Path: utzoo!attcan!utgpu!watserv1!watdragon!rose!ccplumb
From: ccplumb@rose.waterloo.edu (Colin Plumb)
Newsgroups: comp.sys.amiga.tech
Subject: Re: DMA vs Non-DMA
Message-ID: <19078@watdragon.waterloo.edu>
Date: 7 Dec 89 21:36:38 GMT
References: <1989Dec1.152959.5833@dvinci.usask.ca> <1989Dec6.151618.3097@sjsumcs.sjsu.edu>
Sender: daemon@watdragon.waterloo.edu
Reply-To: ccplumb@rose.waterloo.edu (Colin Plumb)
Organization: U. of Waterloo, Ontario
Lines: 19

In article <1989Dec6.151618.3097@sjsumcs.sjsu.edu> 33014-18@sjsumcs.SJSU.EDU (Eduardo Horvath) writes:
>	Is there a bug with the Amiga DMA scheme that causes data to be lost
>	under some obcure conditions, or is that somebody mis-reading 
>	a 2090 review again?

Sigh...

There is no bug in the Amiga's DMA scheme.  There is a bug in the 2090
that causes it to drop bytes when there is severe memory contention
(such as DMA into chip memory with 4-bitplane high-res screens), but it
detects this condition and retries.  The 2090 bug does not cause lost
data, only enormous numbers of retries.

The details are that the 2090 does not do anything with the "buffer full"
indicator on its FIFO... it naively assumes the buffer can be flushed to
memory faster than the disk reads.  So it keeps asking the disk for data
even though it can't accept it.  This causes problems.
-- 
	-Colin