The minimal initramfs was replaced by making initramfs-extra an opt-in
feature. Suggest doing this instead for small boot partitions.

See postmarketOS/pmaports!5636
[ci:skip-build]: already built successfully in CI

This way one doesn't have to conduct shotgun surgery whenever the
suggested approach to this changes.

This was moved to /usr/bin

Nikita Travkin authored 8 months ago and Clayton Craft committed 8 months ago

sd-boot allows loading efi drivers on boot. Add a special directory to
load drivers from.

[ci:skip-build]: already built successfully in CI

19atlas authored 8 months ago and Clayton Craft committed 8 months ago

[ci:skip-build]: already built successfully in CI

This adds a smaller summary at the beginning of CONTRIBUTING.md with
some helpful info for folks who are submitting and/or reviewing patches.

I also added a reference in the README to this doc, to notify/remind
folks (like me) who sometimes don't make it past the readme.

[ci:skip-build] already built successfully in CI

19atlas authored 8 months ago and Clayton Craft committed 8 months ago

[ci:skip-build] already built successfully in CI

Caleb Connolly authored 8 months ago and Clayton Craft committed 8 months ago

Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>

Brady Norander authored 9 months ago and Newbyte committed 9 months ago

Some chromebooks have a small kernel partition which barely fits the
depthcharge image, and in certain cases is too small for the depthcharge
image. Add support for compressing the kernel in the depthcharge image
which fixes this issue. Not every chromebook supports compressed kernels,
so a deviceinfo variable is added to specify the compression type
(either lzma or lz4). If the variable is unset, compression will not be used.

fixes #31

Fixes issues where we try to flash a boot.img that is larger than the
target partition size

methanal authored 11 months ago and Clayton Craft committed 10 months ago

[ci:skip-build]: already built successfully in CI

methanal authored 1 year ago and Clayton Craft committed 10 months ago

The dtbTool and dtbTool-sprd programs require the dtb directory as an
argument. These programs tend to combine multiple devicetrees. We can't
determine for whether the directory has devicetree blobs for a single
device, or multiple devices.

It's safer to copy the target dtb over to a temporary directory, and
pass that directory as an argument to dtbTool.

Syboxez Blank authored 11 months ago and Clayton Craft committed 11 months ago

Stop hardcoding `in-pmbootstrap` to check chroot status in favor of
comparing the current root inode ID to the root inode ID of the `init`
process (PID 1)

Used for sorting multiple entries under esp/loader/entries.
See:
https://uapi-group.org/specifications/specs/boot_loader_specification/#type-1-boot-loader-entry-keys

Newbyte authored 1 year ago and Clayton Craft committed 11 months ago

Taken from https://gitlab.com/postmarketOS/pmaports/-/blob/7aa0c329eded5f198b05daf8bbc32a100d677b82/main/postmarketos-update-depthcharge-kernel/update-depthcharge-kernel.sh

This is to more easily let other distributions re-use boot-deploy. Right
now, boot-deploy has a hard dependency on these external scripts for
certain functionality, which is a problem as they're packaged in
pmaports which isn't a great repository to use as source in packages for
other distributions.

As such, just integrate these scripts directly into boot-deploy. Given
their small size and close semantic function to the rest of boot-deploy
there is no greater benefit in keeping them in separate files.

Newbyte authored 1 year ago and Clayton Craft committed 11 months ago

Taken from https://gitlab.com/postmarketOS/pmaports/-/blob/7aa0c329eded5f198b05daf8bbc32a100d677b82/main/postmarketos-update-kernel/update-kernel.sh

This is to more easily let other distributions re-use boot-deploy. Right
now, boot-deploy has a hard dependency on these external scripts for
certain functionality, which is a problem as they're packaged in
pmaports which isn't a great repository to use as source in packages for
other distributions.

As such, just integrate these scripts directly into boot-deploy. Given
their small size and close semantic function to the rest of boot-deploy
there is no greater benefit in keeping them in separate files.

Closes https://gitlab.com/postmarketOS/boot-deploy/-/issues/3

Newbyte authored 1 year ago and Clayton Craft committed 11 months ago

Future code uses certain echo flags which we want to support.

Fixes generating extlinux config when the system does not use a dtb, by
making sure this variable is initialized. The failure happened later
when this var was used to create the config.

The Linux kernel supports live patching CPU microcode on boot, something
we should definitely support. This patch adds support to grub and
sd-boot / bootloader spec config.

