[PATCH for-rc v2 0/6] Add CM packets missing and harden the proxying

[PATCH for-rc v2 0/6] Add CM packets missing and harden the proxying

[Håkon Bugge]

A high number of MAD packet drops are observed in the mlx4 MAD proxy
system. These are fixed by separating the parameters for the tunnel
vs. wire QPs and by introducing a separate worker-thread for the wire
QPs.

Support for MRA and REJ with its reason being timeout is also added.

Dynamic debug prints adjusted and amended.

    v1->v2:
	* Added commit ("Adjust delayed work when a dup is observed")
	* Minor adjustments in some of the commits

Håkon Bugge (6):
  IB/mlx4: Add and improve logging
  IB/mlx4: Add support for MRA
  IB/mlx4: Separate tunnel and wire bufs parameters
  IB/mlx4: Fix starvation in paravirt mux/demux
  IB/mlx4: Add support for REJ due to timeout
  IB/mlx4: Adjust delayed work when a dup is observed

 drivers/infiniband/hw/mlx4/cm.c      | 148 ++++++++++++++++++++++++-
 drivers/infiniband/hw/mlx4/mad.c     | 158 +++++++++++++++------------
 drivers/infiniband/hw/mlx4/mlx4_ib.h |   8 +-
 3 files changed, 241 insertions(+), 73 deletions(-)

--
2.20.1

[PATCH for-rc v2 1/6] IB/mlx4: Add and improve logging

[Håkon Bugge]

Add missing check for success after call to mlx4_ib_send_to_wire() in
mlx4_ib_multiplex_mad().

Amended the existing pr_debug() in mlx4_ib_multiplex_cm_handler() and
mlx4_ib_demux_cm_handler() with attr_id during a lookup failure.

Removed two noisy pr_debug() in mad.c

Fixes: 3cf69cc8dbeb ("IB/mlx4: Add CM paravirtualization")
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>

---

    v1->v2:
	* Added pr_debug for -EINVAL returns in mlx4_ib_send_to_slave()
	* Amended "Cannot send to"... with MAD type
---
 drivers/infiniband/hw/mlx4/cm.c  |  7 +--
 drivers/infiniband/hw/mlx4/mad.c | 80 ++++++++++++++------------------
 2 files changed, 40 insertions(+), 47 deletions(-)

diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
index b591861934b3..302ea7ec2008 100644
--- a/drivers/infiniband/hw/mlx4/cm.c
+++ b/drivers/infiniband/hw/mlx4/cm.c
@@ -314,8 +314,8 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id
 	}
 
 	if (!id) {
-		pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL!\n",
-			 slave_id, sl_cm_id);
+		pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL! attr_id: 0x%x\n",
+			 slave_id, sl_cm_id, be16_to_cpu(mad->mad_hdr.attr_id));
 		return -EINVAL;
 	}
 
@@ -354,7 +354,8 @@ int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
 	id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1);
 
 	if (!id) {
-		pr_debug("Couldn't find an entry for pv_cm_id 0x%x\n", pv_cm_id);
+		pr_debug("Couldn't find an entry for pv_cm_id 0x%x, attr_id 0x%x\n",
+			 pv_cm_id, be16_to_cpu(mad->mad_hdr.attr_id));
 		return -ENOENT;
 	}
 
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index abe68708d6d6..932786b0689e 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -500,6 +500,13 @@ static int get_gids_from_l3_hdr(struct ib_grh *grh, union ib_gid *sgid,
 					 sgid, dgid);
 }
 
+static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave)
+{
+	int proxy_start = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave;
+
+	return (qpn >= proxy_start && qpn <= proxy_start + 1);
+}
+
 int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
 			  enum ib_qp_type dest_qpt, struct ib_wc *wc,
 			  struct ib_grh *grh, struct ib_mad *mad)
@@ -520,8 +527,10 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
 	u16 cached_pkey;
 	u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 
-	if (dest_qpt > IB_QPT_GSI)
+	if (dest_qpt > IB_QPT_GSI) {
+		pr_debug("dest_qpt (%d) > IB_QPT_GSI\n", dest_qpt);
 		return -EINVAL;
+	}
 
 	tun_ctx = dev->sriov.demux[port-1].tun[slave];
 
@@ -538,12 +547,20 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
 	if (dest_qpt) {
 		u16 pkey_ix;
 		ret = ib_get_cached_pkey(&dev->ib_dev, port, wc->pkey_index, &cached_pkey);
-		if (ret)
+		if (ret) {
+			pr_debug("unable to get %s cached pkey for index %d, ret %d\n",
+				 is_proxy_qp0(dev, wc->src_qp, slave) ? "SMI" : "GSI",
+				 wc->pkey_index, ret);
 			return -EINVAL;
+		}
 
 		ret = find_slave_port_pkey_ix(dev, slave, port, cached_pkey, &pkey_ix);
-		if (ret)
+		if (ret) {
+			pr_debug("unable to get %s pkey ix for pkey 0x%x, ret %d\n",
+				 is_proxy_qp0(dev, wc->src_qp, slave) ? "SMI" : "GSI",
+				 cached_pkey, ret);
 			return -EINVAL;
+		}
 		tun_pkey_ix = pkey_ix;
 	} else
 		tun_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
@@ -715,7 +732,8 @@ static int mlx4_ib_demux_mad(struct ib_device *ibdev, u8 port,
 
