IB/ipath: Performance improvements via mmap of queues

Improve performance of userspace post receive, post SRQ receive, and
poll CQ operations for ipath by allowing userspace to directly mmap()
receive queues and completion queues.  This eliminates the copying
between userspace and the kernel in the data path.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

1 files changed, 103 insertions, 66 deletions

diff --git a/drivers/infiniband/hw/ipath/ipath_ud.c b/drivers/infiniband/hw/ipath/ipath_ud.c
index 3466129af80..82439fcfc2f 100644
--- a/drivers/infiniband/hw/ipath/ipath_ud.c
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c
@@ -34,7 +34,54 @@
 #include <rdma/ib_smi.h>
 
 #include "ipath_verbs.h"
-#include "ipath_common.h"
+#include "ipath_kernel.h"
+
+static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe,
+		    u32 *lengthp, struct ipath_sge_state *ss)
+{
+	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
+	int user = to_ipd(qp->ibqp.pd)->user;
+	int i, j, ret;
+	struct ib_wc wc;
+
+	*lengthp = 0;
+	for (i = j = 0; i < wqe->num_sge; i++) {
+		if (wqe->sg_list[i].length == 0)
+			continue;
+		/* Check LKEY */
+		if ((user && wqe->sg_list[i].lkey == 0) ||
+		    !ipath_lkey_ok(&dev->lk_table,
+				   j ? &ss->sg_list[j - 1] : &ss->sge,
+				   &wqe->sg_list[i], IB_ACCESS_LOCAL_WRITE))
+			goto bad_lkey;
+		*lengthp += wqe->sg_list[i].length;
+		j++;
+	}
+	ss->num_sge = j;
+	ret = 1;
+	goto bail;
+
+bad_lkey:
+	wc.wr_id = wqe->wr_id;
+	wc.status = IB_WC_LOC_PROT_ERR;
+	wc.opcode = IB_WC_RECV;
+	wc.vendor_err = 0;
+	wc.byte_len = 0;
+	wc.imm_data = 0;
+	wc.qp_num = qp->ibqp.qp_num;
+	wc.src_qp = 0;
+	wc.wc_flags = 0;
+	wc.pkey_index = 0;
+	wc.slid = 0;
+	wc.sl = 0;
+	wc.dlid_path_bits = 0;
+	wc.port_num = 0;
+	/* Signal solicited completion event. */
+	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+	ret = 0;
+bail:
+	return ret;
+}
 
 /**
  * ipath_ud_loopback - handle send on loopback QPs
@@ -46,6 +93,8 @@
  *
  * This is called from ipath_post_ud_send() to forward a WQE addressed
  * to the same HCA.
+ * Note that the receive interrupt handler may be calling ipath_ud_rcv()
+ * while this is being called.
  */
 static void ipath_ud_loopback(struct ipath_qp *sqp,
 			      struct ipath_sge_state *ss,
@@ -60,7 +109,11 @@ static void ipath_ud_loopback(struct ipath_qp *sqp,
 	struct ipath_srq *srq;
 	struct ipath_sge_state rsge;
 	struct ipath_sge *sge;
+	struct ipath_rwq *wq;
 	struct ipath_rwqe *wqe;
+	void (*handler)(struct ib_event *, void *);
+	u32 tail;
+	u32 rlen;
 
 	qp = ipath_lookup_qpn(&dev->qp_table, wr->wr.ud.remote_qpn);
 	if (!qp)
@@ -94,6 +147,13 @@ static void ipath_ud_loopback(struct ipath_qp *sqp,
 		wc->imm_data = 0;
 	}
 
+	if (wr->num_sge > 1) {
+		rsge.sg_list = kmalloc((wr->num_sge - 1) *
+					sizeof(struct ipath_sge),
+				       GFP_ATOMIC);
+	} else
+		rsge.sg_list = NULL;
+
 	/*
 	 * Get the next work request entry to find where to put the data.
 	 * Note that it is safe to drop the lock after changing rq->tail
@@ -101,37 +161,52 @@ static void ipath_ud_loopback(struct ipath_qp *sqp,
 	 */
 	if (qp->ibqp.srq) {
 		srq = to_isrq(qp->ibqp.srq);
+		handler = srq->ibsrq.event_handler;
 		rq = &srq->rq;
 	} else {
 		srq = NULL;
+		handler = NULL;
 		rq = &qp->r_rq;
 	}
+
 	spin_lock_irqsave(&rq->lock, flags);
-	if (rq->tail == rq->head) {
-		spin_unlock_irqrestore(&rq->lock, flags);
-		dev->n_pkt_drops++;
-		goto done;
+	wq = rq->wq;
+	tail = wq->tail;
+	while (1) {
+		if (unlikely(tail == wq->head)) {
+			spin_unlock_irqrestore(&rq->lock, flags);
+			dev->n_pkt_drops++;
+			goto bail_sge;
+		}
+		wqe = get_rwqe_ptr(rq, tail);
+		if (++tail >= rq->size)
+			tail = 0;
+		if (init_sge(qp, wqe, &rlen, &rsge))
+			break;
+		wq->tail = tail;
 	}
 	/* Silently drop packets which are too big. */
-	wqe = get_rwqe_ptr(rq, rq->tail);
-	if (wc->byte_len > wqe->length) {
+	if (wc->byte_len > rlen) {
 		spin_unlock_irqrestore(&rq->lock, flags);
 		dev->n_pkt_drops++;
-		goto done;
+		goto bail_sge;
 	}
+	wq->tail = tail;
 	wc->wr_id = wqe->wr_id;
-	rsge.sge = wqe->sg_list[0];
-	rsge.sg_list = wqe->sg_list + 1;
-	rsge.num_sge = wqe->num_sge;
-	if (++rq->tail >= rq->size)
-		rq->tail = 0;
-	if (srq && srq->ibsrq.event_handler) {
+	if (handler) {
 		u32 n;
 
