From mboxrd@z Thu Jan  1 00:00:00 1970
From: Boaz Harrosh <bharrosh@panasas.com>
Subject: Re: [PATCH 07/24] arm: scsi convert to accessors and !use_sg cleanup
Date: Tue, 18 Sep 2007 17:04:31 +0200
Message-ID: <46EFE8FF.3010907@panasas.com>
References: <46E6F92D.2090409@panasas.com> <46E72AE7.4030507@panasas.com> <20070912074205.GB2653@flint.arm.linux.org.uk>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Return-path: <linux-scsi-owner@vger.kernel.org>
Received: from sa7.bezeqint.net ([192.115.104.21]:58230 "EHLO sa7.bezeqint.net"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1757888AbXIRPEj (ORCPT <rfc822;linux-scsi@vger.kernel.org>);
	Tue, 18 Sep 2007 11:04:39 -0400
In-Reply-To: <20070912074205.GB2653@flint.arm.linux.org.uk>
Sender: linux-scsi-owner@vger.kernel.org
List-Id: linux-scsi@vger.kernel.org
To: Russell King <rmk@arm.linux.org.uk>, James Bottomley <James.Bottomley@SteelEye.com>
Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>, Greg Kroah-Hartman <gregkh@suse.de>, Alan Stern <stern@rowland.harvard.edu>, Matthew Dharm <mdharm-usb@one-eyed-alien.net>, Jeff Garzik <jeff@garzik.org>, "David S. Miller" <davem@davemloft.net>, Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>, YOKOTA Hiroshi <yokota@netlab.is.tsukuba.ac.jp>, linux-scsi <linux-scsi@vger.kernel.org>, "Maciej W. Rozycki" <macro@linux-mips.org>, Mark Fortescue <mark@mtfhpc.demon.co.uk>

On Wed, Sep 12 2007 at 10:42 +0300, Russell King <rmk@arm.linux.org.uk> wrote:
> On Wed, Sep 12, 2007 at 02:55:19AM +0300, Boaz Harrosh wrote:
>> -		if (SCpnt->request_bufflen != len)
>> +		if (scsi_bufflen(SCpnt) != len) {
>> +			WARN_ON(1);
> 
> NAK.  The call trace generally doesn't provide any additional information
> on the cause of the error.
> 
In my opinion this can not happen any more. If it does, I want to see that
it is not through the regular scsi-ml .queuecommand mechanism.
But if you insist than sure I will remove it.

>>  			printk(KERN_WARNING "scsi%d.%c: bad request buffer "
>>  			       "length %d, should be %ld\n", SCpnt->device->host->host_no,
>> -			       '0' + SCpnt->device->id, SCpnt->request_bufflen, len);
>> -		SCpnt->request_bufflen = len;
>> +			       '0' + SCpnt->device->id, scsi_bufflen(SCpnt), len);
>> +		}
>>  #endif
>>  	} else {
>> -		SCpnt->SCp.ptr = (unsigned char *)SCpnt->request_buffer;
>> -		SCpnt->SCp.this_residual = SCpnt->request_bufflen;
>> -		SCpnt->SCp.phase = SCpnt->request_bufflen;
>> -	}
>> -
>> -	/*
>> -	 * If the upper SCSI layers pass a buffer, but zero length,
>> -	 * we aren't interested in the buffer pointer.
>> -	 */
>> -	if (SCpnt->SCp.this_residual == 0 && SCpnt->SCp.ptr) {
>> -#if 0 //def BELT_AND_BRACES
>> -		printk(KERN_WARNING "scsi%d.%c: zero length buffer passed for "
>> -		       "command ", SCpnt->host->host_no, '0' + SCpnt->target);
>> -		__scsi_print_command(SCpnt->cmnd);
>> -#endif
>>  		SCpnt->SCp.ptr = NULL;
>> +		SCpnt->SCp.this_residual = 0;
>> +		SCpnt->SCp.phase = 0;
>>  	}
>>  }
> 
> Also NAK.  This was added due to bad behaviour of the SCSI layer and
> was found to be necessary.
> 
No! This check is no longer Relevant. The master if() is on bufflen() now,
and only than do we ever set SCp.ptr. The else will always set both to Zero.
(Which is what you want)

In any way this check is done in scsi-ml, and since 2.6.18 only scsi-ml
can allocate and issue commands. All other sources of commands where removed.
All upper layers issue requests now.

Boaz