net/ipn3ke: support flow API
authorRosen Xu <rosen.xu@intel.com>
Tue, 16 Apr 2019 03:17:43 +0000 (11:17 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Fri, 19 Apr 2019 12:51:54 +0000 (14:51 +0200)
Add Intel FPGA Acceleration NIC IPN3KE Flow of PMD driver.

Signed-off-by: Rosen Xu <rosen.xu@intel.com>
Signed-off-by: Andy Pei <andy.pei@intel.com>
Signed-off-by: Dan Wei <dan.wei@intel.com>
drivers/net/ipn3ke/Makefile
drivers/net/ipn3ke/ipn3ke_ethdev.c
drivers/net/ipn3ke/ipn3ke_ethdev.h
drivers/net/ipn3ke/ipn3ke_flow.c [new file with mode: 0644]
drivers/net/ipn3ke/ipn3ke_flow.h [new file with mode: 0644]
drivers/net/ipn3ke/ipn3ke_representor.c
drivers/net/ipn3ke/ipn3ke_tm.c
drivers/net/ipn3ke/meson.build

index 38d9384..8c3ae37 100644 (file)
@@ -35,5 +35,6 @@ LIBABIVER := 1
 SRCS-$(CONFIG_RTE_LIBRTE_IPN3KE_PMD) += ipn3ke_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_IPN3KE_PMD) += ipn3ke_representor.c
 SRCS-$(CONFIG_RTE_LIBRTE_IPN3KE_PMD) += ipn3ke_tm.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPN3KE_PMD) += ipn3ke_flow.c
 
 include $(RTE_SDK)/mk/rte.lib.mk
index 508ea01..9079b57 100644 (file)
@@ -21,6 +21,7 @@
 #include <ifpga_logs.h>
 
 #include "ipn3ke_rawdev_api.h"
+#include "ipn3ke_flow.h"
 #include "ipn3ke_logs.h"
 #include "ipn3ke_ethdev.h"
 
@@ -266,6 +267,10 @@ ipn3ke_hw_init(struct rte_afu_device *afu_dev,
                if (ret)
                        return ret;
                hw->tm_hw_enable = 1;
+
+               ret = ipn3ke_flow_init(hw);
+               if (ret)
+                       return ret;
                hw->flow_hw_enable = 1;
        }
 
index 36ff2f8..bfda9d5 100644 (file)
@@ -291,6 +291,7 @@ struct ipn3ke_hw {
        uint32_t acc_tm;
        uint32_t acc_flow;
 
+       struct ipn3ke_flow_list flow_list;
        uint32_t flow_max_entries;
        uint32_t flow_num_entries;
 
