- try this on your bash shell:
- `ONEVAR="testing" eval 'bash -c "echo value once $ONEVAR && false && echo value twice $ONEVAR"' '| grep value' '| grep value' ; echo ${PIPESTATUS[*]}`
- Notice how PIPESTATUS has only one element. and it is always true, although we failed explicitly with false in the middle of the bash.
- That is because eval itself is considered a single command, no matter how many pipes you put in there, you'll get a single value, the return code of the LAST pipe.
- Lets export the value of the pipe inside eval so we know outside what happened:
- `ONEVAR="testing" eval 'bash -e -c "echo value once $ONEVAR && false && echo value twice $ONEVAR"' '| grep value' '| grep value' ';EVALPIPE=(${PIPESTATUS[@]})' ; echo ${EVALPIPE[*]}`
- also: pass LC_ALL=C LANG=C to avoid locale warnings
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
CONFIG_DEBUG_INFO_BTF relies on libelf-dev and pahole (provided by dwarves
package).
This commit add dwarves package both as hostdeps and to Dockerfile.
- disabling samba on desktop builds since it delays 1st login and throws out an error in case we have no network (laptop usage)
- disabling broken build target
- now we have proper amd64 repo published!, remove disable
- amd64 does not have dtb package anymore, don't try to install it from repo
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions framework (née "fragments")
- this should actually change nothing at this point, just add capabilities
- the framework is implemented in lib/extensions.sh
- the "if function x exists then call x" replaced with call_extension_method()
- +inline documentation
- +compatibility names
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions framework; meta-extensions: auto-docs and sample extension gen
- 2 extensions dealing with extensibility itself
- detect-unused-extensions: shows which extensions are enabled, but never called.
- gen-sample-extension-docs: generates a sample empty extension & Markdown documentation for extensions
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* new extension methods and features via config variables in core Armbian
- `SKIP_EXTERNAL_TOOLCHAINS=yes` - does not download or use any linaro toolchains, only build host-installed ones
- `SKIP_BOOTSPLASH=yes` - does not patch kernel for splash file
- `EXTRA_BSP_NAME=xyz` - allows for BSP variants, useful for when extensions modify the BSP
- `EXTRA_ROOTFS_MIB_SIZE=x` - add x mib's to rootfs size, for use with very small images
- `KERNEL_EXTRA_TARGETS` - what extra targets to make kernel for, default to "modules dtbs"
- `BOOTCONFIG=none` - does not build nor install u-boot; also doesn't handle bootscripts et al
- `unset KERNELSOURCE` - does not build nor install kernel, nor build initrd, nor build nor install firmware
- `ARMHF_ARCH=skip` - does not add armhf to apt/dpkg, thus pure arm64
- `SKIP_ARMBIAN_REPO=yes` - results in armbian.list.disabled in the final image
- define `APT_EXTRA_DIST_PARAMS` with apt-cacher-ng options and use it for `PACKAGE_LIST_INSTALL/REMOVE` et al
- initial support for targeting x86/amd64 UEFI and BIOS
- some do's/don'ts for x86/amd64, like a different `UBUNTU_MIRROR` default
- GPT/EFI(ESP) partitions (fat, `UEFISIZE=256` to enable, mount `UEFI_MOUNT_POINT=/boot/efi`, first on disk but ends
up at `$uefipart`=15)
- GPT/BIOS partitions (fat, `BIOSSIZE=1` to enable, second on disk but ends up at partition 14)
- `UEFI_FS_LABEL="armbiefi"` - to set the FAT label for the EFI partition, visible in Win/Mac
- hard-requires gdisk package host-side
- add add_host_dependencies() extension method; fill `EXTRA_BUILD_DEPS="pkg pkg2"` to install to host before toolchains
download
- add pre_prepare_partitions() extension method, for custom partition size calculations
- add create_partition_table() extension method, used to do full-custom partitioning if `USE_HOOK_FOR_PARTITION=yes`
- add post_create_partitions() extension method, mostly for easy debugging
- add post_write_sdcard() extension method, where you can also set `SKIP_VERIFY=yes` to skip sdcard verification
- add post_install_kernel_debs() extension method.
- multiple fixes to bsp to avoid spurious errors when files are not where it expects
- v4: detect `update-initramfs` failure and abort build with useful message if it does
- v4: show useful stacktrace in `exit_with_error`
- if `ERROR_DEBUG_SHELL=yes`, drop into a shell before unmounting/deleting everything, so we can inspect what went wrong
- v4: display a message before `apt-get remove PACKAGE_LIST_BOARD_REMOVE` packages, so any errors while removing are easy to understand
- v4: preserve kernel .config's dates when copying
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions framework; refactor tool fetching/building into extensions
- a few examples of core refactoring using extensions
- sunxi-tools extension, enabled by 2 different sunxi family includes ("reuse" example)
- marvel-tools extension, enabled by 2 different mvebu family includes
- rkbin-tools extension, enabled by rockship64_common family include
- amlogic-fip/c2-blobs stuff refactored directly into meson64_common.inc ("single-use" example)
- removed the 'testings' fetch_from_repo completely since not used anywhere.
