Re: [PATCH for-next v4] RDMA/efa: Add EFA query MR support

[Jason Gunthorpe <jgg@nvidia.com>]

On Sun, Jan 07, 2024 at 12:02:56PM +0200, Leon Romanovsky wrote:
> On Thu, Jan 04, 2024 at 09:51:55AM +0000, Michael Margolin wrote:
> > Add EFA driver uapi definitions and register a new query MR method that
> > currently returns the physical interconnects the device is using to
> > reach the MR. Update admin definitions and efa-abi accordingly.
> > 
> > Reviewed-by: Anas Mousa <anasmous@amazon.com>
> > Reviewed-by: Firas Jahjah <firasj@amazon.com>
> > Signed-off-by: Michael Margolin <mrgolin@amazon.com>
> > ---
> >  drivers/infiniband/hw/efa/efa.h               | 12 +++-
> >  .../infiniband/hw/efa/efa_admin_cmds_defs.h   | 33 ++++++++-
> >  drivers/infiniband/hw/efa/efa_com_cmd.c       | 11 ++-
> >  drivers/infiniband/hw/efa/efa_com_cmd.h       | 12 +++-
> >  drivers/infiniband/hw/efa/efa_main.c          |  7 +-
> >  drivers/infiniband/hw/efa/efa_verbs.c         | 71 ++++++++++++++++++-
> >  include/uapi/rdma/efa-abi.h                   | 21 +++++-
> >  7 files changed, 160 insertions(+), 7 deletions(-)
> 
> It is already fourth version of this patch and not a single word about
> the changes. Please add changelog in your next submissions.
> 
> Applied this patch with the following change.
> 
> diff --git a/drivers/infiniband/hw/efa/efa_verbs.c b/drivers/infiniband/hw/efa/efa_verbs.c
> index 8f4435706e4d..2f412db2edcd 100644
> --- a/drivers/infiniband/hw/efa/efa_verbs.c
> +++ b/drivers/infiniband/hw/efa/efa_verbs.c
> @@ -1748,7 +1748,7 @@ static int UVERBS_HANDLER(EFA_IB_METHOD_MR_QUERY)(struct uverbs_attr_bundle *att
>  {
>         struct ib_mr *ibmr = uverbs_attr_get_obj(attrs, EFA_IB_ATTR_QUERY_MR_HANDLE);
>         struct efa_mr *mr = to_emr(ibmr);
> -       u16 ic_id_validity;
> +       u16 ic_id_validity = 0;
>         int ret;
> 
>         ret = uverbs_copy_to(attrs, EFA_IB_ATTR_QUERY_MR_RESP_RECV_IC_ID,
> @@ -1766,12 +1766,12 @@ static int UVERBS_HANDLER(EFA_IB_METHOD_MR_QUERY)(struct uverbs_attr_bundle *att
>         if (ret)
>                 return ret;
> 
> -       ic_id_validity = (mr->ic_info.recv_ic_id_valid ?
> -                         EFA_QUERY_MR_VALIDITY_RECV_IC_ID : 0) |
> -                        (mr->ic_info.rdma_read_ic_id_valid ?
> -                         EFA_QUERY_MR_VALIDITY_RDMA_READ_IC_ID : 0) |
> -                        (mr->ic_info.rdma_recv_ic_id_valid ?
> -                         EFA_QUERY_MR_VALIDITY_RDMA_RECV_IC_ID : 0);
> +       if (mr->ic_info.recv_ic_id_valid)
> +               ic_id_validity |= EFA_QUERY_MR_VALIDITY_RECV_IC_ID;
> +       if (mr->ic_info.rdma_read_ic_id_valid)
> +               ic_id_validity |= EFA_QUERY_MR_VALIDITY_RDMA_READ_IC_ID;
> +       if (mr->ic_info.rdma_recv_ic_id_valid)
> +               ic_id_validity |= EFA_QUERY_MR_VALIDITY_RDMA_RECV_IC_ID;

I was saying in the rdma-core PR that this field shouldn't even
exist..

Jason