#1059870 aboot: Please help testing dropping a.out.h support

Package:
aboot
Source:
aboot
Submitter:
Dimitri John Ledkov
Date:
2024-01-02 21:12:04 UTC
Severity:
normal
#1059870#5
Date:
2024-01-02 17:37:45 UTC
From:
To:
Dear Maintainer,

a.out support has been dropped in upstream kernel. However a.out.h
header is still being using by abootimg tool via objstrip.c. It has
support for both a.out image types and ELF. I have blidly dropped
a.out.h support and made ELF support mandatory in
https://lore.kernel.org/all/20231123180246.750674-2-dimitri.ledkov@canonical.com/
but I have no way to build it for alpha or test if everything still
works. Upgrades, installed systems, and cd-boot.

Could you please consider the attached NMU, build it and test it, and
let me know if everything still works. Cause then a.out.h header can
be removed from upstream linux kernel on all architectures.

Regards,

Dimitri.

#1059870#10
Date:
2024-01-02 17:48:19 UTC
From:
To:
Hi Dimitri!

I have actually already looked into patching aboot myself to deal with the
aout.h issue since I want to make the bootloader fully cross-buildable [1].

I will most likely just copy the relevant definitions out of aout.h, both
the generic and the alpha-specific stuff in order to both be able to drop
reliance on aout.h in the kernel as well as make the package fully cross-
buildable.

I would also prefer to get all relevant changes upstreamed.

Thanks,
Adrian

#1059870#15
Date:
2024-01-02 18:07:41 UTC
From:
To:
On Tue, 2 Jan 2024 at 17:48, John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> wrote:

For cross building, as far as I can tell one needs to build the tool
twice - once for the BUILD architecture and once for HOST
architecture, use one during the build and package the other one into
the deb.

But why? as far as I can tell, the a.out code path is never executed
as it seems like Debian has been using ELF based vmlinuz since like
before 2009. Are you sure that the a.out codepath of objstrip.c is
needed or executed at all?

Or am I missing something, and like objstrip.c portions are executed
against some other a.out formatted things which are not Linux kernel?

Should I provide a patch that adds printfs during a.out codepath
block, to see if it actually is ever executed?

Upstream has policy of not carrying dead code, which in this case its
what it is for kernels built in like last decade.

#1059870#20
Date:
2024-01-02 21:00:54 UTC
From:
To:
That's correct and I am currently pondering over a clever way to do that.

I think it's still reasonable to keep a.out support in the tool because
users might use it for a.out binaries generated from other tools. Besides,
it's just a matter of copying a few header definitions, so not really a
blocker unless I am missing something.

I was talking about aboot upstream which is maintained by Matt Turner from
Gentoo these days [1].

Adrian

#1059870#25
Date:
2024-01-02 21:09:27 UTC
From:
To:
On Tue, 2 Jan 2024 at 21:00, John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> wrote:

Which tools? given that support has been dropped to execute any
binaries like that everywhere. This still sounds very hypothetical,
given that code path is specifically for vmlinux kernel only.

Ok, and I'm talking about linux kernel copy of aboot at
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/alpha/boot

Can you please test the debdiff I have attached, such that I can
confirm with linux kernel upstream that I can drop a.out.h support
there?

I do not have access to alpha, and currently this is the remaining
piece holding up a.out.h in linux kernel upstream on x86 alpha m86k
mips. Which would be nice to drop.