mlx4: Adjust QP1 multiplexing for RoCE/SRIOV
This requires the following modifications:
1. Fix build_mlx4_header to properly fill in the ETH fields
2. Adjust mux and demux QP1 flow to support RoCE.
This commit still assumes only one GID per slave for RoCE.
The commit enabling multiple GIDs is a subsequent commit, and
is done separately because of its complexity.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
index d1f5f1d..b8d9115 100644
--- a/drivers/infiniband/hw/mlx4/cm.c
+++ b/drivers/infiniband/hw/mlx4/cm.c
@@ -315,7 +315,7 @@
}
int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
- struct ib_mad *mad)
+ struct ib_mad *mad)
{
u32 pv_cm_id;
struct id_map_entry *id;
@@ -323,6 +323,9 @@
if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID) {
union ib_gid gid;
+ if (!slave)
+ return 0;
+
gid = gid_from_req_msg(ibdev, mad);
*slave = mlx4_ib_find_real_gid(ibdev, port, gid.global.interface_id);
if (*slave < 0) {
@@ -341,7 +344,8 @@
return -ENOENT;
}
- *slave = id->slave_id;
+ if (slave)
+ *slave = id->slave_id;
set_remote_comm_id(mad, id->sl_cm_id);
if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index f2a3f48..c2e9879 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -467,6 +467,7 @@
int ret = 0;
u16 tun_pkey_ix;
u16 cached_pkey;
+ u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
if (dest_qpt > IB_QPT_GSI)
return -EINVAL;
@@ -509,6 +510,12 @@
* The driver will set the force loopback bit in post_send */
memset(&attr, 0, sizeof attr);
attr.port_num = port;
+ if (is_eth) {
+ ret = mlx4_get_roce_gid_from_slave(dev->dev, port, slave, attr.grh.dgid.raw);
+ if (ret)
+ return ret;
+ attr.ah_flags = IB_AH_GRH;
+ }
ah = ib_create_ah(tun_ctx->pd, &attr);
if (IS_ERR(ah))
return -ENOMEM;
@@ -580,6 +587,41 @@
int err;
int slave;
u8 *slave_id;
+ int is_eth = 0;
+
+ if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
+ is_eth = 0;
+ else
+ is_eth = 1;
+
+ if (is_eth) {
+ if (!(wc->wc_flags & IB_WC_GRH)) {
+ mlx4_ib_warn(ibdev, "RoCE grh not present.\n");
+ return -EINVAL;
+ }
+ if (mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_CM) {
+ mlx4_ib_warn(ibdev, "RoCE mgmt class is not CM\n");
+ return -EINVAL;
+ }
+ if (mlx4_get_slave_from_roce_gid(dev->dev, port, grh->dgid.raw, &slave)) {
+ mlx4_ib_warn(ibdev, "failed matching grh\n");
+ return -ENOENT;
+ }
+ if (slave >= dev->dev->caps.sqp_demux) {
+ mlx4_ib_warn(ibdev, "slave id: %d is bigger than allowed:%d\n",
+ slave, dev->dev->caps.sqp_demux);
+ return -ENOENT;
+ }
+
+ if (mlx4_ib_demux_cm_handler(ibdev, port, NULL, mad))
+ return 0;
+
+ 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",
+ slave, err);
+ return 0;
+ }
/* Initially assume that this mad is for us */
slave = mlx4_master_func_num(dev->dev);
@@ -1260,12 +1302,8 @@
memcpy(&ah.av, &tunnel->hdr.av, sizeof (struct mlx4_av));
ah.ibah.device = ctx->ib_dev;
mlx4_ib_query_ah(&ah.ibah, &ah_attr);
- if ((ah_attr.ah_flags & IB_AH_GRH) &&
- (ah_attr.grh.sgid_index != slave)) {
- mlx4_ib_warn(ctx->ib_dev, "slave:%d accessed invalid sgid_index:%d\n",
- slave, ah_attr.grh.sgid_index);
- return;
- }
+ if (ah_attr.ah_flags & IB_AH_GRH)
+ ah_attr.grh.sgid_index = slave;
mlx4_ib_send_to_wire(dev, slave, ctx->port,
is_proxy_qp0(dev, wc->src_qp, slave) ?
