24 Feb
2006
24 Feb
'06
12:26 p.m.
Brad Parker wrote:
Don North wrote:
That is correct UNIBUS I/O is 18b only, even on all the VAXes that used
UNIBUS. There is no such thing as 22b UNIBUS DMA; sorry to give that
impression. I was just thinking out loud :-)
Besides UNIBUS proper (ie, the cable, or the in/out slots of a UNIBUS
backplane) systems also implemented SPC slots (UNIBUS + other stuff) but
these are 18b only as well.
There are also MUD (Modified UNIBUS Device) slots which are typically
used by non-DMA memory cards. These are only 18b as well.
For 22b memory support, the 11/44 (and 11/24 iirc) implemented four EUD
(Extended UNIBUS Device) slots that are specific for memory cards only
(ie, the MS11-PB 1MB memory card). These provide the extra 4b from the
UNIBUS map to allow a full 22b physcial memory address.
QBUS was originally 18b only as well, but was extended to Q22 to solve
this problem in a much more reasonable way (the UNIBUS map was/is a big
hack, but probably the only reasonable approach given the number of
preexisting 18b interfaces at the time of the 11/70s design).
If Guy really can implement a 22b examine/deposit physical memory in a
pure SPC slot kudos to him. The only way I can think of possibly doing
it (thinking out loud again :-) ) is to steal a UNIBUS map entry and on
each exam/deposit set it up correctly to map the 18b unibus address to
the desired 22b address. I don't recall if the UNIBUS map is accessible
The only problem with SPC slots is they are 18b
UNIBUS only. This will
work on 256KB 18b machines, but is a sticky problem on 4MB 22b machines,
since you need to go thru the UNIBUS map to get to all of memory.
For an 11/44, you could do just a dual slot card that goes in a memory
slot (which has 22b address available) and an SPC slot (to get the DMA
lines) but this would be a hex card. To read/write memory, you would
need to be a 22b DMA device, which I'm not sure the 11/44 can handle
(ie, the CPU may always drive the upper 4 address lines and not provide
for 22b DMA device). It may work, it may not. Certainly not what DEC
designed for.
You scared me there for a second. There are no 22 bit unibus dma
devices, right? only 18 address bits on unibus, right? even on an
11/44, right?
from the UNIBUS itself (if it is, this could work; if
not, then it won't).