Hi,
> -----Original Message-----
> From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Qiming Yang
> Sent: Wednesday, June 12, 2019 3:50 PM
> To: dev@dpdk.org
> Cc: Zhao1, Wei <wei.zh...@intel.com>
> Subject: [dpdk-dev] [PATCH v2 1/3] net/ice: enable switch filter
>
> From: wei zhao <wei.zh...@intel.com>
>
> The patch enables the backend of rte_flow. It transfers
> rte_flow_xxx to device specific data structure and
> configures packet process engine's binary classifier
> (switch) properly.
>
> Signed-off-by: Wei Zhao <wei.zh...@intel.com>
> ---
> drivers/net/ice/Makefile | 1 +
> drivers/net/ice/ice_ethdev.h | 6 +
> drivers/net/ice/ice_switch_filter.c | 502
> ++++++++++++++++++++++++++++++++++++
> drivers/net/ice/ice_switch_filter.h | 28 ++
> drivers/net/ice/meson.build | 3 +-
> 5 files changed, 539 insertions(+), 1 deletion(-)
> create mode 100644 drivers/net/ice/ice_switch_filter.c
> create mode 100644 drivers/net/ice/ice_switch_filter.h
>
> diff --git a/drivers/net/ice/Makefile b/drivers/net/ice/Makefile
> index 0e5c55e..b10d826 100644
> --- a/drivers/net/ice/Makefile
> +++ b/drivers/net/ice/Makefile
> @@ -60,6 +60,7 @@ ifeq ($(CONFIG_RTE_ARCH_X86), y)
> SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_rxtx_vec_sse.c
> endif
>
> +SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_switch_filter.c
> ifeq ($(findstring
> RTE_MACHINE_CPUFLAG_AVX2,$(CFLAGS)),RTE_MACHINE_CPUFLAG_AVX2)
> CC_AVX2_SUPPORT=1
> else
> diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
> index 1385afa..67a358a 100644
> --- a/drivers/net/ice/ice_ethdev.h
> +++ b/drivers/net/ice/ice_ethdev.h
> @@ -234,6 +234,12 @@ struct ice_vsi {
> bool offset_loaded;
> };
>
> +/* Struct to store flow created. */
> +struct rte_flow {
> + TAILQ_ENTRY(rte_flow) node;
> +void *rule;
> +};
> +
> struct ice_pf {
> struct ice_adapter *adapter; /* The adapter this PF associate to */
> struct ice_vsi *main_vsi; /* pointer to main VSI structure */
> diff --git a/drivers/net/ice/ice_switch_filter.c
> b/drivers/net/ice/ice_switch_filter.c
> new file mode 100644
> index 0000000..e679675
> --- /dev/null
> +++ b/drivers/net/ice/ice_switch_filter.c
> @@ -0,0 +1,502 @@
SPDX-License-Identifier missing.
> +#include <sys/queue.h>
> +#include <stdio.h>
> +#include <errno.h>
> +#include <stdint.h>
> +#include <string.h>
> +#include <unistd.h>
> +#include <stdarg.h>
> +
> +#include <rte_debug.h>
> +#include <rte_ether.h>
> +#include <rte_ethdev_driver.h>
> +#include <rte_log.h>
> +#include <rte_malloc.h>
> +#include <rte_eth_ctrl.h>
> +#include <rte_tailq.h>
> +#include <rte_flow_driver.h>
> +
> +#include "ice_logs.h"
> +#include "base/ice_type.h"
> +#include "ice_switch_filter.h"
> +
> +static int
> +ice_parse_switch_filter(
> + const struct rte_flow_item pattern[],
> + const struct rte_flow_action actions[],
> + struct rte_flow_error *error,
> + struct ice_adv_rule_info *rule_info,
> + struct ice_adv_lkup_elem **lkup_list,
> + uint16_t *lkups_num)
> +{
