[TUHS] forgotten versions

Noel Chiappa jnc at mercury.lcs.mit.edu
Thu Jun 23 12:18:58 AEST 2022

    > From: Dan Cross

    > I believe that's actually a menu

Hence the "erroneous _impression_" (emphasis added).

I'm curious as to how they decided which models to run which editions on.
Although V4 _ran_ on the /45, split I+D wasn't supported - for user or kernel
- until V6. (I'm assuming a number of things - both in the kernel, and
applications - started hitting the 64KB limit, which led to its support.)

Speaking of split I+D, there's an interesting little mystery in V6 that at
one point in time I thought involved split I+D - but now that I look closely,
apparently not. The mystery involves a 'tombstone' in the V6 buf.h:

  #define B_RELOC 0200 /* no longer used */

I had created (in my mind) an explanation what this is all about - but now
that I look, it's probably all wrong!

My explanation involves the slightly odd layout of the kernel in physical
memory, with split I+D; data below the code, at physical 0. This actually
makes a lot of sense; it means the virtual address of any data (e.g. a
buffer) is the same as its physical address (needed for DMA). It does require
the oddness of 'sysfix', to invert the order of code+data in the system
binary, plus odd little quirks in the assembler startup (e.g. copying the
code up to make room for BSS).

So I thought that B_RELOC was a hangover from a time, at the start of split
I+D, when data _wasn't_ at physical 0, so a buffer's virtual and phsyical
addresses differed.

But that must be wrong (at least in any simple way). B_RELOC was in buf.h as
of V4 - the first kernel version in C - with no split I+D. So my theory has
to be wrong.

However, I am unable to find any code in the V4 kernel which uses it! So
unless someone who remembers the very early PDP-11 kernel can enlighten us,
its purpose will always remain a mystery!


More information about the TUHS mailing list