From owner-freebsd-hackers Sun Mar 29 20:18:39 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id TAA08647 for freebsd-hackers-outgoing; Sun, 29 Mar 1998 19:45:58 -0800 (PST) (envelope-from owner-freebsd-hackers@FreeBSD.ORG) Received: from nntp1.interworld.net (nntp.interworld.net [206.117.63.7]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id TAA08135 for ; Sun, 29 Mar 1998 19:43:51 -0800 (PST) (envelope-from pete@nntp1.interworld.net) Received: (from pete@localhost) by nntp1.interworld.net (8.8.8/8.8.7) id OAA01765 for hackers@freebsd.org; Sun, 29 Mar 1998 14:17:21 -0800 (PST) From: Pete Carah Message-Id: <199803292217.OAA01765@nntp1.interworld.net> Subject: IRQ conflicts in PCI To: hackers@FreeBSD.ORG Date: Sun, 29 Mar 1998 14:17:20 -0800 (PST) X-Mailer: ELM [version 2.4ME+ PL22 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG I ran into an interesting one trying to figure out why my news server was so slooooow - the config is INN 1.7.2-insync-1.1d (reputed to be fast, and it once was). On a careful inspection of the dmesg, I noticed that both SCSI channels and one of the NIC cards were all on one IRQ, and the other nic card, the vga, and the other scsi channel were on the other. On some boots the scsi channels both ended up on the same IRQ; this seems to be somewhat random... MB is IWILL P55XB2 (with the serial port bug :-( SCSI is a 3940UW with both channels in use. NICs are a SMC 21040 card and a Netgear 21141 card on a private link to the diablo server. I reconfig'd the bios to disable both serials and the printer since this is a dedicated news server anyhow, and it still refused to allocate any of these interrupt numbers to PCI - boooo. At least now I'm to the point where everything is separate except that the 100-base-t card ends up probed first on the same irq as the second scsi channel. Things look as if the first-probed units on the same irq get run polled; is this true? If so, would it be possible for the pci device-register code to queue the vectors so pci interrupt-sharing is possible (presuming that the hardware supports it; I'd hope so :-) (or does it already?) Somehow I get the feeling that there are no more iwill boards in my future, at least for a while (though I don't know if any of the others would be better with this set of cards)... -- Pete To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message