RDMA/cxgb4: Use a mutex for QP and EP state transitions

Move the connection setup/teardown paths to the workq thread removing
spin lock/irq disable requirements for these paths.  This allows calls
down to the LLD for EP and QP state transition actions to be atomic
with respect to processing CPL messages coming up from the HW.
Namely, calls to rdma_init() and rdma_fini() can now be called with
the mutex held avoiding many race conditions with the abort path.

The QP spinlock is still used but only to manipulate the qp state.  This
allows the fastpaths, poll, post_send, and pos_recv, to run in the
irq context.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

1 files changed, 42 insertions, 48 deletions

diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index 7e45f7334282..76a286f88edd 100644
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -35,6 +35,14 @@ static int ocqp_support;
 module_param(ocqp_support, int, 0644);
 MODULE_PARM_DESC(ocqp_support, "Support on-chip SQs (default=0)");
 
+static void set_state(struct c4iw_qp *qhp, enum c4iw_qp_state state)
+{
+	unsigned long flag;
+	spin_lock_irqsave(&qhp->lock, flag);
+	qhp->attr.state = state;
+	spin_unlock_irqrestore(&qhp->lock, flag);
+}
+
 static void dealloc_oc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
 {
 	c4iw_ocqp_pool_free(rdev, sq->dma_addr, sq->memsize);
@@ -949,46 +957,38 @@ static void post_terminate(struct c4iw_qp *qhp, struct t4_cqe *err_cqe,
  * Assumes qhp lock is held.
  */
 static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
-		       struct c4iw_cq *schp, unsigned long *flag)
+		       struct c4iw_cq *schp)
 {
 	int count;
 	int flushed;
+	unsigned long flag;
 
 	PDBG("%s qhp %p rchp %p schp %p\n", __func__, qhp, rchp, schp);
-	/* take a ref on the qhp since we must release the lock */
-	atomic_inc(&qhp->refcnt);
-	spin_unlock_irqrestore(&qhp->lock, *flag);
 
 	/* locking hierarchy: cq lock first, then qp lock. */
-	spin_lock_irqsave(&rchp->lock, *flag);
+	spin_lock_irqsave(&rchp->lock, flag);
 	spin_lock(&qhp->lock);
 	c4iw_flush_hw_cq(&rchp->cq);
 	c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count);
 	flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count);
 	spin_unlock(&qhp->lock);
-	spin_unlock_irqrestore(&rchp->lock, *flag);
+	spin_unlock_irqrestore(&rchp->lock, flag);
 	if (flushed)
 		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
 
 	/* locking hierarchy: cq lock first, then qp lock. */
-	spin_lock_irqsave(&schp->lock, *flag);
+	spin_lock_irqsave(&schp->lock, flag);
 	spin_lock(&qhp->lock);
 	c4iw_flush_hw_cq(&schp->cq);
 	c4iw_count_scqes(&schp->cq, &qhp->wq, &count);
 	flushed = c4iw_flush_sq(&qhp->wq, &schp->cq, count);
 	spin_unlock(&qhp->lock);
-	spin_unlock_irqrestore(&schp->lock, *flag);
+	spin_unlock_irqrestore(&schp->lock, flag);
 	if (flushed)
 		(*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context);
-
-	/* deref */
-	if (atomic_dec_and_test(&qhp->refcnt))
-		wake_up(&qhp->wait);
-
-	spin_lock_irqsave(&qhp->lock, *flag);
 }
 
-static void flush_qp(struct c4iw_qp *qhp, unsigned long *flag)
+static void flush_qp(struct c4iw_qp *qhp)
 {
 	struct c4iw_cq *rchp, *schp;
 
@@ -1002,7 +1002,7 @@ static void flush_qp(struct c4iw_qp *qhp, unsigned long *flag)
 			t4_set_cq_in_error(&schp->cq);
 		return;
 	}
-	__flush_qp(qhp, rchp, schp, flag);
+	__flush_qp(qhp, rchp, schp);
 }
 
 static int rdma_fini(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
@@ -1010,7 +1010,6 @@ static int rdma_fini(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 {
 	struct fw_ri_wr *wqe;
 	int ret;
-	struct c4iw_wr_wait wr_wait;
 	struct sk_buff *skb;
 
