We at Mellanox have started running our regression on this driver. u So far we have found one bug (see attached patch at the end of this mail). xsigo have tested the code with our driver which includes following IB support: 1) CM support 2) MAD query 3) RC connection 4) FMR support. We will continue to run with the attached patch and see if it solves all the problems. Thanks Tzachi p.s. I'm still looking for the correct const for IB_WR_RDMA_READ_WITH_INV. =================================================================== --- Q:/projinf3/trunk/hw/mlx4/kernel/bus/ib/qp.c (revision 5244) +++ Q:/projinf3/trunk/hw/mlx4/kernel/bus/ib/qp.c (revision 5245) @@ -68,9 +68,21 @@ __constant_cpu_to_be32(MLX4_OPCODE_ATOMIC_CS), /* [IB_WR_ATOMIC_CMP_AND_SWP] */ __constant_cpu_to_be32(MLX4_OPCODE_ATOMIC_FA), /* [IB_WR_ATOMIC_FETCH_AND_ADD]*/ __constant_cpu_to_be32(MLX4_OPCODE_LSO | (1 << 6)), /* [IB_WR_LSO] */ + + + __constant_cpu_to_be32(MLX4_OPCODE_SEND_INVAL), /* [IB_WR_SEND_WITH_INV] */ + __constant_cpu_to_be32(MLX4_OPCODE_RDMA_READ), /* [IB_WR_RDMA_READ_WITH_INV] */ + __constant_cpu_to_be32(MLX4_OPCODE_LOCAL_INVAL), /* [IB_WR_LOCAL_INV] */ + __constant_cpu_to_be32(MLX4_OPCODE_FMR), /* [IB_WR_FAST_REG_MR] */ + + + __constant_cpu_to_be32(MLX4_OPCODE_NOP) /* [IB_WR_NOP] */ }; + +//????????????????? IB_WR_RDMA_READ_WITH_INV, //??????????????? + extern inline void rdma_get_ll_mac(struct in6_addr *addr, u8 *mac); static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp)
________________________________ From: Smith, Stan [mailto:[email protected]] Sent: Tuesday, December 08, 2009 8:06 PM To: Tzachi Dar Cc: [email protected]; Ishai Rabinovitz Subject: RE: patch 0/2 Add support for RDMAoEth to the low level driver What level of testing has been applied to the RDMAoE patches to ensure no side effects have been introduced into the IB stack operation? stan. ________________________________ From: [email protected] [mailto:[email protected]] On Behalf Of Tzachi Dar Sent: Wednesday, December 02, 2009 8:21 AM To: [email protected] Subject: [ofw] patch 0/2 Add support for RDMAoEth to the low level driver First I would like to thank James Yang from Xsigo that actually did most of the work, and brought RDMA over Ethernet to a working stage. The first patch enables RDMAoEth to work. It includes code only for the low level drivers. The second patch is small changes needed in two small tests that I have been using in order to test the work. Thanks Tzachi
_______________________________________________ ofw mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ofw
