IB/mlx4: Add support for drain SQ & RQ

This patch follows the logic from ib_core but considers the internal
device state upon executing the involved commands.

Specifically, Upon internal error state modify QP to an error state can
be assumed to be success as each in-progress WR going to be flushed in
error in any case as expected by that modify command.

In addition,
As the drain should never fail the driver makes sure that post_send/recv
will succeed even if the device is already in an internal error state.
As such once the driver will supply the simulated/SW CQEs the CQE for
the drain WR will be handled as well.

In case of an internal error state the CQE for the drain WR may be
completed as part of the main task that handled the error state or by
the task that issued the drain WR.

As the above depends on scheduling the code takes the relevant locks
and actions to make sure that the completion handler for that WR will
always be called after that the post_send/recv were issued but not in
parallel to the other task that handles the error flow.

Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

drivers/infiniband/hw/mlx4/main.c

@@ -2701,6 +2701,8 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
+	ibdev->ib_dev.drain_sq		= mlx4_ib_drain_sq;
+	ibdev->ib_dev.drain_rq		= mlx4_ib_drain_rq;
 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;

drivers/infiniband/hw/mlx4/mlx4_ib.h

@@ -778,6 +778,8 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata);
 int mlx4_ib_destroy_qp(struct ib_qp *qp);
+void mlx4_ib_drain_sq(struct ib_qp *qp);
+void mlx4_ib_drain_rq(struct ib_qp *qp);
 int mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		      int attr_mask, struct ib_udata *udata);
 int mlx4_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,

drivers/infiniband/hw/mlx4/qp.c

@@ -3568,8 +3568,8 @@ static void add_zero_len_inline(void *wqe)
 	inl->byte_count = cpu_to_be32(1 << 31);
 }
 
-int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
-		      struct ib_send_wr **bad_wr)
+static int _mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+			      struct ib_send_wr **bad_wr, bool drain)
 {
 	struct mlx4_ib_qp *qp = to_mqp(ibqp);
 	void *wqe;
@@ -3609,7 +3609,8 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	}
 
 	spin_lock_irqsave(&qp->sq.lock, flags);
-	if (mdev->dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+	if (mdev->dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR &&
+	    !drain) {
 		err = -EIO;
 		*bad_wr = wr;
 		nreq = 0;
@@ -3899,8 +3900,14 @@ out:
 	return err;
 }
 
-int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
-		      struct ib_recv_wr **bad_wr)
+int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		      struct ib_send_wr **bad_wr)
+{
+	return _mlx4_ib_post_send(ibqp, wr, bad_wr, false);
+}
+
+static int _mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+			      struct ib_recv_wr **bad_wr, bool drain)
 {
 	struct mlx4_ib_qp *qp = to_mqp(ibqp);
 	struct mlx4_wqe_data_seg *scat;
@@ -3915,7 +3922,8 @@ int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 	max_gs = qp->rq.max_gs;
 	spin_lock_irqsave(&qp->rq.lock, flags);
 
-	if (mdev->dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+	if (mdev->dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR &&
+	    !drain) {
 		err = -EIO;
 		*bad_wr = wr;
 		nreq = 0;
@@ -3986,6 +3994,12 @@ out:
 	return err;
 }
 
