[comp.unix.xenix] Memory-mapped device IO

compata@cup.portal.com (08/02/88)

I wish to add an IO board to my Xenix system.  The only way to address this
board is via dual-ported memory.  The precise addresses are switch-selectable
on the board.  The board does not generate any interrupts although it is
possible to cause an interrupt to the board by writing to an output port.

I understand that /dev/mem would allow access to all of physical memory.
However, I don't know what physical memory is not already in use.  More
importantly, I can't be sure that the same physical memory will always be
available.  I would like to find a way to address this board in a portable way,
so the same technique could be later used on SysV, and possibly on non-Intel
processor machines.

Any suggestions will be very welcome.  Email gratefully accepted.

Dave Close, Compata, Arlington, Texas - compata@cup.portal.com

jeffh@hpubvwa.HP.COM (Jeff Harrell) (08/06/88)

Memory-mapped I/O is best implemented on the Motorola chips. INTEL
didn't even get that right!

--
DISCLAMER: It's true!! I think...

Jeff (Spectra Software) Harrell
(206) 927-9268

jbayer@ispi.UUCP (id for use with uunet/usenet) (08/07/88)

In article <11650007@hpubvwa.HP.COM>, jeffh@hpubvwa.HP.COM (Jeff Harrell) writes:
> Memory-mapped I/O is best implemented on the Motorola chips. INTEL
> didn't even get that right!
> 

Wrong, wrong, wrong.  This statement shows that you do not know what memory-
mapped I/O is.

This reminds me of an old debate I read in some computer magazine back
in the dark ages when people were still putting computer kits together.  I
saw the discussion when somebody boasted that their SWTP 6800 (South West
Technical Products) had memory mapped I/O and the IMSI 8080 did not.  The
same ignorance applies.

Memory mapped I/O is any I/O which accesses memory addresses instead of using
special instructions to access the device.  Taken to the extreme even access
to normal memory can be considered to be memory mapped I/O.  ANY COMPUTER WHICH
ACCESSES MEMORY CAN HAVE MEMORY MAPPED I/O.  Possibly Intel didn't get some-
thing right, but it wasn't memory-mapped I/O.  

Jonathan Bayer
Intelligent Software Products, Inc.
19 Virginia Ave.
Rockville Centre, NY   11570
uunet!ispi!jbayer

jeffh@weycord.WEYCO.COM (08/12/88)

Johnathan Bayer writes: 
> Possibly Intell didn't get something right, but it wasn't memory-
> mapped I/O. 

Touche'. All I/O is memory mapped in the in the MC680xx processors.
 (but, to say accessing memory is memory-mapped I/O is pure bunk!)

Jeff (Spectra-Software) Harrell
hpubvwa!weycord!jeffh