IB/hfi1: Delay the release of destination mr for TID RDMA WRITE DATA

The reference of destination memory region is first obtained when TID RDMA
WRITE request is first received on the responder side. This reference is
released once all TID RDMA WRITE RESP packets are sent to the requester
side, even though not all TID RDMA WRITE DATA packets may have been
received. This early release will especially be undesired if the software
needs to access the destination memory before the last data packet is
received.

This patch delays the release of the MR until all TID RDMA DATA packets
have been received. A helper function to release the reference is also
created to simplify the code.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

3 files changed, 18 insertions, 29 deletions

diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c
index 5991211d72bd..5ba39a9f65ad 100644
--- a/drivers/infiniband/hw/hfi1/rc.c
+++ b/drivers/infiniband/hw/hfi1/rc.c
@@ -140,10 +140,7 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp,
 	case OP(RDMA_READ_RESPONSE_LAST):
 	case OP(RDMA_READ_RESPONSE_ONLY):
 		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		/* FALLTHROUGH */
 	case OP(ATOMIC_ACKNOWLEDGE):
 		/*
@@ -343,7 +340,8 @@ write_resp:
 			break;
 
 		e->sent = 1;
-		qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
+		/* Do not free e->rdma_sge until all data are received */
+		qp->s_ack_state = OP(ATOMIC_ACKNOWLEDGE);
 		break;
 
 	case TID_OP(READ_RESP):
@@ -2643,10 +2641,7 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data,
 		len = be32_to_cpu(reth->length);
 		if (unlikely(offset + len != e->rdma_sge.sge_length))
 			goto unlock_done;
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		if (len != 0) {
 			u32 rkey = be32_to_cpu(reth->rkey);
 			u64 vaddr = get_ib_reth_vaddr(reth);
@@ -3088,10 +3083,7 @@ send_last:
 			update_ack_queue(qp, next);
 		}
 		e = &qp->s_ack_queue[qp->r_head_ack_queue];
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		reth = &ohdr->u.rc.reth;
 		len = be32_to_cpu(reth->length);
 		if (len) {
@@ -3166,10 +3158,7 @@ send_last:
 			update_ack_queue(qp, next);
 		}
 		e = &qp->s_ack_queue[qp->r_head_ack_queue];
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		/* Process OPFN special virtual address */
 		if (opfn) {
 			opfn_conn_response(qp, e, ateth);
diff --git a/drivers/infiniband/hw/hfi1/rc.h b/drivers/infiniband/hw/hfi1/rc.h
index 8e0935b9bf2a..5ed5e85d5841 100644
--- a/drivers/infiniband/hw/hfi1/rc.h
+++ b/drivers/infiniband/hw/hfi1/rc.h
@@ -41,6 +41,14 @@ static inline u32 restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe,
 	return rvt_restart_sge(ss, wqe, len);
 }
 
+static inline void release_rdma_sge_mr(struct rvt_ack_entry *e)
+{
+	if (e->rdma_sge.mr) {
+		rvt_put_mr(e->rdma_sge.mr);
+		e->rdma_sge.mr = NULL;
+	}
+}
+
 struct rvt_ack_entry *find_prev_entry(struct rvt_qp *qp, u32 psn, u8 *prev,
 				      u8 *prev_ack, bool *scheduled);
 int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, u64 val,
diff --git a/drivers/infiniband/hw/hfi1/tid_rdma.c b/drivers/infiniband/hw/hfi1/tid_rdma.c
index fdda33aca77f..145dd3348fdd 100644
--- a/drivers/infiniband/hw/hfi1/tid_rdma.c
+++ b/drivers/infiniband/hw/hfi1/tid_rdma.c
@@ -2036,10 +2036,7 @@ static int tid_rdma_rcv_error(struct hfi1_packet *packet,
 		if (psn != e->psn || len != req->total_len)
 			goto unlock;
 
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 
 		rkey = be32_to_cpu(reth->rkey);
 		vaddr = get_ib_reth_vaddr(reth);
@@ -2285,10 +2282,7 @@ void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet)
 		update_ack_queue(qp, next);
 	}
 	e = &qp->s_ack_queue[qp->r_head_ack_queue];
-	if (e->rdma_sge.mr) {
-		rvt_put_mr(e->rdma_sge.mr);
-		e->rdma_sge.mr = NULL;
-	}
+	release_rdma_sge_mr(e);
 
 	rkey = be32_to_cpu(reth->rkey);
 	qp->r_len = len;
@@ -3751,10 +3745,7 @@ void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet)
 		goto update_head;
 	}
 
-	if (e->rdma_sge.mr) {
-		rvt_put_mr(e->rdma_sge.mr);
-		e->rdma_sge.mr = NULL;
-	}
+	release_rdma_sge_mr(e);
 
 	/* The length needs to be in multiples of PAGE_SIZE */
 	if (!len || len & ~PAGE_MASK)
@@ -4347,6 +4338,7 @@ void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet)
 		priv->r_tid_ack = priv->r_tid_tail;
 
 	if (opcode == TID_OP(WRITE_DATA_LAST)) {
+		release_rdma_sge_mr(e);
 		for (next = priv->r_tid_tail + 1; ; next++) {
 			if (next > rvt_size_atomic(&dev->rdi))
 				next = 0;