 		err = mlx4_ib_send_to_slave(dev, slave, port, wc->qp->qp_type, wc, grh, mad);
 		if (err)
-			pr_debug("failed sending to slave %d via tunnel qp (%d)\n",
+			pr_debug("failed sending %s to slave %d via tunnel qp (%d)\n",
+				 is_proxy_qp0(dev, wc->src_qp, slave) ? "SMI" : "GSI",
 				 slave, err);
 		return 0;
 	}
@@ -794,7 +812,8 @@ static int mlx4_ib_demux_mad(struct ib_device *ibdev, u8 port,
 
 	err = mlx4_ib_send_to_slave(dev, slave, port, wc->qp->qp_type, wc, grh, mad);
 	if (err)
-		pr_debug("failed sending to slave %d via tunnel qp (%d)\n",
+		pr_debug("failed sending %s to slave %d via tunnel qp (%d)\n",
+			 is_proxy_qp0(dev, wc->src_qp, slave) ? "SMI" : "GSI",
 			 slave, err);
 	return 0;
 }
@@ -807,27 +826,6 @@ static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
 	int err;
 	struct ib_port_attr pattr;
 
-	if (in_wc && in_wc->qp) {
-		pr_debug("received MAD: port:%d slid:%d sqpn:%d "
-			 "dlid_bits:%d dqpn:%d wc_flags:0x%x tid:%016llx cls:%x mtd:%x atr:%x\n",
-			 port_num,
-			 in_wc->slid, in_wc->src_qp,
-			 in_wc->dlid_path_bits,
-			 in_wc->qp->qp_num,
-			 in_wc->wc_flags,
-			 be64_to_cpu(in_mad->mad_hdr.tid),
-			 in_mad->mad_hdr.mgmt_class, in_mad->mad_hdr.method,
-			 be16_to_cpu(in_mad->mad_hdr.attr_id));
-		if (in_wc->wc_flags & IB_WC_GRH) {
-			pr_debug("sgid_hi:0x%016llx sgid_lo:0x%016llx\n",
-				 be64_to_cpu(in_grh->sgid.global.subnet_prefix),
-				 be64_to_cpu(in_grh->sgid.global.interface_id));
-			pr_debug("dgid_hi:0x%016llx dgid_lo:0x%016llx\n",
-				 be64_to_cpu(in_grh->dgid.global.subnet_prefix),
-				 be64_to_cpu(in_grh->dgid.global.interface_id));
-		}
-	}
-
 	slid = in_wc ? ib_lid_cpu16(in_wc->slid) : be16_to_cpu(IB_LID_PERMISSIVE);
 
 	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP && slid == 0) {
@@ -1341,14 +1339,6 @@ static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port,
 	return ret;
 }
 
-static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave)
-{
-	int proxy_start = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave;
-
-	return (qpn >= proxy_start && qpn <= proxy_start + 1);
-}
-
-
 int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
 			 enum ib_qp_type dest_qpt, u16 pkey_index,
 			 u32 remote_qpn, u32 qkey, struct rdma_ah_attr *attr,
@@ -1484,6 +1474,7 @@ static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc
 	u16 vlan_id;
 	u8 qos;
 	u8 *dmac;
+	int sts;
 
 	/* Get slave that sent this packet */
 	if (wc->src_qp < dev->dev->phys_caps.base_proxy_sqpn ||
@@ -1580,13 +1571,17 @@ static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc
 					&vlan_id, &qos))
 		rdma_ah_set_sl(&ah_attr, qos);
 
-	mlx4_ib_send_to_wire(dev, slave, ctx->port,
-			     is_proxy_qp0(dev, wc->src_qp, slave) ?
-			     IB_QPT_SMI : IB_QPT_GSI,
-			     be16_to_cpu(tunnel->hdr.pkey_index),
-			     be32_to_cpu(tunnel->hdr.remote_qpn),
-			     be32_to_cpu(tunnel->hdr.qkey),
-			     &ah_attr, wc->smac, vlan_id, &tunnel->mad);
+	sts = mlx4_ib_send_to_wire(dev, slave, ctx->port,
+				   is_proxy_qp0(dev, wc->src_qp, slave) ?
+				   IB_QPT_SMI : IB_QPT_GSI,
+				   be16_to_cpu(tunnel->hdr.pkey_index),
+				   be32_to_cpu(tunnel->hdr.remote_qpn),
+				   be32_to_cpu(tunnel->hdr.qkey),
+				   &ah_attr, wc->smac, vlan_id, &tunnel->mad);
+	if (sts)
+		pr_debug("failed sending %s to wire on behalf of slave %d (%d)\n",
+			 is_proxy_qp0(dev, wc->src_qp, slave) ? "SMI" : "GSI",
+			 slave, sts);
 }
 
 static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
@@ -1744,9 +1739,6 @@ static void mlx4_ib_tunnel_comp_worker(struct work_struct *work)
 					       "buf:%lld\n", wc.wr_id);
 				break;
 			case IB_WC_SEND:
-				pr_debug("received tunnel send completion:"
-					 "wrid=0x%llx, status=0x%x\n",
-					 wc.wr_id, wc.status);
 				rdma_destroy_ah(tun_qp->tx_ring[wc.wr_id &
 					      (MLX4_NUM_TUNNEL_BUFS - 1)].ah, 0);
 				tun_qp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
-- 
2.20.1

[PATCH for-rc v2 2/6] IB/mlx4: Add support for MRA

[Håkon Bugge]

