Re: [hatari-devel] Problem with SCSI initiator command register

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

On 07/06/2012 18:53, Uwe Seimet wrote:
> Hi,
>
> > Well, what did you expect? One loop iteration will take approximately 5
> > milliseconds (since you're basically waiting for the 200 Hz counter to
> > toggle its sign), and you're looping with d1 = 12000 times ...
> > that makes 12000 * 0.005 s = 60 s = 1 minute.
> >
> > So unless you really expect something different here, I'd say Hatari
> > works here as designed ;-)
>
> This is getting more and more interesting ;-).I have not checked how
> long this loop might take for several years, probably because with a
> real Atari it will always terminate early. (I will have to re-check the
> arbitration timeout value again in any case, but I guess it is OK.)

Hello

just to confirm thomas' math and to clarify this subject, I ran your 
test loop under Hatari and everything looks fine regarding $fffa23 
content and the whole loop is really supposed to take one minute.

Tos starts a 200 Hz timer by using ctrl reg = 6 (prediv = 64) and data 
reg = 192. This effectively gives 2.4576 MHz / 64 / 192 = 200 Hz

My previous mails were not correct on the values you read in data 
register on a real MFP (as well as on Hatari, because we correctly 
emulate those as some programs would lock otherwise).

When the timer is running, you will read values in the data reg between 
1 and the starting value for the timer (192 in that case).

Forget my beq/bne test, there's only one special case when you read a 
"0" in data reg : if you stop the timer at the exact point where it 
would be looping (that is when its value is "1" and would be decremented 
to start again at 192 for example).

So in your case, your loop will read values between 1 and $7f (>0) and 
$80 and $c0 (<0).

"Luckily" for you, as the 200 Hz timer implies a data reg of $c0, your 
code is working and you can detect a timer's loop. But if TOS used some 
other values, it would not work at all (data reg = $50 for example) or 
could be very cpu sensitive (data reg = $82 for example).

With data reg = $c0, 33 % of the read are <0 and 66 % are >0, so this 
gives a large enough margin to account for possible interrupt of your 
loop and still be able to detect the bpl/bmi transition.
But this would not be a correct method to detect a timer's loop in all 
the cases (testing interrupt pending bit or using your own interrupt 
handler to set a "loop flag" would be more accurate).


> So the fact that with a real Atari the loop terminates early even when
> no devices are connected means that the other (non-timer) register
> values differ between Hatari and the TT/Falcon. They are checked in
> order to quit the loop early, depending on the SCSI bus signals.

Some scsi related regs are certainly the cause of your loop exiting 
early, because this loop alone that only reads $fffa23 is correctly 
emulated.

But I never owned an SCSI disk on Atari and did not really checked the 
datasheets for this, so I can't really help on this point.


Nicolas