Hi,

I wonder if someone saw this before or maybe one can help to find the case of the problem: 

I'm using an own io-pkt-v4 instance with an 82574L pci-express network card (devnp-e1000.so driver) for EtherCAT (real 
time ethernet) communication. The SW is basically doing the same sequence every ~1ms (every system interrupt, which is a
 bit faster than 1ms, though that's ok). There is a bpf read() to read the packet of the previous cycle followed by a 
write() to send a new frame. The EtherCAT network will process the frame and it takes around ~35us till an interrupt 
occurs (cause of the returning frame) and the io-pkt process reads the frame (which is read by the application in the 
next cycle...)

Sometimes, about 200 times within 30 minutes, the interrupt is missing and the read call will run into a timeout. 

by looking at the kernel event trace, I can see no errros. The write call returns ok, though the "returning packet" and 
thus the interrupt is really missing. 

anyone saw something like this before? any ideas "where" to search for the problem? 

( it’s a x86 core i7 system with QNX 6.5.0 Sp1 + devnp-e1000.so latest bsp patch drivers, no apic, but multicore) 

Thanks!!
Matthias