[openib-general] oops with multicast patches

[dotanb at dev.mellanox.co.il]

> Roland Dreier wrote:
>> It just seems to attach local QPs to MCGs without talking to the SA at
>> all.  So it's not using the multicast module, but on the other hand I
>> can't see why what it does would have any relevance to the crash.
>
>  From his scenario, I thought that there were two applications running,
> but I
> could be off.  From looking at the attached file, I didn't see a relation
> between that code and any crash.  Dotan, can you clarify what you meant by
> "allocate N multicast groups" and "later application get the mcgs"?
No, there is only one application and a script that executes the SM and
the application (but it takes some time until this application allocates
all the mcgs it will use).

>
>  From your description, does this crash (always?) occur on the node that's
> running the SM?  I have not run this code in the SM node, so I will try
> this.
No.

I executed this test in back2back configuration and both of the machines
hanged. However, most of the time (4/5) the machine that executed the SM
got the crash ...

Dotan