+int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		      struct ib_recv_wr **bad_wr)
+{
+	return _mlx4_ib_post_recv(ibqp, wr, bad_wr, false);
+}
+
 static inline enum ib_qp_state to_ib_qp_state(enum mlx4_qp_state mlx4_state)
 {
 	switch (mlx4_state) {
@@ -4451,3 +4465,131 @@ int mlx4_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
 	kfree(ib_rwq_ind_tbl);
 	return 0;
 }
+
+struct mlx4_ib_drain_cqe {
+	struct ib_cqe cqe;
+	struct completion done;
+};
+
+static void mlx4_ib_drain_qp_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+	struct mlx4_ib_drain_cqe *cqe = container_of(wc->wr_cqe,
+						     struct mlx4_ib_drain_cqe,
+						     cqe);
+
+	complete(&cqe->done);
+}
+
+/* This function returns only once the drained WR was completed */
+static void handle_drain_completion(struct ib_cq *cq,
+				    struct mlx4_ib_drain_cqe *sdrain,
+				    struct mlx4_ib_dev *dev)
+{
+	struct mlx4_dev *mdev = dev->dev;
+
+	if (cq->poll_ctx == IB_POLL_DIRECT) {
+		while (wait_for_completion_timeout(&sdrain->done, HZ / 10) <= 0)
+			ib_process_cq_direct(cq, -1);
+		return;
+	}
+
+	if (mdev->persist->state == MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+		struct mlx4_ib_cq *mcq = to_mcq(cq);
+		bool triggered = false;
+		unsigned long flags;
+
+		spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
+		/* Make sure that the CQ handler won't run if wasn't run yet */
+		if (!mcq->mcq.reset_notify_added)
+			mcq->mcq.reset_notify_added = 1;
+		else
+			triggered = true;
+		spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
+
+		if (triggered) {
+			/* Wait for any scheduled/running task to be ended */
+			switch (cq->poll_ctx) {
+			case IB_POLL_SOFTIRQ:
+				irq_poll_disable(&cq->iop);
+				irq_poll_enable(&cq->iop);
+				break;
+			case IB_POLL_WORKQUEUE:
+				cancel_work_sync(&cq->work);
+				break;
+			default:
+				WARN_ON_ONCE(1);
+			}
+		}
+
+		/* Run the CQ handler - this makes sure that the drain WR will
+		 * be processed if wasn't processed yet.
+		 */
+		mcq->mcq.comp(&mcq->mcq);
+	}
+
+	wait_for_completion(&sdrain->done);
+}
+
+void mlx4_ib_drain_sq(struct ib_qp *qp)
+{
+	struct ib_cq *cq = qp->send_cq;
+	struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
+	struct mlx4_ib_drain_cqe sdrain;
+	struct ib_send_wr *bad_swr;
+	struct ib_rdma_wr swr = {
+		.wr = {
+			.next = NULL,
+			{ .wr_cqe	= &sdrain.cqe, },
+			.opcode	= IB_WR_RDMA_WRITE,
+		},
+	};
+	int ret;
+	struct mlx4_ib_dev *dev = to_mdev(qp->device);
+	struct mlx4_dev *mdev = dev->dev;
+
+	ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
+	if (ret && mdev->persist->state != MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+		WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
+		return;
+	}
+
+	sdrain.cqe.done = mlx4_ib_drain_qp_done;
+	init_completion(&sdrain.done);
+
+	ret = _mlx4_ib_post_send(qp, &swr.wr, &bad_swr, true);
+	if (ret) {
+		WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
+		return;
+	}
+
+	handle_drain_completion(cq, &sdrain, dev);
+}
+
+void mlx4_ib_drain_rq(struct ib_qp *qp)
+{
+	struct ib_cq *cq = qp->recv_cq;
+	struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
+	struct mlx4_ib_drain_cqe rdrain;
+	struct ib_recv_wr rwr = {}, *bad_rwr;
+	int ret;
+	struct mlx4_ib_dev *dev = to_mdev(qp->device);
+	struct mlx4_dev *mdev = dev->dev;
+
+	ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
+	if (ret && mdev->persist->state != MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+		WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
+		return;
+	}
+
+	rwr.wr_cqe = &rdrain.cqe;
+	rdrain.cqe.done = mlx4_ib_drain_qp_done;
+	init_completion(&rdrain.done);
+
+	ret = _mlx4_ib_post_recv(qp, &rwr, &bad_rwr, true);
+	if (ret) {
+		WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
+		return;
+	}
+
+	handle_drain_completion(cq, &rdrain, dev);
+}