Re: [hatari-devel] Changes to STOP/LPSTOP instruction

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 09/11/2017 à 11:26, Toni Wilen a écrit :
> > Wouldn't that stack the new SR without the S bit set in the exception
> > frame, although the instruction itself was executed in supervisor mode,
> > and thus return to user mode, if the handler executes an RTE? That would
> > really be strange.
>
> Yes, that is exactly what happens! (real 68000 confirmed, it is 68000
> undocumented feature)

@Thorsten : here're some real values I measured on my 68000 STF


SR=2700       STOP 555        SR_NEW=2515     SR_STACK=0515 
PC_STACK=STOP+4
SR=2700       STOP 777        SR_NEW=2717     SR_STACK=0717 
PC_STACK=STOP+4

SR=700        STOP 555        SR_NEW=2700     SR_STACK=0700   PC_STACK=STOP
SR=700        STOP 777        SR_NEW=2700     SR_STACK=0700   PC_STACK=STOP
SR=7ff        STOP 777        SR_NEW=271F     SR_STACK=071F   PC_STACK=STOP


The test is to set a privilege violation exception handler (addr $20) 
and to save SR_NEW (value of SR at the start the privilege exception 
handler) as well as the stacked values

We can see that when S=0 before the STOP, we have a "normal" priv 
exception with expected values from the doc ; but when S=1 before the 
STOP but S=0 in the STOP value then the stacked value is the one with 
the S=0 bit, which means that after a RTE in the priv handler you will 
be in user mode, so you might have the feeling that STOP worked and 
ended on the correct interrupt condition, but in fact it was just the 
priv exception that triggered "immediately". Weird behaviour :)