PDP8 / ETOS

[Johnny Billquist]

On 2015-10-19 03:04, Rick Murphy wrote:
> At 08:37 AM 10/18/2015, Johnny Billquist wrote:
>
>> I decided to look into this a couple of days ago, since the basic
>> workings of FRTS includes using interrupts, which is not possible if
>> running under time sharing. That would also imply that it would not be
>> possible to use F4 under RTS-8, which I had some memory of that it
>> actually is possible.
>>
>> To sum things up: The FPP-8 can usually not be used when in
>> timesharing. First of all, you use IOTs to control the FPP, and all
>> IOTs are trapped when in user mode. Second, as you note, the FPP-8
>> uses 15-bit addresses, which would make it impossible to use with
>> virtual memory.
>>
>> Also, the interrupt system is not available when in user mode.
>>
>> However, FRTS actually have code to detect if it is running under
>> RTS-8, and do not use the interrupt system in that case, but adopts.
>> Also, since the FPP-8 IOTs are caught by RTS-8, and do not do
>> anything, FRTS actually believes you are on a system without an FPP-8,
>> even if you actually have one. So, FRTS will always use the FPP-8
>> emulator in this situation.
>>
>> MULTOS-8 hook in to the same functionality, making FRTS believe it is
>> running under RTS-8, which means that F4 programs will actually work
>> under MULTOS-8, including ADVENT. I don't know if ETOS also implements
>> the bits needed to make it look like OS/8 programs are actually under
>> RTS-8. If it does, ADVENT should be possible to run under ETOS as
>> well. Otherwise not.
>
> Sounds like it's possible, if MULTOS-8 can give a user a full 32k to
> work with.

Yes, it does. Not that this really is the important point I was making, 
but anyway...
The point I was making was that FRTS detects if it is running in the 
background of RTS-8, and patches itself to not use the interrupt system 
in this case. And MULTOS-8 made sure to cause FRTS to think that it is 
running in the background of RTS-8, so that the patches are applied, 
thus making it also able to run under MULTOS-8.

Memory requirements and availability is a different story, but yes, 
ADVENT will require the full 32K, as far as I remember.

>> Finally, as I noted, the FPP-8 is not that possible to use in user
>> mode, both because of the IOT instructions being caught, and the
>> 15-bit addressing. However, the FPP-8A have a mode where it only
>> allows memory accesses within the same field, and will trap out if any
>> memory reference goes to another field. With some work, and code, I
>> think it could be possible to actually have access to the FPP-8 from
>> user mode, if you have an FPP-8A, but I have never tried this, as I
>> lack the hardware.
>
> That's not going to help anything using Fortran-IV, of course. The
> library and user code assumes that "virtual" addresses are the same as
> physical ones. And IOTs aren't "not possible" - they just need to be
> trapped and emulated by the OS.

No, you are wrong.
Yes, the IOTs just "needs to be implemented". That is the easy part. The 
memory mapping is not. However, the mode in the FPP-8A that restricts 
the FPP code to only allow access within the current field will then 
trap out if the code tries to access anything outside, and the addresses 
are still 15 bits. This means that the OS can then handle this as a page 
fault (field fault?), and swap in the correct field, and then resume 
FPP-8 execution. However, like I said, I have not tried actually 
implementing anything like this, since I don't even have the hardware. 
And I suspect it might be a headache since I guess addresses are still 
absolute, meaning and field relocations done by the OS needs to somehow 
be done for the FPP-8 code as well. Not sure how you'd do that, but I 
think people with enough creativity could maybe come up with some 
solution. I would have to re-read the documentation to see how well I 
remember this stuff...

> It wouldn't be all that hard to add support for a FPP to a timesharing
> system as a UUO that knows how to map the user's field to something the
> FPP could use, then locking that field in core, as the FPP will continue
> to DMA from it while executing FPP instructions.

That would prevent more than one user using the FPP-8, though. Doable, 
but not really that useful if you want a timesharing system for more 
than one user.

> To support even this simple FPP mode would require the underlying OS to
> handle the housekeeping, as the FPP startup needs to know the physical
> field of the active parameter table, which a timeshared user wouldn't know.

Not to mention you'd have to figure out how to manage if the FPP wants 
to run in field 0, which the OS needs for its own use...

> Seems like a lot of work given the number of FPPs in the wild.

Agreed.

> Also, not just FPP8-A has that feature. The "same field as the APT" bit
> is also implemented by the FPP-12. This makes me wonder what application
> there was for this.

I didn't know the FPP-12 had that. Interesting. Yes, I wonder if 
anything ever used it.

> Of course, you don't need the hardware to experiment with this. The SIMH
> FPP-8A implementation is as compatible to the hardware as I could make
> it - everything but the maintenance mode IOTs are implemented.

Hmm. Good point. Last I looked at this, simh hardly even existed. :-)

	Johnny

-- 
Johnny Billquist                  || "I'm on a bus
                                   ||  on a psychedelic trip
email: bqt at softjar.se             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol