From mboxrd@z Thu Jan 1 00:00:00 1970 From: James Bottomley Subject: RE: [ANNOUNCE]: megaraid driver version 2.20.0.1 Date: 21 Jul 2004 12:06:15 -0500 Sender: linux-scsi-owner@vger.kernel.org Message-ID: <1090429579.1951.13.camel@mulgrave> References: <0E3FA95632D6D047BA649F95DAB60E57033BC89C@exa-atlanta> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: Received: from stat16.steeleye.com ([209.192.50.48]:36525 "EHLO hancock.sc.steeleye.com") by vger.kernel.org with ESMTP id S266545AbUGURGc (ORCPT ); Wed, 21 Jul 2004 13:06:32 -0400 In-Reply-To: <0E3FA95632D6D047BA649F95DAB60E57033BC89C@exa-atlanta> List-Id: linux-scsi@vger.kernel.org To: "Mukker, Atul" Cc: 'Christoph Hellwig' , "'linux-scsi@vger.kernel.org'" On Wed, 2004-07-21 at 08:48, Mukker, Atul wrote: > > The philosophy for scsi drivers for a while has been to > > remove internal > > queueing, which is why yours attracted my attention, so the question I > > was asking is can you use these APIs instead of having to > > maintain this > > internal queue? > The internal pending queue is fed not only from the mid-layer packets but > also the packets from the management module and (forthcoming) sysfs module. > Both of these modules rely on the fact that the commands issued by them > would be issued sooner or later. Hmm, well, in order to re-use the block layer queueing, you can wrap these up into special requests and queue them as well. The actual SCSI implementation does do a head of queue add, so any request so queued is the first to be pulled off again and executed > In addition, in certain situations (logical drive deletion) driver has to > stop sending commands to the FW before these operations can be initiated. > This is also implemented using the pending queue mechanism. Although would > be minor change to move over to scsi_block_requests and scsi_unblock_request > to achieve this. > > This is not to say that I am reluctant to implement SCSI_MLQUEUE_HOST_BUSY, > but just a precursor where additional impacts would be so that reviewers can > focus on those areas as well. Well, OK, just do the changes to get rid of yield, and we can look at fixing the queuing in tree. James