From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jesper Dangaard Brouer Subject: [PATCH 0/2] igb: drop stats due to OS cannot keep up Date: Thu, 07 May 2009 15:35:46 +0200 Message-ID: <1241703346.18487.22.camel@localhost.localdomain> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Cc: "netdev@vger.kernel.org" , "e1000-devel@lists.sourceforge.net" , "Ronciak, John" , "Waskiewicz Jr, Peter P" To: "Kirsher, Jeffrey T" , "David S. Miller" Return-path: Received: from lanfw001a.cxnet.dk ([87.72.215.196]:54178 "EHLO lanfw001a.cxnet.dk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753353AbZEGNfr (ORCPT ); Thu, 7 May 2009 09:35:47 -0400 Sender: netdev-owner@vger.kernel.org List-ID: The following patchset address "drop" stats counters in the igb driver (chips 82576 and 82575). The drop stats addressed here are due to the OS cannot keep up. I have only tested it for the 82576 chip, as my (12 port) 82575 chip based NIC from Hotlava Systems recently died in my testlab :-((( [First patch] Implements reading of the per queue drop stats. These stats only gets incremented if the DROP_EN bit it set (in the SRRCTL register for that queue, or QDE reg is set). [Second patch] Address the case where the DROP_EN it NOT set. In this case a global register records the equivalent drop type. And aggregation of these two types of drops into the device net_stats. I strongly believe that these drops counters should be made available to sysadm's through the normal device stats. These stats are essential when diagnosing server performance issues, as these stats are due to the OS cannot keep up. I strongly believe that it should be stored in the net_stats rx_fifo_errors counter, to indicate performance issues. -- Med venlig hilsen / Best regards Jesper Brouer ComX Networks A/S Linux Network developer Cand. Scient Datalog / MSc. Author of http://adsl-optimizer.dk LinkedIn: http://www.linkedin.com/in/brouer