{net, IB}/mlx5: Manage port association for multiport RoCE
When mlx5_ib_add is called determine if the mlx5 core device being
added is capable of dual port RoCE operation. If it is, determine
whether it is a master device or a slave device using the
num_vhca_ports and affiliate_nic_vport_criteria capabilities.
If the device is a slave, attempt to find a master device to affiliate it
with. Devices that can be affiliated will share a system image guid. If
none are found place it on a list of unaffiliated ports. If a master is
found bind the port to it by configuring the port affiliation in the NIC
vport context.
Similarly when mlx5_ib_remove is called determine the port type. If it's
a slave port, unaffiliate it from the master device, otherwise just
remove it from the unaffiliated port list.
The IB device is registered as a multiport device, even if a 2nd port is
not available for affiliation. When the 2nd port is affiliated later the
GID cache must be refreshed in order to get the default GIDs for the 2nd
port in the cache. Export roce_rescan_device to provide a mechanism to
refresh the cache after a new port is bound.
In a multiport configuration all IB object (QP, MR, PD, etc) related
commands should flow through the master mlx5_core_dev, other commands
must be sent to the slave port mlx5_core_mdev, an interface is provide
to get the correct mdev for non IB object commands.
Signed-off-by: Daniel Jurgens <danielj@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 5fcb2ed..4fbbe4c 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -74,6 +74,23 @@ enum {
MLX5_ATOMIC_SIZE_QP_8BYTES = 1 << 3,
};
+static LIST_HEAD(mlx5_ib_unaffiliated_port_list);
+static LIST_HEAD(mlx5_ib_dev_list);
+/*
+ * This mutex should be held when accessing either of the above lists
+ */
+static DEFINE_MUTEX(mlx5_ib_multiport_mutex);
+
+struct mlx5_ib_dev *mlx5_ib_get_ibdev_from_mpi(struct mlx5_ib_multiport_info *mpi)
+{
+ struct mlx5_ib_dev *dev;
+
+ mutex_lock(&mlx5_ib_multiport_mutex);
+ dev = mpi->ibdev;
+ mutex_unlock(&mlx5_ib_multiport_mutex);
+ return dev;
+}
+
static enum rdma_link_layer
mlx5_port_type_cap_to_rdma_ll(int port_type_cap)
{
@@ -120,7 +137,9 @@ static int mlx5_netdev_event(struct notifier_block *this,
struct mlx5_ib_dev *ibdev;
ibdev = roce->dev;
- mdev = ibdev->mdev;
+ mdev = mlx5_ib_get_native_port_mdev(ibdev, port_num, NULL);
+ if (!mdev)
+ return NOTIFY_DONE;
switch (event) {
case NETDEV_REGISTER:
@@ -175,6 +194,7 @@ static int mlx5_netdev_event(struct notifier_block *this,
break;
}
done:
+ mlx5_ib_put_native_port_mdev(ibdev, port_num);
return NOTIFY_DONE;
}
@@ -183,10 +203,15 @@ static struct net_device *mlx5_ib_get_netdev(struct ib_device *device,
{
struct mlx5_ib_dev *ibdev = to_mdev(device);
struct net_device *ndev;
+ struct mlx5_core_dev *mdev;
- ndev = mlx5_lag_get_roce_netdev(ibdev->mdev);
+ mdev = mlx5_ib_get_native_port_mdev(ibdev, port_num, NULL);
+ if (!mdev)
+ return NULL;
+
+ ndev = mlx5_lag_get_roce_netdev(mdev);
if (ndev)
- return ndev;
+ goto out;
/* Ensure ndev does not disappear before we invoke dev_hold()
*/
@@ -196,9 +221,70 @@ static struct net_device *mlx5_ib_get_netdev(struct ib_device *device,
dev_hold(ndev);
read_unlock(&ibdev->roce[port_num - 1].netdev_lock);
+out:
+ mlx5_ib_put_native_port_mdev(ibdev, port_num);
return ndev;
}
+struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *ibdev,
+ u8 ib_port_num,
+ u8 *native_port_num)
+{
+ enum rdma_link_layer ll = mlx5_ib_port_link_layer(&ibdev->ib_dev,
+ ib_port_num);
+ struct mlx5_core_dev *mdev = NULL;
+ struct mlx5_ib_multiport_info *mpi;
+ struct mlx5_ib_port *port;
+
+ if (native_port_num)
+ *native_port_num = 1;
+
+ if (!mlx5_core_mp_enabled(ibdev->mdev) || ll != IB_LINK_LAYER_ETHERNET)
+ return ibdev->mdev;
+
+ port = &ibdev->port[ib_port_num - 1];
+ if (!port)
+ return NULL;
+
+ spin_lock(&port->mp.mpi_lock);
+ mpi = ibdev->port[ib_port_num - 1].mp.mpi;
+ if (mpi && !mpi->unaffiliate) {
+ mdev = mpi->mdev;
+ /* If it's the master no need to refcount, it'll exist
+ * as long as the ib_dev exists.
