On Wednesday 16 June 2021 17:20:20 Tom Rini wrote:
> On Wed, Jun 16, 2021 at 11:16:32PM +0200, Pali Rohár wrote:
> > On Wednesday 16 June 2021 17:12:13 Tom Rini wrote:
> > > Wait, why is thumb disabled? That's something I'd really like to know
> > > why on as my first guess is there's units with old enough cores that
> > > have some thumb errata maybe?
> >
> > IIRC we disabled at time when N900 support was merged into U-Boot (about
> > 10 years ago) as built binary did not work on real N900 HW. Non-thumb
> > build work fine.
> >
> > I thought it was due to thumb errata but yesterday I was told that thumb
> > errata affects running system only when virtual memory and context
> > switches are used. And correct me if I'm wrong, but u-boot is not using
> > virtual memory mapping and neither multiprocessing context switches.
> >
> > So it looks like that at that time (10 years ago) there was another bug
> > in u-boot which caused that thumb build was non-working.
>
> Thanks for explaining, yes, we should be able to enable thumb and

This needs to be tested on N900 hw if it really works.

> disable EFI and see just how close things are again to the size limit.

EFI is not used on N900 and I'm not aware why it is enabled. Maybe
somebody proactively enabled it and nobody noticed? So very good
candidate for immediate disabling to reduce space.

> Oh, and we have things set such that the hooks to cause link time errors
> on size overflow are enabled here, right?

Seems that link time errors are not enabled. But "integration" test for
N900 which is running on Azure is checking this enforcement. Just
because generating final MTD image (for running test) fails.

Also you can run this test manually on local computer:

./test/nokia_rx51_test.sh

So thanks to Azure CI testing, we are sure if tests passes, N900 code is
working fine (at least in qemu) and also has "correct" size.