From owner-freebsd-current Fri Oct 25 03:55:31 1996 Return-Path: owner-current Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id DAA08847 for current-outgoing; Fri, 25 Oct 1996 03:55:31 -0700 (PDT) Received: from who.cdrom.com (who.cdrom.com [204.216.27.3]) by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id DAA08795 for ; Fri, 25 Oct 1996 03:55:20 -0700 (PDT) Received: from gibbs.bayarea.net (gibbs.bayarea.net [205.219.64.132]) by who.cdrom.com (8.7.5/8.6.11) with ESMTP id XAA15371 for ; Thu, 24 Oct 1996 23:59:32 -0700 (PDT) Received: from localhost (localhost [127.0.0.1]) by gibbs.bayarea.net (8.7.6/8.7.3) with SMTP id XAA00324 for ; Thu, 24 Oct 1996 23:59:59 -0700 (PDT) Message-Id: <199610250659.XAA00324@gibbs.bayarea.net> X-Authentication-Warning: gibbs.bayarea.net: Host localhost [127.0.0.1] didn't use HELO protocol To: current@freebsd.org Subject: New aic7xxx driver drop Date: Thu, 24 Oct 1996 23:59:58 -0700 From: "Justin T. Gibbs" Sender: owner-current@freebsd.org X-Loop: FreeBSD.org Precedence: bulk Hi folks, I just spent a large portion of the last week or so adding memory mapped I/O, dma'ing of SCBs, and greatly improving the performance of SCB paging in the aic7xxx driver. I also took the time to do a full code review and to correct some of the mistakes of my youth. 8-) The main reason I went on this blitz was to try and fix a few problems that have shown up in fast processor (aka P6) machines. There have been a few reports of NMIs and strange traps doing PIO to aic78xx chips on these machines. I was able to find a contact at Adaptec who, although he could not verify any known problems with PIO on these chips, was able to confirm that Adaptec uses DMA to get SCBs down to the adapter and always uses memory mapped I/O. My guess is that they have never seen this problem since they never use PIO. As it turns out, using DMA is a big win, especially when it comes to using SCB paging (something I hope will become the default setting soon). Doing memory mapped I/O is also a big win especially on processors like the P6 where PIO spams its speculative execution foo. I'd like to bring these changes into stable ASAP, but will only do so once I get some feedback on this latest drop of the driver. Please help to test it out. I'd also be interested to hear if this version is faster or slower then previos versions. Thanks, -- Justin T. Gibbs =========================================== FreeBSD: Turning PCs into workstations ===========================================