On Wed, Dec 27, 2017 at 7:15 PM, Leon Romanovsky wrote:
> From: Moni Shoua <monis@xxxxxxxxxxxx>
>
> The QP type IB_QPT_DRIVER doesn't describe the transport or the service that
> the QP provides but those are known only to the hardware driver.
> The extra data that tells more about the QP is in the driver
> channel.
> Take the real QP type from the driver channel and modify the QP initial
> attributes before continuing with create_qp(). Also, define the layout
> of the data structure that comes in the driver channel.
> Downstream patches from this series will add support for both DCI and
> DCT driver QPs.
> +static int validate_driver_qp(struct mlx5_ib_dev *dev,
> + struct ib_qp_init_attr *init_attr,
> + struct mlx5_ib_create_qp *ucmd,
> + struct ib_udata *udata)
> +{
[...]
> + if (ucmd->flags & MLX5_QP_FLAG_TYPE_DCI)
> + init_attr->qp_type = MLX5_IB_QPT_DCI;
> + else
> + init_attr->qp_type = MLX5_IB_QPT_DCT;
> + return 0;
> +}
from maintenance/clarity standpoint, not sure if it is a good idea that
a validate_yyy function does assignment
> +
> struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
> - struct ib_qp_init_attr *init_attr,
> + struct ib_qp_init_attr *verbs_init_attr,
> struct ib_udata *udata)
> {
[...]
> @@ -2153,6 +2195,16 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
> dev = to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
> }
>
> + if (init_attr->qp_type == IB_QPT_DRIVER) {
> + struct mlx5_ib_create_qp ucmd;
> +
> + init_attr = &mlx_init_attr;
> + memcpy(init_attr, verbs_init_attr, sizeof(*verbs_init_attr));
> + err = validate_driver_qp(dev, init_attr, &ucmd, udata);
> + if (err)
> + return ERR_PTR(err);
> + }
> +
> switch (init_attr->qp_type) {
> case IB_QPT_XRC_TGT:
> case IB_QPT_XRC_INI:
> @@ -2214,6 +2266,9 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
> return ERR_PTR(-EINVAL);
> }
>
> + if (verbs_init_attr->qp_type == IB_QPT_DRIVER)
> + qp->driver_qp_type = init_attr->qp_type;
> +
> return &qp->ibqp;
> }
so if the values passed by user space are legal and the FW supports
DC, no error is returned to user-space, right? this doesn't seem bisectable,
if only patches 1...4 are applied, create qp will succeed?
--
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