+ */
+ if (!mpi->is_master)
+ mpi->mdev_refcnt++;
+ }
+ spin_unlock(&port->mp.mpi_lock);
+
+ return mdev;
+}
+
+void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *ibdev, u8 port_num)
+{
+ enum rdma_link_layer ll = mlx5_ib_port_link_layer(&ibdev->ib_dev,
+ port_num);
+ struct mlx5_ib_multiport_info *mpi;
+ struct mlx5_ib_port *port;
+
+ if (!mlx5_core_mp_enabled(ibdev->mdev) || ll != IB_LINK_LAYER_ETHERNET)
+ return;
+
+ port = &ibdev->port[port_num - 1];
+
+ spin_lock(&port->mp.mpi_lock);
+ mpi = ibdev->port[port_num - 1].mp.mpi;
+ if (mpi->is_master)
+ goto out;
+
+ mpi->mdev_refcnt--;
+ if (mpi->unaffiliate)
+ complete(&mpi->unref_comp);
+out:
+ spin_unlock(&port->mp.mpi_lock);
+}
+
static int translate_eth_proto_oper(u32 eth_proto_oper, u8 *active_speed,
u8 *active_width)
{
@@ -3160,12 +3246,11 @@ static void get_ext_port_caps(struct mlx5_ib_dev *dev)
mlx5_query_ext_port_caps(dev, port);
}
-static int get_port_caps(struct mlx5_ib_dev *dev)
+static int get_port_caps(struct mlx5_ib_dev *dev, u8 port)
{
struct ib_device_attr *dprops = NULL;
struct ib_port_attr *pprops = NULL;
int err = -ENOMEM;
- int port;
struct ib_udata uhw = {.inlen = 0, .outlen = 0};
pprops = kmalloc(sizeof(*pprops), GFP_KERNEL);
@@ -3186,22 +3271,21 @@ static int get_port_caps(struct mlx5_ib_dev *dev)
goto out;
}
- for (port = 1; port <= dev->num_ports; port++) {
- memset(pprops, 0, sizeof(*pprops));
- err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
- if (err) {
- mlx5_ib_warn(dev, "query_port %d failed %d\n",
- port, err);
- break;
- }
- dev->mdev->port_caps[port - 1].pkey_table_len =
- dprops->max_pkeys;
- dev->mdev->port_caps[port - 1].gid_table_len =
- pprops->gid_tbl_len;
- mlx5_ib_dbg(dev, "pkey_table_len %d, gid_table_len %d\n",
- dprops->max_pkeys, pprops->gid_tbl_len);
+ memset(pprops, 0, sizeof(*pprops));
+ err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
+ if (err) {
+ mlx5_ib_warn(dev, "query_port %d failed %d\n",
+ port, err);
+ goto out;
}
+ dev->mdev->port_caps[port - 1].pkey_table_len =
+ dprops->max_pkeys;
+ dev->mdev->port_caps[port - 1].gid_table_len =
+ pprops->gid_tbl_len;
+ mlx5_ib_dbg(dev, "port %d: pkey_table_len %d, gid_table_len %d\n",
+ port, dprops->max_pkeys, pprops->gid_tbl_len);
+
out:
kfree(pprops);
kfree(dprops);
@@ -4054,8 +4138,203 @@ mlx5_ib_get_vector_affinity(struct ib_device *ibdev, int comp_vector)
return mlx5_get_vector_affinity(dev->mdev, comp_vector);
}
+/* The mlx5_ib_multiport_mutex should be held when calling this function */
+static void mlx5_ib_unbind_slave_port(struct mlx5_ib_dev *ibdev,
+ struct mlx5_ib_multiport_info *mpi)
+{
+ u8 port_num = mlx5_core_native_port_num(mpi->mdev) - 1;
+ struct mlx5_ib_port *port = &ibdev->port[port_num];
+ int comps;
+ int err;
+ int i;
+
+ spin_lock(&port->mp.mpi_lock);
+ if (!mpi->ibdev) {
+ spin_unlock(&port->mp.mpi_lock);
+ return;
+ }
+ mpi->ibdev = NULL;
+
+ spin_unlock(&port->mp.mpi_lock);
+ mlx5_remove_netdev_notifier(ibdev, port_num);
+ spin_lock(&port->mp.mpi_lock);
+
+ comps = mpi->mdev_refcnt;
+ if (comps) {
+ mpi->unaffiliate = true;
+ init_completion(&mpi->unref_comp);
+ spin_unlock(&port->mp.mpi_lock);
+
+ for (i = 0; i < comps; i++)
+ wait_for_completion(&mpi->unref_comp);
+
+ spin_lock(&port->mp.mpi_lock);
+ mpi->unaffiliate = false;
+ }
+
+ port->mp.mpi = NULL;
+
+ list_add_tail(&mpi->list, &mlx5_ib_unaffiliated_port_list);
+
+ spin_unlock(&port->mp.mpi_lock);
+
+ err = mlx5_nic_vport_unaffiliate_multiport(mpi->mdev);
+
+ mlx5_ib_dbg(ibdev, "unaffiliated port %d\n", port_num + 1);
+ /* Log an error, still needed to cleanup the pointers and add
+ * it back to the list.
