From: Andi Kleen Subject: Re: [dm-devel] [PATCH] DM-CRYPT: Scale to multiple CPUs v2 Date: Tue, 20 Jul 2010 09:26:22 +0200 Message-ID: <20100720072621.GB30592@basil.fritz.box> References: <20100601075259.GA13333@basil.fritz.box> <4C4444C2.6040607@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <4C4444C2.6040607@redhat.com> Sender: linux-kernel-owner@vger.kernel.org To: Milan Broz Cc: device-mapper development , Andi Kleen , agk@redhat.com, linux-kernel@vger.kernel.org, eric.dumazet@gmail.com List-Id: dm-devel.ids On Mon, Jul 19, 2010 at 02:27:46PM +0200, Milan Broz wrote: > On 06/01/2010 09:52 AM, Andi Kleen wrote: > > DM-CRYPT: Scale to multiple CPUs v2 > > > > Updated version with the per CPU access improvements Eric suggested. > Hi, > > seems that this approach is probably the best one we can use now... > (I would better see threads in crypto layer but that's taken of > the programme for today, seems.) > > I did some tests and it works for me, even tries some strange stacked > mapping etc. > Do you have some exact numbers for particular tests btw? I had some, but can't find them right now, sorry. The original motivation was one benchmark on encrypted fs which gave terrible results. But unfortunately I didn't get updated numbers with the patch for that one. The patch also seems to already have a small user base, at least I got a few requests for it. > > Anyway, I have just one comment - please can we make the per-cpu attribute > for IV generic? IOW not add explicit ie (ESSIV) but void * iv_private > - see attached patch. Fine for me. > > (I'll try discuss this with Alasdair to review for next merge > window, I think you requested that already in some private mail...) Yes. Thanks. -Andi -- ak@linux.intel.com -- Speaking for myself only.