First: this is indeed a real bug in the AGC software. However, it did not go unnoticed for the whole program. It was discovered during level 3 testing of SATANCHE, and late development branch of the Command Module software COMANCHE. It was assigned anomaly number L-1D-02, and was fixed between Apollo 14 and 15. There are two known surviving copies of the L-1D-02 anomaly report:

* https://www.ibiblio.org/apollo/Documents/contents_of_luminar...

* https://www.ibiblio.org/apollo/Documents/contents_of_luminar...

The fix described in the article is partially complete, but as noted in the anomaly report there's a little bit more to it. Rather than just adding the two instructions to zero LGYRO, they restructured the code a bit and also cause it to wake up pending jobs. You can compare the relevant sections of the Apollo 14 and Apollo 15 LM software here:

* Apollo 14: https://github.com/virtualagc/virtualagc/blob/master/Luminar...

* Apollo 15: https://github.com/virtualagc/virtualagc/blob/master/Luminar...

The bug would not manifest silently in the way described in the article. For starters, LGYRO is also zeroed in STARTSB2, which is executed via GOPROG2 on any major program change: https://github.com/virtualagc/virtualagc/blob/master/Luminar...

This means that changing from any program to any other program would immediately resolve the issue. This is almost certainly a large part of why it took them so long to notice. Hitting BADEND while actively pulse-torquing is quite rare, and avoided by normal procedure. The scenario presented in the article can't happen since the act of starting P52 will zero LGYRO.

Moreover, in the very specific scenarios in which the bug can be triggered and remain, it results in multiple jobs stacking up attempting to torque the gyros. Eventually the computer runs out of space for new jobs -- similar to what happened on 11 -- and a 31202 (the Apollo 12+ equivalent of 1202) is triggered.

Since the issue was found before the flight of Apollo 14, a further description of how it might occur and what the recovery procedure should be was added to the Apollo 14 Program Notes: https://www.ibiblio.org/apollo/Documents/LUM159_text.pdf#pag...

Some other notes:

> Ken Shirriff has analysed it down to individual gates

I've done the bulk of the gate-level analysis. :)

> the Virtual AGC project runs the software in emulation, having confirmed the recovered source byte-for-byte against the original core rope dumps.

We've only been able to do that in very specific circumstances and only for subsections of assorted programs, but never for a full program. Most AGC software either comes from a program listing, from a core rope dump, or from reconstruction using changelogs and known memory bank checksums. We've disassembled all of the rope dumps into source files that assemble back into the same binary, but the comments and labels will be different from what was in the original listing. And to be extra clear: I've never had the opportunity to dump a module containing Apollo 11 software for either vehicle. Our sole source for both programs is a pair of printouts in the MIT Museum's collection.

> Margaret Hamilton (as “rope mother” for LUMINARY) approved the final flight programs before they were woven into core rope memory.

Jim Kernan was the rope mother for Luminary at least up through Apollo 11. Margaret was the rope mother for Comanche, the CM software, and was later promoted to lead the software division. Their positions at the time of 11 can be seen on this org chart: https://www.ibiblio.org/apollo/Documents/ApolloOrg-1969-02.p...

> Their priority scheduling saved the Apollo 11 landing when the 1202 alarms fired during descent, shedding low-priority tasks under load exactly as designed.

This is a huge topic on its own, but the AGC software was not designed to shed low-priority jobs. Ironically, the lowest priority job during the landing was the landing guidance itself, with high-priority jobs being reserved for things that needed quick response like antenna movements or display updates. If the computer were to shed the lowest-priority jobs, it would shed the landing guidance. This memo contains a list of all jobs active during the landing and their priorities: https://www.ibiblio.org/apollo/Documents/CherryApollo11Exege...

> For example, the ICD for the rendezvous radar specified that two 800 Hz power supplies would be frequency-locked but said nothing about phase synchronisation. The resulting phase drift made the antenna appear to dither, generating roughly 6,400 spurious interrupts per second per angle and consuming roughly 13% of the computer’s capacity during Apollo 11’s descent. This was the underlying cause of the 1202 alarms.

