From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756700Ab0KKNQE (ORCPT ); Thu, 11 Nov 2010 08:16:04 -0500 Received: from mx1.redhat.com ([209.132.183.28]:43971 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755489Ab0KKNQD (ORCPT ); Thu, 11 Nov 2010 08:16:03 -0500 Date: Thu, 11 Nov 2010 15:15:42 +0200 From: "Michael S. Tsirkin" To: Christoph Hellwig Cc: Rusty Russell , Dave Chinner , linux-kernel@vger.kernel.org Subject: Re: [2.6.37-rc1, OOM] virtblk: OOM in do_virtblk_request() Message-ID: <20101111131542.GC25692@redhat.com> References: <20101105013003.GE13830@dastard> <20101110133151.GB2101@infradead.org> <201011111116.48263.rusty@rustcorp.com.au> <20101111125254.GB11194@infradead.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20101111125254.GB11194@infradead.org> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Nov 11, 2010 at 07:52:54AM -0500, Christoph Hellwig wrote: > On Thu, Nov 11, 2010 at 11:16:48AM +1030, Rusty Russell wrote: > > On Thu, 11 Nov 2010 12:01:51 am Christoph Hellwig wrote: > > > Rusty, Michael, > > > > > > any comments? I think Dave's observation is correct, and the lack of > > > a mempool for allocations in the virtio stack is a no-go for virtio_blk. > > > > Interesting. virtio will try to fall back to using direct ring entries > > if it can, but if course if your request is too large it can never do that. > > > > So, we could add a memory pool, or restrict the request size in virtio_blk. IMO size restriction is required anyway as host might not support indirect buffers. So let's do that, and then add mempool if someone has the time to profile and show the benefit. > The mempool looks like the more generic solution. Especially as people > are still talking about swap over nfs, at which point virtio-net will > show the same issue (just even harder to reproduce) I don't think so - virtio-net does it correctly: it won't let you send more packets than can fit in the ring. -- MST