From owner-freebsd-hackers Mon Sep 21 08:03:43 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id IAA23601 for freebsd-hackers-outgoing; Mon, 21 Sep 1998 08:03:43 -0700 (PDT) (envelope-from owner-freebsd-hackers@FreeBSD.ORG) Received: from spinner.netplex.com.au (spinner.netplex.com.au [202.12.86.3]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id IAA23585 for ; Mon, 21 Sep 1998 08:03:37 -0700 (PDT) (envelope-from peter@netplex.com.au) Received: from spinner.netplex.com.au (localhost [127.0.0.1]) by spinner.netplex.com.au (8.9.1/8.9.1/Spinner) with ESMTP id WAA01449; Mon, 21 Sep 1998 22:44:50 +0800 (WST) (envelope-from peter@spinner.netplex.com.au) Message-Id: <199809211444.WAA01449@spinner.netplex.com.au> X-Mailer: exmh version 2.0.2 2/24/98 To: Mike Smith cc: Peter Jeremy , hackers@FreeBSD.ORG Subject: Re: More on the Intel-UNIX standard In-reply-to: Your message of "Sun, 20 Sep 1998 21:32:47 MST." <199809210432.VAA03143@word.smith.net.au> Date: Mon, 21 Sep 1998 22:44:50 +0800 From: Peter Wemm Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG Mike Smith wrote: > > On Sun, 20 Sep 1998 00:26:18 -0500, "Pedro F. Giffuni" wrote: > > >I'm concerned about the SNR in hackers, but I just couldn't resist > > >asking if someone was aware and acting on this > > > http://www.sco.com/udi/ > > > > This was mentioned by some Intel marketroids at the recent AUUG'98 > > conference. There was a fair amount of discussion at a subsequent > > Freenix BOF (which included Greg Lehey and Peter Wemm within the > > FreeBSD group and Robert Hart from Red Hat, as well as assorted > > users from the Linux community and all the *BSD groups). > > > > The almost unanimous concensus(*) was that it was a very bad move and > > the Freenix community should resist it. The major problems seen by > > the group were: > > 1) Binary-only device drivers are a bad idea. It will reduce the chances > > of us getting access to the hardware interface specs, and therefore > > being able to build a device driver that works. > > 2) Binary-only device drivers tie the hardware to the processor. This > > reduces the portability of (eg) PCI cards. > > 3) The difficulty of supporting the kernel services required for a UDI > > driver. > > 4) Increased finger-pointing when a device driver fails. > > Unless there's been a sudden change of direction in the last few months, > UDI is source-level, not a binary-level interface. This invaliates 1) > and 2). 3) is actually less of a problem than many people like to > think; unlike most other driver models (eg. WDM) UDI has been designed > by people that understand the issues. "8. The interfaces exported to drivers must allow for binary compatibility between systems, where appropriate. 8.1. Drivers must not be required to be recompiled to run on different versions of an OS on the same machine. 8.2. Drivers must not be required to be recompiled to run on different OSes which support the same ISA (Instruction Set Architecture), binary file format, and calling conventions." Ie: a UDI binary for an i386 running SCO is supposed to work on all i386 OS's. Having said that, the prospects of being able to grab a pre-existing driver are appealing. Naturally it would be nicest if source was available, but I'd be very suprised if it came from vendors like adaptec etc. Having browsed lightly over the information, I'm not as worried as I was at the conference. I'm beginning to think that it actually might work and work reasonably well. The bulk of the work is going to be in the UDI interface "shims" (for want of a better term) between the UDI modules and the OS proper. It appears to have a lot of callback interfaces, so a UDI-scsi "shim" would probably look a lot like a generic scsi driver with all the "fill me in here" type spots replaced with calls to the other side of the UDI interface. Cheers, -Peter To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message