From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Moyer Subject: Re: [PATCH] dio: Fast-path for page-aligned IOs Date: Wed, 27 Jul 2011 18:15:53 -0400 Message-ID: References: <1308611855-18096-1-git-send-email-dehrenberg@google.com> <20110727210828.GB12714@infradead.org> <20110727141415.a67cafb8.akpm@linux-foundation.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Christoph Hellwig , Dan Ehrenberg , Alexander Viro , linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org To: Andrew Morton Return-path: In-Reply-To: <20110727141415.a67cafb8.akpm@linux-foundation.org> (Andrew Morton's message of "Wed, 27 Jul 2011 14:14:15 -0700") Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-fsdevel.vger.kernel.org Andrew Morton writes: > On Wed, 27 Jul 2011 17:08:28 -0400 > Christoph Hellwig wrote: > >> On Mon, Jun 20, 2011 at 04:17:35PM -0700, Dan Ehrenberg wrote: >> > The fast path does not apply for operations of the wrong size >> > or alignmnent, or for operations on raw drives with 512-byte sectors. >> > It might be possible to make this special case a little more general >> > while maintaining its performance benefits, but I do not believe that >> > the full performance benefits can be achieved without resorting to >> > special handling of simple cases, as is done in this patch. >> >> Did you check how this compares to Andis small optimizations? >> >> Also operations on raw disks are something people with fast devices >> care about a lot. We often hear about benchmark regressions due to >> stupid little things in the direct I/O code. >> >> If we want to special case something that would be a very easy target, >> with a 1:1 mapping of logical to physical blocks and thus no need >> to call the allocator first, and no need for any kind of locking >> or alignment handling. > > Ken did this back in 2006 > (http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=e61c90188b9956edae1105eef361d8981a352fcd) > but we reverted that shortly afterwards for some reason. For this reason: commit b2e895dbd80c420bfc0937c3729b4afe073b3848 Author: Andrew Morton Date: Sat Feb 3 01:14:01 2007 -0800 [PATCH] revert blockdev direct io back to 2.6.19 version Andrew Vasquez is reporting as-iosched oopses and a 65% throughput slowdown due to the recent special-casing of direct-io against blockdevs. We don't know why either of these things are occurring. The patch minimally reverts us back to the 2.6.19 code for a 2.6.20 release.