[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <36D9DB17C6DE9E40B059440DB8D95F52057B3FE4@orsmsx418.amr.corp.intel.com>
Date: Thu, 26 Jun 2008 00:13:49 -0700
From: "Brandeburg, Jesse" <jesse.brandeburg@...el.com>
To: "Herbert Xu" <herbert@...dor.apana.org.au>,
<netdev@...r.kernel.org>
Subject: RE: e1000e "Detected Tx Unit Hang"
Herbert Xu wrote:
> I'm getting these errors after upgrading to 2.6.26-rc8 (previously
> running Red Hat's 2.6.18-53.el5xen which was stable):
>
> Jun 26 08:26:16 gorgoroth kernel: eth0: Link is Up 1000 Mbps Full
> Duplex, Flow Control: RX/TX Jun 26 08:27:28 gorgoroth kernel: eth0:
> TSO is Disabled
<snip>
> Jun 26 08:37:32 gorgoroth kernel: NETDEV WATCHDOG: eth0: transmit
> timed out
> Jun 26 08:37:34 gorgoroth kernel: eth0: Link is Up 1000 Mbps Full
> Duplex, Flow Control: RX/TX
>
> This happens with or without TSO enabled.
Hi Herbert, thanks for the report, can you give some more details of the
hardware and test? We don't know of any outstanding tx hang issues, but
we do have patches pending for upstream, not sure if any will directly
fix your issue without knowing more details.
The most useful information would probably come from a combination of
running the ethregs utility located at e1000.sf.net and running
something like the output from e1000e_dump, which I've attached a patch
that applies to e1000e to do just that (which we could include in the
kernel, since akpm has suggested such in the past)
shortest dump from the debug code will be made by reducing the tx and rx
descriptors to a minimum that will still reproduce your issue
ethtool -G ethX rx 64 tx 64
Jesse
Download attachment "e1000e_dump.patch" of type "application/octet-stream" (13042 bytes)
Powered by blists - more mailing lists