Using CX-3 in virtualized mode, MAD packets are proxied through the PF
driver. However, the handling lacks support of the MRA (Message
Receipt Acknowledgment) packet. When having dynamic debug enabled, we
see tons of:

mlx4_ib_multiplex_cm_handler: id{slave: 7, sl_cm_id: 0x8fcb45a0} is NULL! attr_id: 0x11

Fixes: 3cf69cc8dbeb ("IB/mlx4: Add CM paravirtualization")
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
---
 drivers/infiniband/hw/mlx4/cm.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
index 302ea7ec2008..6f0ffd0906e6 100644
--- a/drivers/infiniband/hw/mlx4/cm.c
+++ b/drivers/infiniband/hw/mlx4/cm.c
@@ -293,8 +293,9 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id
 	int pv_cm_id = -1;
 
 	if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
-			mad->mad_hdr.attr_id == CM_REP_ATTR_ID ||
-			mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
+	    mad->mad_hdr.attr_id == CM_REP_ATTR_ID ||
+	    mad->mad_hdr.attr_id == CM_MRA_ATTR_ID ||
+	    mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
 		sl_cm_id = get_local_comm_id(mad);
 		id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id);
 		if (id)
-- 
2.20.1

[PATCH for-rc v2 3/6] IB/mlx4: Separate tunnel and wire bufs parameters

[Håkon Bugge]

Using CX-3 in virtualized mode, MAD packets are proxied through the PF
driver. The feed is N tunnel QPs, and what is received from the VFs is
multiplexed out on the wire QP. Since this is a many-to-one scenario,
it is better to have separate initialization parameters for the two
usages.

The number of wire and tunnel bufs are yanked up to 2K and 512
respectively. With this set of parameters, a system consisting of
eight physical servers, each with eight VMs and 14 I/O servers (BM),
can run switch fail-over without seeing:

mlx4_ib_demux_mad: failed sending GSI to slave 3 via tunnel qp (-11)

or

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

Fixes: 3cf69cc8dbeb ("IB/mlx4: Add CM paravirtualization")
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>

---
    v1 -> v2:
	* Increased number of tunnel buffers to 512
---
 drivers/infiniband/hw/mlx4/mad.c     | 44 +++++++++++++++-------------
 drivers/infiniband/hw/mlx4/mlx4_ib.h |  3 +-
 2 files changed, 26 insertions(+), 21 deletions(-)

diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 932786b0689e..e1310820352e 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -1391,10 +1391,10 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
 
 	spin_lock(&sqp->tx_lock);
 	if (sqp->tx_ix_head - sqp->tx_ix_tail >=
-	    (MLX4_NUM_TUNNEL_BUFS - 1))
+	    (MLX4_NUM_WIRE_BUFS - 1))
 		ret = -EAGAIN;
 	else
-		wire_tx_ix = (++sqp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS - 1);
+		wire_tx_ix = (++sqp->tx_ix_head) & (MLX4_NUM_WIRE_BUFS - 1);
 	spin_unlock(&sqp->tx_lock);
 	if (ret)
 		goto out;
@@ -1590,19 +1590,20 @@ static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
 	int i;
 	struct mlx4_ib_demux_pv_qp *tun_qp;
 	int rx_buf_size, tx_buf_size;
+	const int nmbr_bufs = is_tun ? MLX4_NUM_TUNNEL_BUFS : MLX4_NUM_WIRE_BUFS;
 
 	if (qp_type > IB_QPT_GSI)
 		return -EINVAL;
 
 	tun_qp = &ctx->qp[qp_type];
 
-	tun_qp->ring = kcalloc(MLX4_NUM_TUNNEL_BUFS,
+	tun_qp->ring = kcalloc(nmbr_bufs,
 			       sizeof(struct mlx4_ib_buf),
 			       GFP_KERNEL);
 	if (!tun_qp->ring)
 		return -ENOMEM;
 
-	tun_qp->tx_ring = kcalloc(MLX4_NUM_TUNNEL_BUFS,
+	tun_qp->tx_ring = kcalloc(nmbr_bufs,
 				  sizeof (struct mlx4_ib_tun_tx_buf),
 				  GFP_KERNEL);
 	if (!tun_qp->tx_ring) {
@@ -1619,7 +1620,7 @@ static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
 		tx_buf_size = sizeof (struct mlx4_mad_snd_buf);
 	}
 
-	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+	for (i = 0; i < nmbr_bufs; i++) {
 		tun_qp->ring[i].addr = kmalloc(rx_buf_size, GFP_KERNEL);
 		if (!tun_qp->ring[i].addr)
 			goto err;
@@ -1633,7 +1634,7 @@ static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
 		}
 	}
 
-	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+	for (i = 0; i < nmbr_bufs; i++) {
 		tun_qp->tx_ring[i].buf.addr =
 			kmalloc(tx_buf_size, GFP_KERNEL);
 		if (!tun_qp->tx_ring[i].buf.addr)
@@ -1664,7 +1665,7 @@ static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
 				    tx_buf_size, DMA_TO_DEVICE);
 		kfree(tun_qp->tx_ring[i].buf.addr);
 	}
-	i = MLX4_NUM_TUNNEL_BUFS;
+	i = nmbr_bufs;
 err:
 	while (i > 0) {
 		--i;
@@ -1685,6 +1686,7 @@ static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
 	int i;
 	struct mlx4_ib_demux_pv_qp *tun_qp;
 	int rx_buf_size, tx_buf_size;
+	const int nmbr_bufs = is_tun ? MLX4_NUM_TUNNEL_BUFS : MLX4_NUM_WIRE_BUFS;
 
 	if (qp_type > IB_QPT_GSI)
 		return;
@@ -1699,13 +1701,13 @@ static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
 	}
 
