Date: Mon, 25 Oct 2004 20:18:26 +0200 From: Matthias Andree <ma@dt.e-technik.uni-dortmund.de> To: Scott Long <scottl@freebsd.org> Cc: Garance A Drosihn <drosih@rpi.edu> Subject: Re: Annoying SCSI waiting... Message-ID: <m3d5z6psp9.fsf@merlin.emma.line.org> In-Reply-To: <41796D6D.7000108@freebsd.org> (Scott Long's message of "Fri, 22 Oct 2004 14:28:29 -0600") References: <417960C2.8040007@freebsd.org> <20041022194008.GA23778@odin.ac.hmc.edu> <41796396.5070804@freebsd.org> <p06110423bd9f1b6312ed@[128.113.24.47]> <41796D6D.7000108@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Scott Long <scottl@freebsd.org> writes: > For just about everyone, a value of 2 seconds or less is just fine. > For those with tape drives, changers, and (less likely) cdroms, a > long delay after reset is still quite necessary. There are two knobs > to adjust this, and I'd like GENERIC to remain compatible. It's 15 > seconds, not 2 minutes. You get a longer delay that that just trying > to set up the inital page tables on a large memory machine! One thing that was suggested for Linux but rejected for 2.6 is probing hardware in parallel. I wonder if a threaded boot-up could be made (not for 5.X though) that lets the kernel proceed with other initialization while the scsi driver waits for the devices to settle. Some scheme for stable device enumeration may then be needed to avoid inconsistencies with cold start vs. fastreboot. One should think when the driver has claimed all the I/O ranges and IRQs and DMA channels it needs, it would be safe to background itself and report back later, at the latest just before launching /etc/rc or /sbin/init or whatever. -- Matthias Andree
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?m3d5z6psp9.fsf>