kernel bug cpsw.c in linux arm 3.8.13-bone28

I encountered a bug while using the latest transmission-bt in a A5B version of BBB. Could you please point me to the bug tracker you are using or have a look and let me know what bug tracker I could use for the following bug:

http://pastebin.com/raw.php?i=reDRxxvE

This is the first time I tried to use a BBB for such task. Running ‘stress’ for couple of hours on the BBB did not reproduced the kernel trace.

Also if I am setting the cpufreq governor to performance the bug is not happening. I suspect there is something wrong with cpsw.c and i2c while encountering a cpu frequency transition.

The effect of the bug is that the eth0 capacity fails completely and pinging the bbb produces a ttl expired in transition. The usb0 is not affected.

Running debian 7.1 and the latest kernel from http://rcn-ee.net/deb/wheezy-armhf/v3.8.13-bone28/

Can you confirm the bug still exists on v3.12?

wget http://rcn-ee.net/deb/wheezy-armhf/v3.12.0-rc7-bone8/install-me.sh
sudo /bin/bash install-me.sh

(little behind, as i haven't pushed v3.12.x final yet..)

Note, this does not support external capes yet..

Regards,

I see you just pushed http://rcn-ee.net/deb/wheezy-armhf/v3.12.0-bone8/ yesterday, is that ok to try that one later ?

Yeap you can, once it hit's "rcn-ee.net" it's officially out.. :wink:

I was just building/testing when I wrote yesterdays reply..

Regards,

Hi Robert

Thanks for this. I’m still seeing a difference with the UARTs loaded from the device-tree overlays in /lib/firmware: I ask for BB-UART1, 2, 4 and 5, I get /dev/ttyO2, ttyO3, ttyO5 and an error in /var/log/syslog - “of_resolve: Could not find symbol ‘uart6’” (I’ve disabled the HDMI cape, and UART5 was working under 3.8).

Also I still get “of_resolve: Could not find symbol ‘ocp’” when I try to load BB-ADC. I presume this is the “missing cape support” mentioned earlier, which I must have missed the first announcement for (only been on the list a few days).

What is working for me are I2C 1 and 2, and USB is infinitely better than 3.8.

Do you have any idea when you’ll be able to roll a 3.12 with cape support and a working UART5?

Cheers… Mike

Bad luck so far, cpufreq is broken in 3.12, as it is not populating /sys/devices/system/cpu/cpu0/cpufreq
Without that the cpu is running at 1GHz no matter what the load is so I cannot really test the bug in cpsw.
Anyway, it would be nice to add a sync operation at the end of the install-me.sh file. I managed to corrupt the FAT partition on a very slow and old usdcard. I thing the eMMC might have the same issue. Otherwise 3.12 is nice and really quick.

Bad luck so far, cpufreq is broken in 3.12, as it is not populating
/sys/devices/system/cpu/cpu0/cpufreq
Without that the cpu is running at 1GHz no matter what the load is so I
cannot really test the bug in cpsw.

Yeah, cpufreq is broken, i need to cherry-pick quite a few commits
from v3.13-rc0...

Anyway, it would be nice to add a sync operation at the end of the
install-me.sh file. I managed to corrupt the FAT partition on a very slow
and old usdcard. I thing the eMMC might have the same issue. Otherwise 3.12
is nice and really quick.

I haven't seen that issue yet, weird, even with some of the crappy
slow microSD card's i've used..

So i just went a little crazy: :wink:

Regards,