From owner-freebsd-bugs Mon Jan 22 08:09:26 1996 Return-Path: owner-bugs Received: (from root@localhost) by freefall.freebsd.org (8.7.3/8.7.3) id IAA23661 for bugs-outgoing; Mon, 22 Jan 1996 08:09:26 -0800 (PST) Received: from proxy.siemens.at (proxy.siemens.at [192.138.228.19]) by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id IAA23647 for ; Mon, 22 Jan 1996 08:08:46 -0800 (PST) Received: from zerberus.hai.siemens.co.at (zerberus.hai.siemens-austria) by proxy.siemens.at with SMTP id AA12609 (5.67a/IDA-1.5 for ); Mon, 22 Jan 1996 17:07:51 +0100 Received: from localhost by zerberus.hai.siemens.co.at (4.1/SMI-4.1) id AA11304; Mon, 22 Jan 96 17:07:48 +0100 Message-Id: <9601221607.AA11304@zerberus.hai.siemens.co.at> To: Bruce Evans Cc: freebsd-bugs@freebsd.org, wirth@zerberus.hai.siemens.co.at Subject: Re: Bug with NCR810 driver: Corrections, Additions and a Solution, see previous message In-Reply-To: Your message of "Tue, 23 Jan 1996 01:00:40 +1100." <199601221400.BAA01173@godzilla.zeta.org.au> Date: Mon, 22 Jan 1996 17:07:47 +0100 From: Helmut Wirth Sender: owner-bugs@freebsd.org Precedence: bulk >The problem may be worse for concurrent opens of the same drive. There is >a known problem with initial concurrent opens. Concurrent initialization >of the slice table is unsafe (previously, concurrent initialization of the >label was unsafe). This problem is usually avoided by initially opening >mosts drives while there is only one active process. fsck -p may cause it. >Bruce I don't know about other side effects, but to give an unit more than one START_SCSI commands tagged in its command cache invites trouble. Some units don't mind like my Quantum ATLAS, but the IBM disks are confused. This is probably a question of the disks firmware; some disks may tolerate it and others not. I think we should get rid of this start unit command. There certainly is a need of such a command while *probing* for the disks. Both IBM disks could be jumpered to spin up with such a start unit command. Maybe the solution is to use it only once at *the first open* and to reset a flag with the *last close*, this regardless if the open is caused by an access via the block device (during mount) or by the character device (with dump, fsck, et.al.). Helmut Wirth