AppleIntelE1000e Hardware Hang - Ethernet port non-responsive until reboot

[The LT]

So for clarity, what I am saying is:

1. For my workflow, the Ethernet port hangs started after upgrading to 10.9.4

2. After extensive troubleshooting, the only driver version that made my port stable for my workflow was v2.4.14

3. I am not saying hnak's drivers are the root cause. The root cause is not yet known.

4. I am not saying that rolling back the driver will resolve the issue for everyone. There is clearly more going on than meets the eye. Fortunately, rolling back the driver resolved the hangs for my workflow at least for now.

Hope that helps

Thanks a lot for summing it up!
Anyways, the issue is really-really strange and somehow, moving to Clover made things a little bit worse. Under Chameleon, I am able to move somewhat more data before the port hang. I am also going to try bandwidth shaping the interface to, say, 950mbps or so and see if it improves things.

Again, the issue does not manifest itself on smallish files. It only manifests itself once more than 300-500Gbytes of data have been moved, so I am guessing it's not THAT common to run into such issues for most of the home users.

JFYI: Same thing happens on various machines, including GA-X58A-UD3R, GA-H87N-WIFI, GA-Z87X-UD7 TH, and others. They run every type of NICs, from Atheros to Realtek & Intel. The hangs are not consistent, sometimes it takes less then 10 gigabytes, sometimes more than 1TB, but it still happens.

ifconfig en0 down
ifconfig en0 up

The above commands fix the issue, so I guess the interface doesn't totally die. Weeks of searching on the net yield little information...

Also, I don't think it is iSCSI related as I can reproduce the issue using just Samba or AFP, with AFP crashing the port faster than Samba.

 

[atlantis852]

Thanks for that. BTW, I am not a "home" user .

Something very ugly and intermittent is going on. Anyone have any thoughts as to the actual root cause?

For those who just started troubleshooting this:

For me on a Maximus VI Extreme, I can not get the port to crash again doing large file transfers multiple times a day after rolling the driver back to v2.4.14. This does not address the root cause, which is not yet known, but it at least lets me get back to work. Hope this workaround can work for you.

 

[atlantis852]

I tried the new AppleIntel1000e v3.1.0 driver in MultiBeast 6.4 but unfortunately the port still hangs during any kind of file transfers. I've rolled back to v2.4.14.

 

[The LT]

Hi,

I've gotten a bit further. It looks like it is not an issue with a specific driver version, it's an issue with TCP Segmentation offload. Disabling TSO in the driver's info.plist file seems to have fixed the issue. I've found no way to disable TSO in the native OS X driver. So far, I've been able to run the latest e1000e driver without any intermittent lockups. I've transferred close to 10 terabytes of data in 1.5TB chunks.

Can you please confirm/deny this?

I've also been able to raise the buffer values to their defaults and beyond for iSCSI use and had no issues.

 

[noodleJam]

Great many thanks to this thread as rolling back to v2.4.14 seems to have worked for me.
I will keep a watchful eye on the thread to follow developments.

 

[atlantis852]

Hi,

I've gotten a bit further. It looks like it is not an issue with a specific driver version, it's an issue with TCP Segmentation offload. Disabling TSO in the driver's info.plist file seems to have fixed the issue. I've found no way to disable TSO in the native OS X driver. So far, I've been able to run the latest e1000e driver without any intermittent lockups. I've transferred close to 10 terabytes of data in 1.5TB chunks.

Can you please confirm/deny this?

I've also been able to raise the buffer values to their defaults and beyond for iSCSI use and had no issues.

Interesting. I will give this a try later today. Please reply with the exact parts of the plist you modified.

- - - Updated - - -

Great many thanks to this thread as rolling back to v2.4.14 seems to have worked for me.
I will keep a watchful eye on the thread to follow developments.

Glad I could help

 

[The LT]

Great many thanks to this thread as rolling back to v2.4.14 seems to have worked for me.
I will keep a watchful eye on the thread to follow developments.

You might try to disable TSO and update to latest driver and see if it helps.

 

[atlantis852]

You might try to disable TSO and update to latest driver and see if it helps.

LT, please post the procedure and values you used. I think this might be helpful to others visiting this thread.

I haven't had a chance yet to test out the TSO option with v3.1 but I will post the results when I do.

 

[The LT]

LT, please post the procedure and values you used. I think this might be helpful to others visiting this thread.

I haven't had a chance yet to test out the TSO option with v3.1 but I will post the results when I do.

To switch TCP Segmentation Offload (TSO) on and off, open Terminal and do:

sudo nano /System/Library/Extensions/IONetworkingFamily.kext/Contents/PlugIns/AppleIntelE1000e.kext/Contents/Info.plist

Scroll down and find a key NETIF_F_TSO and change <true/> below it to <false/>

Hit CTRL-O <enter> or CTRL-X and save & exit nano. You might also edit the plist with TextEdit or other tool of choice.

Don't forget to reboot.

 

[atlantis852]

To switch TCP Segmentation Offload (TSO) on and off, open Terminal and do:

sudo nano /System/Library/Extensions/IONetworkingFamily.kext/Contents/PlugIns/AppleIntelE1000e.kext/Contents/Info.plist

Scroll down and find a key NETIF_F_TSO and change <true/> below it to <false/>

Hit CTRL-O <enter> or CTRL-X and save & exit nano. You might also edit the plist with TextEdit or other tool of choice.

Don't forget to reboot.

Tested it and I still get a port hang with v3.1. In fact, it hangs faster with that plist modified. I am back on 2.4.14