On Friday, April 04/27/18, 2018 at 03:36:18 +0000, Parav Pandit wrote:
Parav,
> Hi Shiraz, Raju,
>
> > -----Original Message-----
> > My validation team reported the same issue on i40iw with 4.17-rc kernels.
> >
> > Some more data. Looks like the failure is because we can't find the cached gid
> > due to the gid idx being wrong in query_gid.
> >
> > rdma_connect
> > cma_connect_iw
> > cma_modify_qp_rtr
> > ib_query_gid
> > ib_get_cached_gid
> > __ib_cache_gid_get (EINVAL)
> >
> This call trace is helpful.
> Can you please run ibv_devinfo -v | grep GID and see that you are getting the expected GID.
> For iWarp we have only one entry GID table. So want to make sure that GID table is build correctly.
>
> From the above call trace, is appears that,
> cma_modify_qp_rtr() contains,
> struct ib_qp_attr qp_attr;
>
> ah_attr from above qp_attr remains uninitialized by iw_cm_init_qp_attr() and iwcm_init_qp_init_attr().
> Before my fix, gid_index was always ignored by the query_gid() callback such as i40iw_query_gid() and c4iw_query_gid().
> So it used to work.
> Now my fix expects all values to be correct; due to uninitialized ib_qp_attr it is likely failing.
>
> So can you please try below hunk and see if ib_query_gid() progresses for you?
> If it works, I will send the proper patch shortly.
>
Its working. Issue is not seen with the below patch.
> diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
> index 8512f63..e119cff 100644
> --- a/drivers/infiniband/core/cma.c
> +++ b/drivers/infiniband/core/cma.c
> @@ -863,7 +863,7 @@ void rdma_destroy_qp(struct rdma_cm_id *id)
> static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
> struct rdma_conn_param *conn_param)
> {
> - struct ib_qp_attr qp_attr;
> + struct ib_qp_attr qp_attr = {};
> int qp_attr_mask, ret;
> union ib_gid sgid;
>
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
