This section gives information about how to extend the functionality already present in Poky, documenting standard tasks such as adding new software packages, extending or customising images or porting poky to new hardware (adding a new machine). It also contains advice about how to manage the process of making changes to Poky to achieve best results.

To add package into Poky you need to write a recipe for it. Writing a recipe means creating a .bb file which sets various variables. The variables useful for recipes are detailed in the recipe reference section along with more detailed information about issues such as recipe naming. Before writing a recipe from scratch it is often useful to check whether someone else has written one already. OpenEmbedded is a good place to look as it has a wider scope and hence a wider range of packages. Poky aims to be compatible with OpenEmbedded so most recipes should just work in Poky. For new packages, the simplest way to add a recipe is to base it on a similar pre-existing recipe. There are some examples below of how to add standard types of packages:

To build an application from a single file stored locally (e.g. under "files/") requires a recipe which has the file listed in the SRC_URI variable. In addition the do_compile and do_install tasks need to be manually written. The S variable defines the directory containing the source code which in this case is set equal to WORKDIR, the directory BitBake uses for the build. DESCRIPTION = "Simple helloworld application" SECTION = "examples" LICENSE = "MIT" PR = "r0" SRC_URI = "file://helloworld.c" S = "${WORKDIR}" do_compile() { ${CC} helloworld.c -o helloworld } do_install() { install -d ${D}${bindir} install -m 0755 helloworld ${D}${bindir} } As a result of the build process "helloworld", "helloworld-dbg" and "hellworld-dev" packages will be built by default. It is possible to customise the packaging process.

Applications which use autotools (autoconf, automake) require a recipe which has a source archive listed in SRC_URI and inherit autotools to instruct BitBake to use the autotools.bbclass which has definitions of all the steps needed to build an autotooled application. The result of the build will be automatically packaged and if the application uses NLS to localise then packages with locale information will be generated (one package per language). Below is one example (hello_2.2.bb) DESCRIPTION = "GNU Helloworld application" SECTION = "examples" LICENSE = "GPLv2+" LIC_FILES_CHKSUM = "file://COPYING;md5=751419260aa954499f7abaabaa882bbe" PR = "r0" SRC_URI = "${GNU_MIRROR}/hello/hello-${PV}.tar.gz" inherit autotools gettext LIC_FILES_CHKSUM is used to track source license change. Autotool based recipe can be quickly created this way like above example.

Applications which use GNU make require a recipe which has the source archive listed in SRC_URI. Adding a do_compile step is not needed as by default BitBake will start the "make" command to compile the application. If there is a need for additional options to make then they should be stored in the EXTRA_OEMAKE variable - BitBake will pass them into the GNU make invocation. A do_install task is required - otherwise BitBake will run an empty do_install task by default. Some applications may require extra parameters to be passed to the compiler, for example an additional header path. This can be done buy adding to the CFLAGS variable, as in the example below: CFLAGS_prepend = "-I ${S}/include " mtd-utils is an example as Makefile-based: DESCRIPTION = "Tools for managing memory technology devices." SECTION = "base" DEPENDS = "zlib lzo e2fsprogs util-linux" HOMEPAGE = "http://www.linux-mtd.infradead.org/" LICENSE = "GPLv2" SRC_URI = "git://git.infradead.org/mtd-utils.git;protocol=git;tag=v${PV}" S = "${WORKDIR}/git/" EXTRA_OEMAKE = "'CC=${CC}' 'CFLAGS=${CFLAGS} -I${S}/include -DWITHOUT_XATTR' \ 'BUILDDIR=${S}'" do_install () { oe_runmake install DESTDIR=${D} SBINDIR=${sbindir} MANDIR=${mandir} \ INCLUDEDIR=${includedir} install -d ${D}${includedir}/mtd/ for f in ${S}/include/mtd/*.h; do install -m 0644 $f ${D}${includedir}/mtd/ done }

