From mboxrd@z Thu Jan 1 00:00:00 1970 From: OGAWA Hirofumi Subject: Re: tg3: regression of netdev stats Date: Sat, 02 Mar 2013 15:10:42 +0900 Message-ID: <87y5e6wcv1.fsf@devron.myhome.or.jp> References: <878v67xnkr.fsf@devron.myhome.or.jp> <1362152836.6212.27.camel@LTIRV-MCHAN1.corp.ad.broadcom.com> Mime-Version: 1.0 Content-Type: text/plain Cc: nsujir@broadcom.com, davem@davemloft.net, netdev@vger.kernel.org To: "Michael Chan" Return-path: Received: from mail.parknet.co.jp ([210.171.160.6]:42418 "EHLO mail.parknet.co.jp" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750756Ab3CBGKv (ORCPT ); Sat, 2 Mar 2013 01:10:51 -0500 In-Reply-To: <1362152836.6212.27.camel@LTIRV-MCHAN1.corp.ad.broadcom.com> (Michael Chan's message of "Fri, 1 Mar 2013 07:47:16 -0800") Sender: netdev-owner@vger.kernel.org List-ID: "Michael Chan" writes: > On Fri, 2013-03-01 at 22:21 +0900, OGAWA Hirofumi wrote: >> Hi, >> >> f4a46d1f46a8fece34edd2023e054072b02e110d (tg3: Prevent spurious tx >> timeout by setting carrier off before tx disable.) seems to add the bug. >> >> On 3.8.1 (I didn't use 3.7.x series), "ifconfig" commands doesn't show >> any stats, i.e. all stats are zero. > > No stats on an active device transmitting and receiving packets? I > don't seem to see the same problem, but we'll look into this further. > Thanks. Yes, /proc/net/dev didn't show any packet traffic stats (always zero). But, e.g. ping/nfs itself works fine. BTW, chip is 02:00.0 Ethernet controller [0200]: Broadcom Corporation NetLink BCM57788 Gigabit Ethernet PCIe [14e4:1691] (rev 01) tg3 0000:02:00.0 eth0: Tigon3 [partno(BCM57788) rev 57780001] (PCI Express) MAC address d4:be:d9:a6:a3:3b tg3 0000:02:00.0 eth0: attached PHY driver [Broadcom BCM57780] (mii_bus:phy_addr=200:01) >> And bisect points the above commit. >> In fact, reverting the above patch fixes the regression of netdev stats. >> >> This patch looks like strange, and I'm not sure what this want to do. >> Any ideas? > > Every time we reset the device (ethtool, MTU change, etc), we need to > turn off all the tx queues before resetting. As soon as we turn off > these queues, if we don't turn off carrier, a tx timeout may be > triggered if it's been more than 5 seconds since the last transmit on > any of the tx queues. These tx timeouts cause unnecessary concern to > the user. Hmm... I'm not reading codes almost, so this might be wrong though. tg3_phy_reset() set off if ->link_up, so this seems to be not saving original state, and I'm not sure what want to do with ->link_up. I felt the following pesudo code might be work and simple... ok = netif_carrier_ok() netif_carrier_off() [reconfigure...] if (ok) netif_carrier_on() Thanks. -- OGAWA Hirofumi