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 <><