The variables PACKAGES and FILES are used to split an application into multiple packages. Below the "libXpm" recipe (libxpm_3.5.7.bb) is used as an example. By default the "libXpm" recipe generates one package which contains the library and also a few binaries. The recipe can be adapted to split the binaries into separate packages. require xorg-lib-common.inc DESCRIPTION = "X11 Pixmap library" LICENSE = "X-BSD" DEPENDS += "libxext libsm libxt" PR = "r3" PE = "1" XORG_PN = "libXpm" PACKAGES =+ "sxpm cxpm" FILES_cxpm = "${bindir}/cxpm" FILES_sxpm = "${bindir}/sxpm" In this example we want to ship the "sxpm" and "cxpm" binaries in separate packages. Since "bindir" would be packaged into the main PN package as standard we prepend the PACKAGES variable so additional package names are added to the start of list. The extra FILES_* variables then contain information to specify which files and directories goes into which package. Files included by earlier package are skipped by latter packages, and thus main PN will not include above listed files

To add a post-installation script to a package, add a pkg_postinst_PACKAGENAME() function to the .bb file where PACKAGENAME is the name of the package to attach the postinst script to. Normally PN can be used which expands to PACKAGENAME automatically. A post-installation function has the following structure: pkg_postinst_PACKAGENAME () { #!/bin/sh -e # Commands to carry out } The script defined in the post installation function gets called when the rootfs is made. If the script succeeds, the package is marked as installed. If the script fails, the package is marked as unpacked and the script will be executed again on the first boot of the image. Sometimes it is necessary that the execution of a post-installation script is delayed until the first boot, because the script needs to be executed on the device itself. To delay script execution until boot time, the post-installation function should have the following structure: pkg_postinst_PACKAGENAME () { #!/bin/sh -e if [ x"$D" = "x" ]; then # Actions to carry out on the device go here else exit 1 fi } The structure above delays execution until first boot because the D variable points to the 'image' directory when the rootfs is being made at build time but is unset when executed on the first boot.

Poky images can be customised to satisfy particular requirements. Several methods are detailed below along with guidelines of when to use them.

One way to get additional software into an image is by creating a custom image. The recipe will contain two lines: IMAGE_INSTALL = "task-poky-x11-base package1 package2" inherit poky-image By creating a custom image, a developer has total control over the contents of the image. It is important to use the correct names of packages in the IMAGE_INSTALL variable. The names must be in the OpenEmbedded notation instead of Debian notation, for example "glibc-dev" instead of "libc6-dev" etc. The other method of creating a new image is by modifying an existing image. For example if a developer wants to add "strace" into "poky-image-sato" the following recipe can be used: require poky-image-sato.bb IMAGE_INSTALL += "strace"

For complex custom images, the best approach is to create a custom task package which is then used to build the image (or images). A good example of a tasks package is meta/packages/tasks/task-poky.bb . The PACKAGES variable lists the task packages to build (along with the complementary -dbg and -dev packages). For each package added, RDEPENDS and RRECOMMENDS entries can then be added each containing a list of packages the parent task package should contain. An example would be: DESCRIPTION = "My Custom Tasks" PACKAGES = "\ task-custom-apps \ task-custom-apps-dbg \ task-custom-apps-dev \ task-custom-tools \ task-custom-tools-dbg \ task-custom-tools-dev \ " RDEPENDS_task-custom-apps = "\ dropbear \ portmap \ psplash" RDEPENDS_task-custom-tools = "\ oprofile \ oprofileui-server \ lttng-control \ lttng-viewer" RRECOMMENDS_task-custom-tools = "\ kernel-module-oprofile" In this example, two task packages are created, task-custom-apps and task-custom-tools with the dependencies and recommended package dependencies listed. To build an image using these task packages, you would then add "task-custom-apps" and/or "task-custom-tools" to IMAGE_INSTALL or other forms of image dependencies as described in other areas of this section.