 
-	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+	for (i = 0; i < nmbr_bufs; i++) {
 		ib_dma_unmap_single(ctx->ib_dev, tun_qp->ring[i].map,
 				    rx_buf_size, DMA_FROM_DEVICE);
 		kfree(tun_qp->ring[i].addr);
 	}
 
-	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+	for (i = 0; i < nmbr_bufs; i++) {
 		ib_dma_unmap_single(ctx->ib_dev, tun_qp->tx_ring[i].buf.map,
 				    tx_buf_size, DMA_TO_DEVICE);
 		kfree(tun_qp->tx_ring[i].buf.addr);
@@ -1785,6 +1787,7 @@ static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx,
 	struct mlx4_ib_qp_tunnel_init_attr qp_init_attr;
 	struct ib_qp_attr attr;
 	int qp_attr_mask_INIT;
+	const int nmbr_bufs = create_tun ? MLX4_NUM_TUNNEL_BUFS : MLX4_NUM_WIRE_BUFS;
 
 	if (qp_type > IB_QPT_GSI)
 		return -EINVAL;
@@ -1795,8 +1798,8 @@ static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx,
 	qp_init_attr.init_attr.send_cq = ctx->cq;
 	qp_init_attr.init_attr.recv_cq = ctx->cq;
 	qp_init_attr.init_attr.sq_sig_type = IB_SIGNAL_ALL_WR;
-	qp_init_attr.init_attr.cap.max_send_wr = MLX4_NUM_TUNNEL_BUFS;
-	qp_init_attr.init_attr.cap.max_recv_wr = MLX4_NUM_TUNNEL_BUFS;
+	qp_init_attr.init_attr.cap.max_send_wr = nmbr_bufs;
+	qp_init_attr.init_attr.cap.max_recv_wr = nmbr_bufs;
 	qp_init_attr.init_attr.cap.max_send_sge = 1;
 	qp_init_attr.init_attr.cap.max_recv_sge = 1;
 	if (create_tun) {
@@ -1858,7 +1861,7 @@ static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx,
 		goto err_qp;
 	}
 
-	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+	for (i = 0; i < nmbr_bufs; i++) {
 		ret = mlx4_ib_post_pv_qp_buf(ctx, tun_qp, i);
 		if (ret) {
 			pr_err(" mlx4_ib_post_pv_buf error"
@@ -1894,8 +1897,8 @@ static void mlx4_ib_sqp_comp_worker(struct work_struct *work)
 			switch (wc.opcode) {
 			case IB_WC_SEND:
 				kfree(sqp->tx_ring[wc.wr_id &
-				      (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
-				sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+				      (MLX4_NUM_WIRE_BUFS - 1)].ah);
+				sqp->tx_ring[wc.wr_id & (MLX4_NUM_WIRE_BUFS - 1)].ah
 					= NULL;
 				spin_lock(&sqp->tx_lock);
 				sqp->tx_ix_tail++;
@@ -1904,13 +1907,13 @@ static void mlx4_ib_sqp_comp_worker(struct work_struct *work)
 			case IB_WC_RECV:
 				mad = (struct ib_mad *) &(((struct mlx4_mad_rcv_buf *)
 						(sqp->ring[wc.wr_id &
-						(MLX4_NUM_TUNNEL_BUFS - 1)].addr))->payload);
+						(MLX4_NUM_WIRE_BUFS - 1)].addr))->payload);
 				grh = &(((struct mlx4_mad_rcv_buf *)
 						(sqp->ring[wc.wr_id &
-						(MLX4_NUM_TUNNEL_BUFS - 1)].addr))->grh);
+						(MLX4_NUM_WIRE_BUFS - 1)].addr))->grh);
 				mlx4_ib_demux_mad(ctx->ib_dev, ctx->port, &wc, grh, mad);
 				if (mlx4_ib_post_pv_qp_buf(ctx, sqp, wc.wr_id &
-							   (MLX4_NUM_TUNNEL_BUFS - 1)))
+							   (MLX4_NUM_WIRE_BUFS - 1)))
 					pr_err("Failed reposting SQP "
 					       "buf:%lld\n", wc.wr_id);
 				break;
@@ -1923,8 +1926,8 @@ static void mlx4_ib_sqp_comp_worker(struct work_struct *work)
 				 ctx->slave, wc.status, wc.wr_id);
 			if (!MLX4_TUN_IS_RECV(wc.wr_id)) {
 				kfree(sqp->tx_ring[wc.wr_id &
-				      (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
-				sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+				      (MLX4_NUM_WIRE_BUFS - 1)].ah);
+				sqp->tx_ring[wc.wr_id & (MLX4_NUM_WIRE_BUFS - 1)].ah
 					= NULL;
 				spin_lock(&sqp->tx_lock);
 				sqp->tx_ix_tail++;
@@ -1964,6 +1967,7 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port,
 {
 	int ret, cq_size;
 	struct ib_cq_init_attr cq_attr = {};
+	const int nmbr_bufs = create_tun ? MLX4_NUM_TUNNEL_BUFS : MLX4_NUM_WIRE_BUFS;
 
 	if (ctx->state != DEMUX_PV_STATE_DOWN)
 		return -EEXIST;
@@ -1988,7 +1992,7 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port,
 		goto err_out_qp0;
 	}
 
