IB/core: add max_send_sge and max_recv_sge attributes

This patch replaces the ib_device_attr.max_sge with max_send_sge and
max_recv_sge. It allows ulps to take advantage of devices that have very
different send and recv sge depths.  For example cxgb4 has a max_recv_sge
of 4, yet a max_send_sge of 16.  Splitting out these attributes allows
much more efficient use of the SQ for cxgb4 with ulps that use the RDMA_RW
API. Consider a large RDMA WRITE that has 16 scattergather entries.
With max_sge of 4, the ulp would send 4 WRITE WRs, but with max_sge of
16, it can be done with 1 WRITE WR.
Acked-by: NSagi Grimberg <sagi@grimberg.me>
Acked-by: NChristoph Hellwig <hch@lst.de>
Acked-by: NSelvin Xavier <selvin.xavier@broadcom.com>
Acked-by: NShiraz Saleem <shiraz.saleem@intel.com>
Acked-by: NDennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NJason Gunthorpe <jgg@mellanox.com>

drivers/infiniband/core/uverbs_cmd.c

@@ -189,7 +189,7 @@ static void copy_query_dev_fields(struct ib_uverbs_file *file,
 	resp->max_qp		= attr->max_qp;
 	resp->max_qp_wr		= attr->max_qp_wr;
 	resp->device_cap_flags	= lower_32_bits(attr->device_cap_flags);
-	resp->max_sge		= attr->max_sge;
+	resp->max_sge		= min(attr->max_send_sge, attr->max_recv_sge);
 	resp->max_sge_rd	= attr->max_sge_rd;
 	resp->max_cq		= attr->max_cq;
 	resp->max_cqe		= attr->max_cqe;

drivers/infiniband/hw/bnxt_re/ib_verbs.c

@@ -166,7 +166,8 @@ int bnxt_re_query_device(struct ib_device *ibdev,
 				    | IB_DEVICE_MEM_WINDOW
 				    | IB_DEVICE_MEM_WINDOW_TYPE_2B
 				    | IB_DEVICE_MEM_MGT_EXTENSIONS;
-	ib_attr->max_sge = dev_attr->max_qp_sges;
+	ib_attr->max_send_sge = dev_attr->max_qp_sges;
+	ib_attr->max_recv_sge = dev_attr->max_qp_sges;
 	ib_attr->max_sge_rd = dev_attr->max_qp_sges;
 	ib_attr->max_cq = dev_attr->max_cq;
 	ib_attr->max_cqe = dev_attr->max_cq_wqes;

drivers/infiniband/hw/cxgb3/iwch_provider.c

@@ -1103,7 +1103,8 @@ static int iwch_query_device(struct ib_device *ibdev, struct ib_device_attr *pro
 	props->max_mr_size = dev->attr.max_mr_size;
 	props->max_qp = dev->attr.max_qps;
 	props->max_qp_wr = dev->attr.max_wrs;
-	props->max_sge = dev->attr.max_sge_per_wr;
+	props->max_send_sge = dev->attr.max_sge_per_wr;
+	props->max_recv_sge = dev->attr.max_sge_per_wr;
 	props->max_sge_rd = 1;
 	props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
 	props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;

drivers/infiniband/hw/cxgb4/provider.c

@@ -343,7 +343,8 @@ static int c4iw_query_device(struct ib_device *ibdev, struct ib_device_attr *pro
 	props->max_mr_size = T4_MAX_MR_SIZE;
 	props->max_qp = dev->rdev.lldi.vr->qp.size / 2;
 	props->max_qp_wr = dev->rdev.hw_queue.t4_max_qp_depth;
-	props->max_sge = T4_MAX_RECV_SGE;
+	props->max_send_sge = min(T4_MAX_SEND_SGE, T4_MAX_WRITE_SGE);
+	props->max_recv_sge = T4_MAX_RECV_SGE;
 	props->max_sge_rd = 1;
 	props->max_res_rd_atom = dev->rdev.lldi.max_ird_adapter;
 	props->max_qp_rd_atom = min(dev->rdev.lldi.max_ordird_qp,

