From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Thu, 6 Jun 2002 16:44:24 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Thu, 6 Jun 2002 16:44:23 -0400 Received: from e31.co.us.ibm.com ([32.97.110.129]:57776 "EHLO e31.co.us.ibm.com") by vger.kernel.org with ESMTP id ; Thu, 6 Jun 2002 16:44:23 -0400 Date: Thu, 6 Jun 2002 13:44:19 -0700 From: Patrick Mansfield To: Shane Walton Cc: Linux Kernel Subject: Re: Stream Lined Booting - SCSI Hold Up Message-ID: <20020606134419.A24456@eng2.beaverton.ibm.com> In-Reply-To: <20020605172240.17081.qmail@web20808.mail.yahoo.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mutt 1.0.1i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Jun 05, 2002 at 10:22:40AM -0700, Shane Walton wrote: > Good day, > > I work with US DoD systems, and one of my requirements > is to boot an ix86 based system with an aic7xxx SCSI > interface in 15 seconds. One of the apparent problems > is the SCSI BIOS and aic7xxx driver take a long time > to find one drive and then probe other empty channels. > Is there a way to inform the aic7xxx driver of > connected hardware to limit the initialization phase? > Thanks for the help. I never thought to disable the reset, getting rid of the SCSI resets probably cuts 10-20 seconds off my boot time. You can turn of the AIC driver reset, and shorten the selection timeout, see linux/drivers/scsi/README.aic7xxx for details. With my hardware, I can also disable the BIOS reset, and limit BIOS scanning from the BIOS config screen (hit ctl-A during boot). On my system, it saves me about 10 seconds to boot with: lilo: linux-1 aic7xxx=no_reset aic7xxx=seltime:3 There are no options in linux scsi to limit the scanning of the target id's or channels, I don't think there are any in the AIC driver either. You could hack the driver and lower the max_id, and/or lower max_channel, or somehow disable unused channels, but I doubt it will save much time (7 id's * 2 channels * 32 ms/per-selection-timeout or so). -- Patrick Mansfield