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, 112 insertions, 44 deletions

diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c
index 44c32d2db990..1ccfc909db1e 100644
--- a/drivers/infiniband/hw/ipath/ipath_qp.c
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c
@@ -35,7 +35,7 @@
 #include <linux/vmalloc.h>
 
 #include "ipath_verbs.h"
-#include "ipath_common.h"
+#include "ipath_kernel.h"
 
 #define BITS_PER_PAGE		(PAGE_SIZE*BITS_PER_BYTE)
 #define BITS_PER_PAGE_MASK	(BITS_PER_PAGE-1)
@@ -44,19 +44,6 @@
 #define find_next_offset(map, off) find_next_zero_bit((map)->page, \
 						      BITS_PER_PAGE, off)
 
-#define TRANS_INVALID	0
-#define TRANS_ANY2RST	1
-#define TRANS_RST2INIT	2
-#define TRANS_INIT2INIT	3
-#define TRANS_INIT2RTR	4
-#define TRANS_RTR2RTS	5
-#define TRANS_RTS2RTS	6
-#define TRANS_SQERR2RTS	7
-#define TRANS_ANY2ERR	8
-#define TRANS_RTS2SQD	9  /* XXX Wait for expected ACKs & signal event */
-#define TRANS_SQD2SQD	10 /* error if not drained & parameter change */
-#define TRANS_SQD2RTS	11 /* error if not drained */
-
 /*
  * Convert the AETH credit code into the number of credits.
  */
@@ -355,8 +342,10 @@ static void ipath_reset_qp(struct ipath_qp *qp)
 	qp->s_last = 0;
 	qp->s_ssn = 1;
 	qp->s_lsn = 0;
-	qp->r_rq.head = 0;
-	qp->r_rq.tail = 0;
+	if (qp->r_rq.wq) {
+		qp->r_rq.wq->head = 0;
+		qp->r_rq.wq->tail = 0;
+	}
 	qp->r_reuse_sge = 0;
 }
 
@@ -410,15 +399,32 @@ void ipath_error_qp(struct ipath_qp *qp)
 	qp->s_hdrwords = 0;
 	qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
 
-	wc.opcode = IB_WC_RECV;
-	spin_lock(&qp->r_rq.lock);
-	while (qp->r_rq.tail != qp->r_rq.head) {
-		wc.wr_id = get_rwqe_ptr(&qp->r_rq, qp->r_rq.tail)->wr_id;
-		if (++qp->r_rq.tail >= qp->r_rq.size)
-			qp->r_rq.tail = 0;
-		ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+	if (qp->r_rq.wq) {
+		struct ipath_rwq *wq;
+		u32 head;
+		u32 tail;
+
+		spin_lock(&qp->r_rq.lock);
+
+		/* sanity check pointers before trusting them */
+		wq = qp->r_rq.wq;
+		head = wq->head;
+		if (head >= qp->r_rq.size)
+			head = 0;
+		tail = wq->tail;
+		if (tail >= qp->r_rq.size)
+			tail = 0;
+		wc.opcode = IB_WC_RECV;
+		while (tail != head) {
+			wc.wr_id = get_rwqe_ptr(&qp->r_rq, tail)->wr_id;
+			if (++tail >= qp->r_rq.size)
+				tail = 0;
+			ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+		}
+		wq->tail = tail;
+
+		spin_unlock(&qp->r_rq.lock);
 	}
-	spin_unlock(&qp->r_rq.lock);
 }
 