-	cq_size = 2 * MLX4_NUM_TUNNEL_BUFS;
+	cq_size = 2 * nmbr_bufs;
 	if (ctx->has_smi)
 		cq_size *= 2;
 
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 6f4ea1067095..a0b0fcb814d9 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -233,7 +233,8 @@ enum mlx4_ib_mad_ifc_flags {
 };
 
 enum {
-	MLX4_NUM_TUNNEL_BUFS		= 256,
+	MLX4_NUM_TUNNEL_BUFS		= 512,
+	MLX4_NUM_WIRE_BUFS		= 2048,
 };
 
 struct mlx4_ib_tunnel_header {
-- 
2.20.1

[PATCH for-rc v2 4/6] IB/mlx4: Fix starvation in paravirt mux/demux

[Håkon Bugge]

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 have sent their N packets as well.

The PF driver has 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 than 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 send-queue. If many remote peers send MAD
packets to the very same VM, the tunnel send-queue destined to the VM
could allegedly be construed to be full by the PF driver.

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.

Fixes: 3cf69cc8dbeb ("IB/mlx4: Add CM paravirtualization")
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>

---
    v1->v2:
	* Massaged commit message
---
 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 e1310820352e..8bd16474708f 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -1297,6 +1297,18 @@ 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)
@@ -1997,7 +2009,8 @@ 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);
@@ -2034,6 +2047,7 @@ 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) {
@@ -2177,7 +2191,7 @@ 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);
@@ -2185,7 +2199,15 @@ 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);
@@ -2196,6 +2218,10 @@ 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;
 
@@ -2243,12 +2269,14 @@ 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 a0b0fcb814d9..8686f8ce7529 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

[PATCH for-rc v2 5/6] IB/mlx4: Add support for REJ due to timeout

[Håkon Bugge]

A CM REJ packet with its reason equal to timeout is a special beast in
the sense that it doesn't have a Remote Communication ID nor does it
have a Remote Port GID.

Using CX-3 virtual functions, either from a bare-metal machine or
pass-through from a VM, MAD packets are proxied through the PF driver.

Since the VF drivers have separate name spaces for MAD Transaction Ids
(TIDs), the PF driver has to re-map the TIDs and keep the book keeping
in a cache.

This proxying doesn't not handle said REJ packets.

If the active side abandons its connection attempt after having sent a
REQ, it will send a REJ with the reason being timeout. This example
can be provoked by a simple user-verbs program, which ends up doing:

    rdma_connect(cm_id, &conn_param);
    rdma_destroy_id(cm_id);

using the async librdmacm API.

Having dynamic debug prints enabled in the mlx4_ib driver, we will
then see:

mlx4_ib_demux_cm_handler: Couldn't find an entry for pv_cm_id 0x0, attr_id 0x12

The solution is to introduce a radix-tree. When a REQ packet is
received and handled in mlx4_ib_demux_cm_handler(), we know the
connecting peer's para-virtual cm_id and the destination slave. We
then insert an entry into the tree with said information. We also
schedule work to remove this entry from the tree and free it, in order
to avoid memory leak.

When a REJ packet with reason timeout is received, we can look up the
slave in the tree, and deliver the packet to the correct slave.

When a duplicate REQ packet is received, the entry is in the tree. In
this case, we adjust the delayed work in order to avoid a too
premature eviction of the entry.

When cleaning up, we simply traverse the tree and modify any delayed
work to use a zero delay. A subsequent flush of the system_wq will
ensure all entries being wiped out.

Fixes: 3cf69cc8dbeb ("IB/mlx4: Add CM paravirtualization")
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>

---
    v1 -> v2:
    * Removed noisy pr_debug()
    * Adjust the delay of the work when a duplicate REQ is observed
    * Re-factor the lookup of the slave to accommodate for a REJ sent
      due to timeout receiving the MRA
    * Added __rcu to "void **slot" in rej_tmout_tree_cleanup()
---
 drivers/infiniband/hw/mlx4/cm.c      | 135 ++++++++++++++++++++++++++-
 drivers/infiniband/hw/mlx4/mlx4_ib.h |   3 +
 2 files changed, 137 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
index 6f0ffd0906e6..b10737f74449 100644
--- a/drivers/infiniband/hw/mlx4/cm.c
+++ b/drivers/infiniband/hw/mlx4/cm.c
@@ -54,11 +54,22 @@ struct id_map_entry {
 	struct delayed_work timeout;
 };
 
+struct rej_tmout_entry {
+	int slave;
+	u32 rem_pv_cm_id;
+	struct delayed_work timeout;
+	struct radix_tree_root *rej_tmout_root;
+	/* Points to the mutex protecting this radix-tree */
+	struct mutex *lock;
+};
+
 struct cm_generic_msg {
 	struct ib_mad_hdr hdr;
 
 	__be32 local_comm_id;
 	__be32 remote_comm_id;
+	unsigned char unused[2];
+	__be16 rej_reason;
 };
 
 struct cm_sidr_generic_msg {
@@ -285,6 +296,7 @@ static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)
 	spin_unlock(&sriov->id_map_lock);
 }
 
