Pull in 3 patches for edk2 to make it build with modern compilers.
The patches have already landed upstream.

Signed-off-by: Bernhard Rosenkränzer <bernhard.rosenkraenzer.ext@huawei.com>
Upstream-Status: Backport

Build Zephyr debug info in DWARF-4 format instead of DWARF-5.
The Zephyr test suite uses tools that can't handle DRAWF-5 properly yet
(py-elfutils), causing tests to fail if Zephyr has been built with
DWARF-5.

Signed-off-by: Bernhard Rosenkränzer <bernhard.rosenkraenzer.ext@huawei.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

I missed this in the earlier patch-set. I've added a note to the RAUC
recipe to allow people to notice this better than by trial and error.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

As we add support for additional boards, doing it by copying and
modifying a file with relatively few parts changing got tedious.

The change adds half a dozen new variables for selecting the boot loader
type, various strings and directories. The defaults are safe, even if
unused, for all kinds of boards. The machine-specific overrides tune the
bootloader used by each board and provide proper identification strings.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

For the moment this allows booting from USB storage. Ideally we would
have one config that works for eMMC, nvme, USB and SATA but we're not
there yet.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

As we add support for additional boards, doing it by copying and
modifying a file with relatively few parts changing got tedious.

The change adds three new variables - RAUC_{COMPAT,SLOT_A,SLOT_B},
and defines them for the boards supported thus far.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

This is useful for the outer squashfs used to deliver a RAUC update
bundle.  The inner squashfs uses stronger compression. The outer is now
using zstd (since a5437196). Having two layers of strong compression
makes no sense, as it cannot offer any improvements, and is much heavier
on CPU and memory.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

The configuration files are very similar to what is available for
Raspberry Pi 4. In general all the files, except for COMPATIBLE
setting can become identical over time, as we move to GPT partition
identifiers to identify the slots.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

This script implements proper A/B update with automatic rollback.
Refer to sysota.grub(7) manual page for details.

The script should become built-into grub EFI executable but that will be
pursued separately.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Those are required by the upcoming grub.cfg boot script.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

This allows using sysota with rauc-hawkbit-updater. It works around the
issue where r-h-u fails to talk to hawkbit whenever the post-install
handler reboots the system.

Set the delay to 180 seconds, which should be enough for the updater to
finish talking to hawkbit, and then some more.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

This brings in initial support for working with GRUB.

The location has changed to Eclipse GitLab instance
with the GO_IMPORT of gitlab.com/zygoon/sysota. See the referenced
pull request for rationale on the latter part of this change.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Shared linking has caused some problems on x86, where sysotactl would
crash on an apparently-stray pointer. This looks like a compiler bug.

Shared linking mode causes Go's standard library to be built into a
large .so file and all applications link to that with the dynamic
linker.

Given that sysota is the only Go application in our image and that is
has only one binary, this does not bring meaningful savings. In fact,
without shared linking, the overall image size is smaller as unused
parts of Go stdlib are removed.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

This also includes a set of additional small tweaks:
1. formatting and small rewarding
2. explicitly use the kitkstone branch for repo init (this avoids
confusion and also saves us from repo edge cases where the default one
is ignored)

Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

Signed-off-by: Pavel Zhukov <pavel.zhukov@huawei.com>

Right now, we do a remove in DISTRO_FEATURES, that, while we don't
neccessarily support them, due to the finality of :remove we force our
end users to not use them either.

By moving DISTRO_FEATURES:remove to DEFAULT_DISTRO_FEATURES:remove, we
allow end users to still use oniro.conf but add in those removed
features.

We also move IMAGE_FEATURES from a distro level var to the image.
IMAGE_FEATURES should generally be put in image recipes. This also gives
end users a way to use Oniro and work around our :remove defaults.

Signed-off-by: Eilís Ní Fhlannagáin <pidge@pidge.org>

Chase Qi authored 3 years ago and Andrei Gherzan committed 2 years ago

Add the recipe to `-base-tests` image for kselftest.

