Hello,

the issue I am describing here happens on a dual-core Atom (without hyperthreading)

It is easy to reproduce with 2.6.32.7+xenomai 2.5.2, which was my initial configuration until
I remembered that Philippe told us that SMP was correctly supported from 2.6.38.8

The fact is that I have been able to reproduce it with 2.6.38.8+xenomai-2.6 as well. Only once, but I did.

I am using CAN with a IXXAT PCI-04 board .
There is a single thread per bus.

With the old kernel, after about 400-500 seconds, and heavy load the communication stops, and after some
analysis, I found out that my process was stuck at :

rtcan_raw.c

  /* Try to pass the guard in order to access the controller */
            ret = rtdm_sem_timeddown(&dev->tx_sem, timeout, NULL);


The Refcount shown in /proc/rtcan/rtcan0/info is 1.

The workaround I found was to set the timeout to a non-zero value with the appropriate ioctl,
and when a timeout issues, to stop and restart the bus, with the effect to destroy and re-recreate the semaphore and
thus to communicate again.

By reading the code, the only reason I can see is that a TX interrupt is lost.

I do not have much more ways to analyze deeper, so any advice would be greatly appreciated

Cheers,
Thierry