diff --git a/drivers/net/ipn3ke/ipn3ke_flow.c b/drivers/net/ipn3ke/ipn3ke_flow.c
new file mode 100644 (file)
index 0000000..e5937df
--- /dev/null
@@ -0,0 +1,1374 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2019 Intel Corporation
+ */
+
+#include <sys/queue.h>
+#include <stdio.h>
+#include <errno.h>
+#include <stdint.h>
+#include <string.h>
+#include <unistd.h>
+#include <stdarg.h>
+
+#include <rte_io.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 "ipn3ke_rawdev_api.h"
+#include "ipn3ke_flow.h"
+#include "ipn3ke_logs.h"
+#include "ipn3ke_ethdev.h"
+
+/** Static initializer for items. */
+#define FLOW_PATTERNS(...) \
+       ((const enum rte_flow_item_type []) { \
+               __VA_ARGS__, RTE_FLOW_ITEM_TYPE_END, \
+       })
+
+enum IPN3KE_HASH_KEY_TYPE {
+       IPN3KE_HASH_KEY_VXLAN,
+       IPN3KE_HASH_KEY_MAC,
+       IPN3KE_HASH_KEY_QINQ,
+       IPN3KE_HASH_KEY_MPLS,
+       IPN3KE_HASH_KEY_IP_TCP,
+       IPN3KE_HASH_KEY_IP_UDP,
+       IPN3KE_HASH_KEY_IP_NVGRE,
+       IPN3KE_HASH_KEY_VXLAN_IP_UDP,
+};
+
+struct ipn3ke_flow_parse {
+       uint32_t mark:1; /**< Set if the flow is marked. */
+       uint32_t drop:1; /**< ACL drop. */
+       uint32_t key_type:IPN3KE_FLOW_KEY_ID_BITS;
+       uint32_t mark_id:IPN3KE_FLOW_RESULT_UID_BITS; /**< Mark identifier. */
+       uint8_t key_len; /**< Length in bit. */
+       uint8_t key[BITS_TO_BYTES(IPN3KE_FLOW_KEY_DATA_BITS)];
+               /**< key1, key2 */
+};
+
+typedef int (*pattern_filter_t)(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser);
+
+
+struct ipn3ke_flow_pattern {
+       const enum rte_flow_item_type *const items;
+
+       pattern_filter_t filter;
+};
+
+/*
+ * @ RTL definition:
+ * typedef struct packed {
+ * logic [47:0]    vxlan_inner_mac;
+ * logic [23:0]    vxlan_vni;
+ * } Hash_Key_Vxlan_t;
+ *
+ * @ flow items:
+ * RTE_FLOW_ITEM_TYPE_VXLAN
+ * RTE_FLOW_ITEM_TYPE_ETH
+ */
+static int
+ipn3ke_pattern_vxlan(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_item_vxlan *vxlan = NULL;
+       const struct rte_flow_item_eth *eth = NULL;
+       const struct rte_flow_item *item;
+
+       for (item = patterns; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (/*!item->spec || item->mask || */item->last) {
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Only support item with 'spec'");
+                       return -rte_errno;
+               }
+
+               switch (item->type) {
+               case RTE_FLOW_ITEM_TYPE_ETH:
+                       eth = item->spec;
+
+                       rte_memcpy(&parser->key[0],
+                                       eth->src.addr_bytes,
+                                       ETHER_ADDR_LEN);
+                       break;
+
+               case RTE_FLOW_ITEM_TYPE_VXLAN:
+                       vxlan = item->spec;
+
+                       rte_memcpy(&parser->key[6], vxlan->vni, 3);
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support item type");
+                       return -rte_errno;
+               }
+       }
+
+       if (vxlan != NULL && eth != NULL) {
+               parser->key_len = 48 + 24;
+               return 0;
+       }
+
+       rte_flow_error_set(error,
+                       EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ITEM,
+                       patterns,
+                       "Missed some patterns");
+       return -rte_errno;
+}
+
+/*
+ * @ RTL definition:
+ * typedef struct packed {
+ * logic [47:0]    eth_smac;
+ * } Hash_Key_Mac_t;
+ *
+ * @ flow items:
+ * RTE_FLOW_ITEM_TYPE_ETH
+ */
+static int
+ipn3ke_pattern_mac(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_item_eth *eth = NULL;
+       const struct rte_flow_item *item;
+
+       for (item = patterns; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (!item->spec || item->mask || item->last) {
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Only support item with 'spec'");
+                       return -rte_errno;
+               }
+
+               switch (item->type) {
+               case RTE_FLOW_ITEM_TYPE_ETH:
+                       eth = item->spec;
+
+                       rte_memcpy(parser->key,
+                                       eth->src.addr_bytes,
+                                       ETHER_ADDR_LEN);
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support item type");
+                       return -rte_errno;
+               }
+       }
+
+       if (eth != NULL) {
+               parser->key_len = 48;
+               return 0;
+       }
+
+       rte_flow_error_set(error,
+                       EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ITEM,
+                       patterns,
+                       "Missed some patterns");
+       return -rte_errno;
+}
+
+/*
+ * @ RTL definition:
+ * typedef struct packed {
+ * logic [11:0]    outer_vlan_id;
+ * logic [11:0]    inner_vlan_id;
+ * } Hash_Key_QinQ_t;
+ *
+ * @ flow items:
+ * RTE_FLOW_ITEM_TYPE_VLAN
+ * RTE_FLOW_ITEM_TYPE_VLAN
+ */
+static int
+ipn3ke_pattern_qinq(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_item_vlan *outer_vlan = NULL;
+       const struct rte_flow_item_vlan *inner_vlan = NULL;
+       const struct rte_flow_item *item;
+       uint16_t tci;
+
+       for (item = patterns; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (!item->spec || item->mask || item->last) {
