9 Jun
2014
9 Jun
'14
3:49 p.m.
At 21:08 Uhr +0100 7.6.2014, Tony Duell wrote:
Again siad books sais 'Based on TI's Nubus' (or somehtign simialr) which
implies to me it is not pure Nubus.
Well, what is "pure Nubus" here? The MIT implementation? Texas Instruments'
1983 specification? IEEE 1196-1987? [3]
Mac Nubus is not
stnadard Nubus (UI have read the specs for both). and so on.
Really? What did Apple do differently? It was close enough that the TI
Nubus interface chips worked fine. The book, alas doesn';t give the
differences other htan the trivial ones (limitation on the size of the
PCB, no -5.2V supply). And yes, it is certainyl close enough to the TI
spec that hte standard ICs will work.
I thought, but can't find the details, I read somewhere tht there was one
fairly obscure thing that Apple did differently and which meant it wasn;t
strictly Nubus. If I come acorss the detials I'll post them.
Assuming you are talking about Apple's "Designing Cards and Drivers" (3rd
ed., 1992) aka DC&D, while they don't summarize in one place Nubus features
the Macintosh does not support, they do talk about them.
(1) Cache coherency is supported, but does not use the Nubus method (DC&D,
ch. 3)
(2) Interrupts - Nubus offers interrupts as write transactions on shared
memory, a shared interrupt line for all slots, or a per-slot interrupt
line. The Macintosh only implements the latter, using a 6522 VIA as
interrupt controller (DC&D ch. 3). The shared memory pattern for
communicating interrupts in multi-processor systems re-appeared with PCI-X
and PCI-E [1].
(3) Block transfers - while later (68040 based) CPUs allow block transfers
between Nubus masters, the mainboard does not support them.
Generally spoken, the Nubus is targeted towards a multi-master system, as
opposed to contemporary buses like ISA, ECB or S100 which are glorified
extensions of CPU buses. Apple tried to go that way with its A/ROSE based
intelligent extension cards, and there were master cards from i?86 to LISP
machines to I/O processors to the Radius Rocket, but the main Nubus
appplication were graphics cards, which did not really fit the Nubus model,
and soon required PDS slots for adequate performance.
So, both the "PC" idea and Apple's view of its Macintosh operating system
[2] were at odds with the underlying concept of Nubus, intelligent
subsystems. Using Nubus for a serial card was just too expensive, compared
to an ISA card.
hauke
[1] <http://mail-index.NetBSD.org/tech-kern/2014/05/30/msg017180.html>
[2] Macintosh New Technical Notes, HW 16 - NuBus Interrupt Latency
(I Was a Teenage DMA Junkie)
[3] Apple DC&D refers to ANSI IEEE 1196-1990 ("Nubus90"), which neither a
web search nor IEEExplore search have surfaced. I assume once Apple decided
to move towards PCI, the update was shelved. Does anyone here have (ever
seen) a 1196-1990 draft?
--
"It's never straight up and down" (DEVO)