Re: Footnote RE: Legacy Terminology RE: DMA across PCI

[dpannell@marvell.com]

It is interesting how history perceives things.  I am not an old man - but I
think I have been around longer than some of those thinking DMA was invented for
the ISA bus.  DMA, which stands for Direct Memory Access, was coined LONG before
the ISA bus used it.  And the 8237 was not the 1st DMA controller chip either.
The original 8237 was called the 9517.  It was designed by AMD.  It was far more
powerful than Intel's DMA controller at the time, which was called the 8217.
Intel acquired the rights to the 9517/8237 along with other AMD IP in exchange
for AMD getting rights to the 8086, 8088 and 80286.  The original 8217 worked
with Intel's 8085 and the older 8080 CPU (the 8080 was the CPU used in the 1st
personal computer kit called an Altair 8800 published in the Jan 1975 issue of
Popular Electronics.  This was years before the Apple I or Apple II).  The 8217
would not work with the 80286 so the 8237 replaced it.  I believe the concept of
DMA was mentioned in Intel databooks on the 8080 since it supported a way to
HOLD the processor so the DMA controller could do its job.  DMA was probably
used as a term well before then.  Maybe someone older than I can enlighten us
all further.

--Don





"Schneider, Dave" <Dave.Schneider@emulex.com> on 12/15/2000 09:46:54 AM

To:   pci-sig@znyx.com
cc:   "PCISIG List (E-mail)" <pci-sig@znyx.com> (bcc: Donald Pannell/Marvell)
Subject:  Footnote  RE: Legacy Terminology   RE: DMA across PCI



> PC industry for more than a decade, so I'm not sure if the term DMA was
> coined earlier than or beside the ISA bus.

As I recall, the Intel 8237 was already off-the-shelf when chosen for the ISA
bus.  It had been developed for Multibus I, a more feature-rich bus than ISA.
It's advantage to PC designers may have been that it didn't need to fetch
instructions (4 level instruction queues don't stay full for long), thus not
interrupting the transfer as often, as well as providing speed matching and
8-to-16 assembly/disassembly unattended.  It also simplifies driver writing.

Of course, when looking at historical designs, I'm partial to 8089-based
solutions myself  :-).

/dps