> + const struct rte_flow_item *item = pattern;
> + enum rte_flow_item_type item_type;
> + const struct rte_flow_item_eth *eth_spec, *eth_mask;
> + const struct rte_flow_item_ipv4 *ipv4_spec, *ipv4_mask;
> + const struct rte_flow_item_ipv6 *ipv6_spec, *ipv6_mask;
> + const struct rte_flow_item_tcp *tcp_spec, *tcp_mask;
> + const struct rte_flow_item_udp *udp_spec, *udp_mask;
> + const struct rte_flow_item_sctp *sctp_spec, *sctp_mask;
> + const struct rte_flow_item_nvgre *nvgre_spec, *nvgre_mask;
> + const struct rte_flow_item_vxlan *vxlan_spec, *vxlan_mask;
> + struct ice_adv_lkup_elem *list;
> + uint16_t i, j, t = 0;
> + uint16_t item_num = 0;
> + enum ice_sw_tunnel_type tun_type = ICE_NON_TUN;
> +
> + for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
> + if (item->type == RTE_FLOW_ITEM_TYPE_ETH ||
> + item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
> + item->type == RTE_FLOW_ITEM_TYPE_IPV6 ||
> + item->type == RTE_FLOW_ITEM_TYPE_UDP ||
> + item->type == RTE_FLOW_ITEM_TYPE_TCP ||
> + item->type == RTE_FLOW_ITEM_TYPE_SCTP ||
> + item->type == RTE_FLOW_ITEM_TYPE_VXLAN ||
> + item->type == RTE_FLOW_ITEM_TYPE_NVGRE)
> + item_num++;
> + }
> +
> + list = rte_zmalloc(NULL, item_num * sizeof(*list), 0);
> + if (!list) {
> + rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ITEM, actions,
> + "no memory malloc");
{RTE_FLOW_ERROR_TYPE_ITEM_NUM, item, "No memory for PMD internal items"} is
more appropriate.
Refer to i40e implementation.
> + goto out;
> + }
> + *lkup_list = list;
> +
> + for (item = pattern, i = 0; item->type !=
> + RTE_FLOW_ITEM_TYPE_END; item++, i++) {
It seems we don't need the "i" variable.
> + item_type = item->type;
> +
> + switch (item_type) {
> + case RTE_FLOW_ITEM_TYPE_ETH:
> + eth_spec = item->spec;
> + eth_mask = item->mask;
> + if (eth_spec && eth_mask) {
> + list[t].type = (tun_type == ICE_NON_TUN) ?
> + ICE_MAC_OFOS : ICE_MAC_IL;
> + for (j = 0; j < RTE_ETHER_ADDR_LEN; j++) {
> + if (eth_mask->src.addr_bytes[j] ==
> + UINT8_MAX) {
> + list[t].h_u.eth_hdr.
> + src_addr[j] =
> + eth_spec->src.addr_bytes[j];
> + list[t].m_u.eth_hdr.
> + src_addr[j] =
> + eth_mask->src.addr_bytes[j];
> + }
> + if (eth_mask->dst.addr_bytes[j] ==
> + UINT8_MAX) {
> + list[t].h_u.eth_hdr.
> + dst_addr[j] =
> + eth_spec->dst.addr_bytes[j];
> + list[t].m_u.eth_hdr.
> + dst_addr[j] =
> + eth_mask->dst.addr_bytes[j];
> + }
> + }
> + if (eth_mask->type == UINT16_MAX) {
> + list[t].h_u.eth_hdr.ethtype_id =
> + rte_be_to_cpu_16(eth_spec->type);
> + list[t].m_u.eth_hdr.ethtype_id =
> + UINT16_MAX;
> + }
> + t++;
A lot of "t++" below, can we move it outside the switch{ } to have only one
"t++"?
> + } else if (!eth_spec && !eth_mask) {
> + list[t].type = (tun_type == ICE_NON_TUN) ?
> + ICE_MAC_OFOS : ICE_MAC_IL;
> + }
> + break;
> +
> + case RTE_FLOW_ITEM_TYPE_IPV4:
> + ipv4_spec = item->spec;
> + ipv4_mask = item->mask;
> + if (ipv4_spec && ipv4_mask) {
> + list[t].type = (tun_type == ICE_NON_TUN) ?
