[PATCH v3 18/19] test/py: android: extend abootimg test

[Safae Ouajih]

On 07/02/2023 20:02, Tom Rini wrote:
> On Mon, Feb 06, 2023 at 12:50:20AM +0100, Safae Ouajih wrote:
>
>> test_abootimg is extended to include the testing of boot images
>> version 4. For this, boot.img and vendor_boot.img have been
>> generated using mkbootimg tool with setting the header
>> version to 4.
>>
>> This tests:
>> - Getting the header version using abootimg
>> - Extracting the load address of the dtb
>> - Extracting the dtb start address in RAM
>>
>> Running test:
>> $ ./test/py/test.py --bd sandbox --build -k test_abootimg
>>
>> Signed-off-by: Safae Ouajih <souajih at baylibre.com>
>> Reviewed-by: Simon Glass <sjg at chromium.org>
>> ---
>>   test/py/tests/test_android/test_abootimg.py | 136 ++++++++++++++++++--
>>   1 file changed, 123 insertions(+), 13 deletions(-)
> Alright, so I don't know where the failure starts, exactly. And to make
> testing this easier, there's currently the
> trini/u-boot-gitlab-ci-runner:jammy-20230126-07Feb2023 container you can
> use to replicate my problem. The problem is that while this test passes
> in CI, with GCC, with Clang it fails, consistently:
> https://source.denx.de/u-boot/u-boot/-/jobs/572239#L284
> and I'm not quite sure why. I hope that building sandbox with clang and
> also just trying these features out interactively will fail too, and so
> debugging this will be less of a problem.
>
Hello Tom,

I have investigated this issue, clang has a strange behavior in:

* abootimg_get_dtb_load_addr() : cmd/abootimg.c
* android_image_get_dtb_by_index() : boot/image-android.c

That is probably linked to some sort of optimization clang does.

However, The fail is not reproducible using clang-15 and clang-16 and also
not reproducible when turning off clang optimizations.

I suggest using clang-15 to run the test or I can remove all optimizations

on the related functions if clang-14 is used.

Best regards,

Safae