+ */
+ if (err)
+ mlx5_ib_err(ibdev, "Failed to unaffiliate port %u\n",
+ port_num + 1);
+
+ ibdev->roce[port_num].last_port_state = IB_PORT_DOWN;
+}
+
+/* The mlx5_ib_multiport_mutex should be held when calling this function */
+static bool mlx5_ib_bind_slave_port(struct mlx5_ib_dev *ibdev,
+ struct mlx5_ib_multiport_info *mpi)
+{
+ u8 port_num = mlx5_core_native_port_num(mpi->mdev) - 1;
+ int err;
+
+ spin_lock(&ibdev->port[port_num].mp.mpi_lock);
+ if (ibdev->port[port_num].mp.mpi) {
+ mlx5_ib_warn(ibdev, "port %d already affiliated.\n",
+ port_num + 1);
+ spin_unlock(&ibdev->port[port_num].mp.mpi_lock);
+ return false;
+ }
+
+ ibdev->port[port_num].mp.mpi = mpi;
+ mpi->ibdev = ibdev;
+ spin_unlock(&ibdev->port[port_num].mp.mpi_lock);
+
+ err = mlx5_nic_vport_affiliate_multiport(ibdev->mdev, mpi->mdev);
+ if (err)
+ goto unbind;
+
+ err = get_port_caps(ibdev, mlx5_core_native_port_num(mpi->mdev));
+ if (err)
+ goto unbind;
+
+ err = mlx5_add_netdev_notifier(ibdev, port_num);
+ if (err) {
+ mlx5_ib_err(ibdev, "failed adding netdev notifier for port %u\n",
+ port_num + 1);
+ goto unbind;
+ }
+
+ return true;
+
+unbind:
+ mlx5_ib_unbind_slave_port(ibdev, mpi);
+ return false;
+}
+
+static int mlx5_ib_init_multiport_master(struct mlx5_ib_dev *dev)
+{
+ int port_num = mlx5_core_native_port_num(dev->mdev) - 1;
+ enum rdma_link_layer ll = mlx5_ib_port_link_layer(&dev->ib_dev,
+ port_num + 1);
+ struct mlx5_ib_multiport_info *mpi;
+ int err;
+ int i;
+
+ if (!mlx5_core_is_mp_master(dev->mdev) || ll != IB_LINK_LAYER_ETHERNET)
+ return 0;
+
+ err = mlx5_query_nic_vport_system_image_guid(dev->mdev,
+ &dev->sys_image_guid);
+ if (err)
+ return err;
+
+ err = mlx5_nic_vport_enable_roce(dev->mdev);
+ if (err)
+ return err;
+
+ mutex_lock(&mlx5_ib_multiport_mutex);
+ for (i = 0; i < dev->num_ports; i++) {
+ bool bound = false;
+
+ /* build a stub multiport info struct for the native port. */
+ if (i == port_num) {
+ mpi = kzalloc(sizeof(*mpi), GFP_KERNEL);
+ if (!mpi) {
+ mutex_unlock(&mlx5_ib_multiport_mutex);
+ mlx5_nic_vport_disable_roce(dev->mdev);
+ return -ENOMEM;
+ }
+
+ mpi->is_master = true;
+ mpi->mdev = dev->mdev;
+ mpi->sys_image_guid = dev->sys_image_guid;
+ dev->port[i].mp.mpi = mpi;
+ mpi->ibdev = dev;
+ mpi = NULL;
+ continue;
+ }
+
+ list_for_each_entry(mpi, &mlx5_ib_unaffiliated_port_list,
+ list) {
+ if (dev->sys_image_guid == mpi->sys_image_guid &&
+ (mlx5_core_native_port_num(mpi->mdev) - 1) == i) {
+ bound = mlx5_ib_bind_slave_port(dev, mpi);
+ }
+
+ if (bound) {
+ dev_dbg(&mpi->mdev->pdev->dev, "removing port from unaffiliated list.\n");
+ mlx5_ib_dbg(dev, "port %d bound\n", i + 1);
+ list_del(&mpi->list);
+ break;
+ }
+ }
+ if (!bound) {
+ get_port_caps(dev, i + 1);
+ mlx5_ib_dbg(dev, "no free port found for port %d\n",
+ i + 1);
+ }
+ }
+
+ list_add_tail(&dev->ib_dev_list, &mlx5_ib_dev_list);