> + ICE_IPV4_OFOS : ICE_IPV4_IL;
> + if (ipv4_mask->hdr.src_addr == UINT32_MAX)
> {
> + list[t].h_u.ipv4_hdr.src_addr =
> + ipv4_spec->hdr.src_addr;
> + list[t].m_u.ipv4_hdr.src_addr =
> + UINT32_MAX;
> + }
> + if (ipv4_mask->hdr.dst_addr == UINT32_MAX)
> {
> + list[t].h_u.ipv4_hdr.dst_addr =
> + ipv4_spec->hdr.dst_addr;
> + list[t].m_u.ipv4_hdr.dst_addr =
> + UINT32_MAX;
> + }
> + if (ipv4_mask->hdr.time_to_live ==
> UINT8_MAX) {
> + list[t].h_u.ipv4_hdr.time_to_live =
> + ipv4_spec->hdr.time_to_live;
> + list[t].m_u.ipv4_hdr.time_to_live =
> + UINT8_MAX;
> + }
> + if (ipv4_mask->hdr.next_proto_id ==
> UINT8_MAX) {
> + list[t].h_u.ipv4_hdr.protocol =
> + ipv4_spec-
> >hdr.next_proto_id;
> + list[t].m_u.ipv4_hdr.protocol =
> + UINT8_MAX;
> + }
> + if (ipv4_mask->hdr.type_of_service ==
> + UINT8_MAX) {
> + list[t].h_u.ipv4_hdr.tos =
> + ipv4_spec-
> >hdr.type_of_service;
> + list[t].m_u.ipv4_hdr.tos = UINT8_MAX;
> + }
> + t++;
> + } else if (!ipv4_spec && !ipv4_mask) {
> + list[t].type = (tun_type == ICE_NON_TUN) ?
> + ICE_IPV4_OFOS : ICE_IPV4_IL;
> + }
> + break;
> +
> + case RTE_FLOW_ITEM_TYPE_IPV6:
> + ipv6_spec = item->spec;
> + ipv6_mask = item->mask;
> + if (ipv6_spec && ipv6_mask) {
> + list[t].type = (tun_type == ICE_NON_TUN) ?
> + ICE_IPV6_OFOS : ICE_IPV6_IL;
> + for (j = 0; j < ICE_IPV6_ADDR_LENGTH; j++) {
> + if (ipv6_mask->hdr.src_addr[j] ==
> + UINT8_MAX) {
> + list[t].h_u.ice_ipv6_ofos_hdr.
> + src_addr[j] =
> + ipv6_spec->hdr.src_addr[j];
> + list[t].m_u.ice_ipv6_ofos_hdr.
> + src_addr[j] =
> + ipv6_mask->hdr.src_addr[j];
> + }
> + if (ipv6_mask->hdr.dst_addr[j] ==
> + UINT8_MAX) {
> + list[t].h_u.ice_ipv6_ofos_hdr.
> + dst_addr[j] =
> + ipv6_spec->hdr.dst_addr[j];
> + list[t].m_u.ice_ipv6_ofos_hdr.
> + dst_addr[j] =
> + ipv6_mask->hdr.dst_addr[j];
> + }
> + }
> + if (ipv6_mask->hdr.proto == UINT8_MAX) {
> +
> list[t].h_u.ice_ipv6_ofos_hdr.next_hdr =
> + ipv6_spec->hdr.proto;
> +
> list[t].m_u.ice_ipv6_ofos_hdr.next_hdr =
> + UINT8_MAX;
> + }
> + if (ipv6_mask->hdr.hop_limits == UINT8_MAX)
> {
> + list[t].h_u.ice_ipv6_ofos_hdr.
> + hop_limit = ipv6_spec-
> >hdr.hop_limits;
> + list[t].m_u.ice_ipv6_ofos_hdr.
> + hop_limit = UINT8_MAX;
> + }
> + t++;
> + } else if (!ipv6_spec && !ipv6_mask) {
> + list[t].type = (tun_type == ICE_NON_TUN) ?
