RDMA/cxgb3: Don't free endpoints early
authorSteve Wise <swise@opengridcomputing.com>
Sun, 6 Sep 2009 03:22:38 +0000 (20:22 -0700)
committerRoland Dreier <rolandd@cisco.com>
Sun, 6 Sep 2009 03:22:38 +0000 (20:22 -0700)
commit6e47fe43502ba6dfe86d556661795d9bb0361309
treebc29f159253320abf8d55ceac0b06dda3ddd3b7d
parentfa0d4c11c4b6eb49708b82b638ceb0761152f46a
RDMA/cxgb3: Don't free endpoints early

- Keep ref on connection request endpoints until either accepted or
  rejected so it doesn't get freed early.

- Endpoint flags now need to be set via atomic bitops because they can
  be set on both the iw_cxgb3 workqueue thread and user disconnect
  threads.

- Don't move out of CLOSING too early due to multiple calls to
  iwch_ep_disconnect.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
drivers/infiniband/hw/cxgb3/iwch_cm.c
drivers/infiniband/hw/cxgb3/iwch_cm.h