The System Call

Howarth's sentence is narrower than it is usually quoted. He writes that entering by extracode “is the only means by which an object program can use extracode control” — and two sentences earlier he writes that “Object programs call the supervisor into action either by causing an interruption and initiating a short interrupt routine, or by initiating an S.E.R.” Faults, overflow, a page that is not there and a peripheral that wants attention all reach the supervisor without any extracode at all. What the 512 entry points bound is the set of places a program may deliberately enter.

Ferranti's extracode manual calls the most significant function digit f0 and numbers a register's bits from the top. The Atlas design description calls the same digit f9 and numbers from the bottom, 23 down to 0. This page builds the extracode control register twice, once from each document's own printed table, and compares all twenty-four bits for every one of the 512 codes. Nothing here would have caught a shared mistake; what it catches is a transcription error, which is the mistake actually available to make.

Both 1962-era papers say so. Howarth: “a possible maximum of 512”. The 1961 paper, describing the machine that existed: “The fixed store contains about 250 subroutines which can be called in from an object program by single instructions called extracodes”. The Computer Conservation Society's published list for its emulator carries 272 distinct codes, which is a later and fuller accounting. This page prints 272 and says whose count it is; it does not reconcile the two. A later list being fuller is a reason for the gap and not a measurement of it, and nothing here separates how much of it is extracodes added over the machine’s life from how much is a difference in what the two counts were counting.

It says main control, extracode control and interrupt control, and each has its own program counter, so there are three and not two. It says object program for what we would call a process, and private store for the region a program may not touch. Those are the words used here. The modern pair is a fair translation and it loses the third control, which is the most privileged of the three.

The 320 and the 192 are derived here from the allocation table, and the design description states the same two numbers independently. The 272 and its breakdown by jump table are transcribed from the Computer Conservation Society's own list and are not derived from anything: what can be checked here is that they are consistent with the A and B split, which they are, and the test parses the archived table and derives them again rather than trusting the transcription.

This page decodes instructions and computes where an extracode would land. It does not simulate the fixed store, the one-level store, the drum, or any extracode's actual work. There is a real Atlas emulator and it is not this: the Computer Conservation Society's runs real Atlas jobs and enters most extracodes by the authentic method. This is an explanation of one bit, checked against the documents.

The Burroughs B5000's operator for entering its control program is documented in 1961, and Stretch's mode changes are earlier still; the Ferranti Orion had a similar mechanism. What Atlas is, is early and unusually well documented, and its documents are open. That is the reason it is the machine on this page rather than a claim about who was first.

An earlier count of how many extracodes the design description lists was reported as a range, 249 to 261, because a line-anchored search over the extracted text was defeated by page breaks in it. Re-parsed, the figure is 259. The range was not a scan problem and it was not honest uncertainty: it was a parse that could not see the whole document reporting the width of its own blind spot as a property of the source. This page does not print that figure at all, and the reason it does not is here rather than nowhere.

Howarth's The Atlas Supervisor Program is July 1962 and is his alone. The Kilburn, Payne and Howarth paper it cites, The Atlas Supervisor, is the Eastern Joint Computer Conference of December 1961, and both the archive's own page and its usual citations date it 1962, which is the proceedings volume's imprint rather than the conference. Howarth's own reference list calls it F.J.C.C., which is a different conference again.

Asked and answered. There is nothing here with a duration except two instruction counts, and both are printed. A click per mode switch would be a decoration of a number already on the screen.