net/mlx5e: Optimize promiscuous mode

Change steering flow to optimize traffic in promiscuous mode. On demand,
add a high priority table containing a catch-all rule. All incoming
packets are caught by this rule and steered directly to the TTC table.
Prior to this change, packets in promiscuous mode may suffer from up to
4 steering hops before reaching TTC table.
In addition, this patch will allow us adding a catch-all rule at the end
of MAC table to serve MAC trap, with no impact on promiscuous mode
performance.

Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit is contained in:
Aya Levin 2021-01-26 15:24:12 -08:00 коммит произвёл Jakub Kicinski
Родитель 241dc15939
Коммит 1c46d7409f
3 изменённых файлов: 100 добавлений и 34 удалений

Просмотреть файл

@ -44,6 +44,11 @@ struct mlx5e_l2_rule {
#define MLX5E_L2_ADDR_HASH_SIZE BIT(BITS_PER_BYTE)
struct mlx5e_promisc_table {
struct mlx5e_flow_table ft;
struct mlx5_flow_handle *rule;
};
struct mlx5e_vlan_table {
struct mlx5e_flow_table ft;
DECLARE_BITMAP(active_cvlans, VLAN_N_VID);
@ -62,7 +67,6 @@ struct mlx5e_l2_table {
struct hlist_head netdev_mc[MLX5E_L2_ADDR_HASH_SIZE];
struct mlx5e_l2_rule broadcast;
struct mlx5e_l2_rule allmulti;
struct mlx5e_l2_rule promisc;
bool broadcast_enabled;
bool allmulti_enabled;
bool promisc_enabled;
@ -126,7 +130,8 @@ struct mlx5e_ttc_table {
/* NIC prio FTS */
enum {
MLX5E_VLAN_FT_LEVEL = 0,
MLX5E_PROMISC_FT_LEVEL,
MLX5E_VLAN_FT_LEVEL,
MLX5E_L2_FT_LEVEL,
MLX5E_TTC_FT_LEVEL,
MLX5E_INNER_TTC_FT_LEVEL,
@ -241,6 +246,7 @@ struct mlx5e_flow_steering {
struct mlx5e_ethtool_steering ethtool;
#endif
struct mlx5e_tc_table tc;
struct mlx5e_promisc_table promisc;
struct mlx5e_vlan_table vlan;
struct mlx5e_l2_table l2;
struct mlx5e_ttc_table ttc;