+#define REJ_REASON(m) be16_to_cpu(((struct cm_generic_msg *)(m))->rej_reason)
 int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id,
 		struct ib_mad *mad)
 {
@@ -295,7 +307,8 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id
 	if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
 	    mad->mad_hdr.attr_id == CM_REP_ATTR_ID ||
 	    mad->mad_hdr.attr_id == CM_MRA_ATTR_ID ||
-	    mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
+	    mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID ||
+	    (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID && REJ_REASON(mad) == IB_CM_REJ_TIMEOUT)) {
 		sl_cm_id = get_local_comm_id(mad);
 		id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id);
 		if (id)
@@ -328,11 +341,88 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id
 	return 0;
 }
 
+static void rej_tmout_timeout(struct work_struct *work)
+{
+	struct delayed_work *delay = to_delayed_work(work);
+	struct rej_tmout_entry *item = container_of(delay, struct rej_tmout_entry, timeout);
+	struct rej_tmout_entry *deleted;
+
+	mutex_lock(item->lock);
+	deleted = radix_tree_delete_item(item->rej_tmout_root, item->rem_pv_cm_id, NULL);
+	mutex_unlock(item->lock);
+
+	if (deleted != item)
+		pr_debug("deleted(%p) != item(%p)\n", deleted, item);
+
+	kfree(item);
+}
+
+static int alloc_rej_tmout(struct mlx4_ib_sriov *sriov, u32 rem_pv_cm_id, int slave)
+{
+	struct rej_tmout_entry *item;
+	int sts;
+
+	mutex_lock(&sriov->rej_tmout_lock);
+	item = radix_tree_lookup(&sriov->rej_tmout_root, (unsigned long)rem_pv_cm_id);
+	mutex_unlock(&sriov->rej_tmout_lock);
+	if (item) {
+		if (IS_ERR(item))
+			return PTR_ERR(item);
+		/* If a retry, adjust delayed work */
+		mod_delayed_work(system_wq, &item->timeout, CM_CLEANUP_CACHE_TIMEOUT);
+		return 0;
+	}
+
+	item = kmalloc(sizeof(*item), GFP_KERNEL);
+	if (!item)
+		return -ENOMEM;
+
+	INIT_DELAYED_WORK(&item->timeout, rej_tmout_timeout);
+	item->slave = slave;
+	item->rem_pv_cm_id = rem_pv_cm_id;
+	item->rej_tmout_root = &sriov->rej_tmout_root;
+	item->lock = &sriov->rej_tmout_lock;
+
+	mutex_lock(&sriov->rej_tmout_lock);
+	sts = radix_tree_insert(&sriov->rej_tmout_root, (unsigned long)rem_pv_cm_id, item);
+	mutex_unlock(&sriov->rej_tmout_lock);
+	if (sts)
+		goto err_insert;
+
+	schedule_delayed_work(&item->timeout, CM_CLEANUP_CACHE_TIMEOUT);
+
+	return 0;
+
+err_insert:
+	kfree(item);
+	return sts;
+}
+
+static int lookup_rej_tmout_slave(struct mlx4_ib_sriov *sriov, u32 rem_pv_cm_id)
+{
+	struct rej_tmout_entry *item;
+
+	mutex_lock(&sriov->rej_tmout_lock);
+	item = radix_tree_lookup(&sriov->rej_tmout_root, (unsigned long)rem_pv_cm_id);
+	mutex_unlock(&sriov->rej_tmout_lock);
+
+	if (!item || IS_ERR(item)) {
+		pr_debug("Could not find slave. rem_pv_cm_id 0x%x error: %d\n",
+			 rem_pv_cm_id, (int)PTR_ERR(item));
+		return !item ? -ENOENT : PTR_ERR(item);
+	}
+
+	return item->slave;
+}
+
 int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
 			     struct ib_mad *mad)
 {
+	struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+	u32 rem_pv_cm_id = get_local_comm_id(mad);
 	u32 pv_cm_id;
 	struct id_map_entry *id;
+	int sts;
 
 	if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
 	    mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
@@ -348,6 +438,13 @@ int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
 				     be64_to_cpu(gid.global.interface_id));
 			return -ENOENT;
 		}
+
+		sts = alloc_rej_tmout(sriov, rem_pv_cm_id, *slave);
+		if (sts)
+			/* Even if this fails, we pass on the REQ to the slave */
+			pr_debug("Could not allocate rej_tmout entry. rem_pv_cm_id 0x%x slave %d status %d\n",
+				 rem_pv_cm_id, *slave, sts);
+
 		return 0;
 	}
 
@@ -355,6 +452,12 @@ int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
 	id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1);
 
 	if (!id) {
+		if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID &&
+		    REJ_REASON(mad) == IB_CM_REJ_TIMEOUT && slave) {
+			*slave = lookup_rej_tmout_slave(sriov, rem_pv_cm_id);
+
+			return (*slave < 0) ? *slave : 0;
+		}
 		pr_debug("Couldn't find an entry for pv_cm_id 0x%x, attr_id 0x%x\n",
 			 pv_cm_id, be16_to_cpu(mad->mad_hdr.attr_id));
 		return -ENOENT;
@@ -377,6 +480,34 @@ void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)
 	INIT_LIST_HEAD(&dev->sriov.cm_list);
 	dev->sriov.sl_id_map = RB_ROOT;
 	xa_init_flags(&dev->sriov.pv_id_table, XA_FLAGS_ALLOC);
+	mutex_init(&dev->sriov.rej_tmout_lock);
+	INIT_RADIX_TREE(&dev->sriov.rej_tmout_root, GFP_KERNEL);
+}
+
+static void rej_tmout_tree_cleanup(struct mlx4_ib_sriov *sriov, int slave)
+{
+	struct radix_tree_iter iter;
+	bool flush_needed = false;
+	__rcu void **slot;
+	int cnt = 0;
+
+	mutex_lock(&sriov->rej_tmout_lock);
+	radix_tree_for_each_slot(slot, &sriov->rej_tmout_root, &iter, 0) {
+		struct rej_tmout_entry *item = *slot;
+
+		if (slave < 0 || slave == item->slave) {
+			mod_delayed_work(system_wq, &item->timeout, 0);
+			flush_needed = true;
+			++cnt;
+		}
+	}
+	mutex_unlock(&sriov->rej_tmout_lock);
+
+	if (flush_needed) {
+		flush_scheduled_work();
+		pr_debug("Deleted %d entries in radix_tree for slave %d during cleanup\n",
+			 slave, cnt);
+	}
 }
 
