Steve thank you for the recollections, that is precisely the sort of story I was hoping to hear regarding your Interdata work. I had found myself quite curious why it would have wound up on a shelf after the work involved, and that makes total sense. That's a shame too, it sounds like the 8/32 could've picked up quite some steam, especially beating the VAX to the punch as a UNIX platform. But hey, it's a good thing so much else precipitated from your work!
Also, those sorts of microarchitectural bugs keep me up at night. For all the good in RISC-V there are also now maaaaany fabs with more license than ever to pump out questionable ICs. Combine that with questionable boards with strange bus architectures, and gee our present time sure does present ripe opportunities to experiment with tackling those sorts of problems in software. Can't say I've had the pleasure but it would be nice to still be able to fix stuff with a wire wrap in the field...
- Matt G.
P.S. TUHS cc as promised, certainly relevant information re: Interdata 8/32 UNIX.
------- Original Message -------
On Friday, August 4th, 2023 at 6:17 PM, scj(a)yaccman.com <scj(a)yaccman.com> wrote:
> Sorry for the year's delay in responding... I wrote the compiler for the Interdata, and Dennis and I did much of the debugging. The Interdata had much easier addressing for storage: the IBM machine made you load a register, and then you had a limited offset from that register that you could use. I think IBM was 10 bits, maybe 12. But all of it way too small to run megabyte-sized programs. The Interdata allowed a larger memory offset and pretty well eliminated the offsets as a problem. I seem to recall some muttering from Dennis and Ken about the I/O structure, which was apparently somewhat strange but much less weird than the IBM.
>
> Also, IBM and Interdata were big-endian, and the PDP was little-endian. This gave Dennis and Ken some problems since it was easy to get the wrong endian, which blew gaskets when executed or copied into the file system. Eventually, we got the machine running, and it was quite nice: true 32-bit computing, it was reasonably fast, and once we got the low-level quirks out (including a famous run-in with the "you are not expected to understand this" code in the kernel, which, it turned out, was a prophecy that came true. On the whole, the project was so successful that we set up a high-level meeting with Interdata to demo and discuss cooperation. And then "the bug" hit. The machine would be running fine, and then Blam! it has lept into low memory and aborted with no hint as to what or where the fault was.
>
> We finally tracked down the problem. The Interdata was a microcode machine. And older Unix system calls would return -1 if they failed. In V7, we fixed this to return 0, but there was still a lot of user code that used the old convention. When the Interdata saw a request to load -1 it first noticed that the integer load was not on an address divisible by 4, and jumped to a location in the microcode and executed a couple of microinstructions. But then it also noticed that the address was out of range and entered the microcode again, overwriting the original address that caused the problem and freezing the machine with no indication of where the problem was. It took us only a day or two to see what the problem was, and it was hardware, and they would need to fix it. We had our meeting with Interdata, gave a pretty good sales pitch on Unix, and then said that the bug we had found was fatal and needed to be fixed or the deal was off. The bottom line, they didn't want to fix the bug in the hardware. They did come out with a Unix port several years later, but I was out of the loop for that one, and the Vax (with the UCB paging code) had become the machine of choice...
>
> ---
>
> On 2023-07-25 16:23, segaloco via COFF wrote:
>
>> So I've been studying the Interdata 32-bit machines a bit more closely lately and I'm wondering if someone who was there at the time has the scoop on what happened to them. The Wikipedia article gives some good info on their history but not really anything about, say, failed follow-ons that tanked their market, significant reasons for avoidance, or anything like that. I also find myself wondering why Bell didn't do anything with the Interdata work after springboarding further portability efforts while several other little streams, even those unreleased like the S/370 and 8086 ports seemed to stick around internally for longer. Were Interdata machines problematic in some sort of way, or was it merely fate, with more popular minis from DEC simply spacing them out of the market? Part of my interest too comes from what influence the legacy of Interdata may have had on Perkin-Elmer, as I've worked with Perkin-Elmer analytical equipment several times in the chemistry-side of my career and am curious if I was ever operating some vague descendent of Interdata designs in the embedded controllers in say one of my mass specs back when.
>>
>> - Matt G.
>>
>> P.S. Looking for more general history hence COFF, but towards a more UNIXy end, if there's any sort of missing scoop on the life and times of the Bell Interdata 8/32 port, for instance, whether it ever saw literally any production use in the System or was only ever on the machines being used for the portability work, I'm sure that could benefit from a CC to TUHS if that history winds up in this thread.
> Most of the time I'd rather not have to care whether the thing
> I'm printing is a string, or a pointer, or an integer, or whatever:
> I just want to see its value.
> Go has %v for exactly this. It's very nice for debugging.
Why so verbose? In Basic, PRINT required no formatting directives at all.
Doug
> From: Clem Cole
> first two subsystems for the 11 that ran out of text space were indeed
> vi and Pascal subsystems
Those were at Berkeley. We've established that S-I&D were in V6 when it was
released in May, 1975 - so my question is 'what was Bell doing in 1975 that
needed more than 64KB?'