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Only support item with 'spec'");
+                       return -rte_errno;
+               }
+
+               switch (item->type) {
+               case RTE_FLOW_ITEM_TYPE_VLAN:
+                       if (!outer_vlan) {
+                               outer_vlan = item->spec;
+
+                               tci = rte_be_to_cpu_16(outer_vlan->tci);
+                               parser->key[0]  = (tci & 0xff0) >> 4;
+                               parser->key[1] |= (tci & 0x00f) << 4;
+                       } else {
+                               inner_vlan = item->spec;
+
+                               tci = rte_be_to_cpu_16(inner_vlan->tci);
+                               parser->key[1] |= (tci & 0xf00) >> 8;
+                               parser->key[2]  = (tci & 0x0ff);
+                       }
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support item type");
+                       return -rte_errno;
+               }
+       }
+
+       if (outer_vlan != NULL && inner_vlan != NULL) {
+               parser->key_len = 12 + 12;
+               return 0;
+       }
+
+       rte_flow_error_set(error,
+                       EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ITEM,
+                       patterns,
+                       "Missed some patterns");
+       return -rte_errno;
+}
+
+/*
+ * @ RTL definition:
+ * typedef struct packed {
+ * logic [19:0]    mpls_label1;
+ * logic [19:0]    mpls_label2;
+ * } Hash_Key_Mpls_t;
+ *
+ * @ flow items:
+ * RTE_FLOW_ITEM_TYPE_MPLS
+ * RTE_FLOW_ITEM_TYPE_MPLS
+ */
+static int
+ipn3ke_pattern_mpls(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_item_mpls *mpls1 = NULL;
+       const struct rte_flow_item_mpls *mpls2 = NULL;
+       const struct rte_flow_item *item;
+
+       for (item = patterns; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (!item->spec || item->mask || item->last) {
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Only support item with 'spec'");
+                       return -rte_errno;
+               }
+
+               switch (item->type) {
+               case RTE_FLOW_ITEM_TYPE_MPLS:
+                       if (!mpls1) {
+                               mpls1 = item->spec;
+
+                               parser->key[0] = mpls1->label_tc_s[0];
+                               parser->key[1] = mpls1->label_tc_s[1];
+                               parser->key[2] = mpls1->label_tc_s[2] & 0xf0;
+                       } else {
+                               mpls2 = item->spec;
+
+                               parser->key[2] |=
+                                       ((mpls2->label_tc_s[0] & 0xf0) >> 4);
+                               parser->key[3] =
+                                       ((mpls2->label_tc_s[0] & 0xf) << 4) |
+                                       ((mpls2->label_tc_s[1] & 0xf0) >> 4);
+                               parser->key[4] =
+                                       ((mpls2->label_tc_s[1] & 0xf) << 4) |
+                                       ((mpls2->label_tc_s[2] & 0xf0) >> 4);
+                       }
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support item type");
+                       return -rte_errno;
+               }
+       }
+
+       if (mpls1 != NULL && mpls2 != NULL) {
+               parser->key_len = 20 + 20;
+               return 0;
+       }
+
+       rte_flow_error_set(error,
+                       EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ITEM,
+                       patterns,
+                       "Missed some patterns");
+       return -rte_errno;
+}
+
+/*
+ * @ RTL definition:
+ * typedef struct packed {
+ * logic [31:0]    ip_sa;
+ * logic [15:0]    tcp_sport;
+ * } Hash_Key_Ip_Tcp_t;
+ *
+ * @ flow items:
+ * RTE_FLOW_ITEM_TYPE_IPV4
+ * RTE_FLOW_ITEM_TYPE_TCP
+ */
+static int
+ipn3ke_pattern_ip_tcp(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_item_ipv4 *ipv4 = NULL;
+       const struct rte_flow_item_tcp *tcp = NULL;
+       const struct rte_flow_item *item;
+
+       for (item = patterns; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (!item->spec || item->mask || item->last) {
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Only support item with 'spec'");
+                       return -rte_errno;
+               }
+
+               switch (item->type) {
+               case RTE_FLOW_ITEM_TYPE_IPV4:
+                       ipv4 = item->spec;
+
+                       rte_memcpy(&parser->key[0], &ipv4->hdr.src_addr, 4);
+                       break;
+
+               case RTE_FLOW_ITEM_TYPE_TCP:
+                       tcp = item->spec;
+
+                       rte_memcpy(&parser->key[4], &tcp->hdr.src_port, 2);
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support item type");
+                       return -rte_errno;
+               }
+       }
+
+       if (ipv4 != NULL && tcp != NULL) {
+               parser->key_len = 32 + 16;
+               return 0;
+       }
+
+       rte_flow_error_set(error,
+                       EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ITEM,
+                       patterns,
+                       "Missed some patterns");
+       return -rte_errno;
+}
+
+/*
+ * @ RTL definition:
+ * typedef struct packed {
+ * logic [31:0]    ip_sa;
+ * logic [15:0]    udp_sport;
+ * } Hash_Key_Ip_Udp_t;
+ *
+ * @ flow items:
