> -----Original Message-----
> From: Dekel Peled <dek...@mellanox.com>
> Sent: Wednesday, October 24, 2018 2:08 PM
> To: Yongseok Koh <ys...@mellanox.com>; Shahaf Shuler
> <shah...@mellanox.com>
> Cc: dev@dpdk.org; Ori Kam <or...@mellanox.com>
> Subject: [PATCH 1/2] net/mlx5: add VXLAN encap decap to Direct Verbs
>
> This patch adds support for VXLAN encap and decap operations, in
> Direct Verbs flow.
>
> Signed-off-by: Dekel Peled <dek...@mellanox.com>
> ---
> drivers/net/mlx5/mlx5_flow.h | 4 +
> drivers/net/mlx5/mlx5_flow_dv.c | 409
> +++++++++++++++++++++++++++++++++++++++-
> 2 files changed, 408 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
> index 38635c9..9c28e50 100644
> --- a/drivers/net/mlx5/mlx5_flow.h
> +++ b/drivers/net/mlx5/mlx5_flow.h
> @@ -92,6 +92,8 @@
> #define MLX5_FLOW_ACTION_DEC_TTL (1u << 19)
> #define MLX5_FLOW_ACTION_SET_MAC_SRC (1u << 20)
> #define MLX5_FLOW_ACTION_SET_MAC_DST (1u << 21)
> +#define MLX5_FLOW_ACTION_VXLAN_ENCAP (1u << 22)
> +#define MLX5_FLOW_ACTION_VXLAN_DECAP (1u << 23)
>
> #define MLX5_FLOW_FATE_ACTIONS \
> (MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
> MLX5_FLOW_ACTION_RSS)
> @@ -181,6 +183,8 @@ struct mlx5_flow_dv {
> #ifdef HAVE_IBV_FLOW_DV_SUPPORT
> struct mlx5dv_flow_action_attr
> actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS];
> /**< Action list. */
> + struct ibv_flow_action *encap_verb; /**< Verbs object of encap. */
> + struct ibv_flow_action *decap_verb; /**< Verbs object of decap. */
Why do we need encap and decap?
> #endif
> int actions_n; /**< number of actions. */
> };
> diff --git a/drivers/net/mlx5/mlx5_flow_dv.c
> b/drivers/net/mlx5/mlx5_flow_dv.c
> index e8f409f..06ecabf 100644
> --- a/drivers/net/mlx5/mlx5_flow_dv.c
> +++ b/drivers/net/mlx5/mlx5_flow_dv.c
> @@ -35,6 +35,16 @@
>
> #ifdef HAVE_IBV_FLOW_DV_SUPPORT
>
> +#define MLX5_UDP 17
> +#define MLX5_TCP 6
> +#define MLX5_GRE 47
Please use already created defines. (IPPROTO_TCP)
> +
> +/*
> + * Encap buf length, max:
> + * Eth:14/VLAN:8/IPv6:40/TCP:36/TUNNEL:20/Eth:14
> + */
> +#define MLX5_ENCAP_LEN 132
> +
> /**
> * Validate META item.
> *
> @@ -97,6 +107,331 @@
> }
>
> /**
> + * Validate the vxlan encap action.
> + *
> + * @param[in] action_flags
> + * Holds the actions detected until now.
> + * @param[in] action
> + * Pointer to the encap action.
> + * @param[in] attr
> + * Pointer to flow attributes
> + * @param[out] error
> + * Pointer to error structure.
> + *
> + * @return
> + * 0 on success, a negative errno value otherwise and rte_errno is set.
> + */
> +static int
> +flow_dv_validate_action_vxlan_encap(uint64_t action_flags,
> + const struct rte_flow_action *action,
> + const struct rte_flow_attr *attr,
> + struct rte_flow_error *error)
> +{
> + const struct rte_flow_action_vxlan_encap *vxlan_encap = action->conf;
> +
> + if (!vxlan_encap)
> + return rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ACTION,
> action,
> + "configuration cannot be null");
> + if (action_flags & MLX5_FLOW_ACTION_DROP)
> + return rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ACTION,
> NULL,
> + "can't drop and encap in same flow");
> + if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP)
> + return rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ACTION,
> NULL,
> + "can't have 2 encap actions in same"
> + " flow");
> + if (attr->ingress)
> + return rte_flow_error_set(error, ENOTSUP,
> +
> RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
> + NULL,
> + "encap action not supported for "
> + "ingress");
> + return 0;
> +}
> +
> +/**
> + * Validate the vxlan decap action.
