Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/17/84; site oasys.UUCP
Path: utzoo!watmath!clyde!cbosgd!ihnp4!qantel!lll-crg!seismo!hao!nbires!oasys!jcg
From: jcg@oasys.UUCP
Newsgroups: net.arch
Subject: DMA controller chips
Message-ID: <171@oasys.UUCP>
Date: Wed, 8-Jan-86 15:01:16 EST
Article-I.D.: oasys.171
Posted: Wed Jan  8 15:01:16 1986
Date-Received: Mon, 13-Jan-86 04:46:38 EST
Distribution: net
Organization: NBI,Inc, Boulder CO
Lines: 53


As part of a design for a 80186-based controller, we've been looking at DMA
controller chips.  We need four DMA channels (in addition to the two in the 
'186) that may all be in operation concurrently.  The peripherals being 
interfaced are 8-bit bus oriented, but the '186 has 16-bit wide RAM so we
want the DMA to do word assembly/disassembly for us.  We have not been able
to find ANY chip on the market which will do the job for us.

In this article, I want to describe the shortcomings of each of the chips we 
have looked at, and generate some discussion of the architectural 'features'
which make these chips useless to us.  Hopefully someone out in net-land will
be able to clue us into a more useful DMA chip.

-Intel 8237:  This is an old 8085 family part which does not fit well into a
16-bit processor design.  It requires a fair amount of 'glue' logic around it
and still performs very slowly -- its maximum clock rate is only 5 MHz so its
bus cycles are very slow.  It is certainly inexpensive, though.  The Intel 8257
is a even older and slower version of the 8237.  These parts have no provision
for word-assembly.  The 8237 is also known as AMD Am9517.

-AMD Am9516 and Zilog Z8016:  These are 2-channel DMA chips which are very
similar to each other.  They have word-assembly and dissassembly but have
only one word former shared between both channels.  The chip performs two
single-byte bus cycles in succession for each DMA request from a peripheral
and the other DMA channel cannot be serviced between them.  These parts are
also a little expensive for us ($20) since we would need two of them to meet
our 4 channel requirement, but this doesn't matter because the channels are
not truly independent.

-68450 from Hitachi (?Motorola):  We haven't looked at this device too hard
because it was quoted to us at $68 which is far too expensive for us.  It is
not clear from the specs we have whether the word-assembly is shared between
channels.

-National NS32203:  This chip is in the right price range ($28) but the data
sheets say "If word assembly/disassembly will be performed ... , after the
first byte is transferred to/from the device, the DMA does not acknowledge o
other channel requests until the second byte is transferred."  In other words,
the channels are AGAIN sharing the word assembly logic.

The conclusions we reach from all this is that the existing DMA chips cannot
do what we want, and that we must build our own or change the requirements. 
This is amazing to us because we did not think our requirements were unusual.

So how about it?  Have we overlooked anything obvious?

Please send any flames or personal replies to oasys!jcg@nbires.UUCP


				Jonathan Griffitts
				NBI, Inc.


(This note states my own opinions and not necessarily those of anyone else)