drivers/infiniband/hw/hfi1/verbs.c

@@ -1410,7 +1410,8 @@ static void hfi1_fill_device_attr(struct hfi1_devdata *dd)
 	rdi->dparms.props.max_fast_reg_page_list_len = UINT_MAX;
 	rdi->dparms.props.max_qp = hfi1_max_qps;
 	rdi->dparms.props.max_qp_wr = hfi1_max_qp_wrs;
-	rdi->dparms.props.max_sge = hfi1_max_sges;
+	rdi->dparms.props.max_send_sge = hfi1_max_sges;
+	rdi->dparms.props.max_recv_sge = hfi1_max_sges;
 	rdi->dparms.props.max_sge_rd = hfi1_max_sges;
 	rdi->dparms.props.max_cq = hfi1_max_cqs;
 	rdi->dparms.props.max_ah = hfi1_max_ahs;

drivers/infiniband/hw/hns/hns_roce_main.c

@@ -206,7 +206,8 @@ static int hns_roce_query_device(struct ib_device *ib_dev,
 	props->max_qp_wr = hr_dev->caps.max_wqes;
 	props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
 				  IB_DEVICE_RC_RNR_NAK_GEN;
-	props->max_sge = max(hr_dev->caps.max_sq_sg, hr_dev->caps.max_rq_sg);
+	props->max_send_sge = hr_dev->caps.max_sq_sg;
+	props->max_recv_sge = hr_dev->caps.max_rq_sg;
 	props->max_sge_rd = 1;
 	props->max_cq = hr_dev->caps.num_cqs;
 	props->max_cqe = hr_dev->caps.max_cqes;

drivers/infiniband/hw/i40iw/i40iw_verbs.c

@@ -71,7 +71,8 @@ static int i40iw_query_device(struct ib_device *ibdev,
 	props->max_mr_size = I40IW_MAX_OUTBOUND_MESSAGE_SIZE;
 	props->max_qp = iwdev->max_qp - iwdev->used_qps;
 	props->max_qp_wr = I40IW_MAX_QP_WRS;
-	props->max_sge = I40IW_MAX_WQ_FRAGMENT_COUNT;
+	props->max_send_sge = I40IW_MAX_WQ_FRAGMENT_COUNT;
+	props->max_recv_sge = I40IW_MAX_WQ_FRAGMENT_COUNT;
 	props->max_cq = iwdev->max_cq - iwdev->used_cqs;
 	props->max_cqe = iwdev->max_cqe;
 	props->max_mr = iwdev->max_mr - iwdev->used_mrs;

drivers/infiniband/hw/mlx4/main.c

@@ -517,8 +517,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
 	props->max_qp		   = dev->dev->quotas.qp;
 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
-	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
-					 dev->dev->caps.max_rq_sg);
+	props->max_send_sge	   = dev->dev->caps.max_sq_sg;
+	props->max_recv_sge	   = dev->dev->caps.max_rq_sg;
 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
 	props->max_cq		   = dev->dev->quotas.cq;
 	props->max_cqe		   = dev->dev->caps.max_cqes;

drivers/infiniband/hw/mlx5/main.c

@@ -888,7 +888,8 @@ static int mlx5_ib_query_device(struct ib_device *ibdev,
 	max_sq_sg = (max_sq_desc - sizeof(struct mlx5_wqe_ctrl_seg) -
 		     sizeof(struct mlx5_wqe_raddr_seg)) /
 		sizeof(struct mlx5_wqe_data_seg);
-	props->max_sge = min(max_rq_sg, max_sq_sg);
+	props->max_send_sge = max_sq_sg;
+	props->max_recv_sge = max_rq_sg;
 	props->max_sge_rd	   = MLX5_MAX_SGE_RD;
 	props->max_cq		   = 1 << MLX5_CAP_GEN(mdev, log_max_cq);
 	props->max_cqe = (1 << MLX5_CAP_GEN(mdev, log_max_cq_sz)) - 1;