> + ICE_IPV4_OFOS : ICE_IPV4_IL;
> + }
> + break;
> +
> + case RTE_FLOW_ITEM_TYPE_UDP:
> + udp_spec = item->spec;
> + udp_mask = item->mask;
> + if (udp_spec && udp_mask) {
> + list[t].type = ICE_UDP_ILOS;
> + if (udp_mask->hdr.src_port == UINT16_MAX)
> {
> + list[t].h_u.l4_hdr.src_port =
> + udp_spec->hdr.src_port;
> + list[t].m_u.l4_hdr.src_port =
> + udp_mask->hdr.src_port;
> + }
> + if (udp_mask->hdr.dst_port == UINT16_MAX)
> {
> + list[t].h_u.l4_hdr.dst_port =
> + udp_spec->hdr.dst_port;
> + list[t].m_u.l4_hdr.dst_port =
> + udp_mask->hdr.dst_port;
> + }
> + t++;
> + } else if (!udp_spec && !udp_mask) {
> + list[t].type = ICE_UDP_ILOS;
> + }
> + break;
> +
> + case RTE_FLOW_ITEM_TYPE_TCP:
> + tcp_spec = item->spec;
> + tcp_mask = item->mask;
> + if (tcp_spec && tcp_mask) {
> + list[t].type = ICE_TCP_IL;
> + if (tcp_mask->hdr.src_port == UINT16_MAX) {
> + list[t].h_u.l4_hdr.src_port =
> + tcp_spec->hdr.src_port;
> + list[t].m_u.l4_hdr.src_port =
> + tcp_mask->hdr.src_port;
> + }
> + if (tcp_mask->hdr.dst_port == UINT16_MAX) {
> + list[t].h_u.l4_hdr.dst_port =
> + tcp_spec->hdr.dst_port;
> + list[t].m_u.l4_hdr.dst_port =
> + tcp_mask->hdr.dst_port;
> + }
> + t++;
> + } else if (!tcp_spec && !tcp_mask) {
> + list[t].type = ICE_TCP_IL;
> + }
> + break;
> +
> + case RTE_FLOW_ITEM_TYPE_SCTP:
> + sctp_spec = item->spec;
> + sctp_mask = item->mask;
> + if (sctp_spec && sctp_mask) {
> + list[t].type = ICE_SCTP_IL;
> + if (sctp_mask->hdr.src_port == UINT16_MAX)
> {
> + list[t].h_u.sctp_hdr.src_port =
> + sctp_spec->hdr.src_port;
> + list[t].m_u.sctp_hdr.src_port =
> + sctp_mask->hdr.src_port;
> + }
> + if (sctp_mask->hdr.dst_port == UINT16_MAX)
> {
> + list[t].h_u.sctp_hdr.dst_port =
> + sctp_spec->hdr.dst_port;
> + list[t].m_u.sctp_hdr.dst_port =
> + sctp_mask->hdr.dst_port;
> + }
> + t++;
> + } else if (!sctp_spec && !sctp_mask) {
> + list[t].type = ICE_SCTP_IL;
> + }
> + break;
> +
> + case RTE_FLOW_ITEM_TYPE_VXLAN:
> + vxlan_spec = item->spec;
> + vxlan_mask = item->mask;
> + tun_type = ICE_SW_TUN_VXLAN;
> + if (vxlan_spec && vxlan_mask) {
> + list[t].type = ICE_VXLAN;
> + if (vxlan_mask->vni[0] == UINT8_MAX &&
> + vxlan_mask->vni[1] == UINT8_MAX
> &&
> + vxlan_mask->vni[2] == UINT8_MAX) {
> + list[t].h_u.tnl_hdr.vni =
> + (vxlan_spec->vni[1] << 8) |
> + vxlan_spec->vni[0];
> + list[t].m_u.tnl_hdr.vni =
> + UINT16_MAX;
vxlan_spec->vni[2] does not need to be put into the list?