-		if (rq->head < rq->tail)
-			n = rq->size + rq->head - rq->tail;
+		/*
+		 * validate head pointer value and compute
+		 * the number of remaining WQEs.
+		 */
+		n = wq->head;
+		if (n >= rq->size)
+			n = 0;
+		if (n < tail)
+			n += rq->size - tail;
 		else
-			n = rq->head - rq->tail;
+			n -= tail;
 		if (n < srq->limit) {
 			struct ib_event ev;
 
@@ -140,12 +215,12 @@ static void ipath_ud_loopback(struct ipath_qp *sqp,
 			ev.device = qp->ibqp.device;
 			ev.element.srq = qp->ibqp.srq;
 			ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
-			srq->ibsrq.event_handler(&ev,
-						 srq->ibsrq.srq_context);
+			handler(&ev, srq->ibsrq.srq_context);
 		} else
 			spin_unlock_irqrestore(&rq->lock, flags);
 	} else
 		spin_unlock_irqrestore(&rq->lock, flags);
+
 	ah_attr = &to_iah(wr->wr.ud.ah)->attr;
 	if (ah_attr->ah_flags & IB_AH_GRH) {
 		ipath_copy_sge(&rsge, &ah_attr->grh, sizeof(struct ib_grh));
@@ -186,7 +261,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp,
 	wc->src_qp = sqp->ibqp.qp_num;
 	/* XXX do we know which pkey matched? Only needed for GSI. */
 	wc->pkey_index = 0;
-	wc->slid = ipath_layer_get_lid(dev->dd) |
+	wc->slid = dev->dd->ipath_lid |
 		(ah_attr->src_path_bits &
 		 ((1 << (dev->mkeyprot_resv_lmc & 7)) - 1));
 	wc->sl = ah_attr->sl;
@@ -196,6 +271,8 @@ static void ipath_ud_loopback(struct ipath_qp *sqp,
 	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), wc,
 		       wr->send_flags & IB_SEND_SOLICITED);
 
+bail_sge:
+	kfree(rsge.sg_list);
 done:
 	if (atomic_dec_and_test(&qp->refcount))
 		wake_up(&qp->wait);
@@ -433,13 +510,9 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 	int opcode;
 	u32 hdrsize;
 	u32 pad;
-	unsigned long flags;
 	struct ib_wc wc;
 	u32 qkey;
 	u32 src_qp;
-	struct ipath_rq *rq;
-	struct ipath_srq *srq;
-	struct ipath_rwqe *wqe;
 	u16 dlid;
 	int header_in_data;
 
@@ -547,19 +620,10 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 
 	/*
 	 * Get the next work request entry to find where to put the data.
-	 * Note that it is safe to drop the lock after changing rq->tail
-	 * since ipath_post_receive() won't fill the empty slot.
 	 */
-	if (qp->ibqp.srq) {
-		srq = to_isrq(qp->ibqp.srq);
-		rq = &srq->rq;
-	} else {
-		srq = NULL;
-		rq = &qp->r_rq;
-	}
-	spin_lock_irqsave(&rq->lock, flags);
-	if (rq->tail == rq->head) {
-		spin_unlock_irqrestore(&rq->lock, flags);
+	if (qp->r_reuse_sge)
+		qp->r_reuse_sge = 0;
+	else if (!ipath_get_rwqe(qp, 0)) {
 		/*
 		 * Count VL15 packets dropped due to no receive buffer.
 		 * Otherwise, count them as buffer overruns since usually,
@@ -573,39 +637,11 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 		goto bail;
 	}
 	/* Silently drop packets which are too big. */
-	wqe = get_rwqe_ptr(rq, rq->tail);
-	if (wc.byte_len > wqe->length) {
-		spin_unlock_irqrestore(&rq->lock, flags);
+	if (wc.byte_len > qp->r_len) {
+		qp->r_reuse_sge = 1;
 		dev->n_pkt_drops++;
 		goto bail;
 	}
-	wc.wr_id = wqe->wr_id;
-	qp->r_sge.sge = wqe->sg_list[0];
-	qp->r_sge.sg_list = wqe->sg_list + 1;
-	qp->r_sge.num_sge = wqe->num_sge;
-	if (++rq->tail >= rq->size)
-		rq->tail = 0;
-	if (srq && srq->ibsrq.event_handler) {
-		u32 n;
-
-		if (rq->head < rq->tail)
-			n = rq->size + rq->head - rq->tail;
-		else
-			n = rq->head - rq->tail;
-		if (n < srq->limit) {
-			struct ib_event ev;
-
-			srq->limit = 0;
-			spin_unlock_irqrestore(&rq->lock, flags);
-			ev.device = qp->ibqp.device;
-			ev.element.srq = qp->ibqp.srq;
-			ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
-			srq->ibsrq.event_handler(&ev,
-						 srq->ibsrq.srq_context);
-		} else
-			spin_unlock_irqrestore(&rq->lock, flags);
-	} else
-		spin_unlock_irqrestore(&rq->lock, flags);
 	if (has_grh) {
 		ipath_copy_sge(&qp->r_sge, &hdr->u.l.grh,
 			       sizeof(struct ib_grh));
@@ -614,6 +650,7 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 		ipath_skip_sge(&qp->r_sge, sizeof(struct ib_grh));
 	ipath_copy_sge(&qp->r_sge, data,
 		       wc.byte_len - sizeof(struct ib_grh));
+	wc.wr_id = qp->r_wr_id;
 	wc.status = IB_WC_SUCCESS;
 	wc.opcode = IB_WC_RECV;
 	wc.vendor_err = 0;