From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ed Swierk Subject: e1000 watchdog timer fails to start after ethool settings change Date: Wed, 15 Apr 2009 21:57:49 -0700 Message-ID: <1239857869.31663.29.camel@localhost.localdomain> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit To: jesse.brandeburg@intel.com, netdev@vger.kernel.org Return-path: Received: from rv-out-0506.google.com ([209.85.198.237]:57934 "EHLO rv-out-0506.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750881AbZDPE5z (ORCPT ); Thu, 16 Apr 2009 00:57:55 -0400 Received: by rv-out-0506.google.com with SMTP id f9so238649rvb.1 for ; Wed, 15 Apr 2009 21:57:54 -0700 (PDT) Sender: netdev-owner@vger.kernel.org List-ID: A recent patch to e1000, intended to avoid a race in the interrupt code, seems to prevent the watchdog timer from resuming properly. It neuters the effect of /* fire a link change interrupt to start the watchdog */ ew32(ICS, E1000_ICS_LSC); in e1000_up() when the __E1000_RESETTING flag is set, for example when invoked by e1000_set_settings(). The result is that running ethtool -s eth0 autoneg on in Qemu with an emulated e1000 NIC causes the kernel to stop noticing link status changes, leaving the link down until I prod the driver with ifconfig eth0 down; ifconfig eth0 up. Here is the patch causing the bug: > From: Jesse Brandeburg > > A nasty bug was found where an MTU change (or anything else that caused a > reset) could race with the interrupt code. The interrupt code was entered > by a shared interrupt during the MTU change. > > This change prevents the interrupt code from running while the driver is in > the middle of its reset path. > > Signed-off-by: Jesse Brandeburg > --- > > drivers/net/e1000/e1000_main.c | 4 ++-- > 1 files changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/net/e1000/e1000_main.c b/drivers/net/e1000/e1000_main.c > index 26474c9..c986978 100644 > --- a/drivers/net/e1000/e1000_main.c > +++ b/drivers/net/e1000/e1000_main.c > @@ -31,7 +31,7 @@ > > char e1000_driver_name[] = "e1000"; > static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver"; > -#define DRV_VERSION "7.3.20-k3-NAPI" > +#define DRV_VERSION "7.3.21-k3-NAPI" > const char e1000_driver_version[] = DRV_VERSION; > static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation."; > > @@ -3712,7 +3712,7 @@ static irqreturn_t e1000_intr(int irq, void *data) > struct e1000_hw *hw = &adapter->hw; > u32 rctl, icr = er32(ICR); > > - if (unlikely(!icr)) > + if (unlikely((!icr) || test_bit(__E1000_RESETTING, &adapter->flags))) > return IRQ_NONE; /* Not our interrupt */ > > /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is > Undoing the patch fixes the problem on a Qemu emulated e1000 NIC: Index: linux-2.6.29.1/drivers/net/e1000/e1000_main.c =================================================================== --- linux-2.6.29.1.orig/drivers/net/e1000/e1000_main.c +++ linux-2.6.29.1/drivers/net/e1000/e1000_main.c @@ -3712,7 +3712,7 @@ static irqreturn_t e1000_intr(int irq, v struct e1000_hw *hw = &adapter->hw; u32 rctl, icr = er32(ICR); - if (unlikely((!icr) || test_bit(__E1000_RESETTING, &adapter->flags))) + if (unlikely(!icr)) return IRQ_NONE; /* Not our interrupt */ /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is --- Of course this leaves the original problem unfixed; hopefully someone can suggest a better solution. --Ed