From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx0a-000f0801.pphosted.com (mx0a-000f0801.pphosted.com [67.231.144.122]) by dpdk.org (Postfix) with ESMTP id B0157559C for ; Mon, 9 Nov 2015 17:32:03 +0100 (CET) Received: from pps.filterd (m0000542.ppops.net [127.0.0.1]) by mx0a-000f0801.pphosted.com (8.15.0.59/8.15.0.59) with SMTP id tA9GSfo3010291; Mon, 9 Nov 2015 08:32:01 -0800 Received: from hq1wp-exmb11.corp.brocade.com ([144.49.131.13]) by mx0a-000f0801.pphosted.com with ESMTP id 1y1bramxjm-1 (version=TLSv1/SSLv3 cipher=AES256-SHA bits=256 verify=NOT); Mon, 09 Nov 2015 08:32:01 -0800 Received: from xeon-e3 (172.16.180.50) by Hq1wp-exmb11.corp.brocade.com (10.70.20.185) with Microsoft SMTP Server (TLS) id 15.0.1104.5; Mon, 9 Nov 2015 08:32:00 -0800 Date: Mon, 9 Nov 2015 08:32:09 -0800 From: Stephen Hemminger To: Harry van Haaren Message-ID: <20151109083209.7c1eb27b@xeon-e3> In-Reply-To: <1447080904-27512-1-git-send-email-harry.van.haaren@intel.com> References: <1447080904-27512-1-git-send-email-harry.van.haaren@intel.com> Organization: Brocade MIME-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit X-Originating-IP: [172.16.180.50] X-ClientProxiedBy: hq1wp-excas13.corp.brocade.com (10.70.36.103) To Hq1wp-exmb11.corp.brocade.com (10.70.20.185) X-Proofpoint-Spam-Reason: safe X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10432:5.15.21, 1.0.33, 0.0.0000 definitions=2015-11-09_13:2015-11-05,2015-11-09,1970-01-01 signatures=0 Cc: dev@dpdk.org Subject: Re: [dpdk-dev] [PATCH] ixgbe: fix crc-strip enable changing rx bytes X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 09 Nov 2015 16:32:04 -0000 On Mon, 9 Nov 2015 14:55:04 +0000 Harry van Haaren wrote: > Fix a consistency issue in ixgbe, that when CRC stripping is enabled, > the CRC bytes are not added to the rx total byte count. When CRC strip > is disabled, these bytes are counted. > > This patch reads the CRC strip register, and when enabled adds 4 bytes > to the total bytes recieved counter for each received packet. > > E-mail discussion thread available: > http://dpdk.org/ml/archives/dev/2015-March/015685.html > > To reproduce, run testpmd with no flags and with --crc-strip, sending a > single packet, and notice that the RX byte counter is 4 bytes smaller > when CRC stripping is enabled. > > Fixes: 29a05247ebad ("ixgbe: configure CRC stripping behaviour of PF") > > Reported-by: Stephen Hemminger > Suggested-by: Konstantin Ananyev > Signed-off-by: Harry van Haaren > --- > doc/guides/rel_notes/release_2_2.rst | 6 ++++++ > drivers/net/ixgbe/ixgbe_ethdev.c | 7 +++++++ > 2 files changed, 13 insertions(+) > > diff --git a/doc/guides/rel_notes/release_2_2.rst b/doc/guides/rel_notes/release_2_2.rst > index 59dda59..6bab51b 100644 > --- a/doc/guides/rel_notes/release_2_2.rst > +++ b/doc/guides/rel_notes/release_2_2.rst > @@ -134,6 +134,12 @@ Drivers > > VF needs the PF interrupt support initialized even if not started. > > +* **ixgbe: Fixed CRC strip enable influencing RX byte counters.** > + > + A workaround has been put in place to ensure consistent presentation > + of total RX bytes. The CRC bytes are now always included in RX bytes, > + providing consistency between RX and TX bytes. > + > * **i40e: Fixed base driver allocation when not using first numa node.** > > Fixed i40e issue that occurred when a DPDK application didn't initialize > diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c > index 0b0bbcf..b311ffb 100644 > --- a/drivers/net/ixgbe/ixgbe_ethdev.c > +++ b/drivers/net/ixgbe/ixgbe_ethdev.c > @@ -2361,6 +2361,13 @@ ixgbe_read_stats_registers(struct ixgbe_hw *hw, struct ixgbe_hw_stats > hw_stats->mrfc += IXGBE_READ_REG(hw, IXGBE_MRFC); > hw_stats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC); > > + /* Workaround for RX byte count not including CRC bytes when CRC > + * strip is enabled. Multiply rx_total_packets by CRC size and add > + * to adjust rx_bytes to the correct value > + */ > + if (IXGBE_READ_REG(hw, IXGBE_HLREG0) & IXGBE_HLREG0_RXCRCSTRP) > + *total_qbrc += (*total_qprc * 4); > + > /* Note that gprc counts missed packets */ > hw_stats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC); > NAK. The CRC should never be part of rx/tx statistics. The problem with putting CRC in statistics is the behavior introduces a device and setting dependency. This is a disaster for generic programs (like vRouter) that want to work across multiple hardware and virtual NIC's. Also, Linux, BSD, and Unix do not include CRC in reported bytes for network interfaces. Please fix up my original patch to work for both settings, or I can do it.