From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Garzik Subject: Re: [PATCH] fix sata_sil quirk Date: Tue, 22 Jun 2004 18:12:38 -0400 Sender: linux-ide-owner@vger.kernel.org Message-ID: <40D8AED6.8050503@pobox.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from parcelfarce.linux.theplanet.co.uk ([195.92.249.252]:5537 "EHLO www.linux.org.uk") by vger.kernel.org with ESMTP id S266022AbUFVWMy (ORCPT ); Tue, 22 Jun 2004 18:12:54 -0400 In-Reply-To: List-Id: linux-ide@vger.kernel.org To: Ricky Beam Cc: linux-ide@vger.kernel.org, Linux Kernel Ricky Beam wrote: > On Tue, 22 Jun 2004, Jeff Garzik wrote: > >> Here's my suggested fix... good catch Ricky. > > > And I don't even know why I looked at max_sectors :-) (I need more Dew.) > >> Yes, unfortunately performance will be dog slow. > > > Well, at least puppy slow... > Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn > sda 1811.65 0.00 9629.85 0 577887 > sdb 1807.15 0.00 9629.60 0 577872 > sdc 1807.25 0.00 9629.86 0 577888 > sdd 1807.05 0.00 9629.86 0 577888 > md_d0 14444.64 0.00 48148.84 0 2889412 > md_d0p2 9629.78 0.00 38519.11 0 2311532 > (over 60sec, 8M O_DIRECT accesses, 128 stripes * 16k RAID0) > > Without the MOD15 hack, the numbers are 2x higher, but they stop after > a few minutes :-) Is this with my patch? If so, I'll go ahead and forward it upstream, since I would certainly like a stabilization fix applied ASAP. >> I've got contacts at Silicon Image, and have been meaning to bug them >> for a "real fix" for a while. It is rumored that there is a much better >> fix, which allows full performance while at the same time not killing >> your SATA drive due to odd-sized SATA frames on the wire. > > > Ask them what they do in their driver? (the linux one and the windows one) > Looking at the linux driver, the mod15 quirk is there, but there doesn't > appear to be any associated device list. (I've already post the single > Maxtor device listed.) FreeBSD detects the stall, resets the chip and > hopes that clears the problem. (People are not happy about that.) The full-speed fix requires splitting affected DMA writes into two separate commands, when the sector count matches "sectors % 15 == 1". Jeff