> + *
> + * @param[in] action_flags
> + * Holds the actions detected until now.
> + * @param[in] action
> + * Pointer to the decap action.
> + * @param[in] attr
> + * Pointer to flow attributes
> + * @param[out] error
> + * Pointer to error structure.
> + *
> + * @return
> + * 0 on success, a negative errno value otherwise and rte_errno is set.
> + */
> +static int
> +flow_dv_validate_action_vxlan_decap(uint64_t action_flags,
> + const struct rte_flow_action *action __rte_unused,
> + const struct rte_flow_attr *attr,
> + struct rte_flow_error *error)
> +{
> + if (action_flags & MLX5_FLOW_ACTION_DROP)
> + return rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ACTION,
> NULL,
> + "can't drop and decap in same flow");
> + if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP)
> + return rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ACTION,
> NULL,
> + "can't encap and decap in same
> flow");
> + if (attr->egress)
> + return rte_flow_error_set(error, ENOTSUP,
> +
> RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
> + NULL,
> + "decap action not supported for "
> + "egress");
Why not EINVAL?
> + return 0;
> +}
> +
> +static size_t item_len[] = {
> + [RTE_FLOW_ITEM_TYPE_VOID] = 0,
> + [RTE_FLOW_ITEM_TYPE_ETH] = sizeof(struct rte_flow_item_eth),
> + [RTE_FLOW_ITEM_TYPE_VLAN] = sizeof(struct rte_flow_item_vlan),
> + [RTE_FLOW_ITEM_TYPE_IPV4] = sizeof(struct rte_flow_item_ipv4),
> + [RTE_FLOW_ITEM_TYPE_IPV6] = sizeof(struct rte_flow_item_ipv6),
> + [RTE_FLOW_ITEM_TYPE_UDP] = sizeof(struct rte_flow_item_udp),
> + [RTE_FLOW_ITEM_TYPE_TCP] = sizeof(struct rte_flow_item_tcp),
> + [RTE_FLOW_ITEM_TYPE_VXLAN] = sizeof(struct rte_flow_item_vxlan),
> + [RTE_FLOW_ITEM_TYPE_GRE] = sizeof(struct rte_flow_item_gre),
> + [RTE_FLOW_ITEM_TYPE_NVGRE] = sizeof(struct rte_flow_item_gre),
> + [RTE_FLOW_ITEM_TYPE_VXLAN_GPE] = sizeof(struct
> rte_flow_item_vxlan_gpe),
> + [RTE_FLOW_ITEM_TYPE_MPLS] = sizeof(struct rte_flow_item_mpls),
> +};
> +
> +/**
> + * Convert the encap action data from rte_flow_item to raw buffer
> + *
> + * @param[in] item
> + * Pointer to rte_flow_item object.
> + * @param[out] buf
> + * Pointer to the output buffer.
> + * @param[out] size
> + * Pointer to the output buffer size.
> + * @param[out] error
> + * Pointer to the error structure.
> + * @param[in] l3_type
> + * ???.
What is ???
> + *
> + * @return
> + * 0 on success, a negative errno value otherwise and rte_errno is set.
> + */
> +static int
> +flow_dv_convert_encap(struct rte_flow_item *item, uint8_t *buf, size_t *size,
> + struct rte_flow_error *error, uint16_t l3_type)
What is the l3_type used for?
> +{
> + struct ether_hdr *eth = NULL;
> + struct vlan_hdr *vlan = NULL;
> + struct ipv4_hdr *ipv4 = NULL;
> + struct ipv6_hdr *ipv6 = NULL;
> + struct udp_hdr *udp = NULL;
> + struct vxlan_hdr *vxlan = NULL;
> + const struct rte_flow_item_vlan *vlan_spec;
Why vlan has dedicated variable?
