From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Miao Subject: Re: [PATCH 6/8] pxa: make lubbock to use smc91x platform data Date: Fri, 20 Jun 2008 09:23:07 +0800 Message-ID: <485B067B.1080207@gmail.com> References: <485A3E4A.6070706@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Cc: linux-netdev , linux-arm-kernel , Magnus Damm To: Nicolas Pitre Return-path: Received: from rv-out-0506.google.com ([209.85.198.234]:51385 "EHLO rv-out-0506.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751475AbYFTBXP (ORCPT ); Thu, 19 Jun 2008 21:23:15 -0400 Received: by rv-out-0506.google.com with SMTP id k40so5446297rvb.1 for ; Thu, 19 Jun 2008 18:23:14 -0700 (PDT) In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: Nicolas Pitre wrote: > On Thu, 19 Jun 2008, Eric Miao wrote: > >> Signed-off-by: Eric Miao > > This (and 7/8 and 8/8) are a bit unfortunate. Maybe not as visible on > Lubbock which has a SMC91C94 which can do 10mbps only, but the others > will have poorer performances because of those patches. Did you compare > performances, let's say with netperf, before and after those patches are > applied? > > I understand your goal of wanting to have a single kernel image that > dynamically can select the right bus access method for all the concerned > platforms. But I think you'll see that the runtime overhead has a > significant impact on network throughput and CPU usage. > > I think that some other solution should be explored such that when only > one machine type (or machine types with the same bus config) is > configured then the compile time optimizations should remain. Something > based on the machine_is_lubbock(), machine_is_Mainstone(), and so on, > which becomes constants with the right kernel configuration. > OK, I know your concern. Let's see if we can work this out in another way. > > Nicolas