[PATCH v3 5/5] test/py: Create a test for launching UEFI binaries from FIT images

[Cristian Ciocaltea <cristian.ciocaltea@gmail.com>]

On Wed, Dec 18, 2019 at 11:56:00AM +0100, Heinrich Schuchardt wrote:
> On 12/18/19 9:22 AM, Cristian Ciocaltea wrote:
> > > > +    };
> > > > +};
> > > > +'''
> > > > +
> > > > + at pytest.mark.boardspec('sandbox')
> > > This test looks ok in principal. But why should we restrict it to the
> > > sandbox?
> > Let me see how this should work on real hardware, I'm going to test
> > on qemu for the moment.
> 
> Device trees cannot be used in conjunction with ACPI tables when booting
> via UEFI. Currently this concerns only x86 and x86_64 but Phytec is
> working on an arm64 board which shall provide an ACPI table. So you
> probably want to check CONFIG_GENERATE_ACPI_TABLE instead of the board
> type, e.g.
> 
> @pytest.mark.notbuildconfigspec('generate_acpi_table')

Thanks for the hint! 

For some strange reason, whatever I put in the 'notbuildconfigspec'
marker causes the test to be skipped:

[-] Section: test_efi_fit
TIME: NOW: 2019/12/18 17:33:14.976576
TIME: SINCE-PREV: 0:00:00.192132
TIME: SINCE-START: 0:00:00.192132
SKIPPED:
('[...]/uboot/test/py/conftest.py', 463, 'Skipped: .config feature "bootm_efi" enabled')

I don't really understand the connection to 'bootm_efi'. If I comment
out the 'bootm_efi' marker, the reported status becomes:

SKIPPED:
('[].../uboot/test/py/conftest.py', 463, 'Skipped: .config feature "cmd_bootefi_hello_compile" enabled')

So it seems 'notbuildconfigspec' gets its parameter from the first
active 'buildconfigspec' statement. This is my current test:

#@pytest.mark.buildconfigspec('bootm_efi')
@pytest.mark.buildconfigspec('cmd_bootefi_hello_compile')
#@pytest.mark.notbuildconfigspec('generate_acpi_table')
@pytest.mark.notbuildconfigspec('fake_item')
@pytest.mark.requiredtool('dtc')

> Best regards
> 
> Heinrich