On 2023-03-07 18:53, Dave McGuire wrote:
On 3/7/23 12:23, Johnny Billquist wrote:The P0/P1/S0/S1 space is in virtual space.The 512M is physical memory. Physical address space is a total of just 1G, with half dedicated to memory, and half to I/O space.So 512M is the absolute maximum the machines can have. And there is some problem with this in NetBSD.This must be model-specific, somehow?
Not really.The 1GB limit is because of the layout of the registers for the memory mapping. It's not possible to map more than 1GB of physical memory. PTE have 21 bits for address, and each page is 512 bytes, so that's 9 bits. Which results in a 30 bit physical address.
This got an update on the NVAX, where a different format PTE was defined, in which you have 25 bits for the address. So theoretical max physical memory for that new scheme was 16G, but NVAX only implemented 32 bits of physical address space (if the new PTE is used).
As for the division between memory and I/O space, I guess it could have been done differently on different implementations, but the complexity it would have implied in software if that had been done was probably not something DEC wanted. So in all machines I've ever looked at, the I/O space starts at 512M. Physical memory is always below that.
One of the machines I know best, which is the 86x0, there is literally 29 address pins on the memory bus, so in theory you could populate that machine with 512M, but because of what memory cards actually was ever made, the practical limit on that machine was 260M. But that already implies that all address pins are being used.
But what is definitely model specific (or at least partially so), is what exists where in I/O space.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt%softjar.se@localhost || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol