[ofa-general] CM sysfs-related oops on device driver reload

[Roland Dreier]

 > Was this function the cause of the problem?  Was the cm port object freed while
 > references were still on one of the cm counter objects?

To be honest, I haven't figured out the exact bug yet.  I made a bunch
of changes to make the counter_groups dynamically allocated, based on
this from Documentation/kobject.txt:

   No structure should EVER have more than one kobject embedded within it.
   If it does, the reference counting for the object is sure to be messed
   up and incorrect, and your code will be buggy.  So do not do this.

but thinking about it further, I think the original way should be OK.
So I want to go back and look for a smaller patch that still fixes
things, and understand what the breakage was.