On Thu, 8 Jul 1999, Jacek Lipkowski wrote:
> you're right. the 8086 has 20 address lines (a0-a19) - so it could access
> 1Mb of memory. the 80286 has 24 of them (a0-a23) - so it can access 16Mb.
> the 'a20 gate' is really an AND gate, with one of the
> inputs connected to the processor a20 gate, and the other connected to one
> of the unused bits in the keyboard status register (or something similar)
> in order to emulate the xt 'wraparound' behavior.
Yes, the keyboard hardware seems to pop up all over the place in these
computers' designs. (For instance it provides a slow way to switch back
to real mode from protected mode. Triple-faulting is faster.)
> btw, you could also get 'memory expansion boards' (or something with a
> similar name). these allowed someone to have >640k in their XT workstation
> :). this is probably the reason people think, that the xt had an a20 gate.
> actually the memory present on the card was accessed using bank-switching
> (the visible bank was put somewhere in a000:0-e800:0).
Right, this usually was accessed using EMS drivers and bank switching.
> ps: i have an unrelated question - can anyone think of any use for
> the math coprocessor (other than doing math of course:) ?
> like to copy memory faster, or something similar...
There was a paper in PLDI '98 about using unused floating point resources
to do integer arithmetic. It was pretty experimental.
pat