Ultimately users may want to add extra image "features" as used by Poky with the IMAGE_FEATURES variable. To create these, the best reference is meta/classes/poky-image.bbclass which illustrates how poky achieves this. In summary, the file looks at the contents of the IMAGE_FEATURES variable and then maps this into a set of tasks or packages. Based on this then the IMAGE_INSTALL variable is generated automatically. Extra features can be added by extending the class or creating a custom class for use with specialised image .bb files.

It is possible to customise image contents by abusing variables used by distribution maintainers in local.conf. This method only allows the addition of packages and is not recommended. To add an "strace" package into the image the following is added to local.conf: DISTRO_EXTRA_RDEPENDS += "strace" However, since the DISTRO_EXTRA_RDEPENDS variable is for distribution maintainers this method does not make adding packages as simple as a custom .bb file. Using this method, a few packages will need to be recreated if they have been created before and then the image is rebuilt. bitbake -c clean task-boot task-base task-poky bitbake poky-image-sato Cleaning task-* packages is required because they use the DISTRO_EXTRA_RDEPENDS variable. There is no need to build them by hand as Poky images depend on the packages they contain so dependencies will be built automatically when building the image. For this reason we don't use the "rebuild" task in this case since "rebuild" does not care about dependencies - it only rebuilds the specified package.

Adding a new machine to Poky is a straightforward process and this section gives an idea of the changes that are needed. This guide is meant to cover adding machines similar to those Poky already supports. Adding a totally new architecture might require gcc/glibc changes as well as updates to the site information and, whilst well within Poky's capabilities, is outside the scope of this section.

A .conf file needs to be added to conf/machine/ with details of the device being added. The name of the file determines the name Poky will use to reference this machine. The most important variables to set in this file are TARGET_ARCH (e.g. "arm"), PREFERRED_PROVIDER_virtual/kernel (see below) and MACHINE_FEATURES (e.g. "kernel26 apm screen wifi"). Other variables like SERIAL_CONSOLE (e.g. "115200 ttyS0"), KERNEL_IMAGETYPE (e.g. "zImage") and IMAGE_FSTYPES (e.g. "tar.gz jffs2") might also be needed. Full details on what these variables do and the meaning of their contents is available through the links. There're lots of existing machine .conf files which can be easily leveraged from meta/conf/machine/.

Poky needs to be able to build a kernel for the machine. You need to either create a new kernel recipe for this machine or extend an existing recipe. There are plenty of kernel examples in the meta/recipes-kernel/linux directory which can be used as references. If creating a new recipe the "normal" recipe writing rules apply for setting up a SRC_URI including any patches and setting S to point at the source code. You will need to create a configure task which configures the unpacked kernel with a defconfig be that through a "make defconfig" command or more usually though copying in a suitable defconfig and running "make oldconfig". By making use of "inherit kernel" and also maybe some of the linux-*.inc files, most other functionality is centralised and the the defaults of the class normally work well. If extending an existing kernel it is usually a case of adding a suitable defconfig file in a location similar to that used by other machine's defconfig files in a given kernel, possibly listing it in the SRC_URI and adding the machine to the expression in COMPATIBLE_MACHINE : COMPATIBLE_MACHINE = '(qemux86|qemumips)'

A formfactor configuration file provides information about the target hardware on which Poky is running, and that Poky cannot obtain from other sources such as the kernel. Some examples of information contained in a formfactor configuration file include framebuffer orientation, whether or not the system has a keyboard, the positioning of the keyboard in relation to the screen, and screen resolution. Sane defaults should be used in most cases, but if customisation is necessary you need to create a machconfig file under meta/packages/formfactor/files/MACHINENAME/ where MACHINENAME is the name for which this infomation applies. For information about the settings available and the defaults, please see meta/packages/formfactor/files/config. Below is one example for qemuarm: HAVE_TOUCHSCREEN=1 HAVE_KEYBOARD=1 DISPLAY_CAN_ROTATE=0 DISPLAY_ORIENTATION=0 #DISPLAY_WIDTH_PIXELS=640 #DISPLAY_HEIGHT_PIXELS=480 #DISPLAY_BPP=16 DISPLAY_DPI=150 DISPLAY_SUBPIXEL_ORDER=vrgb

