[Simh] PDP-8 KL8E XON/XOFF

[Rick Murphy]

At 08:54 PM 6/5/2016, William Wuttke wrote:
>Problem:
>
>When using a teletype on a serial port, XOFF 
>will stop a directory listing (and others), but 
>once that happens, the PDP-8/simh will no longer 
>respond to keyboard input. I’m using a Model 
>33 ASR on a PiDP-8 (raspberry pi) hardware serial port.
>
>Investigation:
>
>All documentation says XON and XOFF should work 
>from the console. Well, it does and it doesn't.

Any documentation that says that about OS/8 is inaccurate.

Many programs simply push character output as 
fast as the native baud rate and never checks for 
any keyboard input (other than perhaps for control-c.)



>If using "screen", it works, but apparently the 
>XON/XOFF functionality lies somewhere other than in the KL8E handler.

That's probably happening within screen, where 
OS/8 never sees the input, with buffering being handled by the native OS.

>  The typing stops when XOFF is sent. Then when 
> XON is sent, you can see some buffered text 
> appear on the screen fast, then the typing 
> slows down to normal speed. (I set the tto unit 
> wait value to 20000 to closely approximate real 
> 110 baud serial output. Otherwise, a directory 
> listing gets buffered up an about 3 seconds, 
> but takes 2-3 minutes or longer to print out on 
> the teletype. This also allows me to Control-C 
> to quit the listing.) It looks like "screen" is buffering the text.

Assuming you're using something like "TYPE 
FILE.TX", which CCL converts to PIP TTY:<FILE.TX. 
So, the KL8E handler gets used here.

>If using the teletype, XOFF stops the typing, 
>but XON does nothing - the machine is locked up 
>in a loop at this point with 7623 in the 
>accumulator (it works as coded). Looking at the 
>KL8E handler, the tty check routine uses KRS to 
>read a character, never resetting the keyboard 
>flag. Apparently, since the KL8E had no 
>buffering, if an incoming character is not read, 
>it would be overwritten by the next incoming 
>character, clearing the keyboard flag during the 
>time that the character was not valid, then 
>setting it after the character was assembled 
>into the keyboard buffer. It looks like the KL8E 
>handler was written depending on this behavior. 
>However, it looks like the simh tti is acting 
>like a modern UART, buffering up the input, and 
>moving through the buffer as characters are 
>read. The KRS instruction is for all purposes a 
>"peek" at the next character in the buffer, 
>which is exactly what the tti is doing. It keeps 
>"peeking" at XOFF (7623 - 7600), which never 
>really gets removed from the buffer. However, it 
>really needs a "read", to advance to the next 
>character. At least, that is what it looks like.

This sounds like a SIMH defect. KRS should read 
the last input character, as there's no buffering on the real hardware.

>Fixes?
>
>I'm going to attempt to mod the KL8E handler to 
>fix this, but might not have much luck since it 
>already pretty much fills up its allotted two 
>pages. It is filled with all sorts of assembly 
>language trickery to fit into two pages. In my 
>way of thinking, the simh KL8E is really 
>“modified” hardware, and OS/8 can be easily 
>adapted to new hardware. I think I have a fix, 
>using KRB instead of KRS - I don’t think that 
>using KRS was a good idea anyway, but on a real 
>PDP-8 it would catch the XON as soon as it happened.
>
>Or, could a mod to simh do the trick? Could the 
>KRS be modified to act like a KRB, but just 
>OR’ing the keyboard buffer with the 
>accumulator instead of clearing the accumulator 
>it first? How would this affect other PDP-8 
>software that relies on KRS without subsequently 
>clearing the keyboard flag? How common is the 
>use of KRS in this way? I guess in this day and 
>age, trying to eliminate all the buffering going 
>on would probably be a real pain. And, the 
>buffering on the input actually eliminates the 
>need use the teletype Reader Control with simh.
>
>Does anyone have any ideas?

If the SIMH KRS doesn't read the last input 
character, it needs fixing. Changes to the KL8E 
handler will only fix TYPE and DIR, and leave 
lots of other problems. Seems like a simple SIMH 
fix to drain the input buffer every time KRS is executed will fix this.
         -Rick