Re: [PATCH] powerpc64: Increase SIGSTKSZ and MINSIGSTKSZ

[Tulio Magno Quites Machado Filho <tuliom@ascii.art.br>]

Rajalakshmi Srinivasaraghavan via Libc-alpha <libc-alpha@sourceware.org>
writes:

> On 12/13/22 10:23 AM, Florian Weimer wrote:

>> This is technically an ABI change, but if you think this is the right
>> forward, you can certainly make that change.  It's not the path some
>> other architectures have taken when they faced similar problems, so I
>> don't think we have much experience in this area.

I do agree this can be interpreted as an ABI change.
However, I used abidiff against the following libraries and it didn't
report any changes:

hesiod/libnss_hesiod.so.2
nis/libnsl.so.1
login/libutil.so.1
libc.so.6
crypt/libcrypt.so.1
locale/libBrokenLocale.so.1
nptl_db/libthread_db.so.1
math/libm.so.6
resolv/libanl.so.1
resolv/libresolv.so.2
resolv/libnss_dns.so.2
nss/libnss_files.so.2
nss/libnss_compat.so.2
nss/libnss_db.so.2
nss/libnss_test1.so.2
nss/libnss_test_errno.so.2
nss/libnss_test2.so.2
dlfcn/libdl.so.2
nptl/libpthread.so.0
rt/librt.so.1
malloc/libc_malloc_debug.so.0

So, I think this change is safe for glibc 2.37.

>> Should we backport this to older releases (or downstream)?
>
> Yes, This should be backported to older releases as well.

IMHO, this is trickier. Not because a glibc binary interface will change,
but because user software might change.
Anyway, the kernel >= 5.19 is already requesting the new sizes, although
it isn't using it to the best of my knowledge.

Maybe we should be extra careful and avoid backports upstream?
That would leave for downstream to apply this patch as necessary
(ideally, before a mass rebuild).

Am I missing anything?

-- 
Tulio Magno