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* .wip's for UEFI arm64 and UEFI/BIOS x86 via new GRUB extension
- v3: added `growroot`-awareness to `armbian-resize-filesystem`
- the partition-growing part of `armbian-resize-filesystem` does not deal correctly with the UEFI layout
- `growroot` is installed on UEFI images by default, that handles growing partition during initramfs
- now `armbian-resize-filesystem` handles `resize2fs` only, and works.
- v4: reworked UEFI board/family/include structure:
- use Distro's `linux-generic` kernel only for `current`
- `edge` now builds it's own pure-mainline `5.15.y` kernel, for both x86 and arm64
- `.config` taken from Ubuntu, probably needs tuning for EXTRAWIFI=yes et al
- v4: introduce `SKIP_KERNEL_SYMLINK=yes`, tested in `builddeb`
- to avoid symlinking kernel; u-boot likes it, but grub and flash-kernel hates it
- v5: many fixes
- v7: more small fixes.
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* .wip for the RaspberryPi 4B via new flash-kernel extension
- this does not build it's own kernel "yet", but uses default linux-raspi kernel from Ubuntu
- flash-kernel is not really a bootloader
- it just prepares kernel et al a FAT partition for booting by the RPi4b bootloader
- flash-kernel is standard Debian package, but has only been tested on Ubuntu releases
- it is really only known-working since Hirsute release.
- Debian's rpi kernel is armhf only, so out of scope here, at least until we add source-built kernels.
- v3: fixed focal rootfs build. untested.
- v3: better variable names, preparing for source-built kernel.
- v5: new edge build with pure mainline kernel.
- v6: many fixes and some hacks for packaging and layout, also firmware (using Ubuntu's)
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* Added first patch to edge x86 related to wifi drivers
* extensions: leave hostapd alone; remove hackish ext; block reentrancy
- package-list-utils does not belong in this PR
- grub or bcm2711 is not the place to remove hostapd
- block recursive enable_extension() calls, for now.
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* gen-sample-extension-docs: fix: avoid counter in generated sample
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions: dependencies: enable_extension() in extensions with a stack
- and better stacktraces, I hope
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* Remove code from package list since we don't have it in repository
Adjust kernel config to disable driver that needs further polishing.
* Allow amd64 to build the same desktops as aarch64. We only have this limit for armhf, where some desktops don't work
* amd64: allow building amd64 on aarch64 with system toolchain
- conditionally add gcc-x86-64-linux-gnu to hostdeps
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* add libelf-dev directly to hostdeps (and Dockerfile), remove extension
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* packaging: remove SKIP_KERNEL_SYMLINK hack, fix the root cause
- which was the missing $image_name for non-arm64 & non-arm, so: x86 for example
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* grub: really obliterate u-boot stuff from BSP
- for now. soon we'll refactor u-boot so not have to do this
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* flash-kernel: really obliterate u-boot stuff from BSP
- for now. soon we'll refactor u-boot so not have to do this
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions: add host_dependencies_ready() hook
- this passes FINAL_HOST_DEPS containing all hostdeps for the run after they're installed
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* Add verification functions for correct selection.
* If UEFI Skip symlink creation
* Do not create dtb package for amd64
* Skip scripts folder cleaning if build process native.
Skip creating postinst prerm scripts for headers.
* Skip applying headers-debian-byteshift.patch if build native
* Fix architecture syntax as x86_64
* Revert "amd64: allow building amd64 on aarch64 with system toolchain"
This reverts commit 0c5ee20bb1.
* Compare architectures before starting compilation.
Signed-off-by: The-going <48602507+The-going@users.noreply.github.com>
* extensions: cleanups after fixes by the-Going
- packaging:
- there is _no need_ anymore for the symlink hack, CONFIG_EFI or no. But check is great, see below
- it's not `amd64` that has no DTB's, it's all UEFI, thus: `is_enabled CONFIG_EFI`, thanks!
- Explicitly disallow "reverse cross compile" in amd64.conf.
- whitespace-only-deletions: revert. we shall shellfmt the whole thing one day, but not today.
- fix a few syntax warnings in newly introduced code (floating `$ARCH` vs `"${ARCH}`) - blame shellcheck
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* packaging: fix: turns out a lot of boards have CONFIG_EFI=y, can't use that for dtb/no-dtb decision.