+ mutex_unlock(&mlx5_ib_multiport_mutex);
+ return err;
+}
+
+static void mlx5_ib_cleanup_multiport_master(struct mlx5_ib_dev *dev)
+{
+ int port_num = mlx5_core_native_port_num(dev->mdev) - 1;
+ enum rdma_link_layer ll = mlx5_ib_port_link_layer(&dev->ib_dev,
+ port_num + 1);
+ int i;
+
+ if (!mlx5_core_is_mp_master(dev->mdev) || ll != IB_LINK_LAYER_ETHERNET)
+ return;
+
+ mutex_lock(&mlx5_ib_multiport_mutex);
+ for (i = 0; i < dev->num_ports; i++) {
+ if (dev->port[i].mp.mpi) {
+ /* Destroy the native port stub */
+ if (i == port_num) {
+ kfree(dev->port[i].mp.mpi);
+ dev->port[i].mp.mpi = NULL;
+ } else {
+ mlx5_ib_dbg(dev, "unbinding port_num: %d\n", i + 1);
+ mlx5_ib_unbind_slave_port(dev, dev->port[i].mp.mpi);
+ }
+ }
+ }
+
+ mlx5_ib_dbg(dev, "removing from devlist\n");
+ list_del(&dev->ib_dev_list);
+ mutex_unlock(&mlx5_ib_multiport_mutex);
+
+ mlx5_nic_vport_disable_roce(dev->mdev);
+}
+
static void mlx5_ib_stage_init_cleanup(struct mlx5_ib_dev *dev)
{
+ mlx5_ib_cleanup_multiport_master(dev);
#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
cleanup_srcu_struct(&dev->mr_srcu);
#endif
@@ -4067,16 +4346,36 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev)
struct mlx5_core_dev *mdev = dev->mdev;
const char *name;
int err;
+ int i;
dev->port = kcalloc(dev->num_ports, sizeof(*dev->port),
GFP_KERNEL);
if (!dev->port)
return -ENOMEM;
- err = get_port_caps(dev);
+ for (i = 0; i < dev->num_ports; i++) {
+ spin_lock_init(&dev->port[i].mp.mpi_lock);
+ rwlock_init(&dev->roce[i].netdev_lock);
+ }
+
+ err = mlx5_ib_init_multiport_master(dev);
if (err)
goto err_free_port;
+ if (!mlx5_core_mp_enabled(mdev)) {
+ int i;
+
+ for (i = 1; i <= dev->num_ports; i++) {
+ err = get_port_caps(dev, i);
+ if (err)
+ break;
+ }
+ } else {
+ err = get_port_caps(dev, mlx5_core_native_port_num(mdev));
+ }
+ if (err)
+ goto err_mp;
+
if (mlx5_use_mad_ifc(dev))
get_ext_port_caps(dev);
@@ -4106,6 +4405,8 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev)
#endif
return 0;
+err_mp:
+ mlx5_ib_cleanup_multiport_master(dev);
err_free_port:
kfree(dev->port);
@@ -4252,16 +4553,16 @@ static int mlx5_ib_stage_roce_init(struct mlx5_ib_dev *dev)
struct mlx5_core_dev *mdev = dev->mdev;
enum rdma_link_layer ll;
int port_type_cap;
- u8 port_num = 0;
+ u8 port_num;
int err;
int i;
+ port_num = mlx5_core_native_port_num(dev->mdev) - 1;
port_type_cap = MLX5_CAP_GEN(mdev, port_type);
ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
if (ll == IB_LINK_LAYER_ETHERNET) {
for (i = 0; i < dev->num_ports; i++) {
- rwlock_init(&dev->roce[i].netdev_lock);
dev->roce[i].dev = dev;
dev->roce[i].native_port_num = i + 1;
dev->roce[i].last_port_state = IB_PORT_DOWN;
@@ -4292,8 +4593,9 @@ static void mlx5_ib_stage_roce_cleanup(struct mlx5_ib_dev *dev)
struct mlx5_core_dev *mdev = dev->mdev;
enum rdma_link_layer ll;