drivers/infiniband/hw/mthca/mthca_provider.c

@@ -96,8 +96,9 @@ static int mthca_query_device(struct ib_device *ibdev, struct ib_device_attr *pr
 	props->page_size_cap       = mdev->limits.page_size_cap;
 	props->max_qp              = mdev->limits.num_qps - mdev->limits.reserved_qps;
 	props->max_qp_wr           = mdev->limits.max_wqes;
-	props->max_sge             = mdev->limits.max_sg;
-	props->max_sge_rd          = props->max_sge;
+	props->max_send_sge        = mdev->limits.max_sg;
+	props->max_recv_sge        = mdev->limits.max_sg;
+	props->max_sge_rd          = mdev->limits.max_sg;
 	props->max_cq              = mdev->limits.num_cqs - mdev->limits.reserved_cqs;
 	props->max_cqe             = mdev->limits.max_cqes;
 	props->max_mr              = mdev->limits.num_mpts - mdev->limits.reserved_mrws;

drivers/infiniband/hw/nes/nes_verbs.c

@@ -436,7 +436,8 @@ static int nes_query_device(struct ib_device *ibdev, struct ib_device_attr *prop
 	props->max_mr_size = 0x80000000;
 	props->max_qp = nesibdev->max_qp;
 	props->max_qp_wr = nesdev->nesadapter->max_qp_wr - 2;
-	props->max_sge = nesdev->nesadapter->max_sge;
+	props->max_send_sge = nesdev->nesadapter->max_sge;
+	props->max_recv_sge = nesdev->nesadapter->max_sge;
 	props->max_cq = nesibdev->max_cq;
 	props->max_cqe = nesdev->nesadapter->max_cqe;
 	props->max_mr = nesibdev->max_mr;

drivers/infiniband/hw/ocrdma/ocrdma_verbs.c

@@ -89,7 +89,8 @@ int ocrdma_query_device(struct ib_device *ibdev, struct ib_device_attr *attr,
 					IB_DEVICE_SYS_IMAGE_GUID |
 					IB_DEVICE_LOCAL_DMA_LKEY |
 					IB_DEVICE_MEM_MGT_EXTENSIONS;
-	attr->max_sge = min(dev->attr.max_send_sge, dev->attr.max_recv_sge);
+	attr->max_send_sge = dev->attr.max_send_sge;
+	attr->max_recv_sge = dev->attr.max_recv_sge;
 	attr->max_sge_rd = dev->attr.max_rdma_sge;
 	attr->max_cq = dev->attr.max_cq;
 	attr->max_cqe = dev->attr.max_cqe;

drivers/infiniband/hw/qedr/verbs.c

@@ -112,7 +112,8 @@ int qedr_query_device(struct ib_device *ibdev,
 	    IB_DEVICE_RC_RNR_NAK_GEN |
 	    IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_MGT_EXTENSIONS;
 
-	attr->max_sge = qattr->max_sge;
+	attr->max_send_sge = qattr->max_sge;
+	attr->max_recv_sge = qattr->max_sge;
 	attr->max_sge_rd = qattr->max_sge;
 	attr->max_cq = qattr->max_cq;
 	attr->max_cqe = qattr->max_cqe;

drivers/infiniband/hw/qib/qib_verbs.c

@@ -1489,7 +1489,8 @@ static void qib_fill_device_attr(struct qib_devdata *dd)
 	rdi->dparms.props.max_mr_size = ~0ULL;
 	rdi->dparms.props.max_qp = ib_qib_max_qps;
 	rdi->dparms.props.max_qp_wr = ib_qib_max_qp_wrs;
-	rdi->dparms.props.max_sge = ib_qib_max_sges;
+	rdi->dparms.props.max_send_sge = ib_qib_max_sges;
+	rdi->dparms.props.max_recv_sge = ib_qib_max_sges;
 	rdi->dparms.props.max_sge_rd = ib_qib_max_sges;
 	rdi->dparms.props.max_cq = ib_qib_max_cqs;
 	rdi->dparms.props.max_cqe = ib_qib_max_cqes;

drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c

@@ -82,7 +82,8 @@ int pvrdma_query_device(struct ib_device *ibdev,
 	props->max_qp = dev->dsr->caps.max_qp;
 	props->max_qp_wr = dev->dsr->caps.max_qp_wr;
 	props->device_cap_flags = dev->dsr->caps.device_cap_flags;
-	props->max_sge = dev->dsr->caps.max_sge;
+	props->max_send_sge = dev->dsr->caps.max_sge;
+	props->max_recv_sge = dev->dsr->caps.max_sge;
 	props->max_sge_rd = PVRDMA_GET_CAP(dev, dev->dsr->caps.max_sge,
 					   dev->dsr->caps.max_sge_rd);
 	props->max_srq = dev->dsr->caps.max_srq;

drivers/infiniband/sw/rdmavt/qp.c

@@ -780,14 +780,15 @@ struct ib_qp *rvt_create_qp(struct ib_pd *ibpd,
 	if (!rdi)
 		return ERR_PTR(-EINVAL);
 
-	if (init_attr->cap.max_send_sge > rdi->dparms.props.max_sge ||
+	if (init_attr->cap.max_send_sge > rdi->dparms.props.max_send_sge ||
 	    init_attr->cap.max_send_wr > rdi->dparms.props.max_qp_wr ||
 	    init_attr->create_flags)
 		return ERR_PTR(-EINVAL);
 
 	/* Check receive queue parameters if no SRQ is specified. */
 	if (!init_attr->srq) {
-		if (init_attr->cap.max_recv_sge > rdi->dparms.props.max_sge ||
+		if (init_attr->cap.max_recv_sge >
+		    rdi->dparms.props.max_recv_sge ||
 		    init_attr->cap.max_recv_wr > rdi->dparms.props.max_qp_wr)
 			return ERR_PTR(-EINVAL);
 

drivers/infiniband/sw/rxe/rxe.c

@@ -91,7 +91,8 @@ static void rxe_init_device_param(struct rxe_dev *rxe)
 	rxe->attr.max_qp			= RXE_MAX_QP;
 	rxe->attr.max_qp_wr			= RXE_MAX_QP_WR;
 	rxe->attr.device_cap_flags		= RXE_DEVICE_CAP_FLAGS;
-	rxe->attr.max_sge			= RXE_MAX_SGE;
+	rxe->attr.max_send_sge			= RXE_MAX_SGE;
+	rxe->attr.max_recv_sge			= RXE_MAX_SGE;
 	rxe->attr.max_sge_rd			= RXE_MAX_SGE_RD;
 	rxe->attr.max_cq			= RXE_MAX_CQ;
 	rxe->attr.max_cqe			= (1 << RXE_MAX_LOG_CQE) - 1;

drivers/infiniband/sw/rxe/rxe_qp.c

@@ -49,9 +49,9 @@ static int rxe_qp_chk_cap(struct rxe_dev *rxe, struct ib_qp_cap *cap,
 		goto err1;
 	}
 
-	if (cap->max_send_sge > rxe->attr.max_sge) {
+	if (cap->max_send_sge > rxe->attr.max_send_sge) {
 		pr_warn("invalid send sge = %d > %d\n",
-			cap->max_send_sge, rxe->attr.max_sge);
+			cap->max_send_sge, rxe->attr.max_send_sge);
 		goto err1;
 	}
 
@@ -62,9 +62,9 @@ static int rxe_qp_chk_cap(struct rxe_dev *rxe, struct ib_qp_cap *cap,
 			goto err1;
 		}
 