The kernel, yeah, it could definitely use S-I&D on a larger system
(especially when you remember that stock V6 didn't play any tricks with
overlays, and also dedicated one segment - the correct term, used in the 1972
-11/45 processor manual - to the user structure, and one to the I/O page,
limiting the non-S-I&D kernel to 48KB). But what user commands?
It happens that I have a complete dump of one of the MIT systems, so I had a
look to see what _we_ were running S-I&D on. Here's the list from /bin (for
some reason that machine doesn't have a /usr/bin):
a68
a86
c86
emacs
lisp
ndd
send
teco
The lisp wasn't a serious use; I think the only thing we ever used it for was
'doctor'. So, two editors, a couple of language tools, an email tool (not
sure why that one got it - maybe for creating large outgoing messages). (The
ndd is probably to allow the biggest possible buffers.)
Nothing in /etc, and in /lib, just lint1 and lint2 (lint, AFAICT, post-dates
V6). Not a lot.
So now I'm really curious what Bell was using S-I&D for. (If I weren't lazy,
I'd pull the V6 distro - which is only available as RK images, and individual
files, alas - and look in /bin and everywhere and see if I can find anything.
I suspect not, though.)
Anyone have any guesses/suggestions? Maybe some custom applications?
Noel
Does anyone know if there are any surviving examples of SVR2 for the PDP-11? Various SVR2 manuals still make mention of the assembler, linker, etc. and the pdp11 variable is present in machid(1)*. And on the note of the later years of the PDP-11, was there any hope for SVR3 on the PDP? I presume the introduction of demand paging was the end of things but I would be curious for anyone's recollections on the final years of System V on the PDP-11.
- Matt G.
P.S. *interesting little 3B5 side note, found as I was checking references that machid(1) in the "System V" branded manual from the initial System V commercial release mentions the pdp11, vax, and u3b machines, the latter being the 3B20S. However, the "Release 5.0" branded manuals also make mention of the u3b5 machine, the 3B5. The System V manuals are from January 1983 and the Release 5.0 manuals are June 1982. There isn't an earlier reference to cite as machid(1) was introduced in Release 5.0, at least from the literature I have available. The 4.x series ran on at least the PDP-11, VAX, and 3B20S computers at least, matching those listed in the System V manual. From what I have available initial 3B5 literature was distributed in the form of small binders a little different from the grey-on-black 1984 DEC Processor SVR2 binders, possibly right on the cusp of the split of p_man from u_man as this 3B5 User's Manual that I have contains sections 1-6 rather than just 1 and 6. They're dark grey with a large orange square in the middle (I believe I've sent a photograph of the manual before).
As a longtime user and lover of ed/ex/vi, I don't know much about emacs,
but lately I've been using it more (as it seems like any self-respecting
lisper, has to at least have a passing acquaintance with it). I recently
went off and got MACLISP running in ITS. As part of that exploration, I
used EMACS, but not just any old emacs, emacs in it's first incarnation
as a set of TECO macros. To me, it just seemed like EMACS. I won't bore
you with the details - imagine lots of control and escape sequences,
many of which are the same today as then. This was late 70's stuff.
My question for the group is - when did emacs arrive in unix and was it
a full fledged text editor when it came or was it sitting on top of some
other subssystem in unix? Was TECO ever on unix?
Will
> From: Clem Cole
> A new hire in 1976, Jeff Mitchell supposedly had a bet with Bill
> Strecker that he could implement an 11 on a single"hex high" CPU board
> if he got rid of the lights and switches. He ran out of room to
> implement seperate I/D, so it became an 11/40 class [and it has an
> 8008-1 that runs the front panel].
I don't know about the Strecker story, but the first PDP-11 CPU on a single
card (a hex card) was the KD11-D:
https://gunkies.org/wiki/KD11-D_CPU
of the -11/04. It didn't have _any_ memory management, though (or a front
panel; to get that, you had to use the KY"11-LB:
https://gunkies.org/wiki/KY11-LB_Programmer%27s_Console
which added another quad card). The first -11 CPU i) on a single card and
ii) with memory management was the FDF11-A:
https://gunkies.org/wiki/KDF11-A_CPU
The first -11 CPU i) on a single card and ii) with split I+D memory
management was the KDJ11-A.
> It was not until 11/44 that DEC was able to make a hex height
> implementation of the 11 that managed to cram a full 11/70 into that
> system.
I'm not sure what your point is here? The KD11-Z CPU of the -11/44:
https://gunkies.org/wiki/KD11-Z_CPU
was a _minimum_ of five hex boards; a little smaller than a KB11-B (15 hex
cards). Floating point was an extra card; CIS was 2 more.