> + size_t len;
> +
> + assert(item);
Why assert on the item? It should be valid and if not return error.
> + *size = 0;
Why not use temp_size and avoid memory access?
> + for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
> + /* TODO: variable length handling: raw, vxlan and nvgre. */
Why TODO?
> + len = item_len[item->type];
This can result in crash if type is not the one you expected.
> + if (len + *size > MLX5_ENCAP_LEN)
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "invalid item length");
> + rte_memcpy((void *)&buf[*size], item->spec, len);
> + switch (item->type) {
> + case RTE_FLOW_ITEM_TYPE_ETH:
> + eth = (void *)&buf[*size];
Why cast to void? Same for all others.
> + break;
> + case RTE_FLOW_ITEM_TYPE_VLAN:
> + vlan_spec = item->spec;
> + vlan = (void *)&buf[*size];
> + if (!eth)
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "eth header not found");
> + vlan->vlan_tci = vlan_spec->tci;
> + vlan->eth_proto = vlan_spec->inner_type;
> + if (!eth->ether_type)
> + eth->ether_type = htons(ETHER_TYPE_VLAN);
Why not use rte_cpu_to_be? Same for all.
> + break;
> + case RTE_FLOW_ITEM_TYPE_IPV4:
> + ipv4 = (void *)&buf[*size];
> + if (!vlan && !eth)
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "neither eth nor vlan header
> found");
> + if (vlan && !vlan->eth_proto)
> + vlan->eth_proto = htons(ETHER_TYPE_IPv4);
> + else if (eth && !eth->ether_type)
> + eth->ether_type = htons(ETHER_TYPE_IPv4);
> + if (!ipv4->version_ihl)
> + ipv4->version_ihl = 0x45;
> + if (!ipv4->time_to_live)
> + ipv4->time_to_live = 0x40;
> + break;
> + case RTE_FLOW_ITEM_TYPE_IPV6:
> + ipv6 = (void *)&buf[*size];
> + if (!vlan && !eth)
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "neither eth nor vlan header
> found");
> + if (vlan && !vlan->eth_proto)
> + vlan->eth_proto = htons(ETHER_TYPE_IPv6);
> + else if (eth && !eth->ether_type)
> + eth->ether_type = htons(ETHER_TYPE_IPv6);
> + if (!ipv6->vtc_flow)
> + ipv6->vtc_flow = htonl(0x60000000);
> + if (!ipv6->hop_limits)
> + ipv6->hop_limits = 0xff;
> + break;
> + case RTE_FLOW_ITEM_TYPE_UDP:
> + udp = (void *)&buf[*size];
> + if (!ipv4 && !ipv6)
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "ip header not found");
> + if (ipv4 && !ipv4->next_proto_id)
> + ipv4->next_proto_id = MLX5_UDP;
> + else if (ipv6 && !ipv6->proto)
> + ipv6->proto = MLX5_UDP;
> + break;
> + case RTE_FLOW_ITEM_TYPE_VXLAN:
> + vxlan = (void *)&buf[*size];
> + if (!udp)
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "udp header not found");
> + if (!udp->dst_port)
> + udp->dst_port =
> htons(MLX5_UDP_PORT_VXLAN);
> + if (!vxlan->vx_flags)
> + vxlan->vx_flags = htonl(0x08000000);
> + break;
> + case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
> + vxlan = (void *)&buf[*size];
> + if (!udp)
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "udp header not found");
> + if (!udp->dst_port)
> + udp->dst_port =
> htons(MLX5_UDP_PORT_VXLAN_GPE);
> + if (!vxlan->vx_flags)
> + vxlan->vx_flags = htonl(0x0c000003);
> + break;
> + case RTE_FLOW_ITEM_TYPE_GRE:
> + case RTE_FLOW_ITEM_TYPE_NVGRE:
> + if (!ipv4 && !ipv6)
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "ip header not found");
> + if (ipv4 && !ipv4->next_proto_id)
> + ipv4->next_proto_id = htons(MLX5_GRE);
> + else if (ipv6 && !ipv6->proto)
> + ipv6->proto = htons(MLX5_GRE);
> + break;
> + case RTE_FLOW_ITEM_TYPE_VOID:
> + break;
> + default:
> + return rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ACTION,
> + (void *)item->type,
> + "unsupported item type");
> + break;
> + }
> + *size += len;
> + }
> + if (l3_type && vlan)
> + vlan->eth_proto = htons(l3_type);
> + else if (l3_type && eth)
> + eth->ether_type = htons(l3_type);
> + return 0;
> +}
> +
> +/**
> + * Convert VXLAN encap action to DV specification.