-		if (cap->max_recv_sge > rxe->attr.max_sge) {
+		if (cap->max_recv_sge > rxe->attr.max_recv_sge) {
 			pr_warn("invalid recv sge = %d > %d\n",
-				cap->max_recv_sge, rxe->attr.max_sge);
+				cap->max_recv_sge, rxe->attr.max_recv_sge);
 			goto err1;
 		}
 	}

drivers/infiniband/ulp/ipoib/ipoib_cm.c

@@ -1068,8 +1068,8 @@ static struct ib_qp *ipoib_cm_create_tx_qp(struct net_device *dev, struct ipoib_
 	struct ib_qp *tx_qp;
 
 	if (dev->features & NETIF_F_SG)
-		attr.cap.max_send_sge =
-			min_t(u32, priv->ca->attrs.max_sge, MAX_SKB_FRAGS + 1);
+		attr.cap.max_send_sge = min_t(u32, priv->ca->attrs.max_send_sge,
+					      MAX_SKB_FRAGS + 1);
 
 	tx_qp = ib_create_qp(priv->pd, &attr);
 	tx->max_send_sge = attr.cap.max_send_sge;

drivers/infiniband/ulp/ipoib/ipoib_verbs.c

@@ -147,7 +147,7 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 		.cap = {
 			.max_send_wr  = ipoib_sendq_size,
 			.max_recv_wr  = ipoib_recvq_size,
-			.max_send_sge = min_t(u32, priv->ca->attrs.max_sge,
+			.max_send_sge = min_t(u32, priv->ca->attrs.max_send_sge,
 					      MAX_SKB_FRAGS + 1),
 			.max_recv_sge = IPOIB_UD_RX_SG
 		},

drivers/infiniband/ulp/isert/ib_isert.c

@@ -136,7 +136,7 @@ isert_create_qp(struct isert_conn *isert_conn,
 	attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS + 1;
 	attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS + 1;
 	attr.cap.max_rdma_ctxs = ISCSI_DEF_XMIT_CMDS_MAX;
-	attr.cap.max_send_sge = device->ib_device->attrs.max_sge;
+	attr.cap.max_send_sge = device->ib_device->attrs.max_send_sge;
 	attr.cap.max_recv_sge = 1;
 	attr.sq_sig_type = IB_SIGNAL_REQ_WR;
 	attr.qp_type = IB_QPT_RC;
@@ -299,7 +299,8 @@ isert_create_device_ib_res(struct isert_device *device)
 	struct ib_device *ib_dev = device->ib_device;
 	int ret;
 
-	isert_dbg("devattr->max_sge: %d\n", ib_dev->attrs.max_sge);
+	isert_dbg("devattr->max_send_sge: %d devattr->max_recv_sge %d\n",
+		  ib_dev->attrs.max_send_sge, ib_dev->attrs.max_recv_sge);
 	isert_dbg("devattr->max_sge_rd: %d\n", ib_dev->attrs.max_sge_rd);
 
 	ret = isert_alloc_comps(device);

drivers/infiniband/ulp/srpt/ib_srpt.c

@@ -1753,13 +1753,15 @@ static int srpt_create_ch_ib(struct srpt_rdma_ch *ch)
 	 */
 	qp_init->cap.max_send_wr = min(sq_size / 2, attrs->max_qp_wr);
 	qp_init->cap.max_rdma_ctxs = sq_size / 2;
-	qp_init->cap.max_send_sge = min(attrs->max_sge, SRPT_MAX_SG_PER_WQE);
+	qp_init->cap.max_send_sge = min(attrs->max_send_sge,
+					SRPT_MAX_SG_PER_WQE);
 	qp_init->port_num = ch->sport->port;
 	if (sdev->use_srq) {
 		qp_init->srq = sdev->srq;
 	} else {
 		qp_init->cap.max_recv_wr = ch->rq_size;
-		qp_init->cap.max_recv_sge = qp_init->cap.max_send_sge;
+		qp_init->cap.max_recv_sge = min(attrs->max_recv_sge,
+						SRPT_MAX_SG_PER_WQE);
 	}
 
