rds: ib: Remove two ib_modify_qp() calls

For some HCAs, ib_modify_qp() is an expensive operation running
virtualized.

For both the active and passive side, the QP returned by the CM has the
state set to RTS, so no need for this excess RTS -> RTS transition. With
IB Core's ability to set the RNR Retry timer, we use this interface to
shave off another ib_modify_qp().

Fixes: ec16227e1414 ("RDS/IB: Infiniband transport")
Link: https://lore.kernel.org/r/1617216194-12890-3-git-send-email-haakon.bugge@oracle.com
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

diff --git a/net/rds/ib_cm.c b/net/rds/ib_cm.c
index f5cbe963cd8f781cdfbb99a59fb51cfae2bd0baf..26b069e1999df9682d6698817dbf66d2ee0c5c96 100644 (file)
--- a/net/rds/ib_cm.c
+++ b/net/rds/ib_cm.c
@@ -67,31 +67,6 @@ static void rds_ib_set_flow_control(struct rds_connection *conn, u32 credits)
        }
 }
 
-/*
- * Tune RNR behavior. Without flow control, we use a rather
- * low timeout, but not the absolute minimum - this should
- * be tunable.
- *
- * We already set the RNR retry count to 7 (which is the
- * smallest infinite number :-) above.
- * If flow control is off, we want to change this back to 0
- * so that we learn quickly when our credit accounting is
- * buggy.
- *
- * Caller passes in a qp_attr pointer - don't waste stack spacv
- * by allocation this twice.
- */
-static void
-rds_ib_tune_rnr(struct rds_ib_connection *ic, struct ib_qp_attr *attr)
-{
-       int ret;
-
-       attr->min_rnr_timer = IB_RNR_TIMER_000_32;
-       ret = ib_modify_qp(ic->i_cm_id->qp, attr, IB_QP_MIN_RNR_TIMER);
-       if (ret)
-               printk(KERN_NOTICE "ib_modify_qp(IB_QP_MIN_RNR_TIMER): err=%d\n", -ret);
-}
-
 /*
  * Connection established.
  * We get here for both outgoing and incoming connection.
@@ -100,7 +75,6 @@ void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_even
 {
        struct rds_ib_connection *ic = conn->c_transport_data;
        const union rds_ib_conn_priv *dp = NULL;
-       struct ib_qp_attr qp_attr;
        __be64 ack_seq = 0;
        __be32 credit = 0;
        u8 major = 0;
@@ -168,14 +142,6 @@ void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_even
         * the posted credit count. */
        rds_ib_recv_refill(conn, 1, GFP_KERNEL);
 
-       /* Tune RNR behavior */
-       rds_ib_tune_rnr(ic, &qp_attr);
-
-       qp_attr.qp_state = IB_QPS_RTS;
-       err = ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE);
-       if (err)
-               printk(KERN_NOTICE "ib_modify_qp(IB_QP_STATE, RTS): err=%d\n", err);
-
        /* update ib_device with this local ipaddr */
        err = rds_ib_update_ipaddr(ic->rds_ibdev, &conn->c_laddr);
        if (err)
@@ -947,6 +913,7 @@ int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
                                  event->param.conn.responder_resources,
                                  event->param.conn.initiator_depth, isv6);
 
+       rdma_set_min_rnr_timer(cm_id, IB_RNR_TIMER_000_32);
        /* rdma_accept() calls rdma_reject() internally if it fails */
        if (rdma_accept(cm_id, &conn_param))
                rds_ib_conn_error(conn, "rdma_accept failed\n");

diff --git a/net/rds/rdma_transport.c b/net/rds/rdma_transport.c
index 5f741e51b4baa46f825b7a676969cde3159b2c21..a9e4ff948a7d69a4c9ade67c482e2f7df25eebdb 100644 (file)
--- a/net/rds/rdma_transport.c
+++ b/net/rds/rdma_transport.c
@@ -87,6 +87,7 @@ static int rds_rdma_cm_event_handler_cmn(struct rdma_cm_id *cm_id,
 
        case RDMA_CM_EVENT_ADDR_RESOLVED:
                rdma_set_service_type(cm_id, conn->c_tos);
+               rdma_set_min_rnr_timer(cm_id, IB_RNR_TIMER_000_32);
                /* XXX do we need to clean up if this fails? */
                ret = rdma_resolve_route(cm_id,
                                         RDS_RDMA_RESOLVE_TIMEOUT_MS);