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index d8f4d1f..c6ef2e7 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1842,9 +1842,9 @@
{
struct ib_device *ib_dev = sqp->qp.ibqp.device;
struct mlx4_wqe_mlx_seg *mlx = wqe;
+ struct mlx4_wqe_ctrl_seg *ctrl = wqe;
struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
- struct net_device *ndev;
union ib_gid sgid;
u16 pkey;
int send_size;
@@ -1868,12 +1868,11 @@
/* When multi-function is enabled, the ib_core gid
* indexes don't necessarily match the hw ones, so
* we must use our own cache */
- sgid.global.subnet_prefix =
- to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
- subnet_prefix;
- sgid.global.interface_id =
- to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
- guid_cache[ah->av.ib.gid_index];
+ err = mlx4_get_roce_gid_from_slave(to_mdev(ib_dev)->dev,
+ be32_to_cpu(ah->av.ib.port_pd) >> 24,
+ ah->av.ib.gid_index, &sgid.raw[0]);
+ if (err)
+ return err;
} else {
err = ib_get_cached_gid(ib_dev,
be32_to_cpu(ah->av.ib.port_pd) >> 24,
@@ -1902,6 +1901,9 @@
sqp->ud_header.grh.flow_label =
ah->av.ib.sl_tclass_flowlabel & cpu_to_be32(0xfffff);
sqp->ud_header.grh.hop_limit = ah->av.ib.hop_limit;
+ if (is_eth)
+ memcpy(sqp->ud_header.grh.source_gid.raw, sgid.raw, 16);
+ else {
if (mlx4_is_mfunc(to_mdev(ib_dev)->dev)) {
/* When multi-function is enabled, the ib_core gid
* indexes don't necessarily match the hw ones, so
@@ -1917,6 +1919,7 @@
be32_to_cpu(ah->av.ib.port_pd) >> 24,
ah->av.ib.gid_index,
&sqp->ud_header.grh.source_gid);
+ }
memcpy(sqp->ud_header.grh.destination_gid.raw,
ah->av.ib.dgid, 16);
}
@@ -1948,17 +1951,19 @@
}
if (is_eth) {
- u8 *smac;
+ u8 smac[6];
+ struct in6_addr in6;
+
u16 pcp = (be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 29) << 13;
mlx->sched_prio = cpu_to_be16(pcp);
memcpy(sqp->ud_header.eth.dmac_h, ah->av.eth.mac, 6);
/* FIXME: cache smac value? */
- ndev = to_mdev(sqp->qp.ibqp.device)->iboe.netdevs[sqp->qp.port - 1];
- if (!ndev)
- return -ENODEV;
- smac = ndev->dev_addr;
+ memcpy(&ctrl->srcrb_flags16[0], ah->av.eth.mac, 2);
+ memcpy(&ctrl->imm, ah->av.eth.mac + 2, 4);
+ memcpy(&in6, sgid.raw, sizeof(in6));
+ rdma_get_ll_mac(&in6, smac);
memcpy(sqp->ud_header.eth.smac_h, smac, 6);
if (!memcmp(sqp->ud_header.eth.smac_h, sqp->ud_header.eth.dmac_h, 6))
mlx->flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
index 7aec6c8..da829f4 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
@@ -788,6 +788,10 @@
MLX4_USE_RR = 1,
};
+struct mlx4_roce_gid_entry {
+ u8 raw[16];
+};
+
struct mlx4_priv {
struct mlx4_dev dev;
@@ -834,6 +838,7 @@
int fs_hash_mode;
u8 virt2phys_pkey[MLX4_MFUNC_MAX][MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS];
__be64 slave_node_guids[MLX4_MFUNC_MAX];
+ struct mlx4_roce_gid_entry roce_gids[MLX4_MAX_PORTS][MLX4_ROCE_MAX_GIDS];
atomic_t opreq_count;
struct work_struct opreq_task;
diff --git a/drivers/net/ethernet/mellanox/mlx4/port.c b/drivers/net/ethernet/mellanox/mlx4/port.c
index a58bcbf..9c063d6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/port.c
+++ b/drivers/net/ethernet/mellanox/mlx4/port.c
@@ -927,3 +927,37 @@
*stats_bitmap |= MLX4_STATS_ERROR_COUNTERS_MASK;
}
EXPORT_SYMBOL(mlx4_set_stats_bitmap);
+
+int mlx4_get_slave_from_roce_gid(struct mlx4_dev *dev, int port, u8 *gid, int *slave_id)
+{
+ struct mlx4_priv *priv = mlx4_priv(dev);
+ int i, found_ix = -1;
+
+ if (!mlx4_is_mfunc(dev))
+ return -EINVAL;
+
+ for (i = 0; i < MLX4_ROCE_MAX_GIDS; i++) {
+ if (!memcmp(priv->roce_gids[port - 1][i].raw, gid, 16)) {
+ found_ix = i;
+ break;
+ }
+ }
+
+ if (found_ix >= 0)
+ *slave_id = found_ix;
+
+ return (found_ix >= 0) ? 0 : -EINVAL;
+}
+EXPORT_SYMBOL(mlx4_get_slave_from_roce_gid);
+
+int mlx4_get_roce_gid_from_slave(struct mlx4_dev *dev, int port, int slave_id, u8 *gid)
+{
+ struct mlx4_priv *priv = mlx4_priv(dev);
+
+ if (!mlx4_is_master(dev))
+ return -EINVAL;
+
+ memcpy(gid, priv->roce_gids[port - 1][slave_id].raw, 16);
+ return 0;
+}
+EXPORT_SYMBOL(mlx4_get_roce_gid_from_slave);