Add kernel-selftest.bbappend to install the kselftest runner scripts
and also make it compatible with libc-musl.

Signed-off-by: Chase Qi <chase.qi@linaro.org>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Read-only Oniro images now have space to provision rauc-hawkbit-updater
configuration file. The startup of the update service is conditional on
the presence of that file. The default configuration file is stored as
an example in the read-only image.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

This fixes a crash on NULL pointer on startup.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Signed-off-by: Davide Gardenal <davide.gardenal@huawei.com>

Version 0.28 will come with these patches merged in. This will happen as
soon as the relevant upstream merge gets in[1].

[1] meta-zephyr!1



Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

Zygmunt Krynicki authored 3 years ago and Andrei Gherzan committed 3 years ago

This saves half of the space used by compressed disk artifacts.

I ran into the issue when creating grubenv recovery logic. To my surprise
the execution flow ended up in the case where both slots are available
and timestamp on the kernel is used to break the tie.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

Make sure we explicitly set up the WKS file for both C61 targets
after the target rename.

Signed-off-by: Marta Rybczynska <marta.rybczynska@huawei.com>
Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

Move compiler security options from only linux to all kernels

Signed-off-by: Davide Gardenal <davide.gardenal@huawei.com>

Fix go hardcoding glibc's dynamic linker location into musl binaries

Signed-off-by: Bernhard Rosenkränzer <bernhard.rosenkraenzer.ext@huawei.com>

Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

sysota is a go project using go modules. Due to how this support is
currently implemented in the core, the recipe will download the
dependent go packages during compile.

On the other hand, lately, the core also defaulted to a restricted
network environment in the compile task breaking (but for good reasons)
the compilation of sysota. This situation was done as a forcing function
for the community to react and improve the current implementation of go
mods in the core. For now, we workaround it as this doesn't change the
current limitations/behaviour while we investigate better options in the
core.

Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

Upstream moved the default definition of TCLIBC as part of:

commit 57aa60ef6422568b425b6ccc4451567efc578469
Author: Martin Jansa <Martin.Jansa@gmail.com>
Date:   Wed Sep 16 15:28:52 2020 +0200
    bitbake.conf: use ${TCMODE}-${TCLIBC} directory for CACHE

This moved TCLIBC from defaultsetup.conf to bitbake.conf but the problem
is that it did it before the place were bitbake.conf includes the distro
configuration changing its behaviour. With this change, a ?= in the
distro configuration will be ignored as the first one is already
defined in bitbake.conf.

Because of the above, we ended up with defaulting to glibc. We revert
this by forcing the value of TCLIBC accordingly. This will have a side
effect that shell environment variable will be ignored. So doing that
through BB_ENV_EXTRAWHITE will be broken from now on. If we want that
ability in the future, we can try to address.

Fixes: https://booting.oniroproject.org/distro/oniro/-/issues/252



Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>

The oniro-grub-bootconf recipe provides the virtual-grub-bootconf
package which in turn contains a stub grub.cfg and empty grubenv file.

The change is spread around a considerable number of areas. Specifically
the kick-start file needs to explicitly reference the grub.cfg file which
must be placed in the wic/ directory. Since we need the same file in
grub/files directory, so that it can be installed into the image,
a symbolic link is employed.

The distribution configuration file, oniro-linux.conf sets
IMAGE_EFI_BOOT_FILES so that the grubenv file is installed into the boot
partition. This is required since grub cannot create the file by itself,
it can only edit the file, in-place, if it is present there already.

The actual configuration file for GRUB is a stub. Appropriate file with
some non-trivial logic will be proposed separately. The stub file boots
the first slot all the time. The kernel is loaded from the rootfs.

This test-boots fine in x86 and x86-64 qemu builds.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>

The COMPATIBLE strings are set to "QEMU x86" and "QEMU x86-64"
respectively. The values are not standardized, but since the idea is
that bundles with a matching COMPATIBLE string can be used for updates,
picking a generic value is most appropriate.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>