From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jesse Barnes Subject: Re: [PATCH] use mmiowb in tg3_poll Date: Tue, 31 May 2005 09:45:36 -0700 Message-ID: <200505310945.36903.jesse.barnes@intel.com> References: <200410211628.06906.jbarnes@engr.sgi.com> <1117544022.6134.54.camel@localhost.localdomain> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-6" Content-Transfer-Encoding: 7bit Cc: Arthur Kepner , Lennert Buytenhek , netdev@oss.sgi.com, gnb@sgi.com Return-path: To: hadi@cyberus.ca In-Reply-To: <1117544022.6134.54.camel@localhost.localdomain> Content-Disposition: inline Sender: netdev-bounce@oss.sgi.com Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org On Tuesday, May 31, 2005 5:53 am, jamal wrote: > Did i read correctly that your reads are about 2x more expensive than > the writes? The test above just measures the difference between using a full PIO read to ensure write ordering vs. the lighter weight mmiowb call. As for your question though, non-relaxed reads are very expensive on the Altix platform due to its highly distributed NUMA I/O architecture. The _relaxed variants can be quite fast however, and can be used anytime a PIO read doesn't imply anything about previous DMA transactions. Jesse