> + }
> + t++;
> + } else if (!vxlan_spec && !vxlan_mask) {
> + list[t].type = ICE_VXLAN;
> + }
> + break;
> +
> + case RTE_FLOW_ITEM_TYPE_NVGRE:
> + nvgre_spec = item->spec;
> + nvgre_mask = item->mask;
> + tun_type = ICE_SW_TUN_NVGRE;
> + if (nvgre_spec && nvgre_mask) {
> + list[t].type = ICE_NVGRE;
> + if (nvgre_mask->tni[0] == UINT8_MAX &&
> + nvgre_mask->tni[1] == UINT8_MAX
> &&
> + nvgre_mask->tni[2] == UINT8_MAX) {
> + list[t].h_u.nvgre_hdr.tni =
> + (nvgre_spec->tni[1] << 8) |
> + nvgre_spec->tni[0];
> + list[t].m_u.nvgre_hdr.tni =
> + UINT16_MAX;
> + }
> + t++;
> + } else if (!nvgre_spec && !nvgre_mask) {
> + list[t].type = ICE_NVGRE;
> + }
> + break;
> +
> + case RTE_FLOW_ITEM_TYPE_VOID:
> + case RTE_FLOW_ITEM_TYPE_END:
> + break;
> +
> + default:
> + rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ITEM, actions,
> + "Invalid pattern item.");
> + goto out;
> + }
> + }
> +
> + rule_info->tun_type = tun_type;
> + *lkups_num = t;
> +
> + return 0;
> +out:
We may need to free the allocated list before return.
> + return -rte_errno;
> +}
> +
> +/* By now ice switch filter action code implement only
> +* supports QUEUE or DROP.
> +*/
> +static int
> +ice_parse_switch_action(struct ice_pf *pf,
> + const struct rte_flow_action *actions,
> + struct rte_flow_error *error,
> + struct ice_adv_rule_info *rule_info)
> +{
> + struct ice_hw *hw = ICE_PF_TO_HW(pf);
> + struct ice_vsi *vsi = pf->main_vsi;
> + const struct rte_flow_action *act;
> + const struct rte_flow_action_queue *act_q;
> + uint16_t base_queue, index = 0;
> + uint32_t reg;
> +
> + /* Check if the first non-void action is QUEUE or DROP. */
> + NEXT_ITEM_OF_ACTION(act, actions, index);
> + if (act->type != RTE_FLOW_ACTION_TYPE_QUEUE &&
> + act->type != RTE_FLOW_ACTION_TYPE_DROP) {
> + rte_flow_error_set(error, EINVAL,
> RTE_FLOW_ERROR_TYPE_ACTION,
> + act, "Not supported action.");
> + return -rte_errno;
> + }
> + reg = ICE_READ_REG(hw, PFLAN_RX_QALLOC);
> + if (reg & PFLAN_RX_QALLOC_VALID_M) {
> + base_queue = reg & PFLAN_RX_QALLOC_FIRSTQ_M;
Can we get this register info earlier in dev_start() or somewhere else? Then we
can use the base_queue directly.
> + } else {
> + rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ACTION,
> + act, "Invalid queue register");
> + return -rte_errno;
> + }
> + if (act->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
> + act_q = act->conf;
> + rule_info->sw_act.fltr_act = ICE_FWD_TO_Q;
> + rule_info->sw_act.fwd_id.q_id = base_queue + act_q->index;
> + if (act_q->index >= pf->dev_data->nb_rx_queues) {
> + rte_flow_error_set(error, EINVAL,
> + RTE_FLOW_ERROR_TYPE_ACTION,
> + act, "Invalid queue ID for"
> + " switch filter.");
> + return -rte_errno;
> + }
> + } else {
> + rule_info->sw_act.fltr_act = ICE_DROP_PACKET;
> + }
> +
> + rule_info->sw_act.vsi_handle = vsi->idx;
> + rule_info->rx = 1;
> + rule_info->sw_act.src = vsi->idx;
> +
> + /* Check if the next non-void item is END */
> + index++;
> + NEXT_ITEM_OF_ACTION(act, actions, index);
> + if (act->type != RTE_FLOW_ACTION_TYPE_END) {
> + rte_flow_error_set(error, EINVAL,
> RTE_FLOW_ERROR_TYPE_ACTION,
> + act, "Not supported action.");
> + return -rte_errno;
> + }
> +
> + return 0;
> +}
> +
> +static int
> +ice_switch_rule_set(struct ice_pf *pf,
> + struct ice_adv_lkup_elem *list,
> + uint16_t lkups_cnt,
> + struct ice_adv_rule_info *rule_info,
> + struct rte_flow *flow)
> +{
> + struct ice_hw *hw = ICE_PF_TO_HW(pf);
> + int ret;
> + struct ice_rule_query_data rule_added = {0};
> + struct ice_rule_query_data *filter_ptr;
> +
> + if (lkups_cnt > ICE_MAX_CHAIN_WORDS) {
> + PMD_DRV_LOG(ERR, "item number too large for rule");
Why not rte_flow_error_set() to report error?
