Comment # 7 on bug 1196465 from Fabian Vogt

(In reply to Cyril Hrubis from comment #6)
> > > Ah right, looking at the ARM specification it indeed does return 0. And for
> > > floating point there is a FPCR register where you can turn on/off if
> > > exceptions are generated in that case and it looks like it's off by default
> > > and has to be enabled. Also looks like QEMU does not even implement these
> > > registers. So I guess the best we can do is to directly raise() the SIGFPE
> > > on ARM.
> > 
> > IMO that needs to be done on all architectures. Relying on undefined
> > behaviour is just broken.
> 
> As far as I can tell it's not undefined as long as division by zero causes
> exception, which is well defined per architecture. How else can we test that
> the kernel handler for this interrupt does work as expected?

You can use float instead of int, that should be supported better.