 	PDBG("%s qhp %p qid 0x%x tid %u\n", __func__, qhp, qhp->wq.sq.qid,
@@ -1029,15 +1028,15 @@ static int rdma_fini(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 	wqe->flowid_len16 = cpu_to_be32(
 		FW_WR_FLOWID(ep->hwtid) |
 		FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe, 16)));
-	wqe->cookie = (unsigned long) &wr_wait;
+	wqe->cookie = (unsigned long) &ep->com.wr_wait;
 
 	wqe->u.fini.type = FW_RI_TYPE_FINI;
-	c4iw_init_wr_wait(&wr_wait);
+	c4iw_init_wr_wait(&ep->com.wr_wait);
 	ret = c4iw_ofld_send(&rhp->rdev, skb);
 	if (ret)
 		goto out;
 
-	ret = c4iw_wait_for_reply(&rhp->rdev, &wr_wait, qhp->ep->hwtid,
+	ret = c4iw_wait_for_reply(&rhp->rdev, &ep->com.wr_wait, qhp->ep->hwtid,
 			     qhp->wq.sq.qid, __func__);
 out:
 	PDBG("%s ret %d\n", __func__, ret);
@@ -1072,7 +1071,6 @@ static int rdma_init(struct c4iw_dev *rhp, struct c4iw_qp *qhp)
 {
 	struct fw_ri_wr *wqe;
 	int ret;
-	struct c4iw_wr_wait wr_wait;
 	struct sk_buff *skb;
 
 	PDBG("%s qhp %p qid 0x%x tid %u\n", __func__, qhp, qhp->wq.sq.qid,
@@ -1092,7 +1090,7 @@ static int rdma_init(struct c4iw_dev *rhp, struct c4iw_qp *qhp)
 		FW_WR_FLOWID(qhp->ep->hwtid) |
 		FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe, 16)));
 
-	wqe->cookie = (unsigned long) &wr_wait;
+	wqe->cookie = (unsigned long) &qhp->ep->com.wr_wait;
 
 	wqe->u.init.type = FW_RI_TYPE_INIT;
 	wqe->u.init.mpareqbit_p2ptype =
@@ -1129,13 +1127,13 @@ static int rdma_init(struct c4iw_dev *rhp, struct c4iw_qp *qhp)
 	if (qhp->attr.mpa_attr.initiator)
 		build_rtr_msg(qhp->attr.mpa_attr.p2p_type, &wqe->u.init);
 
-	c4iw_init_wr_wait(&wr_wait);
+	c4iw_init_wr_wait(&qhp->ep->com.wr_wait);
 	ret = c4iw_ofld_send(&rhp->rdev, skb);
 	if (ret)
 		goto out;
 
-	ret = c4iw_wait_for_reply(&rhp->rdev, &wr_wait, qhp->ep->hwtid,
-			     qhp->wq.sq.qid, __func__);
+	ret = c4iw_wait_for_reply(&rhp->rdev, &qhp->ep->com.wr_wait,
+				  qhp->ep->hwtid, qhp->wq.sq.qid, __func__);
 out:
 	PDBG("%s ret %d\n", __func__, ret);
 	return ret;
@@ -1148,7 +1146,6 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 {
 	int ret = 0;
 	struct c4iw_qp_attributes newattr = qhp->attr;
-	unsigned long flag;
 	int disconnect = 0;
 	int terminate = 0;
 	int abort = 0;
@@ -1159,7 +1156,7 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 	     qhp, qhp->wq.sq.qid, qhp->wq.rq.qid, qhp->ep, qhp->attr.state,
 	     (mask & C4IW_QP_ATTR_NEXT_STATE) ? attrs->next_state : -1);
 
-	spin_lock_irqsave(&qhp->lock, flag);
+	mutex_lock(&qhp->mutex);
 
 	/* Process attr changes if in IDLE */
 	if (mask & C4IW_QP_ATTR_VALID_MODIFY) {
@@ -1210,7 +1207,7 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 			qhp->attr.mpa_attr = attrs->mpa_attr;
 			qhp->attr.llp_stream_handle = attrs->llp_stream_handle;
 			qhp->ep = qhp->attr.llp_stream_handle;
-			qhp->attr.state = C4IW_QP_STATE_RTS;
+			set_state(qhp, C4IW_QP_STATE_RTS);
 
