Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!rutgers!ames!ucbcad!ucbvax!decvax!tektronix!sequent!mntgfx!dclemans
From: dclemans@mntgfx.MENTOR.COM (Dave Clemans)
Newsgroups: comp.sys.atari.st
Subject: Re: Bug report: 'The Russian Doll'  sp. nova (?)
Message-ID: <640@dclemans.mntgfx.MENTOR.COM>
Date: Tue, 12-May-87 11:51:59 EDT
Article-I.D.: dclemans.640
Posted: Tue May 12 11:51:59 1987
Date-Received: Sat, 16-May-87 01:28:52 EDT
References: <8705081618.AA15013@ucbvax.Berkeley.EDU> <13104@watmath.UUCP>
Organization: Mentor Graphics, Beaverton OR
Lines: 33
Summary: floppy disk driver bug report


There is a bug in the floppy driver in the current TOS ROM's that can
very easily cause symptoms like those described in the article that
this message references.

Basically, the problem is that while the WD1772 floppy controller does
have a built-in read multiple sectors command, that command is somewhat
brain damaged; the only official way to terminate the read is to reset
the controller (and taking any time hit that that reset causes).

However, if you look at the floppy disk driver code you find that the
driver is using the "read multiple sectors" command of the 1772 for
all (all that I've found so far at least) floppy reads.  And under
"normal" circumstances the driver tries to never reset the controller;
i.e. it puts the read length into the DMA controller, and depends on
the floppy controller timing out because the DMA controller stops
talking to it.

This works "most" of the time...  But if your really trying to push
data through the system "fast" you have a non-zero chance of seeing
a timing problem pop up.

This problem does not affect floppy writes; only floppy reads.  The
best way to work around it, and still use "fast" formats, is to write
the disk in whatever way you desire, but NEVER try to read more than
one sector at a time (or to start out reading multiple sectors, but
switch to single sectors if an error appears).  An example of this
is the "slowread" routine in the twister restore code.

Unfortunately you don't always have control over the sizes of read
requests made by various packages...

dgc