[ofa-general] rdma_create_qp fails with -12

[Shipman, Galen M.]

We are seeing failures setting up a QP using rdma_create_qp.


This only occurs when:

init_qp_attr.cap.max_send_wr

init_qp_attr.cap.max_recv_wr

Totals to more than 16K.

I have queried the device attributes and I have found:

ib_device_attr.max_qp_wr = 65535


The CQ is setup to have: init_qp_attr.cap.max_send_wr  +  
init_qp_attr.cap.max_recv_wr CQE's  and we are not seeing any  
failures creating the CQ with ib_create_cq.


I have tried replicating this problem using user level verbs but I am  
unable to do so using ibv_create_qp. I used the test program located  
here:

http://lists.openfabrics.org/pipermail/general/2006-February/017104.html



For reference ibv_devinfo reports:

hca_id: mthca0
         fw_ver:                         4.7.600
         node_guid:                      0008:f104:0397:b6b8
         sys_image_guid:                 0008:f104:0397:b6bb
         vendor_id:                      0x08f1
         vendor_part_id:                 25208
         hw_ver:                         0xA0
         board_id:                       VLT0040010001
         phys_port_cnt:                  2
                 port:   1
                         state:                  PORT_ACTIVE (4)
                         max_mtu:                2048 (4)
                         active_mtu:             2048 (4)
                         sm_lid:                 1
                         port_lid:               1
                         port_lmc:               0x00

                 port:   2
                         state:                  PORT_DOWN (1)
                         max_mtu:                2048 (4)
                         active_mtu:             512 (2)
                         sm_lid:                 0
                         port_lid:               0
                         port_lmc:               0x00




And lspci:

04:00.0 InfiniBand: Mellanox Technologies MT25208 InfiniHost III Ex  
(Tavor compatibility mode) (rev 20)


Any ideas here?

Thanks,

Galen