> + *
> + * @param[in] dev
> + * Pointer to rte_eth_dev structure.
> + * @param[in] action
> + * Pointer to action structure.
> + * @param[out] error
> + * Pointer to the error structure.
> + *
> + * @return
> + * Pointer to action on success, NULL otherwise and rte_errno is set.
> + */
> +static struct ibv_flow_action *
> +flow_dv_create_vxlan_encap(struct rte_eth_dev *dev,
> + const struct rte_flow_action *action,
> + struct rte_flow_error *error)
> +{
> + struct ibv_flow_action *encap_verb = NULL;
> + const struct rte_flow_action_vxlan_encap *encap_data;
> + struct priv *priv = dev->data->dev_private;
> + uint8_t buf[MLX5_ENCAP_LEN];
> + size_t size = 0;
> + int convert_result;
> +
> + encap_data = (const struct rte_flow_action_vxlan_encap *)action-
> >conf;
> + convert_result = flow_dv_convert_encap(encap_data->definition,
> + buf, &size, error, 0);
> + if (convert_result)
> + return NULL;
> + encap_verb = mlx5_glue->dv_create_flow_action_packet_reformat
> + (priv->ctx, size, (size ? buf : NULL),
> +
> MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
> + MLX5DV_FLOW_TABLE_TYPE_NIC_TX);
> + if (!encap_verb)
> + rte_flow_error_set(error, EINVAL,
> RTE_FLOW_ERROR_TYPE_ACTION,
> + NULL, "cannot create vxlan encap action");
> + return encap_verb;
> +}
> +
> +/**
> + * Convert VXLAN decap action to DV specification.
> + *
> + * @param[in] dev
> + * Pointer to rte_eth_dev structure.
> + * @param[in] action
> + * Pointer to action structure.
> + * @param[out] error
> + * Pointer to the error structure.
> + *
> + * @return
> + * Pointer to action on success, NULL otherwise and rte_errno is set.
> + */
> +static struct ibv_flow_action *
> +flow_dv_create_vxlan_decap(struct rte_eth_dev *dev,
> + const struct rte_flow_action *action __rte_unused,
> + struct rte_flow_error *error)
> +{
> + struct ibv_flow_action *decap_verb = NULL;
> + struct priv *priv = dev->data->dev_private;
> +
> + decap_verb = mlx5_glue->dv_create_flow_action_packet_reformat
> + (priv->ctx, 0, NULL,
> +
> MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
> + MLX5DV_FLOW_TABLE_TYPE_NIC_RX);
> + if (!decap_verb)
> + rte_flow_error_set(error, EINVAL,
> RTE_FLOW_ERROR_TYPE_ACTION,
> + NULL, "cannot create decap action");
> + return decap_verb;
> +}
> +
> +/**
> * Verify the @p attributes will be correctly understood by the NIC and store
> * them in the @p flow if everything is correct.
