-----BEGIN PGP SIGNATURE-----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=tAJh
-----END PGP SIGNATURE-----
On mer. 13 août 2025 09:04:22, Aaron Rainbolt wrote:
> On Wed, Aug 13, 2025 at 7:39 AM Simon Chopin <simon.chopin@canonical.com> wrote:
[snip]
> > Now, if I'm wrong about lack of major breakage, you want to watch out
> > for the following symptoms in your autopkgtests, build logs, or bug
> > reports:
> >
> > * FTBFS due to missing termio.h
> > Your package is using some really outdated interfaces that were
> > already obsolete when the Berlin Wall fell. Please look into
> > termios.h.
> > * ARM issues only on modern ARM64 processors (e.g. Apple M4), around coroutine
> > or exception-handling code. Tobias, do we even support running on M4
> > hardware!?
>
> Does "M4 hardware" mean running directly on Apple's M4 CPU bare metal,
> or does it also include running inside a virtual machine under
> Parallels, VMware Fusion, UTM, etc.? If this is a problem inside
> virtual machines, I personally don't think that would be acceptable.
> If it's a baremetal-only issue, then meh, probably not that big of a
> deal. Just my 2 cents on the matter.
Don't worry too much about this. I'm not saying those systems are
broken, just that *if* you see issues looking like that on those systems
then this upload might be the cause.
The details are a bit beyond my knowledge, but basically the SME matrix
storage is disabled when calling setjmp() for consistency with
longjmp(). It's unlikely to affect Ubuntu since we don't generate those
opcodes by default and even then, why would you call setjmp() in the
middle of matrix operations, but OTOH it's the kind of thing that's a
nightmare to debug, which is why I figured I would mention it.
As to whether or not your VMs are affected, I wouldn't know. Check
/proc/cpuinfo to see if you have SME in your VM?
Cheers,
Simon