The frequency-lock prevents phase drift, so the phase is essentially fixed once the power supplies are up. Ironically, however, the bigger issue is that one reference was 28V while the other was 15V. Initial testing on actual Apollo hardware suggests that at least for Apollo 11, this voltage difference was the key contributor rather than the phase difference: https://www.youtube.com/watch?v=dT33c70EIYk

It's a lot cheaper and more accessible in this form, and much easier to integrate into projects that need AGC stand-ins.

Aside from the FPGA design though, yeah, all of the AGC software and the assembler is in the VirtualAGC repository.

Comments URL: https://news.ycombinator.com/item?id=29462190

Points: 1

# Comments: 0

This is a work-in-progress disassembly of the core rope modules we dumped at the MIT Museum, so apologies for the less friendly formatting!

[1] https://github.com/virtualagc/virtualagc/blob/master/Luminar...

[2] http://www.ibiblio.org/apollo/Documents/LUM156_text.pdf -- look for PCN 1037

[1] https://github.com/virtualagc/virtualagc

https://archive.org/details/Comanche55J2k60

https://archive.org/details/Luminary99001J2k60

He's been an invaluable resource for us at the VirtualAGC project [1]. He provided over half (!) of the AGC programs we've made available, including LM software for Apollo 5, 9, 10, 11, 12, 13, and 15-17, plus several other ground development programs. He also was the source of a majority of the schematics that we used in the AGC restoration [2] -- I can say with 100% confidence that it wouldn't have been possible to get that computer working again without his help.

He's got a book out now, Sunburst and Luminary [3], that is very good and goes into a lot of technical detail on the LM software, without getting too much into the weeds. I highly recommend it if you're interested in that sort of thing!

[1] http://www.ibiblio.org/apollo/

[2] https://www.youtube.com/watch?v=2KSahAoOLdU&list=PL-_93BVApb...

[3] https://www.sunburstandluminary.com/SLhome.html

CDU theory of operation (starting PDF page 15): http://www.ibiblio.org/apollo/Documents/HSI-208435-003.pdf

CDU coarse module schematic: https://archive.org/stream/apertureCardBox462NARASW_images#p...

Grumman memo (from 1968!) describing the problem, and mentioning it is due to the reference switching to a 15V 800Hz source: https://www.ibiblio.org/apollo/Documents/Memo-GAEC_LMO_541_1...

Excerpt from the LM-8 Systems Handbook showing the reference voltage RR switch wiring: https://i.imgur.com/fMsQ7RI.png

Don Eyles describes the software side best in his book Sunburst and Luminary (which I highly recommend) but he also talks about it in some detail on his website: https://doneyles.com/LM/Tales.html

It wasn't a switch misconfiguration; the Apollo 11 astronauts were flying to the checklist, and did as they had simulated. The Rendezvous Radar switch has three settings -- LGC, AUTO TRACK, and SLEW. In LGC mode, the AGC controls the positioning of the antenna; in AUTO TRACK, the radar automatically tracks the CSM based on return strength; and in SLEW it is automatically positioned.

The trouble came from how the trunnion and shaft angles of the antenna were measured. They used "resolvers", which are sort of like variable transfomers. Resolvers look like motors, and attached to the shaft there are two windings positioned 90 degrees apart from each other. An AC "reference voltage" is applied to an outer winding in the case, and that voltage couples onto the two inner windings with a magnitude proportional to the angle on the shaft. One winding (the "sine" winding) produces an output equal to Vrefsin(theta) and the other ("cosine") winding produces an output equal to Vrefcos(theta), where Vref is the reference voltage and theta is the angle of the shaft. The voltage and phase of both windings can be used to determine exactly what the theta was that produced them.

