From owner-freebsd-hackers Wed Jun 21 6:58:36 2000 Delivered-To: freebsd-hackers@freebsd.org Received: from chuggalug.clues.com (chuggalug.clues.com [194.217.82.22]) by hub.freebsd.org (Postfix) with ESMTP id 280E437BF9D for ; Wed, 21 Jun 2000 06:58:24 -0700 (PDT) (envelope-from geoffb@chuggalug.clues.com) Received: (from geoffb@localhost) by chuggalug.clues.com (8.9.3/8.9.3) id OAA21041; Wed, 21 Jun 2000 14:58:20 +0100 (BST) (envelope-from geoffb) Date: Wed, 21 Jun 2000 14:58:20 +0100 From: Geoff Buckingham To: les@safety.net Cc: freebsd-hackers@FreeBSD.ORG Subject: Re: Conflict between Intel 82558/9 and VIA MVP4? Message-ID: <20000621145820.A20728@chuggalug.clues.com> References: <200006141603.JAA80849@ns3.safety.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mutt 0.95.4i In-Reply-To: <200006141603.JAA80849@ns3.safety.net>; from Les Biffle on Wed, Jun 14, 2000 at 09:03:24AM -0700 Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On Wed, Jun 14, 2000 at 09:03:24AM -0700, Les Biffle wrote: > We're having problems with the Intel EtherExpress 10/100 NICs in our > product platform. We suspect unfavorable interaction between the 82558 > and 82559 Intel parts and our motherboard chipset. Here are some > specifics: > > We're using 3.4-STABLE, with the "latest" fxp driver code: > > $FreeBSD: src/sys/pci/if_fxp.c,v 1.59.2.7 2000/04/01 19:04:21 dg Exp $ > $FreeBSD: src/sys/pci/if_fxpreg.h,v 1.13.2.3 1999/12/06 20:11:53 peter Exp $ > $FreeBSD: src/sys/pci/if_fxpvar.h,v 1.6.2.2 2000/04/01 19:04:22 dg Exp $ > > The platform is a small PC designed for the point of sale folks, and uses > the VIA Apollo MVP4 chipset. From dmesg: > > chip0: rev 0x02 on pci0.0.0 > chip1: rev 0x00 on pci0.1.0 > chip2: rev 0x14 on pci0.7.0 > chip3: rev 0x10 on pci0.7.4 > > We use an AMD K6-2 at 350 or 450 Mhz, 32MB of RAM and boot from Compact Flash. > > The two PCI slots are on a riser card. On the riser card is a RealTEK > 8139 10/100 interface which works quite well: > > rl0: rev 0x10 int a irq 12 on pci0.13.0 > > We can install other RealTEK-based NICs in either or both riser card PCI > slots, and they work well, as do WAN cards. The problem comes when we > install a NIC based on the Intel 82558 or 82559 parts. > > When the NIC is in the "top" slot on the riser (pci0.1.19), the kernel > panics in if_fxp.c at fxp_add_rfabuf + 0xc4. The backtrace says > fxp_add_rfabuf was called from fxp_intr. > > With the NIC in the "bottom" slot (pci0.1.17), there is no panic, but the > card gets choked up and seems not to listen reliably. For example, it > will hear an ARP reply if it sent the ARP request, but will ignore an > ARP request inbound. My sniffer shows the packets on the link, but there > is no indication in a "netstat -i" that the NIC saw them. > Excuse me if I have missed something obviouse ( I havn't had a lot to do with PC hardware recently.) Is your 'riser' a customer part of the board or something that just plugs into an existing PCI slot. If the later I **ASSUME** (and may be wrong) that it is a PCI to PCI bridge itself. Leading to the questions does the fxp work in the PCI slot without the riser? and Are others successfully using that PCI-PCI bridge with an fxp -- GeoffB To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message