 /* slave = -1 ==> all slaves */
@@ -446,4 +577,6 @@ void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
 		list_del(&map->list);
 		kfree(map);
 	}
+
+	rej_tmout_tree_cleanup(sriov, slave);
 }
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 8686f8ce7529..2ab83ed3a097 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -495,6 +495,9 @@ struct mlx4_ib_sriov {
 	spinlock_t id_map_lock;
 	struct rb_root sl_id_map;
 	struct list_head cm_list;
+	/* Protects the radix-tree */
+	struct mutex rej_tmout_lock;
+	struct radix_tree_root rej_tmout_root;
 };
 
 struct gid_cache_context {
-- 
2.20.1

Re: [PATCH for-rc v2 5/6] IB/mlx4: Add support for REJ due to timeout

[kernel test robot]

[-- Attachment #1: Type: text/plain, Size: 2462 bytes --]

Hi "Håkon,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on rdma/for-next]
[also build test WARNING on v5.8 next-20200803]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/H-kon-Bugge/Add-CM-packets-missing-and-harden-the-proxying/20200803-142244
base:   https://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma.git for-next
config: x86_64-randconfig-s021-20200804 (attached as .config)
compiler: gcc-9 (Debian 9.3.0-14) 9.3.0
reproduce:
        # apt-get install sparse
        # sparse version: v0.6.2-117-g8c7aee71-dirty
        # save the attached .config to linux build tree
        make W=1 C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' ARCH=x86_64 

If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>


sparse warnings: (new ones prefixed by >>)

>> drivers/infiniband/hw/mlx4/cm.c:496:48: sparse: sparse: incorrect type in initializer (different address spaces) @@     expected struct rej_tmout_entry *item @@     got void [noderef] __rcu * @@
>> drivers/infiniband/hw/mlx4/cm.c:496:48: sparse:     expected struct rej_tmout_entry *item
   drivers/infiniband/hw/mlx4/cm.c:496:48: sparse:     got void [noderef] __rcu *

vim +496 drivers/infiniband/hw/mlx4/cm.c

   486	
   487	static void rej_tmout_tree_cleanup(struct mlx4_ib_sriov *sriov, int slave)
   488	{
   489		struct radix_tree_iter iter;
   490		bool flush_needed = false;
   491		__rcu void **slot;
   492		int cnt = 0;
   493	
   494		mutex_lock(&sriov->rej_tmout_lock);
   495		radix_tree_for_each_slot(slot, &sriov->rej_tmout_root, &iter, 0) {
 > 496			struct rej_tmout_entry *item = *slot;
   497	
   498			if (slave < 0 || slave == item->slave) {
   499				mod_delayed_work(system_wq, &item->timeout, 0);
   500				flush_needed = true;
   501				++cnt;
   502			}
   503		}
   504		mutex_unlock(&sriov->rej_tmout_lock);
   505	
   506		if (flush_needed) {
   507			flush_scheduled_work();
   508			pr_debug("Deleted %d entries in radix_tree for slave %d during cleanup\n",
   509				 slave, cnt);
   510		}
   511	}
   512	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

[-- Attachment #2: config.gz --]
[-- Type: application/gzip, Size: 38108 bytes --]

[PATCH for-rc v2 6/6] IB/mlx4: Adjust delayed work when a dup is observed

[Håkon Bugge]

When scheduling delayed work to clean up the cache, if the entry
already has been scheduled for deletion, we adjust the delay.

Fixes: 3cf69cc8dbeb ("IB/mlx4: Add CM paravirtualization")
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
---
 drivers/infiniband/hw/mlx4/cm.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
index b10737f74449..0ce4b5a3ffa7 100644
--- a/drivers/infiniband/hw/mlx4/cm.c
+++ b/drivers/infiniband/hw/mlx4/cm.c
@@ -291,6 +291,9 @@ static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)
 	if (!sriov->is_going_down && !id->scheduled_delete) {
 		id->scheduled_delete = 1;
 		schedule_delayed_work(&id->timeout, CM_CLEANUP_CACHE_TIMEOUT);
+	} else if (id->scheduled_delete) {
+		/* Adjust timeout if already scheduled */
+		mod_delayed_work(system_wq, &id->timeout, CM_CLEANUP_CACHE_TIMEOUT);
 	}
 	spin_unlock_irqrestore(&sriov->going_down_lock, flags);
 	spin_unlock(&sriov->id_map_lock);
-- 
2.20.1

Re: [PATCH for-rc v2 0/6] Add CM packets missing and harden the proxying

[Håkon Bugge]

A friendly reminder.