We recognise that people will want to extend/configure/optimise Poky for their specific uses, especially due to the extreme configurability and flexibility Poky offers. To ensure ease of keeping pace with future changes in Poky we recommend making changes to Poky in a controlled way. Poky supports the idea of "layers" which when used properly can massively ease future upgrades and allow segregation between the Poky core and a given developer's changes. Some other advice on managing changes to Poky is also given in the following section.

Often, people want to extend Poky either through adding packages or overriding files contained within Poky to add their own functionality. Bitbake has a powerful mechanism called layers which provides a way to handle this extension in a fully supported and non-invasive fashion. The Poky tree includes several additional layers which demonstrate this functionality, such as meta-emenlow and meta-extras. The meta-emenlow layer is an example layer enabled by default. The meta-extras repostory is not enabled by default but enabling any layer is as easy as adding the layers path to the BBLAYERS variable in your bblayers.conf. this is how meta-extras are enabled in Poky builds: LCONF_VERSION = "1" BBFILES ?= "" BBLAYERS = " \ /path/to/poky/meta \ /path/to/poky/meta-moblin \ /path/to/poky/meta-emenlow \ /path/to/poky/meta-extras \ " Bitbake parses the conf/layer.conf of each of the layers in BBLAYERS to add the layers packages, classes and configuration to Poky. To create your own layer, independent of the main Poky repository, you need only create a directory with a conf/layer.conf file and add the directory to your bblayers.conf. The meta-emenlow/conf/layer.conf demonstrates the required syntax: # We have a conf and classes directory, add to BBPATH BBPATH := "${BBPATH}:${LAYERDIR}" # We have a packages directory, add to BBFILES BBFILES := "${BBFILES} ${LAYERDIR}/packages/*/*.bb \ ${LAYERDIR}/packages/*/*.bbappend" BBFILE_COLLECTIONS += "emenlow" BBFILE_PATTERN_emenlow := "^${LAYERDIR}/" BBFILE_PRIORITY_emenlow = "6" As can be seen, the layers recipes are added to BBFILES. The BBFILE_COLLECTIONS variable is then appended to with the layer name. The BBFILE_PATTERN variable is immediately expanded with a regular expression used to match files from BBFILES into a particular layer, in this case by using the base pathname. The BBFILE_PRIORITY variable then assigns different priorities to the files in different layers. This is useful in situations where the same package might appear in multiple layers and allows you to choose which layer should 'win'. Note the use of LAYERDIR with the immediate expansion operator. LAYERDIR expands to the directory of the current layer and requires use of the immediate expansion operator so that Bitbake does not lazily expand the variable when it's parsing a different directory. Additional bbclass and configuration files can be locationed by bitbake through the addition to the BBPATH environment variable. In this case, the first file with the matching name found in BBPATH is the one that is used, just like the PATH variable for binaries. It is therefore recommended that you use unique bbclass and configuration file names in your custom layer. The recommended approach for custom layers is to store them in a git repository of the format meta-prvt-XXXX and have this repository cloned alongside the other meta directories in the Poky tree. This way you can keep your Poky tree and it's configuration entirely inside POKYBASE.

Modifications to Poky are often managed under some kind of source revision control system. The policy for committing to such systems is important as some simple policy can significantly improve usability. The tips below are based on the policy followed for the Poky core. It helps to use a consistent style for commit messages when committing changes. We've found a style where the first line of a commit message summarises the change and starts with the name of any package affected work well. Not all changes are to specific packages so the prefix could also be a machine name or class name instead. If a change needs a longer description this should follow the summary: bitbake/data.py: Add emit_func() and generate_dependencies() functions These functions allow generation of dependency data between funcitons and variables allowing moves to be made towards generating checksums and allowing use of the dependency information in other parts of bitbake. Signed-off-by: Richard Purdie rpurdie@linux.intel.com Any commit should be self contained in that it should leave the metadata in a consistent state, buildable before and after the commit. This helps ensure the autobuilder test results are valid but is good practice regardless.

