Re: Q:040/060 MMU and Snooping

From: Michael White <michael_at_fastlane.net>
Date: Wed, 10 Mar 1999 19:45:13 -0600 (CST)

>> If the Afterburner Tool Kit that Doug assembled (and the
>> apps that Overscan provided with the A/B) enable the MMU
>> 'after' the machine has booted, does this mean the MMU is
>> disabled (MDIS LOW) by the Afterburner during reset and
>> system initialization, then initializes it by software
>> during the boot sequence (ala the autofolder)?
>>
>> According to the users manual;
>> "MMU Disable disables the translation mechanism of the
>> MMU's"
>>
>> If the MMU is disabled during initialization, does this mean
>> that if TOS sends 030 MMU instructions to the 040/060, that
>> they will be ignored and no error will occur?
>
>I asserted MDIS continually last night (tonight/this morning:).
>This made no difference with the 060 on the diagnostic cart, it booted
>as usual. The 040 however, would not come up under TOS, only the
>diagnostic cart. An error I am not sure I mentioned before was a
>diagnostic error T4. The combel is not generating an address for the
>screen display (with the 060 installed).
>So I guess that the MDIS only affects the translation mechanism, but not
>the MMU "tree".

Correct. The MMU disable line only affects the MMU co-processor
and has no effect whatsoever on the software.

Note, though, that there are minor differences between the 040
and 060 MMU, but not as large as between the 030 and 040. Also,
the system should probably be able to init without the MMU on.
As far as I can see, the only reason the MMU is used on the
Afterburner is to allow patching of TOS. Other than that, the
tables are pretty much pass through.

Which brings be to a question: the Afterburner MMU tables are cyclic,
in that, after a while, they basically repeat the first 16 MBytes
of memory. Why is this? Also, when attempting use the data transparent
translation register to access memory way outside any valid range
(say, 0x80100000, with the local address base in the TT register
set to 0x80), my Falcon invariably locks up. I expected a bus error.
What gives?

Thanks.

Michael White (michael_at_fastlane.net)
Received on fr. mars 12 1999 - 17:12:22 CET