IB/mlx4: Reset flow support for IB kernel ULPs

The driver exposes interfaces that directly relate to HW state. Upon fatal
error, consumers of these interfaces (ULPs) that rely on completion of
all their posted work-request could hang, thereby introducing dependencies
in shutdown order.  To prevent this from happening, we manage the
relevant resources (CQs, QPs) that are used by the device. Upon a fatal error,
we now generate simulated completions for outstanding WQEs that were not
completed at the time the HW was reset.

It includes invoking the completion event handler for all involved CQs so that
the ULPs will poll those CQs. When polled we return simulated CQEs with
IB_WC_WR_FLUSH_ERR return code enabling ULPs to clean up their resources and
not wait forever for completions upon receiving remove_one.

The above change requires an extra check in the data path to make sure that when
device is in error state, the simulated CQEs will be returned and no further
WQEs will be posted.

Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 53 insertions, 6 deletions

diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 792f9dc86ada..dfc6ca128a7e 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -46,6 +46,11 @@
 #include "mlx4_ib.h"
 #include "user.h"
 
+static void mlx4_ib_lock_cqs(struct mlx4_ib_cq *send_cq,
+			     struct mlx4_ib_cq *recv_cq);
+static void mlx4_ib_unlock_cqs(struct mlx4_ib_cq *send_cq,
+			       struct mlx4_ib_cq *recv_cq);
+
 enum {
 	MLX4_IB_ACK_REQ_FREQ	= 8,
 };
@@ -618,6 +623,8 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 	struct mlx4_ib_sqp *sqp;
 	struct mlx4_ib_qp *qp;
 	enum mlx4_ib_qp_type qp_type = (enum mlx4_ib_qp_type) init_attr->qp_type;
+	struct mlx4_ib_cq *mcq;
+	unsigned long flags;
 
 	/* When tunneling special qps, we use a plain UD qp */
 	if (sqpn) {
@@ -828,6 +835,24 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 	qp->mqp.event = mlx4_ib_qp_event;
 	if (!*caller_qp)
 		*caller_qp = qp;
+
+	spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
+	mlx4_ib_lock_cqs(to_mcq(init_attr->send_cq),
+			 to_mcq(init_attr->recv_cq));
+	/* Maintain device to QPs access, needed for further handling
+	 * via reset flow
+	 */
+	list_add_tail(&qp->qps_list, &dev->qp_list);
+	/* Maintain CQ to QPs access, needed for further handling
+	 * via reset flow
+	 */
+	mcq = to_mcq(init_attr->send_cq);
+	list_add_tail(&qp->cq_send_list, &mcq->send_qp_list);
+	mcq = to_mcq(init_attr->recv_cq);
+	list_add_tail(&qp->cq_recv_list, &mcq->recv_qp_list);
+	mlx4_ib_unlock_cqs(to_mcq(init_attr->send_cq),
+			   to_mcq(init_attr->recv_cq));
+	spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
 	return 0;
 
 err_qpn:
@@ -886,13 +911,13 @@ static void mlx4_ib_lock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv
 	__acquires(&send_cq->lock) __acquires(&recv_cq->lock)
 {
 	if (send_cq == recv_cq) {
-		spin_lock_irq(&send_cq->lock);
+		spin_lock(&send_cq->lock);
 		__acquire(&recv_cq->lock);
 	} else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
-		spin_lock_irq(&send_cq->lock);
+		spin_lock(&send_cq->lock);
 		spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING);
 	} else {
-		spin_lock_irq(&recv_cq->lock);
+		spin_lock(&recv_cq->lock);
 		spin_lock_nested(&send_cq->lock, SINGLE_DEPTH_NESTING);
 	}
 }
@@ -902,13 +927,13 @@ static void mlx4_ib_unlock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *re
 {
 	if (send_cq == recv_cq) {
 		__release(&recv_cq->lock);
-		spin_unlock_irq(&send_cq->lock);
+		spin_unlock(&send_cq->lock);
 	} else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
 		spin_unlock(&recv_cq->lock);
-		spin_unlock_irq(&send_cq->lock);
+		spin_unlock(&send_cq->lock);
 	} else {
 		spin_unlock(&send_cq->lock);
-		spin_unlock_irq(&recv_cq->lock);
+		spin_unlock(&recv_cq->lock);
 	}
 }
 
@@ -953,6 +978,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 			      int is_user)
 {
 	struct mlx4_ib_cq *send_cq, *recv_cq;
+	unsigned long flags;
 
 	if (qp->state != IB_QPS_RESET) {
 		if (mlx4_qp_modify(dev->dev, NULL, to_mlx4_state(qp->state),
@@ -984,8 +1010,13 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 
 	get_cqs(qp, &send_cq, &recv_cq);
 
+	spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
 	mlx4_ib_lock_cqs(send_cq, recv_cq);
 
+	/* del from lists under both locks above to protect reset flow paths */
+	list_del(&qp->qps_list);
+	list_del(&qp->cq_send_list);
+	list_del(&qp->cq_recv_list);
 	if (!is_user) {
 		__mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn,
 				 qp->ibqp.srq ? to_msrq(qp->ibqp.srq): NULL);
@@ -996,6 +1027,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 	mlx4_qp_remove(dev->dev, &qp->mqp);
 
 	mlx4_ib_unlock_cqs(send_cq, recv_cq);
+	spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
 
 	mlx4_qp_free(dev->dev, &qp->mqp);
 
@@ -2618,8 +2650,15 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	__be32 uninitialized_var(lso_hdr_sz);
 	__be32 blh;
 	int i;
+	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 
 	spin_lock_irqsave(&qp->sq.lock, flags);
+	if (mdev->dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+		err = -EIO;
+		*bad_wr = wr;
+		nreq = 0;
+		goto out;
+	}
 
 	ind = qp->sq_next_wqe;
 
@@ -2917,10 +2956,18 @@ int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 	int ind;
 	int max_gs;
 	int i;
+	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 
 	max_gs = qp->rq.max_gs;
 	spin_lock_irqsave(&qp->rq.lock, flags);
 
+	if (mdev->dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+		err = -EIO;
+		*bad_wr = wr;
+		nreq = 0;
+		goto out;
+	}
+
 	ind = qp->rq.head & (qp->rq.wqe_cnt - 1);
 
 	for (nreq = 0; wr; ++nreq, wr = wr->next) {