[openib-general] Infiniband on Debian etch RC1

[Diego Guella]

> If you are using OFED 1.1, then you should use the source RPM for
> OpenSM. There was one patch on the list found with Debian for a stack
> smashing issue with osm_helper.c.

The SM which is currently running (on SuSE 9.3) is the one included in 
OFED-1.0.
Should I migrate to OFED-1.1 or can I build opensm from the OFED-1.0 source 
RPM?


>> I added a line with "ib_ipoib" to /etc/modules.
>>
>> So now I think I have to configure 2 new devices (MHES28 has 2 ports) in
>> /etc/network/interfaces.
>>
>> I added 2 devices named ib0 and ib1, and I configured them to have static 
>> IP
>> addresses, just like a normal ethernet device.
>>
>> ifconfig shows they are up, one has the attribute "RUNNING" too, the 
>> other
>> not (I think this is because one has the cable plugged, the other not).
>>
>> All this is done on server PE1950
>>
>> Now, that cable goes to the other server, a PE2850, which has a SM 
>> running.
>>
>> I try to ping that server, using the IP address of the infiniband IPoIB
>> interace, but I get "destination unreachable".
>
> I presume the two machines are on the same IPoIB subnet. Are there any
> errors in the OpenSM log ?
>
Yes, my Ethernet subnet is 192.168.200.0/255 and my Infiniband IPoIB subnet 
is 193.168.200.0/255, this is the same on all the machines.

I opened /var/log/osm.log for the first time now and (apart for the log 
size - 31MB!)
there is this error, that is repeating every 10 seconds from June, 26 (the 
date when I installed OFED-1.0) till today:
-----
Nov 21 13:00:22 438727 [42003960] -> __osm_sm_state_mgr_signal_error: ERR 
3207: Invalid signal OSM_SM_SIGNAL_DISCOVER in state 
IB_SMINFO_STATE_DISCOVERING
Nov 21 13:00:32 441056 [0000] -> SM port is down
-----

I want to point out that I have another system, a desktop, with installed 
SuSE 9.3 and OFED-1.0 and a MHES14 card, and that works fine, IPoIB, SDP, 
RDMA, all the features are OK.



Thanks,
Diego