From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from moutng.kundenserver.de (moutng.kundenserver.de [212.227.126.183]) by ozlabs.org (Postfix) with ESMTP id C832367B55 for ; Wed, 11 Oct 2006 19:25:45 +1000 (EST) From: Arnd Bergmann To: Geoff Levand Subject: Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing Date: Wed, 11 Oct 2006 11:25:09 +0200 References: <20061010204946.GW4381@austin.ibm.com> <452C2AAA.5070001@austin.ibm.com> <452C4CE0.5010607@am.sony.com> In-Reply-To: <452C4CE0.5010607@am.sony.com> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Message-Id: <200610111125.10902.arnd@arndb.de> Cc: akpm@osdl.org, jeff@garzik.org, netdev@vger.kernel.org, James K Lewis , linux-kernel@vger.kernel.org, linuxppc-dev@ozlabs.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Wednesday 11 October 2006 03:46, Geoff Levand wrote: > >=20 > > The others look good, but this one complicates the code and doesn't hav= e any benefit. =C2=A020=20 > > for 21 isn't bad. >=20 > Is the motivation for this change to improve performance by reducing the = overhead > of the mapping calls? =C2=A0If so, there may be some benefit for some sys= tems. =C2=A0Could > you please elaborate? >=20 =46rom what I understand, this patch drastically reduces the number of I/O PTEs that are needed in the iommu. With the current static IOMMU mapping, it should only make a difference during initialization, but any platform that uses a dynamic mapping of iommu entries will benefit a lot from it, because: =2D the card can do better prefetching of consecutive memory =2D there are more I/O ptes available for other drivers. Arnd <>< From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1750826AbWJKJZc (ORCPT ); Wed, 11 Oct 2006 05:25:32 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1750791AbWJKJZc (ORCPT ); Wed, 11 Oct 2006 05:25:32 -0400 Received: from moutng.kundenserver.de ([212.227.126.183]:54776 "EHLO moutng.kundenserver.de") by vger.kernel.org with ESMTP id S1750785AbWJKJZb convert rfc822-to-8bit (ORCPT ); Wed, 11 Oct 2006 05:25:31 -0400 From: Arnd Bergmann To: Geoff Levand Subject: Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing Date: Wed, 11 Oct 2006 11:25:09 +0200 User-Agent: KMail/1.9.4 Cc: jschopp , Linas Vepstas , akpm@osdl.org, jeff@garzik.org, netdev@vger.kernel.org, James K Lewis , linux-kernel@vger.kernel.org, linuxppc-dev@ozlabs.org References: <20061010204946.GW4381@austin.ibm.com> <452C2AAA.5070001@austin.ibm.com> <452C4CE0.5010607@am.sony.com> In-Reply-To: <452C4CE0.5010607@am.sony.com> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8BIT Content-Disposition: inline Message-Id: <200610111125.10902.arnd@arndb.de> X-Provags-ID: kundenserver.de abuse@kundenserver.de login:c48f057754fc1b1a557605ab9fa6da41 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Wednesday 11 October 2006 03:46, Geoff Levand wrote: > > > > The others look good, but this one complicates the code and doesn't have any benefit.  20 > > for 21 isn't bad. > > Is the motivation for this change to improve performance by reducing the overhead > of the mapping calls?  If so, there may be some benefit for some systems.  Could > you please elaborate? > >>From what I understand, this patch drastically reduces the number of I/O PTEs that are needed in the iommu. With the current static IOMMU mapping, it should only make a difference during initialization, but any platform that uses a dynamic mapping of iommu entries will benefit a lot from it, because: - the card can do better prefetching of consecutive memory - there are more I/O ptes available for other drivers. Arnd <>< From mboxrd@z Thu Jan 1 00:00:00 1970 From: Arnd Bergmann Subject: Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing Date: Wed, 11 Oct 2006 11:25:09 +0200 Message-ID: <200610111125.10902.arnd@arndb.de> References: <20061010204946.GW4381@austin.ibm.com> <452C2AAA.5070001@austin.ibm.com> <452C4CE0.5010607@am.sony.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: jschopp , Linas Vepstas , akpm@osdl.org, jeff@garzik.org, netdev@vger.kernel.org, James K Lewis , linux-kernel@vger.kernel.org, linuxppc-dev@ozlabs.org Return-path: Received: from moutng.kundenserver.de ([212.227.126.183]:54776 "EHLO moutng.kundenserver.de") by vger.kernel.org with ESMTP id S1750785AbWJKJZb convert rfc822-to-8bit (ORCPT ); Wed, 11 Oct 2006 05:25:31 -0400 To: Geoff Levand In-Reply-To: <452C4CE0.5010607@am.sony.com> Content-Disposition: inline Sender: netdev-owner@vger.kernel.org List-Id: netdev.vger.kernel.org On Wednesday 11 October 2006 03:46, Geoff Levand wrote: > >=20 > > The others look good, but this one complicates the code and doesn't= have any benefit. =C2=A020=20 > > for 21 isn't bad. >=20 > Is the motivation for this change to improve performance by reducing = the overhead > of the mapping calls? =C2=A0If so, there may be some benefit for some= systems. =C2=A0Could > you please elaborate? >=20 =46rom what I understand, this patch drastically reduces the number of I/O PTEs that are needed in the iommu. With the current static IOMMU mapping, it should only make a difference during initialization, but any platform that uses a dynamic mapping of iommu entries will benefit a lot from it, because: - the card can do better prefetching of consecutive memory - there are more I/O ptes available for other drivers. Arnd <><