A new function is added to list out ucode in a directory (currently just
the output dir, which is almost always /boot). It's very basic at the
moment, it makes some pretty big assumptions about the name of the ucode
patch file (i.e. currently expects it to match *ucode.img) This if fine
for the ucode fw filenames that Alpine Linux / pmOS support. These
things can/should be improved later if we need to support more
situations.

For context/background, see: https://gitlab.com/postmarketOS/pmaports/-/issues/2370

Earlier versions of this patch tried to preserve writing the whole
configs for each bootloader in one call to `cat`, e.g. by generating the
list of lines to add ucode files, but it was a bit unruly (extraneous
newlines and formatting was inconsistent)

This fixes a problem where the kernel-dtb check in bootimg testing
failed, because the work dir only had artifacts from mkinitfs and not
from boot-deploy.

This is already tested implicitly by the bootimg test

If find doesn't turn up any files, then this check was always passing,
which is obviously wrong :)

There's a comment about using find to support globbing, but shell
globing works and using `test -e` is simpler than trying to get `find`
to do what we want.

Caleb Connolly authored 1 year ago and Clayton Craft committed 1 year ago

Since MR 50 vmlinuz-dtb is treated as ephemeral and is only accessible
in the workdir, adjust the path for our validation.

Signed-off-by: Caleb Connolly <caleb@connolly.tech>

Caleb Connolly authored 1 year ago and Clayton Craft committed 1 year ago

For some kernels in postmarketOS it will be useful to support
systemd-booting (via a chainloaded bootloader like u-boot) as well as
supporting the existing bootloader which only supports a GZIP compressed
kernel image.

To enable this usecase, check to see if a version of the kernel file
with the `.efi` extension exists and prioritise using it for
systemd-boot.

Signed-off-by: Caleb Connolly <caleb@connolly.tech>

Caleb Connolly authored 1 year ago and Clayton Craft committed 1 year ago

When appending the dtb to the kernel, we would also copy the resulting
file to /boot, however there is no use for this and it takes up precious
space (the default size of /boot is 128M). Stop copying it there.

Signed-off-by: Caleb Connolly <caleb@connolly.tech>

Caleb Connolly authored 1 year ago and Clayton Craft committed 1 year ago

Add tests that run boot-deploy in the context of a real rootfs and
validates the behaviour.

Signed-off-by: Caleb Connolly <caleb@connolly.tech>

methanal authored 1 year ago and Clayton Craft committed 1 year ago

Add support for generation of dt.img in boot-deploy.

The type is defined in `deviceinfo_bootimg_qcdt_type'. Valid values
are "qcom", "exynos" and "sprd".

Nikita Travkin authored 1 year ago and Clayton Craft committed 1 year ago

Some devices may have multiple possible bootloaders where it may be
benefitial to ship both an androidboot image and some kind of filesystem
based config. In this case the dtb should be deployed despite being
appended to the kernel blob.

Nikita Travkin authored 1 year ago and Clayton Craft committed 1 year ago

To allow for user-modifiable deviceinfo values, the source_deviceinfo
was introduced to the system. Make use of it, if it exists.

Ben Westover authored 1 year ago and Clayton Craft committed 1 year ago

My Amazon Echo Dot (2nd gen) has a boot.img with no header on the ramdisk.
These changes account for that situation by allowing the kernel and ramdisk
header labels to be specified separately instead of being hardcoded together.

methanal authored 1 year ago and Clayton Craft committed 1 year ago

Hooks can be used to run scripts and binaries after source_*
functions are called.

Hooks are probed in two directories: /usr/share/boot-deploy/hooks and
/etc/boot-deploy/hooks

This adds support for systemd-boot, by combining it with support for
gummiboot (the gummiboot project turned into "systemd-boot"...). This
also installs all available bootloaders under the relevant systemd-boot
and gummiboot directories. That's useful because it allows something
outside of boot-deploy to control what boot-deploy installs, without
having to add a bunch of new logic + new deviceinfo var for it.

The `generate_gummiboot` var was renamed to `generate_systemd_boot`, one
turned into the other, and the internal logic in boot-deploy will now
search dirs that both gummiboot and systemd-boot are installed in. It
prefers systemd-boot over gummiboot, in case both options are located on
the system.

deviceinfo changes:

- ADDED: deviceinfo_generate_systemd_boot
        - triggers installing support for systemd-boot

- DROP: deviceinfo_generate_gummiboot
        - Use deviceinfo_generate_systemd_boot to trigger gummiboot
          support