 	if (ch->using_rdma_cm) {

drivers/nvme/target/rdma.c

@@ -874,7 +874,7 @@ static int nvmet_rdma_create_queue_ib(struct nvmet_rdma_queue *queue)
 	qp_attr.cap.max_send_wr = queue->send_queue_size + 1;
 	qp_attr.cap.max_rdma_ctxs = queue->send_queue_size;
 	qp_attr.cap.max_send_sge = max(ndev->device->attrs.max_sge_rd,
-					ndev->device->attrs.max_sge);
+					ndev->device->attrs.max_send_sge);
 
 	if (ndev->srq) {
 		qp_attr.srq = ndev->srq;

fs/cifs/smbdirect.c

@@ -1661,9 +1661,16 @@ static struct smbd_connection *_smbd_get_connection(
 	info->max_receive_size = smbd_max_receive_size;
 	info->keep_alive_interval = smbd_keep_alive_interval;
 
-	if (info->id->device->attrs.max_sge < SMBDIRECT_MAX_SGE) {
-		log_rdma_event(ERR, "warning: device max_sge = %d too small\n",
-			info->id->device->attrs.max_sge);
+	if (info->id->device->attrs.max_send_sge < SMBDIRECT_MAX_SGE) {
+		log_rdma_event(ERR,
+			"warning: device max_send_sge = %d too small\n",
+			info->id->device->attrs.max_send_sge);
+		log_rdma_event(ERR, "Queue Pair creation may fail\n");
+	}
+	if (info->id->device->attrs.max_recv_sge < SMBDIRECT_MAX_SGE) {
+		log_rdma_event(ERR,
+			"warning: device max_recv_sge = %d too small\n",
+			info->id->device->attrs.max_recv_sge);
 		log_rdma_event(ERR, "Queue Pair creation may fail\n");
 	}
 

include/rdma/ib_verbs.h

@@ -345,7 +345,8 @@ struct ib_device_attr {
 	int			max_qp;
 	int			max_qp_wr;
 	u64			device_cap_flags;
-	int			max_sge;
+	int			max_send_sge;
+	int			max_recv_sge;
 	int			max_sge_rd;
 	int			max_cq;
 	int			max_cqe;

net/rds/ib.c

@@ -143,7 +143,7 @@ static void rds_ib_add_one(struct ib_device *device)
 	INIT_WORK(&rds_ibdev->free_work, rds_ib_dev_free);
 
 	rds_ibdev->max_wrs = device->attrs.max_qp_wr;
-	rds_ibdev->max_sge = min(device->attrs.max_sge, RDS_IB_MAX_SGE);
+	rds_ibdev->max_sge = min(device->attrs.max_send_sge, RDS_IB_MAX_SGE);
 
 	has_fr = (device->attrs.device_cap_flags &
 		  IB_DEVICE_MEM_MGT_EXTENSIONS);

net/sunrpc/xprtrdma/svc_rdma_transport.c

@@ -476,7 +476,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 
 	/* Qualify the transport resource defaults with the
 	 * capabilities of this particular device */
-	newxprt->sc_max_send_sges = dev->attrs.max_sge;
+	newxprt->sc_max_send_sges = dev->attrs.max_send_sge;
 	/* transport hdr, head iovec, one page list entry, tail iovec */
 	if (newxprt->sc_max_send_sges < 4) {
 		pr_err("svcrdma: too few Send SGEs available (%d)\n",

net/sunrpc/xprtrdma/verbs.c

@@ -508,7 +508,7 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
 	unsigned int max_sge;
 	int rc;
 
-	max_sge = min_t(unsigned int, ia->ri_device->attrs.max_sge,
+	max_sge = min_t(unsigned int, ia->ri_device->attrs.max_send_sge,
 			RPCRDMA_MAX_SEND_SGES);
 	if (max_sge < RPCRDMA_MIN_SEND_SGES) {
 		pr_warn("rpcrdma: HCA provides only %d send SGEs\n", max_sge);