]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
net/mlx5e: Add GRE protocol offloading
authorOz Shlomo <ozsh@mellanox.com>
Mon, 29 Oct 2018 06:54:42 +0000 (08:54 +0200)
committerSaeed Mahameed <saeedm@mellanox.com>
Mon, 10 Dec 2018 23:53:04 +0000 (15:53 -0800)
Add HW offloading support for TC flower filters configured on
gretap/ip6gretap net devices.

Signed-off-by: Oz Shlomo <ozsh@mellanox.com>
Reviewed-by: Eli Britstein <elibr@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun.c
drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun.h

index eaa43477e0eae62e0b7eb4069d48af33e4bf82cd..c1515f0135018558913e0f563f9f5d5bbec18eb4 100644 (file)
@@ -2,6 +2,7 @@
 /* Copyright (c) 2018 Mellanox Technologies. */
 
 #include <net/vxlan.h>
+#include <net/gre.h>
 #include "lib/vxlan.h"
 #include "en/tc_tun.h"
 
@@ -109,6 +110,30 @@ static int mlx5e_gen_vxlan_header(char buf[], struct ip_tunnel_key *tun_key)
        return 0;
 }
 
+static int mlx5e_gen_gre_header(char buf[], struct ip_tunnel_key *tun_key)
+{
+       __be32 tun_id = tunnel_id_to_key32(tun_key->tun_id);
+       int hdr_len;
+       struct gre_base_hdr *greh = (struct gre_base_hdr *)(buf);
+
+       /* the HW does not calculate GRE csum or sequences */
+       if (tun_key->tun_flags & (TUNNEL_CSUM | TUNNEL_SEQ))
+               return -EOPNOTSUPP;
+
+       greh->protocol = htons(ETH_P_TEB);
+
+       /* GRE key */
+       hdr_len = gre_calc_hlen(tun_key->tun_flags);
+       greh->flags = gre_tnl_flags_to_gre_flags(tun_key->tun_flags);
+       if (tun_key->tun_flags & TUNNEL_KEY) {
+               __be32 *ptr = (__be32 *)(((u8 *)greh) + hdr_len - 4);
+
+               *ptr = tun_id;
+       }
+
+       return 0;
+}
+
 static int mlx5e_gen_ip_tunnel_header(char buf[], __u8 *ip_proto,
                                      struct mlx5e_encap_entry *e)
 {
@@ -118,6 +143,9 @@ static int mlx5e_gen_ip_tunnel_header(char buf[], __u8 *ip_proto,
        if (e->tunnel_type == MLX5E_TC_TUNNEL_TYPE_VXLAN) {
                *ip_proto = IPPROTO_UDP;
                err = mlx5e_gen_vxlan_header(buf, key);
+       } else if  (e->tunnel_type == MLX5E_TC_TUNNEL_TYPE_GRETAP) {
+               *ip_proto = IPPROTO_GRE;
+               err = mlx5e_gen_gre_header(buf, key);
        } else {
                pr_warn("mlx5: Cannot generate tunnel header for tunnel type (%d)\n"
                        , e->tunnel_type);
@@ -358,6 +386,9 @@ int mlx5e_tc_tun_get_type(struct net_device *tunnel_dev)
 {
        if (netif_is_vxlan(tunnel_dev))
                return MLX5E_TC_TUNNEL_TYPE_VXLAN;
+       else if (netif_is_gretap(tunnel_dev) ||
+                netif_is_ip6gretap(tunnel_dev))
+               return MLX5E_TC_TUNNEL_TYPE_GRETAP;
        else
                return MLX5E_TC_TUNNEL_TYPE_UNKNOWN;
 }
@@ -370,6 +401,9 @@ bool mlx5e_tc_tun_device_to_offload(struct mlx5e_priv *priv,
        if (tunnel_type == MLX5E_TC_TUNNEL_TYPE_VXLAN &&
            MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
                return true;
+       else if (tunnel_type == MLX5E_TC_TUNNEL_TYPE_GRETAP &&
+                MLX5_CAP_ESW(priv->mdev, nvgre_encap_decap))
+               return true;
        else
                return false;
 }
@@ -394,6 +428,9 @@ int mlx5e_tc_tun_init_encap_attr(struct net_device *tunnel_dev,
                }
                e->reformat_type = MLX5_REFORMAT_TYPE_L2_TO_VXLAN;
                e->tunnel_hlen = VXLAN_HLEN;
+       } else if (e->tunnel_type == MLX5E_TC_TUNNEL_TYPE_GRETAP) {
+               e->reformat_type = MLX5_REFORMAT_TYPE_L2_TO_NVGRE;
+               e->tunnel_hlen = gre_calc_hlen(e->tun_info.key.tun_flags);
        } else {
                e->reformat_type = -1;
                e->tunnel_hlen = -1;
@@ -472,6 +509,53 @@ static int mlx5e_tc_tun_parse_vxlan(struct mlx5e_priv *priv,
        return 0;
 }
 
