From owner-freebsd-bugs  Tue Dec  9 13:00:05 1997
Return-Path: <owner-freebsd-bugs>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.7/8.8.7) id NAA05071
          for bugs-outgoing; Tue, 9 Dec 1997 13:00:05 -0800 (PST)
          (envelope-from owner-freebsd-bugs)
Received: (from gnats@localhost)
          by hub.freebsd.org (8.8.7/8.8.7) id NAA05043;
          Tue, 9 Dec 1997 13:00:02 -0800 (PST)
          (envelope-from gnats)
Date: Tue, 9 Dec 1997 13:00:02 -0800 (PST)
Message-Id: <199712092100.NAA05043@hub.freebsd.org>
To: freebsd-bugs
Cc: 
From: J Wunsch <j@uriah.heep.sax.de>
Subject: Re: kern/5231: Mounted MS-DOS floppy disk writes unreliable/reads can crash system
Reply-To: J Wunsch <j@uriah.heep.sax.de>
Sender: owner-freebsd-bugs@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

The following reply was made to PR kern/5231; it has been noted by GNATS.

From: J Wunsch <j@uriah.heep.sax.de>
To: gram@cdsec.com
Cc: freebsd-gnats-submit@FreeBSD.ORG
Subject: Re: kern/5231: Mounted MS-DOS floppy disk writes unreliable/reads can crash system
Date: Tue, 9 Dec 1997 21:43:18 +0100

 As gram@cdsec.com wrote:
 
 >  This is with an Adaptec 1542B, although
 > I also have problems with my home PC which uses a motherboard-based FDC.
 > I haven't tried reading a corrupt diskette on the IDE machine as yet,
 > but I suspect the behaviour will be similar. I suspect that the corruption
 > on write is due to the PC services other interrupts while the write is
 > in progress; i.e. is timing related.
 
 No, interrupts are interrupts, floppy transfers are done by DMA.
 
 Does it also happen when _writing_ floppies on the IDE machine?  Note
 that the FDC itself is totally unrelated, but the concurrent use of
 the AHA154x probably is.
 
 The AHA154x series has been known to hog the bus in DMA mode for too
 long for a long time.  The floppy driver silently retries all
 transfers that have been aborted by the FDC with an error flag of
 ``DMA overrun'', until the transfer finally succeeds.  When reading,
 this only reduces your effective data rate (drastically, btw.).  When
 writing however, this means you've got half-written sectors all over
 the place, which have to be retried.  This can cause subtle side-
 effects.
 
 If the above is indeed your problem, i'm afraid we can't do very much
 about it.  DOS doesn't suffer from the problem solely since SCSI and
 floppy transfers never happen by the same time.  Serializing all disk
 IO requests in Unix just for the purpose of supporting misbehaving
 hardware is probably a little much of work...
 
 I can try to reproduce and verify this on my scratch machine, although
 the floppy driver there is largely unused...  That's with an AHA1540A
 adapter, and i know it suffers from the DMA overrun syndrome.  (The
 initial ``retry all DMA overrun aborted transfers'' has been developed
 and tested on this machine.)
 
 -- 
 cheers, J"org
 
 joerg_wunsch@uriah.heep.sax.de -- http://www.sax.de/~joerg/ -- NIC: JW11-RIPE
 Never trust an operating system you don't have sources for. ;-)