[ofw] Problem with "Avoid the SM" patch

[Fab Tillier]

Hi Xalex,

>From: Alex Naslednikov [mailto:xalex at mellanox.co.il]
>Sent: Wednesday, September 10, 2008 9:56 AM
>
>Hello Fab,
>Finally, we found the problem and continue in order to fix it.
>Here is the description.
>
>2. Today we found, that this is not the original problem.
>The problem occurs when one restarts (kill and run) opensm. New instance
>of opensm initalize to zero some fields in AV.

I'm confused about this - how does the SM initialize AV fields?

>We found with IB analyser sends with REMOTE_LID==0 right after
>restarting the opensm, that caused to PING to fail.

Did the ARP request get sent properly?  What about the ARP response?  Where the contents of these packets 'sane'?

>Also, it's not related to the kind of connection, we tried it on
>back-to-back connection as well as on switch connection

Ok, that's good I suppose.

>3. We continue to debug in order to provide the solution. Please, let us
>know if you have some proposal to resolve this issue

Did you see this running on top of the 1486 revision, or against the head?  I worry that with all the recent changes to IPoIB that a change might have been lost.

-Fab