[PATCHv2 for-next 1/2] RDMA/rtrs: Use new shared CQ mechanism

[PATCHv2 for-next 1/2] RDMA/rtrs: Use new shared CQ mechanism

[Jack Wang]

Has the driver use shared CQs providing ~10%-20% improvement during
test.
Instead of opening a CQ for each QP per connection, a CQ for each QP
will be provided by the RDMA core driver that will be shared between
the QPs on that core reducing interrupt overhead.

Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/ulp/rtrs/rtrs-clt.c | 10 +++++-----
 drivers/infiniband/ulp/rtrs/rtrs-pri.h |  1 +
 drivers/infiniband/ulp/rtrs/rtrs-srv.c | 10 +++++-----
 drivers/infiniband/ulp/rtrs/rtrs.c     |  9 +++++----
 4 files changed, 16 insertions(+), 14 deletions(-)

diff --git a/drivers/infiniband/ulp/rtrs/rtrs-clt.c b/drivers/infiniband/ulp/rtrs/rtrs-clt.c
index 0a08b4b742a3..4e9cf06cc17a 100644
--- a/drivers/infiniband/ulp/rtrs/rtrs-clt.c
+++ b/drivers/infiniband/ulp/rtrs/rtrs-clt.c
@@ -325,7 +325,7 @@ static void rtrs_rdma_error_recovery(struct rtrs_clt_con *con)
 
 static void rtrs_clt_fast_reg_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	struct rtrs_clt_con *con = cq->cq_context;
+	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
 
 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
 		rtrs_err(con->c.sess, "Failed IB_WR_REG_MR: %s\n",
@@ -345,7 +345,7 @@ static void rtrs_clt_inv_rkey_done(struct ib_cq *cq, struct ib_wc *wc)
 {
 	struct rtrs_clt_io_req *req =
 		container_of(wc->wr_cqe, typeof(*req), inv_cqe);
-	struct rtrs_clt_con *con = cq->cq_context;
+	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
 
 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
 		rtrs_err(con->c.sess, "Failed IB_WR_LOCAL_INV: %s\n",
@@ -586,7 +586,7 @@ static int rtrs_post_recv_empty_x2(struct rtrs_con *con, struct ib_cqe *cqe)
 
 static void rtrs_clt_rdma_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	struct rtrs_clt_con *con = cq->cq_context;
+	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
 	struct rtrs_clt_sess *sess = to_clt_sess(con->c.sess);
 	u32 imm_type, imm_payload;
 	bool w_inval = false;
@@ -2241,7 +2241,7 @@ static int init_conns(struct rtrs_clt_sess *sess)
 
 static void rtrs_clt_info_req_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	struct rtrs_clt_con *con = cq->cq_context;
+	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
 	struct rtrs_clt_sess *sess = to_clt_sess(con->c.sess);
 	struct rtrs_iu *iu;
 
@@ -2323,7 +2323,7 @@ static int process_info_rsp(struct rtrs_clt_sess *sess,
 
 static void rtrs_clt_info_rsp_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	struct rtrs_clt_con *con = cq->cq_context;
+	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
 	struct rtrs_clt_sess *sess = to_clt_sess(con->c.sess);
 	struct rtrs_msg_info_rsp *msg;
 	enum rtrs_clt_state state;
diff --git a/drivers/infiniband/ulp/rtrs/rtrs-pri.h b/drivers/infiniband/ulp/rtrs/rtrs-pri.h
index 8caad0a2322b..1b31bda9ca78 100644
--- a/drivers/infiniband/ulp/rtrs/rtrs-pri.h
+++ b/drivers/infiniband/ulp/rtrs/rtrs-pri.h
@@ -91,6 +91,7 @@ struct rtrs_con {
 	struct ib_cq		*cq;
 	struct rdma_cm_id	*cm_id;
 	unsigned int		cid;
+	u16                     cq_size;
 };
 