> + return -ENOTSUP;
> + }
> + if (!list) {
> + PMD_DRV_LOG(ERR, "lookup list should not be NULL");
Ditto.
> + return -ENOTSUP;
> + }
> +
> + ret = ice_add_adv_rule(hw, list, lkups_cnt, rule_info, &rule_added);
> +
> + if (!ret) {
> + filter_ptr = rte_zmalloc("ice_switch_filter",
> + sizeof(struct ice_rule_query_data), 0);
> + if (!filter_ptr) {
> + PMD_DRV_LOG(ERR, "failed to allocate memory");
> + return -EINVAL;
> + }
> + flow->rule = filter_ptr;
> + rte_memcpy(filter_ptr,
> + &rule_added,
> + sizeof(struct ice_rule_query_data));
> + }
> +
> + return ret;
> +}
> +
> +int
> +ice_create_switch_filter(struct ice_pf *pf,
> + const struct rte_flow_item pattern[],
> + const struct rte_flow_action actions[],
> + struct rte_flow *flow,
> + struct rte_flow_error *error)
> +{
> + int ret = 0;
> + struct ice_adv_rule_info rule_info = {0};
> + struct ice_adv_lkup_elem *list = NULL;
> + uint16_t lkups_num = 0;
> +
> + ret = ice_parse_switch_filter(pattern, actions, error,
> + &rule_info, &list, &lkups_num);
> + if (ret)
> + goto out;
> +
> + ret = ice_parse_switch_action(pf, actions, error, &rule_info);
> + if (ret)
> + goto out;
> +
> + ret = ice_switch_rule_set(pf, list, lkups_num, &rule_info, flow);
> + if (ret)
> + goto out;
> +
> + rte_free(list);
> + return 0;
> +
> +out:
> + rte_free(list);
> +
> + return -rte_errno;
> +}
> +
> +int
> +ice_destroy_switch_filter(struct ice_pf *pf,
> + struct rte_flow *flow)
> +{
> + struct ice_hw *hw = ICE_PF_TO_HW(pf);
> + int ret;
> + struct ice_rule_query_data *filter_ptr;
> + struct ice_rule_query_data rule_added;
> +
> + filter_ptr = (struct ice_rule_query_data *)
> + flow->rule;
> + rte_memcpy(&rule_added, filter_ptr,
> + sizeof(struct ice_rule_query_data));
> +
> + if (!filter_ptr) {
> + PMD_DRV_LOG(ERR, "no such flow"
> + " create by switch filter");
> + return -EINVAL;
> + }
We should do this check at least before rte_memcpy.
> +
> + ret = ice_rem_adv_rule_by_id(hw, &rule_added);
We can use filter_ptr directly for the switch rule delete.
> +
> + rte_free(filter_ptr);
> +
> + return ret;
> +}
> +
> +void
> +ice_free_switch_filter_rule(void *rule)
> +{
> + struct ice_rule_query_data *filter_ptr;
> +
> + filter_ptr = (struct ice_rule_query_data *)rule;
> +
> + rte_free(filter_ptr);
> +}
> diff --git a/drivers/net/ice/ice_switch_filter.h
> b/drivers/net/ice/ice_switch_filter.h
> new file mode 100644
> index 0000000..957d0d1
> --- /dev/null
> +++ b/drivers/net/ice/ice_switch_filter.h
> @@ -0,0 +1,28 @@
Also a license is needed in new file.
BRs,
Xiao
> +#ifndef _ICE_SWITCH_FILTER_H_
> +#define _ICE_SWITCH_FILTER_H_