> *
> @@ -347,6 +682,24 @@
> action_flags |= MLX5_FLOW_ACTION_COUNT;
> ++actions_n;
> break;
> + case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
> + ret =
> flow_dv_validate_action_vxlan_encap(action_flags,
> + actions, attr,
> + error);
> + if (ret < 0)
> + return ret;
> + action_flags |= MLX5_FLOW_ACTION_VXLAN_ENCAP;
> + ++actions_n;
> + break;
> + case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
> + ret =
> flow_dv_validate_action_vxlan_decap(action_flags,
> + actions, attr,
> + error);
> + if (ret < 0)
> + return ret;
> + action_flags |= MLX5_FLOW_ACTION_VXLAN_DECAP;
> + ++actions_n;
> + break;
> default:
> return rte_flow_error_set(error, ENOTSUP,
>
> RTE_FLOW_ERROR_TYPE_ACTION,
> @@ -1056,14 +1409,23 @@
> /**
> * Store the requested actions in an array.
> *
> + * @param[in] dev
> + * Pointer to rte_eth_dev structure.
> * @param[in] action
> * Flow action to translate.
> * @param[in, out] dev_flow
> * Pointer to the mlx5_flow.
> + * @param[out] error
> + * Pointer to the error structure.
> + *
> + * @return
> + * 0 on success, a negative errno value otherwise and rte_errno is set.
> */
> -static void
> -flow_dv_create_action(const struct rte_flow_action *action,
> - struct mlx5_flow *dev_flow)
> +static int
> +flow_dv_create_action(struct rte_eth_dev *dev,
> + const struct rte_flow_action *action,
> + struct mlx5_flow *dev_flow,
> + struct rte_flow_error *error)
> {
> const struct rte_flow_action_queue *queue;
> const struct rte_flow_action_rss *rss;
> @@ -1110,10 +1472,35 @@
> /* Added to array only in apply since we need the QP */
> flow->actions |= MLX5_FLOW_ACTION_RSS;
> break;
> + case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
> + dev_flow->dv.actions[actions_n].type =
> + MLX5DV_FLOW_ACTION_IBV_FLOW_ACTION;
> + dev_flow->dv.actions[actions_n].action =
> + flow_dv_create_vxlan_encap(dev, action,
> error);
> + if (!(dev_flow->dv.actions[actions_n].action))
> + return -rte_errno;
> + dev_flow->dv.encap_verb =
> + dev_flow->dv.actions[actions_n].action;
> + flow->actions |= MLX5_FLOW_ACTION_VXLAN_ENCAP;
> + actions_n++;
> + break;
> + case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
> + dev_flow->dv.actions[actions_n].type =
> + MLX5DV_FLOW_ACTION_IBV_FLOW_ACTION;
> + dev_flow->dv.actions[actions_n].action =
> + flow_dv_create_vxlan_decap(dev, action,
> error);
> + if (!(dev_flow->dv.actions[actions_n].action))
> + return -rte_errno;
> + dev_flow->dv.decap_verb =
> + dev_flow->dv.actions[actions_n].action;
> + flow->actions |= MLX5_FLOW_ACTION_VXLAN_DECAP;
> + actions_n++;
> + break;
> default:
> break;
> }
> dev_flow->dv.actions_n = actions_n;
> + return 0;
> }
>
> static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
> @@ -1279,8 +1666,10 @@
> matcher.egress = attr->egress;
> if (flow_dv_matcher_register(dev, &matcher, dev_flow, error))
> return -rte_errno;
> - for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
> - flow_dv_create_action(actions, dev_flow);
> + for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
> + if (flow_dv_create_action(dev, actions, dev_flow, error))
> + return -rte_errno;
> + }
> return 0;
> }
>
> @@ -1465,6 +1854,16 @@
> LIST_REMOVE(dev_flow, next);
> if (dev_flow->dv.matcher)
> flow_dv_matcher_release(dev, dev_flow);
> + if (dev_flow->dv.encap_verb) {
> + claim_zero(mlx5_glue->destroy_flow_action
> + (dev_flow->dv.encap_verb));
> + dev_flow->dv.encap_verb = NULL;
> + }
> + if (dev_flow->dv.decap_verb) {
> + claim_zero(mlx5_glue->destroy_flow_action
> + (dev_flow->dv.decap_verb));
> + dev_flow->dv.decap_verb = NULL;
> + }
> rte_free(dev_flow);
> }
> }
> --
> 1.8.3.1
Thanks,
Ori
