Talk:KS11 Memory Protection and Relocation option

Ken Thompson's memory of it

Ken was nice enough to respond to a query asking for help from old-timers; here is his brief reply:

i did all the programming for it.

it was based on the pdp-10 memory map. it had 2 pages growing toward each other and some horse crap to let the har[d]ware multiply unit shine through.

sorry, i dont remember too many details.

which, alas, does not really add much to what little we know. Jnc (talk) 04:16, 17 May 2023 (CEST)

And from Bob Supnik

Bob's reply to my query:

The KS11 created a PDP-10 (KA) style memory management system for the 11/20: that is, high-segment (instructions, in theory, and shareable, in theory) and low-segment (data). It did create a user vs monitor mode. It [did] support memory expansion, although that may have required a separate option, the MX11. It did not, IIRC, do IO mapping (IO scatter/gather). Because the high and low segments were physically contiguous, scatter/gather wasn't needed, only a "simple" relocation and bounds check on DMA transfers.

The high/low segmentation wasn't automatic; that is, it didn't function like I/D space in official memory management. Rather, the software writer had to organize a program into two .psects - a "pure" instruction segment and an "impure" data segment. The stack was always impure; the 11/20 didn't have the MARK instruction to screw that up.

So still some important open questions, but this adds a bit more. Jnc (talk) 04:41, 17 May 2023 (CEST)

And my axe

Clem Cole wrote me:

One of my old friends from DEC that worked in CSS, validated our belief that KS11 is supposed to have been modeled after the KA10 MMU.

and

we only have some information from UNIX2 and 3 when run on an 11/20 with a KS11 - what we don't know is how much physical memory was installed. We know it worked as a bus repeater, so it's possible it could have supported 18 bits of Unibus.