RDMA/mlx5: Add a umr recovery flow

When a UMR fails, the UMR QP state changes to an error state. Therefore,
all the further UMR operations will fail too.

Add a recovery flow to the UMR QP, and repost the flushed WQEs.

Link: https://lore.kernel.org/r/6cc24816cca049bd8541317f5e41d3ac659445d3.1652588303.git.leonro@nvidia.com
Signed-off-by: Aharon Landau <aharonl@nvidia.com>
Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

drivers/infiniband/hw/mlx5/cq.c

@@ -523,6 +523,10 @@ static int mlx5_poll_one(struct mlx5_ib_cq *cq,
 			    "Requestor" : "Responder", cq->mcq.cqn);
 		mlx5_ib_dbg(dev, "syndrome 0x%x, vendor syndrome 0x%x\n",
 			    err_cqe->syndrome, err_cqe->vendor_err_synd);
+		if (wc->status != IB_WC_WR_FLUSH_ERR &&
+		    (*cur_qp)->type == MLX5_IB_QPT_REG_UMR)
+			dev->umrc.state = MLX5_UMR_STATE_RECOVER;
+
 		if (opcode == MLX5_CQE_REQ_ERR) {
 			wq = &(*cur_qp)->sq;
 			wqe_ctr = be16_to_cpu(cqe64->wqe_counter);

drivers/infiniband/hw/mlx5/mlx5_ib.h

@@ -717,13 +717,23 @@ struct mlx5_ib_umr_context {
 	struct completion	done;
 };
 
+enum {
+	MLX5_UMR_STATE_ACTIVE,
+	MLX5_UMR_STATE_RECOVER,
+	MLX5_UMR_STATE_ERR,
+};
+
 struct umr_common {
 	struct ib_pd	*pd;
 	struct ib_cq	*cq;
 	struct ib_qp	*qp;
-	/* control access to UMR QP
+	/* Protects from UMR QP overflow
 	 */
 	struct semaphore	sem;
+	/* Protects from using UMR while the UMR is not active
+	 */
+	struct mutex lock;
+	unsigned int state;
 };
 
 struct mlx5_cache_ent {

drivers/infiniband/hw/mlx5/umr.c

@@ -176,6 +176,7 @@ int mlx5r_umr_resource_init(struct mlx5_ib_dev *dev)
 	dev->umrc.pd = pd;
 
 	sema_init(&dev->umrc.sem, MAX_UMR_WR);
+	mutex_init(&dev->umrc.lock);
 
 	return 0;
 
@@ -195,6 +196,31 @@ void mlx5r_umr_resource_cleanup(struct mlx5_ib_dev *dev)
 	ib_dealloc_pd(dev->umrc.pd);
 }
 
+static int mlx5r_umr_recover(struct mlx5_ib_dev *dev)
+{
+	struct umr_common *umrc = &dev->umrc;
+	struct ib_qp_attr attr;
+	int err;
+
+	attr.qp_state = IB_QPS_RESET;
+	err = ib_modify_qp(umrc->qp, &attr, IB_QP_STATE);
+	if (err) {
+		mlx5_ib_dbg(dev, "Couldn't modify UMR QP\n");
+		goto err;
+	}
+
+	err = mlx5r_umr_qp_rst2rts(dev, umrc->qp);
+	if (err)
+		goto err;
+
+	umrc->state = MLX5_UMR_STATE_ACTIVE;
+	return 0;
+
+err:
+	umrc->state = MLX5_UMR_STATE_ERR;
+	return err;
+}
+
 static int mlx5r_umr_post_send(struct ib_qp *ibqp, u32 mkey, struct ib_cqe *cqe,
 			       struct mlx5r_umr_wqe *wqe, bool with_data)
 {
@@ -231,7 +257,7 @@ static int mlx5r_umr_post_send(struct ib_qp *ibqp, u32 mkey, struct ib_cqe *cqe,
 
 	id.ib_cqe = cqe;
 	mlx5r_finish_wqe(qp, ctrl, seg, size, cur_edge, idx, id.wr_id, 0,
-			 MLX5_FENCE_MODE_NONE, MLX5_OPCODE_UMR);
+			 MLX5_FENCE_MODE_INITIATOR_SMALL, MLX5_OPCODE_UMR);
 
 	mlx5r_ring_db(qp, 1, ctrl);
 
@@ -270,17 +296,49 @@ static int mlx5r_umr_post_send_wait(struct mlx5_ib_dev *dev, u32 mkey,
 	mlx5r_umr_init_context(&umr_context);
 
 	down(&umrc->sem);
-	err = mlx5r_umr_post_send(umrc->qp, mkey, &umr_context.cqe, wqe,
-				  with_data);
-	if (err)
-		mlx5_ib_warn(dev, "UMR post send failed, err %d\n", err);
-	else {
-		wait_for_completion(&umr_context.done);
-		if (umr_context.status != IB_WC_SUCCESS) {
-			mlx5_ib_warn(dev, "reg umr failed (%u)\n",
-				     umr_context.status);
+	while (true) {
+		mutex_lock(&umrc->lock);
+		if (umrc->state == MLX5_UMR_STATE_ERR) {
+			mutex_unlock(&umrc->lock);
 			err = -EFAULT;
+			break;
+		}
+
+		if (umrc->state == MLX5_UMR_STATE_RECOVER) {
+			mutex_unlock(&umrc->lock);
+			usleep_range(3000, 5000);
+			continue;
+		}
+
+		err = mlx5r_umr_post_send(umrc->qp, mkey, &umr_context.cqe, wqe,
+					  with_data);
+		mutex_unlock(&umrc->lock);
+		if (err) {
+			mlx5_ib_warn(dev, "UMR post send failed, err %d\n",
+				     err);
+			break;
 		}
+
+		wait_for_completion(&umr_context.done);
+
+		if (umr_context.status == IB_WC_SUCCESS)
+			break;
+
+		if (umr_context.status == IB_WC_WR_FLUSH_ERR)
+			continue;
+
+		WARN_ON_ONCE(1);
+		mlx5_ib_warn(dev,
+			"reg umr failed (%u). Trying to recover and resubmit the flushed WQEs\n",
+			umr_context.status);
+		mutex_lock(&umrc->lock);
+		err = mlx5r_umr_recover(dev);
+		mutex_unlock(&umrc->lock);
+		if (err)
+			mlx5_ib_warn(dev, "couldn't recover UMR, err %d\n",
+				     err);
+		err = -EFAULT;
+		break;
 	}
 	up(&umrc->sem);
 	return err;