Re: e1000e regression (interface hang) with latest -git

[Ingo Molnar <mingo@elte.hu>]

* Allan, Bruce W <bruce.w.allan@intel.com> wrote:

> >-----Original Message-----
> >From: Ingo Molnar [mailto:mingo@elte.hu]
> >Sent: Wednesday, January 21, 2009 2:29 AM
> >To: linux-kernel@vger.kernel.org; Kirsher, Jeffrey T; Brandeburg, Jesse;
> >Allan, Bruce W; Waskiewicz Jr, Peter P; e1000-devel@lists.sourceforge.net;
> >netdev@vger.kernel.org
> >Cc: Rafael J. Wysocki
> >Subject: e1000e regression (interface hang) with latest -git
> >
> >
> >I've got a Nehalem testbox that developed a new e1000e problem in this
> >merge window: after a few minutes of uptime the network interface goes
> >dead - no rx and no tx. If i ifdown/ifup the interface it comes back. If i
> >wait too long then even ifdown/ifup does not help anymore - only a reboot.
> >
> >Other e1000e using testboxes i have are working just fine - so the problem
> >is specific to this hw.
> >
> >Is this a known problem?
> >
> >I have this hw:
> >
> > 01:00.0 Ethernet controller: Intel Corporation 82575EB Gigabit Network
> >Connection (rev 02)
> > 01:00.1 Ethernet controller: Intel Corporation 82575EB Gigabit Network
> >Connection (rev 02)
> >
> >If this is a new problem, what kind of other info do you need from me to
> >debug and fix this?
> >
> >I started seeing this very early in the merge window, so candidates would
> >be one of these early commits:
> >
> >eb14f01: Merge branch 'master' of
> >master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
> >cb7b48f: igb/e1000e: Naming interrupt vectors
> >5b9ab2e: Merge branch 'master' of
> >master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
> >e243455: e1000e: check return code from NVM accesses and fix bank
> >detection
> >a20e4cf: e1000e: fix incorrect link status when switch module pulled
> >8452759: e1000e: store EEPROM version number to prevent unnecessary NVM
> >reads
> >0285c8d: e1000e: cosmetic newline in debug message
> >5c48ef3: e1000e: sync change flow control variables with ixgbe
> >8f12fe8: e1000e: link up/down messages must follow a specific format
> >75eb0fa: e1000e: ESB2 config after link up
> >438b365: e1000e: check return of pci_save_state
> >1605927: e1000e: update comments listing supported parts for each MAC
> >family
> >63dcf3d: e1000e: 82571 check for link fix on 82571 serdes
> >5aa49c8: e1000e: commit speed/duplex changes for m88 PHY
> >005cbdf: e1000e: disable correctable errors for quad ports while going to
> >D3
> >0082982: netdev: add more functions to netdevice ops
> >651c246: e1000e: convert to net_device_ops
> >198d6ba: Merge branch 'master' of
> >master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
> >6ea7ae1: e1000e: enable ECC correction on 82571 silicon
> >4cf1653: netdevice: safe convert to netdev_priv() #part-2
> >babcda7: drivers/net: Kill now superfluous ->last_rx stores.
> >7c510e4: net: convert more to %pM
> >
> >If you suspect a specific list of commits i can test their revert. (But
> >the box is a slow booter and the problem can take up to 15 minutes to
> >trigger so i'd rather not spend half a day bisecting it, if it can be
> >avoided.)
> >
> >Thanks,
> >
> >	Ingo
> 
> 82575EB is not supported by e1000e, it is supported by igb.  Are you 
> sure that is the correct device?  Please send the system log and output 
> of:

yeah - sorry, i got confused - it is using IGB:

 CONFIG_IGB=y
 CONFIG_IGB_LRO=y

The same point remains though: it worked before (and works fine with a 
distro kernel, 2.6.28-0.140.rc9.git1.fc11.x86_64) and doesnt work with 
latest -git.

I'll capture some debug data and send it to you.

	Ingo