> if you look at the link line of sys/run the 45 does not have -i
Split I+D for the kernel was not supported by the linker in V6; a combination
of 'sysfix' (a special post-processor, which took as input a relocatable
linked image) and code in m45.s was needed.
https://gunkies.org/wiki/Upgrading_UNIX_Sixth_Editionhttps://gunkies.org/wiki/UNIX_V6_memory_layout
The code in m45.s to handle split I+D in the kernel:
https://minnie.tuhs.org/cgi-bin/utree.pl?file=V6/usr/sys/conf/m45.s
starts at 'start:' and is adequately commented to tell you what it's doing
when it plays around with kernel memory.
> From: Will Senn
> with I/D, you can use 64k for I and 64k for D. Was that it, or were
> there other tricks to get even more allocated
I have this vague memory that someone (Berkeley, I think?) added support for
automatic code overlays in user processes. The programmer had to decide which
modules went in which overlays, but after that it was all automagic. There
was a 4xx code allocated to them.
I think the support for that (in e.g. the linker) was somehow involved with
the use of overlays in the kernel, but I don't know the details (nothing
after V6 is at all interesting to me).
> didn't the 11 max out at 256k)?
You need to distinguish between i) the amount of memory the machine could
handle, and ii) the amount of memory that running code could 'see' at any
instant. The latter was always either 64KB, or 64KB+64KB (with split I+D
turned on, on CPUs which supported it).
The former, it's complicated. Mostly, on UNIBUS only machines, it was 256KB.
(Although there was the Able ENABLE:
https://gunkies.org/wiki/Able_ENABLE
which added an Extended UNIBUS, and could take them up to 4MB.) The -11/70,
as mentioned, had a Main Memory Bus, and could handle up to 4MB. The -11/44
had an Extended UNIBUS, and could also handle up to 4MB (but only after the
MS11-P became available; there were only 4 main memory slots in the
backplane, and MS11-M cards were only 256KB.) On QBUS achines, after the
KB11-A (Revision A), which only suppported 256 KB, all later revisions and
CPUs could also handle up to 4MB.
Noel
not quite split i and d but i do have a memory of some code which could run in three parts as overlays.
this could have been through exec’ing the next overlay with an appropriate argv and a pipe or file of data, or, perhaps there was some kernel support for overlays in early unix.
anyone seen evidence of this?
sadly i cannot remember where i saw it, i want to say it was a versatex printer driver but i am pretty sure that is rubbish.
-Steve
> From: Will Senn
> Does unix (v7) know about the PDP-11 45's split I/D space through
> configuration or is it convention and programmer's responsibility to
> know and manage what's actually available?
There are two different cases: i) support of split I+D in the kernel, and
ii) support of split I+D in user processes. Both arrived with V6; the
V5 source:
https://minnie.tuhs.org/cgi-bin/utree.pl?file=V5/usr/sys/conf/mch.shttps://minnie.tuhs.org/cgi-bin/utree.pl?file=V5/usr/sys/ken/main.c
(former for kernel; later for users) shows no sign of it.
> From: Kenneth Goodwin <kennethgoodwin56(a)gmail.com>
> 1. I don't think the 11/45 had split I & d.
> But I could be wrong.
> That did not appear until the 11/70
You are wrong.
The chief differences between the KB11-A&-D of the -11/45 and the -B&-C of
the -11/70 were i) the latter had a cache, and ii) the latter used the 32-bit
wide Main Memory Bus, which also allowed up to 4 Mbytes of main memory.
Detail here:
https://gunkies.org/wiki/PDP-11/70
along with a couple of lesser differences.
> From: "Ronald Natalie"
> with only 8 segment registers combined for code, data, and stack
I think you meant for code, data, and user block.
> The 55 (just a tweaked 45)
The /50 and /55 had the identical KB11-A&-D of the /45; the difference was
that they came pre-configured with Fastbus memory.
> In addition the 23/24/J-11 and those derived processors did.
No; the F-11 processors did not support I&D, the J-11 did.
Noel
> And I think even V7 make supported what you described, as well as implicit rules for compiling .c into a .o or into a binary.
>
> Warner Losh
You're right, I just tried it out. Been avoiding that pattern for years because I swear some make implementation I used at one point was very unhappy with that, but if V7 does it, then whatever implementation that was is probably not what I want to be using anyway.
Also shows how little I've used specifics of BSD, I've never made a Makefile using bsd.prog.mk, although I have this desire for a write-once-build-everywhere Makefile that the preponderance of build systems that generate them imply is an exercise in futility...
On that note, one quirk I've found with the implicit .c.o rule on older UNIX, just tested on V7 and System III, is that they render:
cc -c $<
rather than
cc -c -o $@ $<
If you have an object list with objects in several different directories, it spits them all out in the CWD, causing problems if you have similarly named files in multiple directories, and then outright failing on the final compilation if it's something like $(CC) -o $(BIN) $(OBJS) because $(OBJS) is a string of object pathnames with the full nested path, not just the resulting *.o files.
Granted, I could be anti-patterning here for all I know, I haven't worked closely with a whole lot of Make-based projects that aren't my own. Maybe I just haven't read these darn papers I'm always hunting down enough.
- Matt G.