The circuitry to do this is a bit involved though and lived outside the computer, in a device called the CDU, or Coupling Data Unit. The CDU constantly maintained its own idea of what the angle ("psi") in a digital register. It translated the incoming sine and cosine voltages into a digital representation by mechanizing the equation +-sin(theta-psi) = +-sin(theta)cos(psi) -+ cos(theta)sin(psi). It did so by using the bits of its digital register containing psi to switch on and off resistor dividers that effected cos(psi) and sin(psi) onto the incoming signals, which were then added together with a summing amplifier. The goal of the CDU is to zero this sum; to accomplish this, it "counts" the angle register up or down to reduce the magnitude of the sum. As it counts, switches are changed, which switch out resistors in the circuit, which in turn change cos(psi) and sin(psi) in the above equation. And also, with every other increment, a pulse is transmitted to the AGC to indicate that the angle has changed slightly.

The problem comes in because in addition to the above, the CDU also, for many angles, added to the sum some fraction of the reference voltage directly. This is fine when the switch is in the LGC position; the resolvers are supplied with the same 28V, 800Hz reference voltage that is used inside the CDU. However, when the switch is put in either of the other two positions, the reference voltage for the RR resolvers is switched to an unrelated 15V rail. Critically, this 15V reference has no defined phase relationship with the CDU's 28V 800Hz reference. The phasing is locked in by the exact millisecond at which you power up your subsystems.

So when the switch is changed, the sine and cosine outputs from the resolver are suddenly derived from the 15V reference -- they are much lower before and at a random phase. The CDU doesn't know that this has happened, and still tries to perform the summing as before. However, for many theta/phase relationships, it becomes impossible for the CDU to actually null the sum. In these cases, the CDU becomes "manic", and starts seeking back and forth, frantically changing switches to try to figure out what the angle is, but never succeeding.

This causes a huge flurry of +1 and -1 pulses to the AGC. In order to minimize circuitry, the AGC implemented what was called "unprogrammed" or "cycle-stealing" instructions. The computer only contains a single adder, and adding or subtracting 1 from the current angle requires use of that adder and a memory cycle. Rather than generating a full interrupt, which would require many memory cycles and instructions to handle, the computer simply transparently inserts a single-cycle instruction in between two "programmed" instructions that performs the addition or subtraction. This is totally transparent to software, normally. But with a manic CDU that is incessantly seeking on both RR angles, the AGC receives something close to 12,800 pulses per second, which translates into something around 15% of its total computational time. The landing software had only been designed with a margin of 10% or so.

The 1202s were also a lot less benign than is often reported. They occurred because of the fixed two-second guidance cycle in the landing software. That is, once every two seconds, a job called the SERVICER would start. SERVICER had many tasks during the landing. In order: navigation, guidance, commanding throttle, commanding attitude, and updating displays. With an excessive load as caused by the CDU, new SERVICERs were starting before old ones could finish. Eventually there would be two many old SERVICERs hanging around, and when the time came to start a new one, there would be no slots for new jobs available. When this happened, the EXECUTIVE (job scheduler) would issue a 1201 or 1202 alarm and cause a soft restart of the computer. Every job and task was flushed, and the computer started up fresh, resuming from its last checkpoint. It was essentially a full-on crash and restart, rather than a graceful cancellation of a few jobs. And unlike is often said, the computer wasn't dropping low-priority things; it was failing to complete the most critical job of the landing, the SERVICER.

Luckily, the load was light enough that of the SERVICER's duties, the old SERVICER was usually in the final display updating code when it got preempted by a new SERVICER. This caused times in the descent when the display stopped updating entirely, but the flight proceeded mostly as usual. However, with slightly more load, it was fully possible that the SERVICER could have been preempted in the attitude control portion of the code, or worse yet, the throttle control portion. Since each SERVICER shared the same memory location as the last one (since there was only ever supposed to be one running at a time), this could lead to violent attitude or throttle excursions, which would have certainly called for an abort. Luckily, this didn't happen -- and the flight controllers didn't abort the mission not because 1202s were always safe, but because they didn't understand just how bad it could be, were the load just a tiny bit higher.

I think the Computer History Museum's might have Sprague caps in it, though.

Comments URL: https://news.ycombinator.com/item?id=18447910

Points: 3

# Comments: 0