[TUHS] DH11 flaw Was: pre-more pager?

[TUHS] DH11 flaw Was: pre-more pager?

[Noel Chiappa]

    From: Ron Natalie

    > We actually still had some real DEC DH's on our system.
    > ...
    > At least the DZ doesn't loop on the ready bit like the kernel printf

This reminds me of something I recall reading about John McNamara (designer of
the DH11) admitting that he'd screwed up a bit in the DH design; IIRC it was
that if you set the input silo alarm (interrupt) level to something greater
than 1 character, and someone types one character, and then nothing
else... you never get an input interrupt!

(Which is why some Unix DH driver which sets the silo alarm level > 1 - to get
more efficient processing by reducing the number of interrupts _if possible_ -
has to call a 'input characters ready from the DH' routine in the clock
interrupt code.)

IIRC McNamara said he should have included a timeout, so that if the silo
count was non-zero,and stayed that way for a while, it should have caused
a timeout, and an interrupt.

I was just looking for this, but couldn't find it. I thought it was here:

  http://woffordwitch.com/McNamaraDH11.asp

but it doesn't seem to be. Does anyone recall seeing this anywhere, and if so,
where? Web search engines didn't turn anything up, alas...

       Noel

[TUHS] DH11 flaw Was: pre-more pager?

[Clem Cole]

I've forgotten the details now, so I'll have look at the prints when I get
home, but think i remember that Ken had a strap on the DHDM that corrected
this.  But I do remember the discussion.   I have not idea where it was.
 I have a file of 'coms stuff' at home (paper of course).   I'll look to
see if it's in there.

Reading the McNamara interview, sadly he (like many folks) considered the
DZ to have modem control.   It had >>partial<< support, while the DH (with
the DM option) had full.   The difference is important.

Dave Cane (HW lead on the 750) used the DZ as his model for the original
MC-500 serial ports -- what a mess.   Here it is be had a real-time machine
and could not keep the clock straight for the first couple of versions of
the OS because the HW was dorked.  [Later systems did it properly].

Clem

On Wed, Nov 8, 2017 at 9:27 PM, Noel Chiappa <jnc at mercury.lcs.mit.edu>
wrote:

>     From: Ron Natalie
>
>     > We actually still had some real DEC DH's on our system.
>     > ...
>     > At least the DZ doesn't loop on the ready bit like the kernel printf
>
> This reminds me of something I recall reading about John McNamara
> (designer of
> the DH11) admitting that he'd screwed up a bit in the DH design; IIRC it
> was
> that if you set the input silo alarm (interrupt) level to something greater
> than 1 character, and someone types one character, and then nothing
> else... you never get an input interrupt!
>
> (Which is why some Unix DH driver which sets the silo alarm level > 1 - to
> get
> more efficient processing by reducing the number of interrupts _if
> possible_ -
> has to call a 'input characters ready from the DH' routine in the clock
> interrupt code.)
>
> IIRC McNamara said he should have included a timeout, so that if the silo
> count was non-zero,and stayed that way for a while, it should have caused
> a timeout, and an interrupt.
>
> I was just looking for this, but couldn't find it. I thought it was here:
>
>   http://woffordwitch.com/McNamaraDH11.asp
>
> but it doesn't seem to be. Does anyone recall seeing this anywhere, and if
> so,
> where? Web search engines didn't turn anything up, alas...
>
>        Noel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://minnie.tuhs.org/pipermail/tuhs/attachments/20171108/75403f86/attachment.html>