 struct rtrs_sess {
diff --git a/drivers/infiniband/ulp/rtrs/rtrs-srv.c b/drivers/infiniband/ulp/rtrs/rtrs-srv.c
index d071809e3ed2..37ba121564a2 100644
--- a/drivers/infiniband/ulp/rtrs/rtrs-srv.c
+++ b/drivers/infiniband/ulp/rtrs/rtrs-srv.c
@@ -199,7 +199,7 @@ static void rtrs_srv_wait_ops_ids(struct rtrs_srv_sess *sess)
 
 static void rtrs_srv_reg_mr_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	struct rtrs_srv_con *con = cq->cq_context;
+	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
 	struct rtrs_sess *s = con->c.sess;
 	struct rtrs_srv_sess *sess = to_srv_sess(s);
 
@@ -720,7 +720,7 @@ static void rtrs_srv_stop_hb(struct rtrs_srv_sess *sess)
 
 static void rtrs_srv_info_rsp_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	struct rtrs_srv_con *con = cq->cq_context;
+	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
 	struct rtrs_sess *s = con->c.sess;
 	struct rtrs_srv_sess *sess = to_srv_sess(s);
 	struct rtrs_iu *iu;
@@ -862,7 +862,7 @@ static int process_info_req(struct rtrs_srv_con *con,
 
 static void rtrs_srv_info_req_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	struct rtrs_srv_con *con = cq->cq_context;
+	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
 	struct rtrs_sess *s = con->c.sess;
 	struct rtrs_srv_sess *sess = to_srv_sess(s);
 	struct rtrs_msg_info_req *msg;
@@ -1110,7 +1110,7 @@ static void rtrs_srv_inv_rkey_done(struct ib_cq *cq, struct ib_wc *wc)
 {
 	struct rtrs_srv_mr *mr =
 		container_of(wc->wr_cqe, typeof(*mr), inv_cqe);
-	struct rtrs_srv_con *con = cq->cq_context;
+	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
 	struct rtrs_sess *s = con->c.sess;
 	struct rtrs_srv_sess *sess = to_srv_sess(s);
 	struct rtrs_srv *srv = sess->srv;
@@ -1167,7 +1167,7 @@ static void rtrs_rdma_process_wr_wait_list(struct rtrs_srv_con *con)
 
 static void rtrs_srv_rdma_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	struct rtrs_srv_con *con = cq->cq_context;
+	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
 	struct rtrs_sess *s = con->c.sess;
 	struct rtrs_srv_sess *sess = to_srv_sess(s);
 	struct rtrs_srv *srv = sess->srv;
diff --git a/drivers/infiniband/ulp/rtrs/rtrs.c b/drivers/infiniband/ulp/rtrs/rtrs.c
index d13aff0aa816..64f2fd63e9c5 100644
--- a/drivers/infiniband/ulp/rtrs/rtrs.c
+++ b/drivers/infiniband/ulp/rtrs/rtrs.c
@@ -218,14 +218,15 @@ static int create_cq(struct rtrs_con *con, int cq_vector, u16 cq_size,
 	struct rdma_cm_id *cm_id = con->cm_id;
 	struct ib_cq *cq;
 
-	cq = ib_alloc_cq(cm_id->device, con, cq_size,
-			 cq_vector, poll_ctx);
+	cq = ib_cq_pool_get(cm_id->device, cq_size,
+			    cq_vector, poll_ctx);
 	if (IS_ERR(cq)) {
 		rtrs_err(con->sess, "Creating completion queue failed, errno: %ld\n",
 			  PTR_ERR(cq));
 		return PTR_ERR(cq);
 	}
 	con->cq = cq;
+	con->cq_size = cq_size;
 
 	return 0;
 }
@@ -273,7 +274,7 @@ int rtrs_cq_qp_create(struct rtrs_sess *sess, struct rtrs_con *con,
 	err = create_qp(con, sess->dev->ib_pd, max_send_wr, max_recv_wr,
 			max_send_sge);
 	if (err) {
-		ib_free_cq(con->cq);
+		ib_cq_pool_put(con->cq, con->cq_size);
 		con->cq = NULL;
 		return err;
 	}
@@ -290,7 +291,7 @@ void rtrs_cq_qp_destroy(struct rtrs_con *con)
 		con->qp = NULL;
 	}
 	if (con->cq) {
-		ib_free_cq(con->cq);
+		ib_cq_pool_put(con->cq, con->cq_size);
 		con->cq = NULL;
 	}
 }
-- 
2.25.1

[PATCHv2 for-next 2/2] RDMA/rtrs-clt: Use rdma_event_msg in log

[Jack Wang]

It's easier to understand string instead of enum.

Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/ulp/rtrs/rtrs-clt.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/ulp/rtrs/rtrs-clt.c b/drivers/infiniband/ulp/rtrs/rtrs-clt.c
index 4e9cf06cc17a..f95955fc2992 100644
--- a/drivers/infiniband/ulp/rtrs/rtrs-clt.c
+++ b/drivers/infiniband/ulp/rtrs/rtrs-clt.c
@@ -1853,12 +1853,14 @@ static int rtrs_clt_rdma_cm_handler(struct rdma_cm_id *cm_id,
 	case RDMA_CM_EVENT_UNREACHABLE:
 	case RDMA_CM_EVENT_ADDR_CHANGE:
 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
-		rtrs_wrn(s, "CM error event %d\n", ev->event);
+		rtrs_wrn(s, "CM error (CM event: %s, err: %d)\n",
+			 rdma_event_msg(ev->event), ev->status);
 		cm_err = -ECONNRESET;
 		break;
 	case RDMA_CM_EVENT_ADDR_ERROR:
 	case RDMA_CM_EVENT_ROUTE_ERROR:
-		rtrs_wrn(s, "CM error event %d\n", ev->event);
+		rtrs_wrn(s, "CM error (CM event: %s, err: %d)\n",
+			 rdma_event_msg(ev->event), ev->status);
 		cm_err = -EHOSTUNREACH;
 		break;
 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
@@ -1868,7 +1870,8 @@ static int rtrs_clt_rdma_cm_handler(struct rdma_cm_id *cm_id,
 		rtrs_clt_close_conns(sess, false);
 		return 0;
 	default:
-		rtrs_err(s, "Unexpected RDMA CM event (%d)\n", ev->event);
+		rtrs_err(s, "Unexpected RDMA CM error (CM event: %s, err: %d)\n",
+			 rdma_event_msg(ev->event), ev->status);
 		cm_err = -ECONNRESET;
 		break;
 	}
-- 
2.25.1

Re: [PATCHv2 for-next 1/2] RDMA/rtrs: Use new shared CQ mechanism

[Jason Gunthorpe]

On Mon, Feb 22, 2021 at 03:15:50PM +0100, Jack Wang wrote:
> Has the driver use shared CQs providing ~10%-20% improvement during
> test.
> Instead of opening a CQ for each QP per connection, a CQ for each QP
> will be provided by the RDMA core driver that will be shared between
> the QPs on that core reducing interrupt overhead.
> 
> Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
> Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
> ---
>  drivers/infiniband/ulp/rtrs/rtrs-clt.c | 10 +++++-----
>  drivers/infiniband/ulp/rtrs/rtrs-pri.h |  1 +
>  drivers/infiniband/ulp/rtrs/rtrs-srv.c | 10 +++++-----
>  drivers/infiniband/ulp/rtrs/rtrs.c     |  9 +++++----
>  4 files changed, 16 insertions(+), 14 deletions(-)

Applied to for-next, thanks

Jason

Re: [PATCHv2 for-next 2/2] RDMA/rtrs-clt: Use rdma_event_msg in log

[Jason Gunthorpe]

On Mon, Feb 22, 2021 at 03:15:51PM +0100, Jack Wang wrote:
> It's easier to understand string instead of enum.
> 
> Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
> Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
> ---
>  drivers/infiniband/ulp/rtrs/rtrs-clt.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)

Applied to for-next, thanks

Jason