 			/*
 			 * Ref the endpoint here and deref when we
@@ -1219,15 +1216,13 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 			 * transition.
 			 */
 			c4iw_get_ep(&qhp->ep->com);
-			spin_unlock_irqrestore(&qhp->lock, flag);
 			ret = rdma_init(rhp, qhp);
-			spin_lock_irqsave(&qhp->lock, flag);
 			if (ret)
 				goto err;
 			break;
 		case C4IW_QP_STATE_ERROR:
-			qhp->attr.state = C4IW_QP_STATE_ERROR;
-			flush_qp(qhp, &flag);
+			set_state(qhp, C4IW_QP_STATE_ERROR);
+			flush_qp(qhp);
 			break;
 		default:
 			ret = -EINVAL;
@@ -1238,39 +1233,38 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 		switch (attrs->next_state) {
 		case C4IW_QP_STATE_CLOSING:
 			BUG_ON(atomic_read(&qhp->ep->com.kref.refcount) < 2);
-			qhp->attr.state = C4IW_QP_STATE_CLOSING;
+			set_state(qhp, C4IW_QP_STATE_CLOSING);
 			ep = qhp->ep;
 			if (!internal) {
 				abort = 0;
 				disconnect = 1;
-				c4iw_get_ep(&ep->com);
+				c4iw_get_ep(&qhp->ep->com);
 			}
-			spin_unlock_irqrestore(&qhp->lock, flag);
 			ret = rdma_fini(rhp, qhp, ep);
-			spin_lock_irqsave(&qhp->lock, flag);
 			if (ret) {
-				c4iw_get_ep(&ep->com);
+				if (internal)
+					c4iw_get_ep(&qhp->ep->com);
 				disconnect = abort = 1;
 				goto err;
 			}
 			break;
 		case C4IW_QP_STATE_TERMINATE:
-			qhp->attr.state = C4IW_QP_STATE_TERMINATE;
+			set_state(qhp, C4IW_QP_STATE_TERMINATE);
 			if (qhp->ibqp.uobject)
 				t4_set_wq_in_error(&qhp->wq);
 			ep = qhp->ep;
-			c4iw_get_ep(&ep->com);
 			if (!internal)
 				terminate = 1;
 			disconnect = 1;
+			c4iw_get_ep(&qhp->ep->com);
 			break;
 		case C4IW_QP_STATE_ERROR:
-			qhp->attr.state = C4IW_QP_STATE_ERROR;
+			set_state(qhp, C4IW_QP_STATE_ERROR);
 			if (!internal) {
 				abort = 1;
 				disconnect = 1;
 				ep = qhp->ep;
-				c4iw_get_ep(&ep->com);
+				c4iw_get_ep(&qhp->ep->com);
 			}
 			goto err;
 			break;
@@ -1286,8 +1280,8 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 		}
 		switch (attrs->next_state) {
 		case C4IW_QP_STATE_IDLE:
-			flush_qp(qhp, &flag);
-			qhp->attr.state = C4IW_QP_STATE_IDLE;
+			flush_qp(qhp);
+			set_state(qhp, C4IW_QP_STATE_IDLE);
 			qhp->attr.llp_stream_handle = NULL;
 			c4iw_put_ep(&qhp->ep->com);
 			qhp->ep = NULL;
@@ -1309,7 +1303,7 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 			ret = -EINVAL;
 			goto out;
 		}
-		qhp->attr.state = C4IW_QP_STATE_IDLE;
+		set_state(qhp, C4IW_QP_STATE_IDLE);
 		break;
 	case C4IW_QP_STATE_TERMINATE:
 		if (!internal) {
@@ -1335,13 +1329,13 @@ err:
 	if (!ep)
 		ep = qhp->ep;
 	qhp->ep = NULL;
-	qhp->attr.state = C4IW_QP_STATE_ERROR;
+	set_state(qhp, C4IW_QP_STATE_ERROR);
 	free = 1;
 	wake_up(&qhp->wait);
 	BUG_ON(!ep);
-	flush_qp(qhp, &flag);
+	flush_qp(qhp);
 out:
-	spin_unlock_irqrestore(&qhp->lock, flag);
+	mutex_unlock(&qhp->mutex);
 
 	if (terminate)
 		post_terminate(qhp, NULL, internal ? GFP_ATOMIC : GFP_KERNEL);
@@ -1363,7 +1357,6 @@ out:
 	 */
 	if (free)
 		c4iw_put_ep(&ep->com);
-
 	PDBG("%s exit state %d\n", __func__, qhp->attr.state);
 	return ret;
 }
@@ -1478,6 +1471,7 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs,
 	qhp->attr.max_ord = 1;
 	qhp->attr.max_ird = 1;
 	spin_lock_init(&qhp->lock);
+	mutex_init(&qhp->mutex);
 	init_waitqueue_head(&qhp->wait);
 	atomic_set(&qhp->refcnt, 1);