Thxs, Håkon


> On 3 Aug 2020, at 08:19, Håkon Bugge <haakon.bugge@oracle.com> wrote:
> 
> A high number of MAD packet drops are observed in the mlx4 MAD proxy
> system. These are fixed by separating the parameters for the tunnel
> vs. wire QPs and by introducing a separate worker-thread for the wire
> QPs.
> 
> Support for MRA and REJ with its reason being timeout is also added.
> 
> Dynamic debug prints adjusted and amended.
> 
>    v1->v2:
> 	* Added commit ("Adjust delayed work when a dup is observed")
> 	* Minor adjustments in some of the commits
> 
> Håkon Bugge (6):
>  IB/mlx4: Add and improve logging
>  IB/mlx4: Add support for MRA
>  IB/mlx4: Separate tunnel and wire bufs parameters
>  IB/mlx4: Fix starvation in paravirt mux/demux
>  IB/mlx4: Add support for REJ due to timeout
>  IB/mlx4: Adjust delayed work when a dup is observed
> 
> drivers/infiniband/hw/mlx4/cm.c      | 148 ++++++++++++++++++++++++-
> drivers/infiniband/hw/mlx4/mad.c     | 158 +++++++++++++++------------
> drivers/infiniband/hw/mlx4/mlx4_ib.h |   8 +-
> 3 files changed, 241 insertions(+), 73 deletions(-)
> 
> --
> 2.20.1
> 

Re: [PATCH for-rc v2 0/6] Add CM packets missing and harden the proxying

[Leon Romanovsky]

On Mon, Aug 10, 2020 at 01:20:43PM +0200, Håkon Bugge wrote:
> A friendly reminder.

We are in merge window.

BTW, the patches are in our regression all that time and everything works
as expected.

Thanks

>
>
> Thxs, Håkon
>
>
> > On 3 Aug 2020, at 08:19, Håkon Bugge <haakon.bugge@oracle.com> wrote:
> >
> > A high number of MAD packet drops are observed in the mlx4 MAD proxy
> > system. These are fixed by separating the parameters for the tunnel
> > vs. wire QPs and by introducing a separate worker-thread for the wire
> > QPs.
> >
> > Support for MRA and REJ with its reason being timeout is also added.
> >
> > Dynamic debug prints adjusted and amended.
> >
> >    v1->v2:
> > 	* Added commit ("Adjust delayed work when a dup is observed")
> > 	* Minor adjustments in some of the commits
> >
> > Håkon Bugge (6):
> >  IB/mlx4: Add and improve logging
> >  IB/mlx4: Add support for MRA
> >  IB/mlx4: Separate tunnel and wire bufs parameters
> >  IB/mlx4: Fix starvation in paravirt mux/demux
> >  IB/mlx4: Add support for REJ due to timeout
> >  IB/mlx4: Adjust delayed work when a dup is observed
> >
> > drivers/infiniband/hw/mlx4/cm.c      | 148 ++++++++++++++++++++++++-
> > drivers/infiniband/hw/mlx4/mad.c     | 158 +++++++++++++++------------
> > drivers/infiniband/hw/mlx4/mlx4_ib.h |   8 +-
> > 3 files changed, 241 insertions(+), 73 deletions(-)
> >
> > --
> > 2.20.1
> >
>

Re: [PATCH for-rc v2 0/6] Add CM packets missing and harden the proxying

[Gal Pressman]

On 10/08/2020 14:46, Leon Romanovsky wrote:
> On Mon, Aug 10, 2020 at 01:20:43PM +0200, Håkon Bugge wrote:
>> A friendly reminder.
> 
> We are in merge window.

The merge window shouldn't affect bug fixes submissions, no?

Re: [PATCH for-rc v2 0/6] Add CM packets missing and harden the proxying

[Leon Romanovsky]

On Mon, Aug 10, 2020 at 05:10:44PM +0300, Gal Pressman wrote:
> On 10/08/2020 14:46, Leon Romanovsky wrote:
> > On Mon, Aug 10, 2020 at 01:20:43PM +0200, Håkon Bugge wrote:
> >> A friendly reminder.
> >
> > We are in merge window.
>
> The merge window shouldn't affect bug fixes submissions, no?

It is hard to call bug fixes, according to Fixes line and description
the code is broken from day one. There is no urgency to merge it now.

Thanks

Re: [PATCH for-rc v2 0/6] Add CM packets missing and harden the proxying

[Jason Gunthorpe]

On Mon, Aug 03, 2020 at 08:19:35AM +0200, Håkon Bugge wrote:
> A high number of MAD packet drops are observed in the mlx4 MAD proxy
> system. These are fixed by separating the parameters for the tunnel
> vs. wire QPs and by introducing a separate worker-thread for the wire
> QPs.
> 
> Support for MRA and REJ with its reason being timeout is also added.
> 
> Dynamic debug prints adjusted and amended.
> 
>     v1->v2:
> 	* Added commit ("Adjust delayed work when a dup is observed")
> 	* Minor adjustments in some of the commits
> 
> Håkon Bugge (6):
>   IB/mlx4: Add and improve logging
>   IB/mlx4: Add support for MRA
>   IB/mlx4: Separate tunnel and wire bufs parameters
>   IB/mlx4: Fix starvation in paravirt mux/demux
>   IB/mlx4: Add support for REJ due to timeout
>   IB/mlx4: Adjust delayed work when a dup is observed

Applied to for-next, this does not look like -rc material

Thanks,
Jason