problem with 82599 controller

[Alexander Fiveg]

Hello, 

the following problem I've faced while working with 82599-controller (ixgbe
driver):

- During packet capturing, after the number of received packets exceeds all
allocated descriptors (ixgbe_rxd * ixgbe_num_queues), the next new incoming
packets will be sometimes DMA'ed into the RAM incorrectly. 

Output from my tcpdump session: 

% tcpdump -i ix1
...
15:36:54.970343 IP 10.0.0.2.discard > 12.0.0.160.2200: UDP, length 58
15:36:55.070357 IP 10.0.0.2.discard > 12.0.0.161.2200: UDP, length 58
15:36:55.170373 c7:49:54:a8:00:0c (oui Unknown) > 35:c5:66:d7:df:e8 (oui 
Unknown), ethertype Unknown (0xd53b), length 100:
        0x0000:  4d98 ed85 7537 3b6b 3b8f 7102 4b1c 2cd4  M...u7;k;.q.K.,.
        0x0010:  41a8 2f3d 4faa fc8a a039 0fe2 5960 fad5  A./=O....9..Y`..
        0x0020:  c8b0 964b b0e0 2213 6aa2 330c ef93 80a9  ...K..".j.3.....
        0x0030:  6ac8 071b a9bd 0d51 ecca 94ba ac9c 873b  j......Q.......;
        0x0040:  a83f aeb0 20f4 cfd9 d1fa 93f3 795c 7d20  .?..........y\}.
        0x0050:  2993                                     ).
15:36:55.270388 IP 10.0.0.2.discard > 12.0.0.163.2200: UDP, length 58
...

For packets generating I am using Linux kernel packet generator. The receive- 
and send-interfaces are connected directly (without any hops in the middle). 
By the each new generated packet the destination-IP-address will be 
incremented, so I can proof the correctness of received traffic.  This error 
appears in both -STABLE and -CURRENT

I guess it is not a software bug and I would like to check out whether the
controller is in order.  

Are there any test setups for 8259x controllers that I could use ?

Thanks,
Alex