On Mon, 29 Dec 2025 11:28:20 +0100
Theo Buehler <[email protected]> wrote:

> On Sun, Dec 28, 2025 at 11:44:05PM +0100, Dennis Camera wrote:
> > >Synopsis:  OpenSSH 10.2: Illegal instruction in AES-256-GCM cipher on i686
> > >Category:  user
> > >Environment:  
> >     System      : OpenBSD 7.8
> >     Details     : OpenBSD 7.8 (GENERIC) #113: Sun Oct 12 15:23:27 MDT 2025
> >                      
> > [email protected]:/usr/src/sys/arch/i386/compile/GENERIC
> > 
> >     Architecture: OpenBSD.i386
> >     Machine     : i386  
> > >Description:  
> >     When the OpenSSH client is used on i686 (Intel Pentium 2)
> >     the ssh process crashes due to an Illegal instruction.
> > 
> >     A similar problem also occurs when a client tries to
> >     connect to an OpenSSH server running on i686 using the
> >     [email protected] cipher.  
> 
> > cpu0: Intel Celeron ("GenuineIntel" 686-class, 256KB L2 cache) 367 MHz, 
> > 06-06-0a, patch 0000000d
> > cpu0: 
> > FPU,V86,DE,PSE,TSC,MSR,PAE,MCE,CX8,SEP,MTRR,PGE,MCA,CMOV,PSE36,MMX,FXSR,PERF,MELTDOWN
> >   
> 
> This is a pure MMX machine without SSE and therefore is the same bug as
> reported here two weeks ago:
> 
> https://marc.info/?l=libressl&m=176581546511021&w=2
> 
> Recent logic changes made a mistake due to misleading naming of the i386
> perlasm routines where the MMX routines also use SSE instructions.
> The fix is to use this code only for machines with MMX + SSE. My
> understanding is that jsing has a fix in mind, which he should land.

Hi Theo,

thank you for this immediate response.

It sounds very plausible that the bug you linked is also the cause of
my problem.

In this case I think my report is a duplicate and can be closed.

Best regards,
Dennis

Reply via email to