OSDN Git Service

RDMA/iw_cxgb4: initiate CLOSE when entering TERM
authorKrishnamraju Eraparaju <krishna2@chelsio.com>
Tue, 4 Feb 2020 09:12:30 +0000 (14:42 +0530)
committerJason Gunthorpe <jgg@mellanox.com>
Tue, 11 Feb 2020 18:28:00 +0000 (14:28 -0400)
commitd219face9059f38ad187bde133451a2a308fdb7c
tree83aa8a9dbfca1ab683728290fe970ff7ce9b1f23
parent10189e8e6fe8dcde13435f9354800429c4474fb1
RDMA/iw_cxgb4: initiate CLOSE when entering TERM

As per draft-hilland-iwarp-verbs-v1.0, sec 6.2.3, always initiate a CLOSE
when entering into TERM state.

In c4iw_modify_qp(), disconnect operation should only be performed when
the modify_qp call is invoked from ib_core. And all other internal
modify_qp calls(invoked within iw_cxgb4) that needs 'disconnect' should
call c4iw_ep_disconnect() explicitly after modify_qp. Otherwise, deadlocks
like below can occur:

 Call Trace:
  schedule+0x2f/0xa0
  schedule_preempt_disabled+0xa/0x10
  __mutex_lock.isra.5+0x2d0/0x4a0
  c4iw_ep_disconnect+0x39/0x430    => tries to reacquire ep lock again
  c4iw_modify_qp+0x468/0x10d0
  rx_data+0x218/0x570              => acquires ep lock
  process_work+0x5f/0x70
  process_one_work+0x1a7/0x3b0
  worker_thread+0x30/0x390
  kthread+0x112/0x130
  ret_from_fork+0x35/0x40

Fixes: d2c33370ae73 ("RDMA/iw_cxgb4: Always disconnect when QP is transitioning to TERMINATE state")
Link: https://lore.kernel.org/r/20200204091230.7210-1-krishna2@chelsio.com
Signed-off-by: Krishnamraju Eraparaju <krishna2@chelsio.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
drivers/infiniband/hw/cxgb4/cm.c
drivers/infiniband/hw/cxgb4/qp.c