 /**
@@ -544,7 +550,7 @@ int ipath_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	attr->dest_qp_num = qp->remote_qpn;
 	attr->qp_access_flags = qp->qp_access_flags;
 	attr->cap.max_send_wr = qp->s_size - 1;
-	attr->cap.max_recv_wr = qp->r_rq.size - 1;
+	attr->cap.max_recv_wr = qp->ibqp.srq ? 0 : qp->r_rq.size - 1;
 	attr->cap.max_send_sge = qp->s_max_sge;
 	attr->cap.max_recv_sge = qp->r_rq.max_sge;
 	attr->cap.max_inline_data = 0;
@@ -597,13 +603,23 @@ __be32 ipath_compute_aeth(struct ipath_qp *qp)
 	} else {
 		u32 min, max, x;
 		u32 credits;
-
+		struct ipath_rwq *wq = qp->r_rq.wq;
+		u32 head;
+		u32 tail;
+
+		/* sanity check pointers before trusting them */
+		head = wq->head;
+		if (head >= qp->r_rq.size)
+			head = 0;
+		tail = wq->tail;
+		if (tail >= qp->r_rq.size)
+			tail = 0;
 		/*
 		 * Compute the number of credits available (RWQEs).
 		 * XXX Not holding the r_rq.lock here so there is a small
 		 * chance that the pair of reads are not atomic.
 		 */
-		credits = qp->r_rq.head - qp->r_rq.tail;
+		credits = head - tail;
 		if ((int)credits < 0)
 			credits += qp->r_rq.size;
 		/*
@@ -680,27 +696,37 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 	case IB_QPT_UD:
 	case IB_QPT_SMI:
 	case IB_QPT_GSI:
-		qp = kmalloc(sizeof(*qp), GFP_KERNEL);
+		sz = sizeof(*qp);
+		if (init_attr->srq) {
+			struct ipath_srq *srq = to_isrq(init_attr->srq);
+
+			sz += sizeof(*qp->r_sg_list) *
+				srq->rq.max_sge;
+		} else
+			sz += sizeof(*qp->r_sg_list) *
+				init_attr->cap.max_recv_sge;
+		qp = kmalloc(sz, GFP_KERNEL);
 		if (!qp) {
-			vfree(swq);
 			ret = ERR_PTR(-ENOMEM);
-			goto bail;
+			goto bail_swq;
 		}
 		if (init_attr->srq) {
+			sz = 0;
 			qp->r_rq.size = 0;
 			qp->r_rq.max_sge = 0;
 			qp->r_rq.wq = NULL;
+			init_attr->cap.max_recv_wr = 0;
+			init_attr->cap.max_recv_sge = 0;
 		} else {
 			qp->r_rq.size = init_attr->cap.max_recv_wr + 1;
 			qp->r_rq.max_sge = init_attr->cap.max_recv_sge;
-			sz = (sizeof(struct ipath_sge) * qp->r_rq.max_sge) +
+			sz = (sizeof(struct ib_sge) * qp->r_rq.max_sge) +
 				sizeof(struct ipath_rwqe);
-			qp->r_rq.wq = vmalloc(qp->r_rq.size * sz);
+			qp->r_rq.wq = vmalloc_user(sizeof(struct ipath_rwq) +
+					      qp->r_rq.size * sz);
 			if (!qp->r_rq.wq) {
-				kfree(qp);
-				vfree(swq);
 				ret = ERR_PTR(-ENOMEM);
-				goto bail;
+				goto bail_qp;
 			}
 		}
 
@@ -726,12 +752,10 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 		err = ipath_alloc_qpn(&dev->qp_table, qp,
 				      init_attr->qp_type);
 		if (err) {
-			vfree(swq);
-			vfree(qp->r_rq.wq);
-			kfree(qp);
 			ret = ERR_PTR(err);
-			goto bail;
+			goto bail_rwq;
 		}
+		qp->ip = NULL;
 		ipath_reset_qp(qp);
 
 		/* Tell the core driver that the kernel SMA is present. */
@@ -748,8 +772,51 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 
 	init_attr->cap.max_inline_data = 0;
 
+	/*
+	 * Return the address of the RWQ as the offset to mmap.
+	 * See ipath_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		struct ipath_mmap_info *ip;
+		__u64 offset = (__u64) qp->r_rq.wq;
+		int err;
+
+		err = ib_copy_to_udata(udata, &offset, sizeof(offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_rwq;
+		}
+
+		if (qp->r_rq.wq) {
+			/* Allocate info for ipath_mmap(). */
+			ip = kmalloc(sizeof(*ip), GFP_KERNEL);
+			if (!ip) {
+				ret = ERR_PTR(-ENOMEM);
+				goto bail_rwq;
+			}
+			qp->ip = ip;
+			ip->context = ibpd->uobject->context;
+			ip->obj = qp->r_rq.wq;
+			kref_init(&ip->ref);
+			ip->mmap_cnt = 0;
+			ip->size = PAGE_ALIGN(sizeof(struct ipath_rwq) +
+					      qp->r_rq.size * sz);
+			spin_lock_irq(&dev->pending_lock);
+			ip->next = dev->pending_mmaps;
+			dev->pending_mmaps = ip;
+			spin_unlock_irq(&dev->pending_lock);
+		}
+	}
+
 	ret = &qp->ibqp;
+	goto bail;
 
+bail_rwq:
+	vfree(qp->r_rq.wq);
+bail_qp:
+	kfree(qp);
+bail_swq:
+	vfree(swq);
 bail:
 	return ret;
 }
@@ -773,11 +840,9 @@ int ipath_destroy_qp(struct ib_qp *ibqp)
 	if (qp->ibqp.qp_type == IB_QPT_SMI)
 		ipath_layer_set_verbs_flags(dev->dd, 0);
 
-	spin_lock_irqsave(&qp->r_rq.lock, flags);
-	spin_lock(&qp->s_lock);
+	spin_lock_irqsave(&qp->s_lock, flags);
 	qp->state = IB_QPS_ERR;
-	spin_unlock(&qp->s_lock);
-	spin_unlock_irqrestore(&qp->r_rq.lock, flags);
+	spin_unlock_irqrestore(&qp->s_lock, flags);
 
 	/* Stop the sending tasklet. */
 	tasklet_kill(&qp->s_task);
@@ -798,8 +863,11 @@ int ipath_destroy_qp(struct ib_qp *ibqp)
 	if (atomic_read(&qp->refcount) != 0)
 		ipath_free_qp(&dev->qp_table, qp);
 
+	if (qp->ip)
+		kref_put(&qp->ip->ref, ipath_release_mmap_info);
+	else
+		vfree(qp->r_rq.wq);
 	vfree(qp->s_wq);
-	vfree(qp->r_rq.wq);
 	kfree(qp);
 	return 0;
 }