Просмотреть файл

@ -46,7 +46,6 @@ static void mlx5e_del_l2_flow_rule(struct mlx5e_priv *priv,
enum {
MLX5E_FULLMATCH = 0,
MLX5E_ALLMULTI = 1,
MLX5E_PROMISC = 2,
};
enum {
@ -596,6 +595,83 @@ static void mlx5e_handle_netdev_addr(struct mlx5e_priv *priv)
mlx5e_apply_netdev_addr(priv);
}
#define MLX5E_PROMISC_GROUP0_SIZE BIT(0)
#define MLX5E_PROMISC_TABLE_SIZE MLX5E_PROMISC_GROUP0_SIZE
static int mlx5e_add_promisc_rule(struct mlx5e_priv *priv)
{
struct mlx5_flow_table *ft = priv->fs.promisc.ft.t;
struct mlx5_flow_destination dest = {};
struct mlx5_flow_handle **rule_p;
MLX5_DECLARE_FLOW_ACT(flow_act);
struct mlx5_flow_spec *spec;
int err = 0;
spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
if (!spec)
return -ENOMEM;
dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
dest.ft = priv->fs.ttc.ft.t;
rule_p = &priv->fs.promisc.rule;
*rule_p = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
if (IS_ERR(*rule_p)) {
err = PTR_ERR(*rule_p);
*rule_p = NULL;
netdev_err(priv->netdev, "%s: add promiscuous rule failed\n", __func__);
}
kvfree(spec);
return err;
}
static int mlx5e_create_promisc_table(struct mlx5e_priv *priv)
{
struct mlx5e_flow_table *ft = &priv->fs.promisc.ft;
struct mlx5_flow_table_attr ft_attr = {};
int err;
ft_attr.max_fte = MLX5E_PROMISC_TABLE_SIZE;
ft_attr.autogroup.max_num_groups = 1;
ft_attr.level = MLX5E_PROMISC_FT_LEVEL;
ft_attr.prio = MLX5E_NIC_PRIO;
ft->t = mlx5_create_auto_grouped_flow_table(priv->fs.ns, &ft_attr);
if (IS_ERR(ft->t)) {
err = PTR_ERR(ft->t);
netdev_err(priv->netdev, "fail to create promisc table err=%d\n", err);
return err;
}
err = mlx5e_add_promisc_rule(priv);
if (err)
goto err_destroy_promisc_table;
return 0;
err_destroy_promisc_table:
mlx5_destroy_flow_table(ft->t);
ft->t = NULL;
return err;
}
static void mlx5e_del_promisc_rule(struct mlx5e_priv *priv)
{
if (WARN(!priv->fs.promisc.rule, "Trying to remove non-existing promiscuous rule"))
return;
mlx5_del_flow_rules(priv->fs.promisc.rule);
priv->fs.promisc.rule = NULL;
}
static void mlx5e_destroy_promisc_table(struct mlx5e_priv *priv)
{
if (WARN(!priv->fs.promisc.ft.t, "Trying to remove non-existing promiscuous table"))
return;
mlx5e_del_promisc_rule(priv);
mlx5_destroy_flow_table(priv->fs.promisc.ft.t);
priv->fs.promisc.ft.t = NULL;
}
void mlx5e_set_rx_mode_work(struct work_struct *work)
{
struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
@ -615,14 +691,15 @@ void mlx5e_set_rx_mode_work(struct work_struct *work)
bool disable_allmulti = ea->allmulti_enabled && !allmulti_enabled;
bool enable_broadcast = !ea->broadcast_enabled && broadcast_enabled;
bool disable_broadcast = ea->broadcast_enabled && !broadcast_enabled;
int err;
if (enable_promisc) {
if (!priv->channels.params.vlan_strip_disable)
err = mlx5e_create_promisc_table(priv);
if (err)
enable_promisc = false;
if (!priv->channels.params.vlan_strip_disable && !err)
netdev_warn_once(ndev,
"S-tagged traffic will be dropped while C-tag vlan stripping is enabled\n");
mlx5e_add_l2_flow_rule(priv, &ea->promisc, MLX5E_PROMISC);
if (!priv->fs.vlan.cvlan_filter_disabled)
mlx5e_add_any_vid_rules(priv);
}
if (enable_allmulti)
mlx5e_add_l2_flow_rule(priv, &ea->allmulti, MLX5E_ALLMULTI);
@ -635,11 +712,8 @@ void mlx5e_set_rx_mode_work(struct work_struct *work)
mlx5e_del_l2_flow_rule(priv, &ea->broadcast);
if (disable_allmulti)
mlx5e_del_l2_flow_rule(priv, &ea->allmulti);
if (disable_promisc) {
if (!priv->fs.vlan.cvlan_filter_disabled)
mlx5e_del_any_vid_rules(priv);
mlx5e_del_l2_flow_rule(priv, &ea->promisc);
}
if (disable_promisc)
mlx5e_destroy_promisc_table(priv);
ea->promisc_enabled = promisc_enabled;
ea->allmulti_enabled = allmulti_enabled;
@ -1306,9 +1380,6 @@ static int mlx5e_add_l2_flow_rule(struct mlx5e_priv *priv,
mc_dmac[0] = 0x01;
mv_dmac[0] = 0x01;
break;
case MLX5E_PROMISC:
break;
}
ai->rule = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
@ -1324,13 +1395,11 @@ static int mlx5e_add_l2_flow_rule(struct mlx5e_priv *priv,
return err;
}
#define MLX5E_NUM_L2_GROUPS 3
#define MLX5E_L2_GROUP1_SIZE BIT(0)
#define MLX5E_L2_GROUP2_SIZE BIT(15)
#define MLX5E_L2_GROUP3_SIZE BIT(0)
#define MLX5E_NUM_L2_GROUPS 2
#define MLX5E_L2_GROUP1_SIZE BIT(15)
#define MLX5E_L2_GROUP2_SIZE BIT(0)
#define MLX5E_L2_TABLE_SIZE (MLX5E_L2_GROUP1_SIZE +\
MLX5E_L2_GROUP2_SIZE +\
MLX5E_L2_GROUP3_SIZE)
MLX5E_L2_GROUP2_SIZE)
static int mlx5e_create_l2_table_groups(struct mlx5e_l2_table *l2_table)
{
int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
@ -1353,20 +1422,11 @@ static int mlx5e_create_l2_table_groups(struct mlx5e_l2_table *l2_table)
mc = MLX5_ADDR_OF(create_flow_group_in, in, match_criteria);
mc_dmac = MLX5_ADDR_OF(fte_match_param, mc,
outer_headers.dmac_47_16);
/* Flow Group for promiscuous */
MLX5_SET_CFG(in, start_flow_index, ix);
ix += MLX5E_L2_GROUP1_SIZE;
MLX5_SET_CFG(in, end_flow_index, ix - 1);
ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
if (IS_ERR(ft->g[ft->num_groups]))
goto err_destroy_groups;
ft->num_groups++;
/* Flow Group for full match */
eth_broadcast_addr(mc_dmac);
MLX5_SET_CFG(in, match_criteria_enable, MLX5_MATCH_OUTER_HEADERS);
MLX5_SET_CFG(in, start_flow_index, ix);
ix += MLX5E_L2_GROUP2_SIZE;
ix += MLX5E_L2_GROUP1_SIZE;
MLX5_SET_CFG(in, end_flow_index, ix - 1);
ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
if (IS_ERR(ft->g[ft->num_groups]))
@ -1377,7 +1437,7 @@ static int mlx5e_create_l2_table_groups(struct mlx5e_l2_table *l2_table)
eth_zero_addr(mc_dmac);
mc_dmac[0] = 0x01;
MLX5_SET_CFG(in, start_flow_index, ix);
ix += MLX5E_L2_GROUP3_SIZE;
ix += MLX5E_L2_GROUP2_SIZE;
MLX5_SET_CFG(in, end_flow_index, ix - 1);
ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
if (IS_ERR(ft->g[ft->num_groups]))

Просмотреть файл

@ -105,8 +105,8 @@
#define ETHTOOL_PRIO_NUM_LEVELS 1
#define ETHTOOL_NUM_PRIOS 11
#define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
/* Vlan, mac, ttc, inner ttc, {aRFS/accel and esp/esp_err} */
#define KERNEL_NIC_PRIO_NUM_LEVELS 6
/* Promiscuous, Vlan, mac, ttc, inner ttc, {aRFS/accel and esp/esp_err} */
#define KERNEL_NIC_PRIO_NUM_LEVELS 7
#define KERNEL_NIC_NUM_PRIOS 1
/* One more level for tc */
#define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)