So far, W=1 is used because W=2 warns about duplicated unit addresses
due to USI / I2C / serial / SPI as per
https://lore.kernel.org/all/166636173496.3593878.13529338596528679973.robh@kernel.org/



I've added a small filter to filter-out these warnings specifically,
and we can enable W=2 by default now, as W=2 is very useful generally.
This filter is very specific on purpose, assuming order of warnings
doesn't change from time to time, so that we can catch any (erroneous)
changes immediately.

One can still override this to W=1 on the make command line if so
desired (but that shouldn't be needed - if there's a genuine need, we'd
better update the filter list).

Implementation-wise, we have to redirect kbuild's stderr (which
contains the DT warnings) to stdout, so that these can be piped to the
script and filtered, and then redirect that filtered output back to
stderr, without mixing the original kbuild's stdout with the filtered
output.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

Seems Rob's DT validation scripts when sending binding patches also
explicitly run this kbuild target.

Unfortunately, this target is not run by default during
dt_binding_check, and additionally this target doesn't error in case of
(new) invalid documentation references.

Add a wrapper here that filters out warnings at this point in time and
errors in case of new warnings, hopefully allowing us to easily spot
any new warnings that we might add, preventing us from sending
erroneous patches.

This should get some better integration with this build system here so
as to make it easier to run all the presubmit checks without having to
remember the growing list of make targets to invoke manually, but for
now it's at least something.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

This is probably not useful as-is, as the kernel still has thousands of
undocumented compatibles, but one could manually generate a list of
before and after any changes and then compare them.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

Rather than having to add (and remember!) bindings one-by-one, we
should probably have something more clever, e.g. based on commits
touching files below Documentation/devicetree/bindings/.

For now, just adding something that matches using this hard-coded list
is the easiest way forward, though.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

For compile testing. arc usually triggeres other errors that are useful
to catch early.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

Signed-off-by: André Draszik <andre.draszik@linaro.org>

We don't need Image.gz - we only need Image, which we post-process by
compressing using lz4 and then package the result of that.

This shaves about 250ms off each build.

While we could specify Image.lz4 here, I've opted not to do that, as
Kbuild uses compression level -9 whereas we use --fast, which is about
1.2 seconds faster in my environment. The size difference between the
two is less than 6MiB, which results in a barely measurable difference
in time when uploading to the device using fastboot boot.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

If .config doesn't change between two runs, Kbuild keeps the old
version (i.e. the old time stamp).

This causes this build system to try to rebuild it forever, as the
kernel repository's time stamp $(KERNEL_STAMP) will stay newer than
.config.

To avoid that and the time Kbuild spends on doing that, we can update
.config's time stamp.

This should be safe, as once we decided to build a kernel (i.e. because
$(KERNEL_STAMP) has changed), we know that .config is up-to-date, and
can therefore update the time stamp. This 'touch' isn't part of the
$(KERNEL_DOTCONFIG) recipe, because that would unconditionally force
this build system to rebuild the kernel by invoking the $(BUILD_KERNEL)
recipe, which we don't want. But if this $(BUILD_KERNEL) recipe is
invoked, there is no harm (I believe) in also updating the .config time
stamp. Kbuild is clever enough to not do additional work even with
.config's time stamp change.

This (only) helps on subsequent runs if the kernel sources haven't
actually changed, e.g. if you run 'make run' instead of 'make rerun' to
boot the device again, without any kernel changes.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

Since commit f7ef7980 ("Makefile: avoid copying of
$(OUT)/linux/arch/arm64/boot/Image") we don't necessarily 'touch' our
build time stamp anymore.

Stop using $@ and be explicit.

Fixes: f7ef7980 ("Makefile: avoid copying of $(OUT)/linux/arch/arm64/boot/Image")
Signed-off-by: André Draszik <andre.draszik@linaro.org>

We can simply reference the built $(OUT)/linux/arch/arm64/boot/Image
when compressing it, instead of copying it first.

This uses a rule with grouped targets, and lets make know explicitly
how to generate $(OUT)/linux/arch/arm64/boot/Image rather than it being
an implicit side-effect of the $(BUILD_STAMP) rule.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

Fixes: 2217b043 ("Makefile: fully drop DTBO support")
Signed-off-by: André Draszik <andre.draszik@linaro.org>

We now know how to make it work (using some custom decon init code),
and can therefore enable CONFIG_FB_SIMPLE.

Note that this shouldn't reach upstream, as this really is deprecated
and DRM should be used instead. For the time being, this gives us a
working framebuffer, though.

As a side-effect, the Linux framebuffer will turn the display mostly
black, resulting in huge energy savings (and a much cooler phone).

Signed-off-by: André Draszik <andre.draszik@linaro.org>

This speeds them up a lot, and we don't have to specify -j on the
command line explicitly anymore.

It is safe to do so I think since commit 0b17d6a1 ("Makefile:
support DT checks in parallel (using -j)")

Signed-off-by: André Draszik <andre.draszik@linaro.org>

* Rename previously oriole-specific makefile targets to raviole where
  it makes sense:
    dtbs_check_oriole -> dtbs_check_raviole
* check all gs101-*.dts files, to cover both, oriole & raven

Signed-off-by: André Draszik <andre.draszik@linaro.org>

This is as alternative to Pixel 6 (Oriole), only one can be specified
at a time.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

We will need to support more than just gs101-oriole.dtb (in particular
gs101-raven.dtb), and for that we need to tweak things slightly so that
either can be specified.

