[PATCH for-rc 4/5] IB/mlx4: Fix starvation in paravirt mux/demux

["Håkon Bugge" <haakon.bugge@oracle.com>]

The mlx4 driver will proxy MAD packets through the PF driver. A VM or
an instantiated VF will send its MAD packets to the PF driver using
loop-back. The PF driver will be informed by an interrupt, but defer
the handling and polling of CQEs to a worker thread running on an
ordered work-queue.

Consider the following scenario: the VMs will in short proximity in
time, for example due to a network event, send many MAD packets to the
PF driver. Lets say there are K VMs, each sending N packets.

The interrupt from the first VM will start the worker thread, which
will poll N CQEs. A common case here is where the PF driver will
multiplex the packets received from the VMs out on the wire QP.

But before the wire QP has returned a send CQE and associated interrupt,
the other K - 1 VMs has sent their N packets.

The PF driver will have to multiplex K * N packets out on the wire
QP. But the send-queue on the wire QP has a finite capacity.

So, in this scenario, if K * N is larger that the send-queue capacity
of the wire QP, we will get MAD packets dropped on the floor with this
dynamic debug message:

mlx4_ib_multiplex_mad: failed sending GSI to wire on behalf of slave 2 (-11)

and this despite the fact that the wire send-queue could have
capacity, but the PF driver isn't aware, because the wire send CQEs
have not yet been polled.

We can also have a similar scenario inbound, with a wire recv-queue
larger than the tunnel QP's recv queue. If many remote peers sends MAD
packets to the very same VM, the tunnel send-queue destined to the VM
could overflow.

This starvation is fixed by introducing separate work queues for the
wire QPs vs. the tunnel QPs

With this fix, using a dual ported HCA, 8 VFs instantiated, we could
run cmtime on each of the 18 interfaces towards a similar configured
peer, each cmtime instance with 800 QPs (all in all 14400 QPs) without
a single CM packet getting lost.

Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
---
 drivers/infiniband/hw/mlx4/mad.c     | 34 +++++++++++++++++++++++++---
 drivers/infiniband/hw/mlx4/mlx4_ib.h |  2 ++
 2 files changed, 33 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 336b894de7cf..28bf8ddb019e 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -1278,6 +1278,18 @@ static void mlx4_ib_tunnel_comp_handler(struct ib_cq *cq, void *arg)
 	spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
 }
 
+static void mlx4_ib_wire_comp_handler(struct ib_cq *cq, void *arg)
+{
+	unsigned long flags;
+	struct mlx4_ib_demux_pv_ctx *ctx = cq->cq_context;
+	struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
+
+	spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+	if (!dev->sriov.is_going_down && ctx->state == DEMUX_PV_STATE_ACTIVE)
+		queue_work(ctx->wi_wq, &ctx->work);
+	spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
 static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx,
 				  struct mlx4_ib_demux_pv_qp *tun_qp,
 				  int index)
@@ -1986,7 +1998,8 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port,
 		cq_size *= 2;
 
 	cq_attr.cqe = cq_size;
-	ctx->cq = ib_create_cq(ctx->ib_dev, mlx4_ib_tunnel_comp_handler,
+	ctx->cq = ib_create_cq(ctx->ib_dev,
+			       create_tun ? mlx4_ib_tunnel_comp_handler : mlx4_ib_wire_comp_handler,
 			       NULL, ctx, &cq_attr);
 	if (IS_ERR(ctx->cq)) {
 		ret = PTR_ERR(ctx->cq);
@@ -2023,6 +2036,7 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port,
 		INIT_WORK(&ctx->work, mlx4_ib_sqp_comp_worker);
 
 	ctx->wq = to_mdev(ibdev)->sriov.demux[port - 1].wq;
+	ctx->wi_wq = to_mdev(ibdev)->sriov.demux[port - 1].wi_wq;
 
 	ret = ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP);
 	if (ret) {
@@ -2166,7 +2180,7 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev,
 		goto err_mcg;
 	}
 
-	snprintf(name, sizeof name, "mlx4_ibt%d", port);
+	snprintf(name, sizeof(name), "mlx4_ibt%d", port);
 	ctx->wq = alloc_ordered_workqueue(name, WQ_MEM_RECLAIM);
 	if (!ctx->wq) {
 		pr_err("Failed to create tunnelling WQ for port %d\n", port);
@@ -2174,7 +2188,15 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev,
 		goto err_wq;
 	}
 
-	snprintf(name, sizeof name, "mlx4_ibud%d", port);
+	snprintf(name, sizeof(name), "mlx4_ibwi%d", port);
+	ctx->wi_wq = alloc_ordered_workqueue(name, WQ_MEM_RECLAIM);
+	if (!ctx->wi_wq) {
+		pr_err("Failed to create wire WQ for port %d\n", port);
+		ret = -ENOMEM;
+		goto err_wiwq;
+	}
+
+	snprintf(name, sizeof(name), "mlx4_ibud%d", port);
 	ctx->ud_wq = alloc_ordered_workqueue(name, WQ_MEM_RECLAIM);
 	if (!ctx->ud_wq) {
 		pr_err("Failed to create up/down WQ for port %d\n", port);
@@ -2185,6 +2207,10 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev,
 	return 0;
 
 err_udwq:
+	destroy_workqueue(ctx->wi_wq);
+	ctx->wi_wq = NULL;
+
+err_wiwq:
 	destroy_workqueue(ctx->wq);
 	ctx->wq = NULL;
 
@@ -2232,12 +2258,14 @@ static void mlx4_ib_free_demux_ctx(struct mlx4_ib_demux_ctx *ctx)
 				ctx->tun[i]->state = DEMUX_PV_STATE_DOWNING;
 		}
 		flush_workqueue(ctx->wq);
+		flush_workqueue(ctx->wi_wq);
 		for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
 			destroy_pv_resources(dev, i, ctx->port, ctx->tun[i], 0);
 			free_pv_object(dev, i, ctx->port);
 		}
 		kfree(ctx->tun);
 		destroy_workqueue(ctx->ud_wq);
+		destroy_workqueue(ctx->wi_wq);
 		destroy_workqueue(ctx->wq);
 	}
 }
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 8999fecb045b..20cfa69d0aed 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -455,6 +455,7 @@ struct mlx4_ib_demux_pv_ctx {
 	struct ib_pd *pd;
 	struct work_struct work;
 	struct workqueue_struct *wq;
+	struct workqueue_struct *wi_wq;
 	struct mlx4_ib_demux_pv_qp qp[2];
 };
 
@@ -462,6 +463,7 @@ struct mlx4_ib_demux_ctx {
 	struct ib_device *ib_dev;
 	int port;
 	struct workqueue_struct *wq;
+	struct workqueue_struct *wi_wq;
 	struct workqueue_struct *ud_wq;
 	spinlock_t ud_lock;
 	atomic64_t subnet_prefix;
-- 
2.20.1