int port_type_cap;
- u8 port_num = 0;
+ u8 port_num;
+ port_num = mlx5_core_native_port_num(dev->mdev) - 1;
port_type_cap = MLX5_CAP_GEN(mdev, port_type);
ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
@@ -4443,6 +4745,8 @@ static void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
ib_dealloc_device((struct ib_device *)dev);
}
+static void *mlx5_ib_add_slave_port(struct mlx5_core_dev *mdev, u8 port_num);
+
static void *__mlx5_ib_add(struct mlx5_core_dev *mdev,
const struct mlx5_ib_profile *profile)
{
@@ -4457,7 +4761,8 @@ static void *__mlx5_ib_add(struct mlx5_core_dev *mdev,
return NULL;
dev->mdev = mdev;
- dev->num_ports = MLX5_CAP_GEN(mdev, num_ports);
+ dev->num_ports = max(MLX5_CAP_GEN(mdev, num_ports),
+ MLX5_CAP_GEN(mdev, num_vhca_ports));
for (i = 0; i < MLX5_IB_STAGE_MAX; i++) {
if (profile->stage[i].init) {
@@ -4520,15 +4825,81 @@ static const struct mlx5_ib_profile pf_profile = {
NULL),
};
+static void *mlx5_ib_add_slave_port(struct mlx5_core_dev *mdev, u8 port_num)
+{
+ struct mlx5_ib_multiport_info *mpi;
+ struct mlx5_ib_dev *dev;
+ bool bound = false;
+ int err;
+
+ mpi = kzalloc(sizeof(*mpi), GFP_KERNEL);
+ if (!mpi)
+ return NULL;
+
+ mpi->mdev = mdev;
+
+ err = mlx5_query_nic_vport_system_image_guid(mdev,
+ &mpi->sys_image_guid);
+ if (err) {
+ kfree(mpi);
+ return NULL;
+ }
+
+ mutex_lock(&mlx5_ib_multiport_mutex);
+ list_for_each_entry(dev, &mlx5_ib_dev_list, ib_dev_list) {
+ if (dev->sys_image_guid == mpi->sys_image_guid)
+ bound = mlx5_ib_bind_slave_port(dev, mpi);
+
+ if (bound) {
+ rdma_roce_rescan_device(&dev->ib_dev);
+ break;
+ }
+ }
+
+ if (!bound) {
+ list_add_tail(&mpi->list, &mlx5_ib_unaffiliated_port_list);
+ dev_dbg(&mdev->pdev->dev, "no suitable IB device found to bind to, added to unaffiliated list.\n");
+ } else {
+ mlx5_ib_dbg(dev, "bound port %u\n", port_num + 1);
+ }
+ mutex_unlock(&mlx5_ib_multiport_mutex);
+
+ return mpi;
+}
+
static void *mlx5_ib_add(struct mlx5_core_dev *mdev)
{
+ enum rdma_link_layer ll;
+ int port_type_cap;
+
+ port_type_cap = MLX5_CAP_GEN(mdev, port_type);
+ ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
+
+ if (mlx5_core_is_mp_slave(mdev) && ll == IB_LINK_LAYER_ETHERNET) {
+ u8 port_num = mlx5_core_native_port_num(mdev) - 1;
+
+ return mlx5_ib_add_slave_port(mdev, port_num);
+ }
+
return __mlx5_ib_add(mdev, &pf_profile);
}
static void mlx5_ib_remove(struct mlx5_core_dev *mdev, void *context)
{
- struct mlx5_ib_dev *dev = context;
+ struct mlx5_ib_multiport_info *mpi;
+ struct mlx5_ib_dev *dev;
+ if (mlx5_core_is_mp_slave(mdev)) {
+ mpi = context;
+ mutex_lock(&mlx5_ib_multiport_mutex);
+ if (mpi->ibdev)
+ mlx5_ib_unbind_slave_port(mpi->ibdev, mpi);
+ list_del(&mpi->list);
+ mutex_unlock(&mlx5_ib_multiport_mutex);
+ return;
+ }
+
+ dev = context;
__mlx5_ib_remove(dev, dev->profile, MLX5_IB_STAGE_MAX);
}