From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S935282Ab1JETzm (ORCPT ); Wed, 5 Oct 2011 15:55:42 -0400 Received: from 173-166-109-252-newengland.hfc.comcastbusiness.net ([173.166.109.252]:55660 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S934696Ab1JETzk (ORCPT ); Wed, 5 Oct 2011 15:55:40 -0400 Message-Id: <20111005195403.407628164@bombadil.infradead.org> User-Agent: quilt/0.48-1 Date: Wed, 05 Oct 2011 15:54:03 -0400 From: Christoph Hellwig To: Rusty Russell Cc: Chris Wright , Jens Axboe , Stefan Hajnoczi , kvm@vger.kernel.org, linux-kernel@vger.kernel.org Subject: [PATCH 0/5] RFC: ->make_request support for virtio-blk X-SRS-Rewrite: SMTP reverse-path rewritten from by bombadil.infradead.org See http://www.infradead.org/rpr.html Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org This patchset allows the virtio-blk driver to support much higher IOP rates which can be driven out of modern PCI-e flash devices. At this point it really is just a RFC due to various issues. The first four patches are infrastructure that could go in fairly soon as far as I'm concerned. Patch 5 implements the actual ->make_request support and still has a few issues, see there for more details. With it I can driver my PCI-e test devices to 85-90% of the native IOPS and bandwith, but be warned that this is still a fairly low end setup as far as expensive flash storage is concerned. One big downside that is has is that it current exposes a nasty race in the qemu virtqueue code - just running xfstests inside a guest using the new virtio-blk driver (even on a slow device) will trigger it and lead to a filesystem shutdown. I've tracked it down to getting data I/O segments overwritten with status s/g list entries, but got lost at that point. I can start a separate thread on it. Besides that it is missing a few features, and we have to decided how to select which mode to use in virtio-blk - either a module option, sysfs attribute or something that the host communicates. Or maybe decide that just going with ->make_request alone is fine, even on my cheap laptop SSD it actually is just as fast if not slightly faster than the request based variant on my laptop. There are a few other bottlenecks in virtio that this exposes. The first one is the low queue length of just 128 entries in the virtio-blk queue - to drive higher IOPs with a deep queue we absolutely need to increment that. Comments welcome!