From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jes Sorensen Subject: Re: [patch] qla1280 update for 2.6.1-mm4 Date: 19 Jan 2004 16:18:42 -0500 Sender: linux-scsi-owner@vger.kernel.org Message-ID: References: <16395.51364.833229.536552@gargle.gargle.HOWL> <1074531798.1894.17.camel@mulgrave> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from jaguar.mkp.net ([192.139.46.146]:8898 "EHLO jaguar.mkp.net") by vger.kernel.org with ESMTP id S263596AbUASVSo (ORCPT ); Mon, 19 Jan 2004 16:18:44 -0500 In-Reply-To: <1074531798.1894.17.camel@mulgrave> List-Id: linux-scsi@vger.kernel.org To: James Bottomley Cc: SCSI Mailing List , Andrew Morton , jeremy@sgi.com >>>>> "James" == James Bottomley writes: James> Something to think about (I'm not asking you to change it) is James> whether you really want to use 64 bit descriptors on all 64 bit James> machines. The way the qla1280 single issue queue is organised James> it looks like we eat up slots (and, worse, overflow slots) James> quite a bit faster in 64 bit addressing mode. Therefore, it James> seems like the driver would operate more efficiently in 32 bit James> mode, so it might be worth a runtime check even on 64 bit James> machines to see whether the addressing mode is really necessary James> (i.e. no physical memory > 4GB for instance). Hi James, Certainly a valid concern, however it would probably more likely be based on available IOMMU slots, ie. with loads of 1280's in a box or other devices causing a lot of packets going through the system, it will probably be preferred to use 64 bit addressing where on smaller boxes I think you suggestion will be a pretty decent win. The driver also needs to handle it's busy situation a lot better, the current busy wait loop in qla1280_req_pkt is ehm ... no comment. Even if we decide to do dynamic switching between 64 and 32 bit mappings then I think it will be desirable to merge the two queuecommand versions for cache locality. Cheers, Jes