensa_.tar select XTENSA_CUSTOM, set config parameter CT_ARCH_XTENSA_CUSTOM_NAME to CORE_NAME and CT_ARCH_XTENSA_CUSTOM_OVERLAY_LOCATION to PATH. The fsf target architecture variant is the configuration provided by toolchain components by default. It is present only for build-testing toolchain components and is in no way special or universal.