Note that we can only embed one DTB into dtb.img, as the bootloader
will pick the DTB based on SoC-ID, which is the same in both cases. The
bootloader additionally supports picking a DTBO overlay based on
board-id and board-revision, but with upstream there is no
infrastructure to build those.

Therefore, we have to build a dtb.img for the specific board that we
want to boot, but at least everything else can stay the same for either
boot target.

Introduce the make variables
    RUNTARGET= and ORIOLE=
which allow selecting which board to boot.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

This will make it easier to add support for building / booting for more
than one target board, e.g. gs101-oriole.dtb and gs101-raven.dtb

Signed-off-by: André Draszik <andre.draszik@linaro.org>

python packages need to be updated every now and then, add a little
note about that.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

Signed-off-by: André Draszik <andre.draszik@linaro.org>

At least one publicly available boot loader works fine now with the
dtbo partition empty, there is no need for this to work-around broken
boot loaders, and the README has been updated as well.

Drop all support for DTBO generation and flashing from here.

This will also simplify addition of support for further DTBs.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

slider-15.2-12357666 certainly works fine, and has been released
publicly via Android 15 QPR2 Beta 1.

Update the README with this information.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

pip appears to be silly enough to break package dependencies of
installed packages if not all installed packages are listed on the
command line.

In particular, dtschema requires jsonschema 4.17.3, while newer
versions are available. If everything except jsonschema is up-to-date,
the command line will only mention jsonschema, which pip treats as
'please update jsonschema unconditionally and break all my other
installed packages', which is surprising considering how package
managers in Linux normally work.

Update the command line to always mention the packages we actually care
about, to hopefully avoid this issue.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

Using
    make -j$(nproc) validate_dt
currently doesn't work, as kbuild end up generating
processed-schema.json from multiple threads and corrupting the output
file.

Teach make about that by specifying processed-schema.json as dependency
where required.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

When a non-default umask is in effect, e.g. 0027, building the package
fails:
    dpkg-deb: error: control directory has bad permissions 750 (must be >=0755 and <=0775)

Adjust the permission accordingly.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

While still assuming arm64, this change allows us to at least build
images for different platforms, e.g. by passing in a make variable or
from the environment.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

With this, we can build a debian package to pull in (and keep)
dependencies of this build environment.

Simply invoke
    ./buildme
    sudo dpkg -i ./pixel-build-deps*.deb
    sudo apt install -f

Signed-off-by: André Draszik <andre.draszik@linaro.org>

We now support sparse checking, so pull it in in the description of the
environment setup.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

By default, if sparse is available in PATH, enable it in the kernel
make command invocation using default arguments:
    C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__ -fmax-errors=unlimited -fmax-warnings=unlimited'

This can be overridden here on this build system's command line by
specifying an alternative sparse command line using
    SPARSE='alternative flags'
as well as the C=... flag.

To completely disable sparse checking, SPARSE can also be specified as
empty:
    make SPARSE= images

Signed-off-by: André Draszik <andre.draszik@linaro.org>
Suggested-by: Tudor Ambarus <tudor.ambarus@linaro.org>

Signed-off-by: Peter Griffin <gpeter@google.com>

Signed-off-by: Peter Griffin <gpeter@google.com>

To start things off lets add a section on clock name mangling,
the rationale behind it and the sed reg ex.

Signed-off-by: Peter Griffin <gpeter@google.com>

It avoids the following errors running xfstests encrypt tests

[ 1379.558414][T15925] xfs_io (pid 15925) is setting deprecated v1 encryption policy; recommend upgrading to v2.
[ 1379.703072][T15656] fscrypt (sda26, inode 260609): Missing crypto API support for Adiantum (API name: "adiantum(xchacha12,aes)")
[ 1379.710001][T15656] fscrypt (sda26, inode 260609): Missing crypto API support for Adiantum (API name: "adiantum(xchacha12,aes)")

Signed-off-by: Peter Griffin <gpeter@google.com>

This reverts commit 68053fd9.

The main reason for this revert is to re-enable all other DWC3 glues as
we want to at least compile-test them if / when changing framework
APIs.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

This reverts commit 13674a3e.

Should never be active, this commit just existed to document what's
necessary for downstream drivers.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

These are just for testing the downstream drivers, and should never be
merged

Signed-off-by: André Draszik <andre.draszik@linaro.org>

* build DWC3 as module (instead of the default built-in)
* disable non-exynos DWC3 glues
* disable various other USB host controller drivers
* build samsung usb phy as module (instead of default built-in)

Note that for testing the downstream drivers (i.e. if copying the
downstream drivers into the upstream kernel tree), the following would
be needed:
    # downstream USB
    CONFIG_PHY_EXYNOS_USBDRD=m
    CONFIG_PHY_SAMSUNG_USB_GEN2_V4=y
    # downstream exynos DWC3 glue
    CONFIG_USB_DWC3_EXYNOS_GS=m
    CONFIG_TYPEC_MAX77759=m

    CONFIG_GOOGLE_VOTABLE=y

Signed-off-by: André Draszik <andre.draszik@linaro.org>

We only need the maxim one, but let's enable all to be able to at least
compile-test them if / when changing framework APIs.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

COMPILE_TEST implies WERROR, and is broader as is compiles more code.

Signed-off-by: André Draszik <andre.draszik@linaro.org>

should have done that long ago...

Signed-off-by: André Draszik <andre.draszik@linaro.org>