Re: [PATCH rdma-next] RDMA/uapi: Fix and re-organize the usage of rdma_driver_id

[Jason Gunthorpe <jgg@mellanox.com>]

On Thu, Oct 10, 2019 at 03:05:41PM +0300, Leon Romanovsky wrote:
> From: Yishai Hadas <yishaih@mellanox.com>
> 
> Fix 'enum rdma_driver_id' to preserve other driver values before that
> RDMA_DRIVER_CXGB3 was deleted.  As this value is UAPI we can't affect
> other values as of a deletion of one driver id.
> 
> In addition,
> - Expose 'enum rdma_driver_id' to user applications by moving it to
>   ib_user_ioctl_verbs.h which is exposed in rdma-core to applications.
> 
> - Drop the dependency of ib_user_ioctl_verbs.h on ib_user_verbs.h which
>   is not really required.

One change per patch when it is a fixes..

> Fixes: 30e0f6cf5acb ("RDMA/iw_cxgb3: Remove the iw_cxgb3 module from kernel")
> Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
> Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
>  include/uapi/rdma/ib_user_ioctl_verbs.h  | 47 +++++++++++++++++++++++-
>  include/uapi/rdma/ib_user_verbs.h        | 25 -------------
>  include/uapi/rdma/rdma_user_ioctl_cmds.h | 21 -----------
>  3 files changed, 46 insertions(+), 47 deletions(-)
> 
> diff --git a/include/uapi/rdma/ib_user_ioctl_verbs.h b/include/uapi/rdma/ib_user_ioctl_verbs.h
> index 72c7fc75f960..3d703be40012 100644
> +++ b/include/uapi/rdma/ib_user_ioctl_verbs.h
> @@ -35,7 +35,6 @@
>  #define IB_USER_IOCTL_VERBS_H
>  
>  #include <linux/types.h>
> -#include <rdma/ib_user_verbs.h>
>  
>  #ifndef RDMA_UAPI_PTR
>  #define RDMA_UAPI_PTR(_type, _name)	__aligned_u64 _name
> @@ -167,10 +166,56 @@ enum ib_uverbs_advise_mr_flag {
>  	IB_UVERBS_ADVISE_MR_FLAG_FLUSH = 1 << 0,
>  };
>  
> +struct ib_uverbs_query_port_resp {
> +	__u32 port_cap_flags;		/* see ib_uverbs_query_port_cap_flags */
> +	__u32 max_msg_sz;
> +	__u32 bad_pkey_cntr;
> +	__u32 qkey_viol_cntr;
> +	__u32 gid_tbl_len;
> +	__u16 pkey_tbl_len;
> +	__u16 lid;
> +	__u16 sm_lid;
> +	__u8  state;
> +	__u8  max_mtu;
> +	__u8  active_mtu;
> +	__u8  lmc;
> +	__u8  max_vl_num;
> +	__u8  sm_sl;
> +	__u8  subnet_timeout;
> +	__u8  init_type_reply;
> +	__u8  active_width;
> +	__u8  active_speed;
> +	__u8  phys_state;
> +	__u8  link_layer;
> +	__u8  flags;			/* see ib_uverbs_query_port_flags */
> +	__u8  reserved;
> +};

Probably better to move ib_uverbs_query_port_resp_ex, since it is
really more of write() style interface

Jason