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* grub: remove debug
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* firmware: allow installing `armbian-firmware-full`; make it really full
- can now use `BOARD_FIRMWARE_INSTALL="-full"` to install full firmware for the board. enable for UEFI.
- don't rely on KERNELSOURCE for firmware-related decisions. introduce `INSTALL_ARMBIAN_FIRMWARE` which defaults to `yes`
- rpi4b/flash-kernel: disable Armbian firmware; we need linux-firmware-raspi2, which conflicts.
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions: log to /${LOG_SUBPATH}/ instead of fixed /debug/
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions: introduce cleanup_extension_manager() called by build-all-ng's unset_all()
- to reset/unset everything done by the the initializer, so build can run again
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions: remove 'global' logging, for use with build_all_ng
- enable_extensions() will have to live on without logging to file. it's just too early.
- now init EXTENSION_MANAGER_TMP_DIR in initialize_extension_manager()
- now init EXTENSION_MANAGER_LOG_FILE in initialize_extension_manager()
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* extensions: build-all-ng.sh bugfix due to extension's debug to stdout
- extensions (among other things) can produce output to stdout when activated
- fix: check_hash() produced "idential" (sic, now changed to IDENTICAL) to stdout as a trigger
- debugging output got mixed with "idential", rendering hash cache void for families that used extensions
- eg: sunxi, others
- fix is to send stdout to the bitbucket when sourcing the board & arch config files
- proper fix would be stop using stdout in this case and use return code for check_hash()
- one day soon
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* Add CI build targets
Co-authored-by: Igor Pecovnik <igor.pecovnik@gmail.com>
Co-authored-by: The-going <48602507+The-going@users.noreply.github.com>
This helps when we have no time to work on certain branch patches right now, while all others builds without a problem.
We create a file this way:
c0c6fd3a02
- try this on your bash shell:
- `ONEVAR="testing" eval 'bash -c "echo value once $ONEVAR && false && echo value twice $ONEVAR"' '| grep value' '| grep value' ; echo ${PIPESTATUS[*]}`
- Notice how PIPESTATUS has only one element. and it is always true, although we failed explicitly with false in the middle of the bash.
- That is because eval itself is considered a single command, no matter how many pipes you put in there, you'll get a single value, the return code of the LAST pipe.
- Lets export the value of the pipe inside eval so we know outside what happened:
- `ONEVAR="testing" eval 'bash -e -c "echo value once $ONEVAR && false && echo value twice $ONEVAR"' '| grep value' '| grep value' ';EVALPIPE=(${PIPESTATUS[@]})' ; echo ${EVALPIPE[*]}`
- also: pass LC_ALL=C LANG=C to avoid locale warnings
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* Add new function apply_patch_series
* Apply a series of patches if a series.conf file exists
* Add all megous orange-pi-5.15 branch patches as series
* Add an Exception for version 5.15 for sanxi
* Adjust the switch for sunxi
* Fix megous patches series for v5.15.3
* Switch version to 5.15.3 for sunxi
* Check the exception for 5.15 sunxi
* Store changelog for upstream changes in the cache/hash/
* Cleanup
* Calculate urls
* Use URL
Don't merge since it doesn't work properly yet
* Check if OLDHASHTARGET commit exists otherwise use oldest
* Fix log updating
* Improve log format, add better support for sunxi
* introduce DEB_COMPRESS for dpkg-deb+kernel's KDEB_COMPRESS
- Avoids zstd which is now default in Ubuntu Impish, but can't be read by Debian dpkg
- Allows for xz, gzip, or none; defaults to xz; use gzip for faster, and none for much faster.
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
* use KDEB_COMPRESS in packages/armbian/builddep for 5.12+ kernels too
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
- Escape $TERM, this is meant to be resolved at runtime and not in armbian/build's context
- This was very hard to track down, if building manually $TERM is actually valid (builder's $TERM)
- When building without a tty this shows up as "unknown"
Signed-off-by: Ricardo Pardini <ricardo@pardini.net>
That lead to glued packagelist when build images in a row.
[ o.k. ] Installing the desktop packages for [ Armbian ]
Reading package lists...
Building dependency tree...
Reading state information...
E: Unable to locate package remminaanacron
Host machines with 4Gb physical memory are unable to build all images. During workiin on CI improvements it was noticed that runners with just 4Gb of memory are unable to create rootfs cache or make an images even that should be sufficient. Memory reservations are set very conservatively and we can keep that since adding a swap into the equation proves to be enough. If host machine is running Armbian (which will be in most cases) our ZRAM / ZSWAP memory management takes care of securing that extra needed memory.
