[openib-general] ibv_cmd_create_qp() question

[Roland Dreier]

 > I've uncovered a problem while debugging my rhel4u4 chelsio backport and
 > I have question for you.  libcxgb3's iwch_create_qp() calls
 > ibv_cmd_create_qp().  If that returns success, it then mmap()s the wq
 > memory.  If mmap() fails (which is what's happening on rhel4u4), it then
 > calls ibv_cmd_destroy_qp().  However the qp->context field isn't
 > initialized at this point so ibv_cmd_destroy_qp() faults.  
 > 
 > Q: Should ibv_cmd_create_qp() set the context field since
 > ibv_cmd_destroy_qp() requires it for destroying the object?

Yes, I see the problem.

Hmm, setting context in the create commands looks like the best way to
handle this.  Can you make a patch to do that (we should fix all the
create commands and move setting context from verbs.c to cmd.c)?

Thanks...