Re: [PATCH v2 0/4] Bug 24431 Fix analysis of v4.19 aarch64 kernels

[Dodji Seketeli <dodji@seketeli.org>]

Hello Matthias,

maennich@google.com a écrit:

> This series addresses an issue with reading the __ksymtab of v4.19+ aarch64
> kernels that has been introduced post libabigail-1.6.
>
> The minimal patch would be to cast the symbol addresses acquired by
> read_int_from_array_of_bytes to int32_t. But treating that values as a
> 32bit value already when reading allows to remove the architecture specific
> treatment and the need to deal with overflows.

The patch series looks good to me, and I have checked it in as it seems
to pass on the the few kernel binaries I am watching.

As a side note, this makes me think that we need a regression test
harness for the kernel (and for all huge binaries), a bit like what we
have when we run make check today.  That new harness would have to be
separate from the libabigail tarball/repository though, as kernel
binaries are big :-) I am thinking maybe something that would be hosted
on github or somewhere, would either store (kernel) binaries or know how
to build them, and store the expected output of the various libabigail
tools on those binaries.  The harness would then run the tool on the
binary and compare the outcome with the expected one.  This is key to
ensure we keep functionalities working as we move forward.

Parties (like you) who have specific binaries/areas of interests, would
submit data to that repository to ensure that subsequent changes to the
libabigail doesn't break the particular functionality they are
interested in.

But that would be an effort for another time.  I just thought I'd
write this down for future reference :-)

Thanks!

-- 
		Dodji