perftest: Fix seg-fault during clean-up for cm retry flow

[SherrinZhou]

Hi, @sshaulnv.
Follow the last comment on PR #368 . This is the fix for the clean up routine that I mentioned.

[SherrinZhou]

With this fix, for the manual rejection issued by the server , server side print will be like:

************************************
* Waiting for client to connect... *
************************************

[DEBUG] Intentionally rejecting connection request #1

[DEBUG] Intentionally rejecting connection request #2
---------------------------------------------------------------------------------------
                    RDMA_Write BW Test
 Dual-port       : OFF          Device         : rsnc_0
 Number of qps   : 1            Transport type : IB
 Connection type : RC           Using SRQ      : OFF
 PCIe relax order: ON           Lock-free      : OFF
 ibv_wr* API     : OFF          Using Enhanced Reorder      : OFF
 CQ Moderation   : 1
 CQE Poll Batch  : Dynamic
 Mtu             : 1024[B]
 Link type       : Ethernet
 GID index       : 1
 Max inline data : 0[B]
 rdma_cm QPs     : ON
 Data ex. method : rdma_cm
---------------------------------------------------------------------------------------
 Waiting for client rdma_cm QP to connect
 Please run the same command with the IB/RoCE interface IP
---------------------------------------------------------------------------------------
 local address: LID 0000 QPN 0x002f PSN 0x2691e6
 GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:10:01:01
 remote address: LID 0000 QPN 0x0049 PSN 0x207bd6
 GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:10:01:02
---------------------------------------------------------------------------------------
 #bytes     #iterations    BW peak[MiB/sec]    BW average[MiB/sec]   MsgRate[Mpps]
 65536      5000             3107.19            3107.15              0.049714
---------------------------------------------------------------------------------------

Client side print will be like:

RDMA CM event error:
Event: RDMA_CM_EVENT_REJECTED; error: 28.

ERRNO: Operation not supported.
Failed to handle RDMA CM event.
ERRNO: Operation not supported.
Failed to connect RDMA CM events.
ERRNO: Operation not supported.
RDMA CM event error:
Event: RDMA_CM_EVENT_REJECTED; error: 28.

ERRNO: Operation not supported.
Failed to handle RDMA CM event.
ERRNO: Operation not supported.
Failed to connect RDMA CM events.
ERRNO: Operation not supported.
---------------------------------------------------------------------------------------
                    RDMA_Write BW Test
 Dual-port       : OFF          Device         : rsnc_0
 Number of qps   : 1            Transport type : IB
 Connection type : RC           Using SRQ      : OFF
 PCIe relax order: ON           Lock-free      : OFF
 ibv_wr* API     : OFF          Using Enhanced Reorder      : OFF
 TX depth        : 128
 CQ Moderation   : 1
 CQE Poll Batch  : Dynamic
 Mtu             : 1024[B]
 Link type       : Ethernet
 GID index       : 1
 Max inline data : 0[B]
 rdma_cm QPs     : ON
 Data ex. method : rdma_cm
---------------------------------------------------------------------------------------
 local address: LID 0000 QPN 0x0049 PSN 0x207bd6
 GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:10:01:02
 remote address: LID 0000 QPN 0x002f PSN 0x2691e6
 GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:10:01:01
---------------------------------------------------------------------------------------
 #bytes     #iterations    BW peak[MiB/sec]    BW average[MiB/sec]   MsgRate[Mpps]
Conflicting CPU frequency values detected: 3029.045000 != 4061.956000. CPU Frequency is not max.
 65536      5000             3107.19            3107.15              0.049714
---------------------------------------------------------------------------------------

Now the CM will correctly retry the connection attempt with no resource leak.

[sshaulnv]

@SherrinZhou , tested it and it look great, thanks!