IB/hfi1: Prepare resource waits for dual leg

Current implementation allows each qp to have only one send engine.  As
such, each qp has only one list to queue prebuilt packets when send engine
resources are not available. To improve performance, it is desired to
support multiple send engines for each qp.

This patch creates the framework to support two send engines
(two legs) for each qp for the TID RDMA protocol, which can be easily
extended to support more send engines. It achieves the goal by creating a
leg specific struct, iowait_work in the iowait struct, to hold the
work_struct and the tx_list as well as a pointer to the parent iowait
struct.

The hfi1_pkt_state now has an additional field to record the current legs
work structure and that is now passed to all egress waiters to determine
the leg that needs to wait via a new iowait helper.  The APIs are adjusted
to use the new leg specific struct as required.

Many new and modified helpers are added to support this change.

Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@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>

1 files changed, 50 insertions, 17 deletions

diff --git a/drivers/infiniband/hw/hfi1/qp.c b/drivers/infiniband/hw/hfi1/qp.c
index b1044a205ab6..126e9739e44f 100644
--- a/drivers/infiniband/hw/hfi1/qp.c
+++ b/drivers/infiniband/hw/hfi1/qp.c
@@ -66,7 +66,7 @@ MODULE_PARM_DESC(qp_table_size, "QP table size");
 static void flush_tx_list(struct rvt_qp *qp);
 static int iowait_sleep(
 	struct sdma_engine *sde,
-	struct iowait *wait,
+	struct iowait_work *wait,
 	struct sdma_txreq *stx,
 	unsigned int seq,
 	bool pkts_sent);
@@ -134,15 +134,13 @@ const struct rvt_operation_params hfi1_post_parms[RVT_OPERATION_MAX] = {
 
 };
 
-static void flush_tx_list(struct rvt_qp *qp)
+static void flush_list_head(struct list_head *l)
 {
-	struct hfi1_qp_priv *priv = qp->priv;
-
-	while (!list_empty(&priv->s_iowait.tx_head)) {
+	while (!list_empty(l)) {
 		struct sdma_txreq *tx;
 
 		tx = list_first_entry(
-			&priv->s_iowait.tx_head,
+			l,
 			struct sdma_txreq,
 			list);
 		list_del_init(&tx->list);
@@ -151,6 +149,14 @@ static void flush_tx_list(struct rvt_qp *qp)
 	}
 }
 
+static void flush_tx_list(struct rvt_qp *qp)
+{
+	struct hfi1_qp_priv *priv = qp->priv;
+
+	flush_list_head(&iowait_get_ib_work(&priv->s_iowait)->tx_head);
+	flush_list_head(&iowait_get_tid_work(&priv->s_iowait)->tx_head);
+}
+
 static void flush_iowait(struct rvt_qp *qp)
 {
 	struct hfi1_qp_priv *priv = qp->priv;
@@ -336,7 +342,7 @@ int hfi1_setup_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe, bool *call_send)
  * It is only used in the post send, which doesn't hold
  * the s_lock.
  */
-void _hfi1_schedule_send(struct rvt_qp *qp)
+bool _hfi1_schedule_send(struct rvt_qp *qp)
 {
 	struct hfi1_qp_priv *priv = qp->priv;
 	struct hfi1_ibport *ibp =
@@ -344,10 +350,10 @@ void _hfi1_schedule_send(struct rvt_qp *qp)
 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
 	struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device);
 
-	iowait_schedule(&priv->s_iowait, ppd->hfi1_wq,
-			priv->s_sde ?
-			priv->s_sde->cpu :
-			cpumask_first(cpumask_of_node(dd->node)));
+	return iowait_schedule(&priv->s_iowait, ppd->hfi1_wq,
+			       priv->s_sde ?
+			       priv->s_sde->cpu :
+			       cpumask_first(cpumask_of_node(dd->node)));
 }
 
 static void qp_pio_drain(struct rvt_qp *qp)
@@ -375,12 +381,32 @@ static void qp_pio_drain(struct rvt_qp *qp)
  *
  * This schedules qp progress and caller should hold
  * the s_lock.
+ * @return true if the first leg is scheduled;
+ * false if the first leg is not scheduled.
  */
-void hfi1_schedule_send(struct rvt_qp *qp)
+bool hfi1_schedule_send(struct rvt_qp *qp)
 {
 	lockdep_assert_held(&qp->s_lock);
-	if (hfi1_send_ok(qp))
+	if (hfi1_send_ok(qp)) {
 		_hfi1_schedule_send(qp);
+		return true;
+	}
+	if (qp->s_flags & HFI1_S_ANY_WAIT_IO)
+		iowait_set_flag(&((struct hfi1_qp_priv *)qp->priv)->s_iowait,
+				IOWAIT_PENDING_IB);
+	return false;
+}
+
+static void hfi1_qp_schedule(struct rvt_qp *qp)
+{
+	struct hfi1_qp_priv *priv = qp->priv;
+	bool ret;
+
+	if (iowait_flag_set(&priv->s_iowait, IOWAIT_PENDING_IB)) {
+		ret = hfi1_schedule_send(qp);
+		if (ret)
+			iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_IB);
+	}
 }
 
 void hfi1_qp_wakeup(struct rvt_qp *qp, u32 flag)
@@ -391,16 +417,22 @@ void hfi1_qp_wakeup(struct rvt_qp *qp, u32 flag)
 	if (qp->s_flags & flag) {
 		qp->s_flags &= ~flag;
 		trace_hfi1_qpwakeup(qp, flag);
-		hfi1_schedule_send(qp);
+		hfi1_qp_schedule(qp);
 	}
 	spin_unlock_irqrestore(&qp->s_lock, flags);
 	/* Notify hfi1_destroy_qp() if it is waiting. */
 	rvt_put_qp(qp);
 }
 
+void hfi1_qp_unbusy(struct rvt_qp *qp, struct iowait_work *wait)
+{
+	if (iowait_set_work_flag(wait) == IOWAIT_IB_SE)
+		qp->s_flags &= ~RVT_S_BUSY;
+}
+
 static int iowait_sleep(
 	struct sdma_engine *sde,
-	struct iowait *wait,
+	struct iowait_work *wait,
 	struct sdma_txreq *stx,
 	uint seq,
 	bool pkts_sent)
@@ -441,7 +473,7 @@ static int iowait_sleep(
 			rvt_get_qp(qp);
 		}
 		write_sequnlock(&dev->iowait_lock);
-		qp->s_flags &= ~RVT_S_BUSY;
+		hfi1_qp_unbusy(qp, wait);
 		spin_unlock_irqrestore(&qp->s_lock, flags);
 		ret = -EBUSY;
 	} else {
@@ -640,6 +672,7 @@ void *qp_priv_alloc(struct rvt_dev_info *rdi, struct rvt_qp *qp)
 		&priv->s_iowait,
 		1,
 		_hfi1_do_send,
+		NULL,
 		iowait_sleep,
 		iowait_wakeup,
 		iowait_sdma_drained);
@@ -689,7 +722,7 @@ void stop_send_queue(struct rvt_qp *qp)
 {
 	struct hfi1_qp_priv *priv = qp->priv;
 
-	cancel_work_sync(&priv->s_iowait.iowork);
+	iowait_cancel_work(&priv->s_iowait);
 }
 
 void quiesce_qp(struct rvt_qp *qp)