If a committed change will result in changing the package output then the value of the PR variable needs to be increased (commonly referred to as 'bumped') as part of that commit. Only integer values are used and PR = "r0" should be added into new recipes as, while this is the default value, not having the variable defined in a recipe makes it easy to miss incrementing it when updating the recipe. When upgrading the version of a package (PV), the PR variable should be reset to "r0". The aim is that the package version will only ever increase. If for some reason PV will change and but not increase, the PE (Package Epoch) can be increased (it defaults to '0'). The version numbers aim to follow the Debian Version Field Policy Guidelines which define how versions are compared and hence what "increasing" means. There are two reasons for doing this, the first is to ensure that when a developer updates and rebuilds, they get all the changes to the repository and don't have to remember to rebuild any sections. The second is to ensure that target users are able to upgrade their devices via their package manager such as with the opkg upgrade commands (or similar for dpkg/apt or rpm based systems). The aim is to ensure Poky has upgradable packages in all cases.

It may not be immediately clear how Poky can work in a team environment, or scale to a large team of developers. The specifics of any situation will determine the best solution and poky offers immense flexibility in that aspect but there are some practises that experience has shown to work well. The core component of any development effort with Poky is often an automated build testing framework and image generation process. This can be used to check that the metadata is buildable, highlight when commits break the builds and provide up to date images allowing people to test the end result and use them as a base platform for further development. Experience shows that buildbot is a good fit for this role and that it works well to configure it to make two types of build - incremental builds and 'from scratch'/full builds. The incremental builds can be tied to a commit hook which triggers them each time a commit is made to the metadata and are a useful acid test of whether a given commit breaks the build in some serious way. They catch lots of simple errors and whilst they won't catch 100% of failures, the tests are fast so developers can get feedback on their changes quickly. The full builds are builds that build everything from the ground up and test everything. They usually happen at preset times such as at night when the machine load isn't high from the incremental builds. poky autobuilder is an example implementation with buildbot. Most teams have pieces of software undergoing active development. It is of significant benefit to put these under control of a source control system compatible with Poky such as git or svn. The autobuilder can then be set to pull the latest revisions of these packages so the latest commits get tested by the builds allowing any issues to be highlighted quickly. Poky easily supports configurations where there is both a stable known good revision and a floating revision to test. Poky can also only take changes from specific source control branches giving another way it can be used to track/test only specified changes. Perhaps the hardest part of setting this up is the policy that surrounds the different source control systems, be them software projects or the Poky metadata itself. The circumstances will be different in each case but this is one of Poky's advantages - the system itself doesn't force any particular policy unlike a lot of build systems, allowing the best policy to be chosen for the circumstances.

Often, rather than reflashing a new image you might wish to install updated packages into an existing running system. This can be done by sharing the tmp/deploy/ipk/ directory through a web server and then on the device, changing /etc/opkg/base-feeds.conf to point at this server, for example by adding: src/gz all http://www.mysite.com/somedir/deploy/ipk/all src/gz armv7a http://www.mysite.com/somedir/deploy/ipk/armv7a src/gz beagleboard http://www.mysite.com/somedir/deploy/ipk/beagleboard

Poky is usually used to build software rather than modifying it. However, there are ways Poky can be used to modify software. During building, the sources are available in WORKDIR directory. Where exactly this is depends on the type of package and the architecture of target device. For a standard recipe not related to MACHINE it will be tmp/work/PACKAGE_ARCH-poky-TARGET_OS/PN-PV-PR/. Target device dependent packages use MACHINE instead of PACKAGE_ARCH in the directory name. Check the package recipe sets the S variable to something other than standard WORKDIR/PN-PV/ value. After building a package, a user can modify the package source code without problem. The easiest way to test changes is by calling the "compile" task: bitbake -c compile -f NAME_OF_PACKAGE "-f" or "--force" is used to force re-execution of the specified task. Other tasks may also be called this way. But note that all the modifications in WORKDIR are gone once you executes "-c clean" for a package.