+static int mlx5e_tc_tun_parse_gretap(struct mlx5e_priv *priv,
+                                    struct mlx5_flow_spec *spec,
+                                    struct tc_cls_flower_offload *f,
+                                    void *outer_headers_c,
+                                    void *outer_headers_v)
+{
+       void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
+                                   misc_parameters);
+       void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
+                                   misc_parameters);
+
+       if (!MLX5_CAP_ESW(priv->mdev, nvgre_encap_decap)) {
+               NL_SET_ERR_MSG_MOD(f->common.extack,
+                                  "GRE HW offloading is not supported");
+               netdev_warn(priv->netdev, "GRE HW offloading is not supported\n");
+               return -EOPNOTSUPP;
+       }
+
+       MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, ip_protocol);
+       MLX5_SET(fte_match_set_lyr_2_4, outer_headers_v,
+                ip_protocol, IPPROTO_GRE);
+
+       /* gre protocol*/
+       MLX5_SET_TO_ONES(fte_match_set_misc, misc_c, gre_protocol);
+       MLX5_SET(fte_match_set_misc, misc_v, gre_protocol, ETH_P_TEB);
+
+       /* gre key */
+       if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
+               struct flow_dissector_key_keyid *mask = NULL;
+               struct flow_dissector_key_keyid *key = NULL;
+
+               mask = skb_flow_dissector_target(f->dissector,
+                                                FLOW_DISSECTOR_KEY_ENC_KEYID,
+                                                f->mask);
+               MLX5_SET(fte_match_set_misc, misc_c,
+                        gre_key.key, be32_to_cpu(mask->keyid));
+
+               key = skb_flow_dissector_target(f->dissector,
+                                               FLOW_DISSECTOR_KEY_ENC_KEYID,
+                                               f->key);
+               MLX5_SET(fte_match_set_misc, misc_v,
+                        gre_key.key, be32_to_cpu(key->keyid));
+       }
+
+       return 0;
+}
+
 int mlx5e_tc_tun_parse(struct net_device *filter_dev,
                       struct mlx5e_priv *priv,
                       struct mlx5_flow_spec *spec,
@@ -486,6 +570,9 @@ int mlx5e_tc_tun_parse(struct net_device *filter_dev,
        if (tunnel_type == MLX5E_TC_TUNNEL_TYPE_VXLAN) {
                err = mlx5e_tc_tun_parse_vxlan(priv, spec, f,
                                               headers_c, headers_v);
+       } else if (tunnel_type == MLX5E_TC_TUNNEL_TYPE_GRETAP) {
+               err = mlx5e_tc_tun_parse_gretap(priv, spec, f,
+                                               headers_c, headers_v);
        } else {
                netdev_warn(priv->netdev,
                            "decapsulation offload is not supported for %s net device (%d)\n",
index ad4fc93b17a1f6ec55071053c7a686c1ec63a78c..706ce7bf15e7f19f88ac4fe6afe7375ff5445c13 100644 (file)
@@ -13,7 +13,8 @@
 
 enum {
        MLX5E_TC_TUNNEL_TYPE_UNKNOWN,
-       MLX5E_TC_TUNNEL_TYPE_VXLAN
+       MLX5E_TC_TUNNEL_TYPE_VXLAN,
+       MLX5E_TC_TUNNEL_TYPE_GRETAP
 };
 
 int mlx5e_tc_tun_init_encap_attr(struct net_device *tunnel_dev,