| Message ID | 20210219115019.38981-1-jinpu.wang@cloud.ionos.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [1/2] RDMA/rtrs: Use new shared CQ mechanism | expand |
On Fri, Feb 19, 2021 at 12:50:18PM +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> > --- > 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 | 11 +++++++---- > 4 files changed, 18 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..d5ec78280937 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; > } > @@ -273,8 +274,9 @@ 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; > + con->cq_size = 0; It is better do not clear fields that not used, it hides bugs. Other than that. Thanks, Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
On Sun, Feb 21, 2021 at 7:23 AM Leon Romanovsky <leon@kernel.org> wrote: > > On Fri, Feb 19, 2021 at 12:50:18PM +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> > > --- > > 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 | 11 +++++++---- > > 4 files changed, 18 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..d5ec78280937 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; > > } > > @@ -273,8 +274,9 @@ 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; > > + con->cq_size = 0; > > It is better do not clear fields that not used, it hides bugs. > Other than that. I feel rewinding on the error path by resetting the cq_size is the right thing to do. > > Thanks, > Reviewed-by: Leon Romanovsky <leonro@nvidia.com> Thanks for the review!
On Mon, Feb 22, 2021 at 11:31:55AM +0100, Jinpu Wang wrote: > On Sun, Feb 21, 2021 at 7:23 AM Leon Romanovsky <leon@kernel.org> wrote: > > > > On Fri, Feb 19, 2021 at 12:50:18PM +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> > > > --- > > > 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 | 11 +++++++---- > > > 4 files changed, 18 insertions(+), 14 deletions(-) <...> > > > 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; > > > + con->cq_size = 0; > > > > It is better do not clear fields that not used, it hides bugs. > > Other than that. > I feel rewinding on the error path by resetting the cq_size is the > right thing to do. It is the right thing to do if down the road you have an access to cq_size with if (..) check. Other than that, it is not right thing to do. Thanks
On Mon, Feb 22, 2021 at 2:17 PM Leon Romanovsky <leon@kernel.org> wrote: > > On Mon, Feb 22, 2021 at 11:31:55AM +0100, Jinpu Wang wrote: > > On Sun, Feb 21, 2021 at 7:23 AM Leon Romanovsky <leon@kernel.org> wrote: > > > > > > On Fri, Feb 19, 2021 at 12:50:18PM +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> > > > > --- > > > > 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 | 11 +++++++---- > > > > 4 files changed, 18 insertions(+), 14 deletions(-) > > <...> > > > > > 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; > > > > + con->cq_size = 0; > > > > > > It is better do not clear fields that not used, it hides bugs. > > > Other than that. > > I feel rewinding on the error path by resetting the cq_size is the > > right thing to do. > > It is the right thing to do if down the road you have an access to > cq_size with if (..) check. Other than that, it is not right thing to > do. > > Thanks Double checked, will remove the cq_size clear lines in v2. Thanks!
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..d5ec78280937 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; } @@ -273,8 +274,9 @@ 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; + con->cq_size = 0; return err; } con->sess = sess; @@ -290,8 +292,9 @@ 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; + con->cq_size = 0; } } EXPORT_SYMBOL_GPL(rtrs_cq_qp_destroy);
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> --- 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 | 11 +++++++---- 4 files changed, 18 insertions(+), 14 deletions(-)