By default Poky uses quilt to manage patches in do_patch task. It is a powerful tool which can be used to track all modifications done to package sources. Before modifying source code it is important to notify quilt so it will track changes into new patch file: quilt new NAME-OF-PATCH.patch Then add all files which will be modified into that patch: quilt add file1 file2 file3 Now start editing. At the end quilt needs to be used to generate final patch which will contain all modifications: quilt refresh The resulting patch file can be found in the patches/ subdirectory of the source (S) directory. For future builds it should be copied into Poky metadata and added into SRC_URI of a recipe: SRC_URI += "file://NAME-OF-PATCH.patch" This also requires a bump of PR value in the same recipe as we changed resulting packages.

The license of one upstream project may change in the future, and Poky provides one mechanism to track such license change - LIC_FILES_CHKSUM variable.

LIC_FILES_CHKSUM = "file://COPYING; md5=xxxx \ file://licfile1.txt; beginline=5; endline=29;md5=yyyy \ file://licfile2.txt; endline=50;md5=zzzz \ ..." S is the default directory for searching files listed in LIC_FILES_CHKSUM. Relative path could be used too: LIC_FILES_CHKSUM = "file://src/ls.c;startline=5;endline=16;\ md5=bb14ed3c4cda583abc85401304b5cd4e" LIC_FILES_CHKSUM = "file://../license.html;md5=5c94767cedb5d6987c902ac850ded2c6" The first line locates a file in S/src/ls.c, and the second line refers to a file in WORKDIR, which is the parent of S

This parameter lists all the important files containing the text of licenses for the source code. It is also possible to specify on which line the license text starts and on which line it ends within that file using the "beginline" and "endline" parameters. If the "beginline" parameter is not specified then license text begins from the 1st line is assumed. Similarly if "endline" parameter is not specified then the license text ends at the last line in the file is assumed. So if a file contains only licensing information, then there is no need to specify "beginline" and "endline" parameters. The "md5" parameter stores the md5 checksum of the license text. So if the license text changes in any way from a file, then its md5 sum will differ and will not match with the previously stored md5 checksum. This mismatch will trigger build failure, notifying developer about the license text md5 mismatch, and allowing the developer to review the license text changes. Also note that if md5 checksum is not matched while building, the correct md5 checksum is printed in the build log which can be easily copied to .bb file. There is no limit on how many files can be specified on this parameter. But generally every project would need specifying of just one or two files for license tracking. Many projects would have a "COPYING" file which will store all the license information for all the source code files. If the "COPYING" file is valid then tracking only that file would be enough. 1. If you specify empty or invalid "md5" parameter; then while building the package, bitbake will give md5 not matched error, and also show the correct "md5" parameter value both on the screen and in the build log 2. If the whole file contains only license text, then there is no need to specify "beginline" and "endline" parameters.

Poky implements a distro_check task which automatically connects to major distributions and checks whether they contains same package. Sometimes the same package has different names in different distributions, which results in a mismatch from distro_check task This can be solved by defining per distro recipe name alias - DISTRO_PN_ALIAS

DISTRO_PN_ALIAS_pn-PACKAGENAME = "distro1=package_name_alias1 \ distro2=package_name_alias2 \ distro3=package_name_alias3 \ ..." Use space as the delimiter if there're multiple distro aliases The current code can check if the src package for a recipe exists in the latest releases of these distributions automatically. Fedora, OpenSuSE, Debian, Ubuntu, Mandriva For example, this command will generate a report, listing which linux distros include the sources for each of the poky recipe. bitbake world -f -c distro_check The results will be stored in the build/tmp/log/distro_check-${DATETIME}.results file.