+ * RTE_FLOW_ITEM_TYPE_IPV4
+ * RTE_FLOW_ITEM_TYPE_UDP
+ */
+static int
+ipn3ke_pattern_ip_udp(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_item_ipv4 *ipv4 = NULL;
+       const struct rte_flow_item_udp *udp = NULL;
+       const struct rte_flow_item *item;
+
+       for (item = patterns; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (!item->spec || item->mask || item->last) {
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Only support item with 'spec'");
+                       return -rte_errno;
+               }
+
+               switch (item->type) {
+               case RTE_FLOW_ITEM_TYPE_IPV4:
+                       ipv4 = item->spec;
+
+                       rte_memcpy(&parser->key[0], &ipv4->hdr.src_addr, 4);
+                       break;
+
+               case RTE_FLOW_ITEM_TYPE_UDP:
+                       udp = item->spec;
+
+                       rte_memcpy(&parser->key[4], &udp->hdr.src_port, 2);
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support item type");
+                       return -rte_errno;
+               }
+       }
+
+       if (ipv4 != NULL && udp != NULL) {
+               parser->key_len = 32 + 16;
+               return 0;
+       }
+
+       rte_flow_error_set(error,
+                       EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ITEM,
+                       patterns,
+                       "Missed some patterns");
+       return -rte_errno;
+}
+
+/*
+ * @ RTL definition:
+ * typedef struct packed {
+ * logic [31:0]    ip_sa;
+ * logic [15:0]    udp_sport;
+ * logic [23:0]    vsid;
+ * } Hash_Key_Ip_Nvgre_t;
+ *
+ * @ flow items:
+ * RTE_FLOW_ITEM_TYPE_IPV4
+ * RTE_FLOW_ITEM_TYPE_UDP
+ * RTE_FLOW_ITEM_TYPE_NVGRE
+ */
+static int
+ipn3ke_pattern_ip_nvgre(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_item_nvgre *nvgre = NULL;
+       const struct rte_flow_item_ipv4 *ipv4 = NULL;
+       const struct rte_flow_item_udp *udp = NULL;
+       const struct rte_flow_item *item;
+
+       for (item = patterns; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (!item->spec || item->mask || item->last) {
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Only support item with 'spec'");
+                       return -rte_errno;
+               }
+
+               switch (item->type) {
+               case RTE_FLOW_ITEM_TYPE_IPV4:
+                       ipv4 = item->spec;
+
+                       rte_memcpy(&parser->key[0], &ipv4->hdr.src_addr, 4);
+                       break;
+
+               case RTE_FLOW_ITEM_TYPE_UDP:
+                       udp = item->spec;
+
+                       rte_memcpy(&parser->key[4], &udp->hdr.src_port, 2);
+                       break;
+
+               case RTE_FLOW_ITEM_TYPE_NVGRE:
+                       nvgre = item->spec;
+
+                       rte_memcpy(&parser->key[6], nvgre->tni, 3);
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support item type");
+                       return -rte_errno;
+               }
+       }
+
+       if (ipv4 != NULL && udp != NULL && nvgre != NULL) {
+               parser->key_len = 32 + 16 + 24;
+               return 0;
+       }
+
+       rte_flow_error_set(error,
+                       EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ITEM,
+                       patterns,
+                       "Missed some patterns");
+       return -rte_errno;
+}
+
+/*
+ * @ RTL definition:
+ * typedef struct packed{
+ * logic [23:0]    vxlan_vni;
+ * logic [31:0]    ip_sa;
+ * logic [15:0]    udp_sport;
+ * } Hash_Key_Vxlan_Ip_Udp_t;
+ *
+ * @ flow items:
+ * RTE_FLOW_ITEM_TYPE_VXLAN
+ * RTE_FLOW_ITEM_TYPE_IPV4
+ * RTE_FLOW_ITEM_TYPE_UDP
+ */
+static int
+ipn3ke_pattern_vxlan_ip_udp(const struct rte_flow_item patterns[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_item_vxlan *vxlan = NULL;
+       const struct rte_flow_item_ipv4 *ipv4 = NULL;
+       const struct rte_flow_item_udp *udp = NULL;
+       const struct rte_flow_item *item;
+
+       for (item = patterns; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (!item->spec || item->mask || item->last) {
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Only support item with 'spec'");
+                       return -rte_errno;
+               }
+
+               switch (item->type) {
+               case RTE_FLOW_ITEM_TYPE_VXLAN:
+                       vxlan = item->spec;
+
+                       rte_memcpy(&parser->key[0], vxlan->vni, 3);
+                       break;
+
+               case RTE_FLOW_ITEM_TYPE_IPV4:
+                       ipv4 = item->spec;
+
+                       rte_memcpy(&parser->key[3], &ipv4->hdr.src_addr, 4);
+                       break;
+
+               case RTE_FLOW_ITEM_TYPE_UDP:
+                       udp = item->spec;
+
+                       rte_memcpy(&parser->key[7], &udp->hdr.src_port, 2);
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support item type");
+                       return -rte_errno;
+               }
+       }
+
+       if (vxlan != NULL && ipv4 != NULL && udp != NULL) {
+               parser->key_len = 24 + 32 + 16;
+               return 0;
+       }
+
+       rte_flow_error_set(error,
+                       EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ITEM,
+                       patterns,
+                       "Missed some patterns");
+       return -rte_errno;
+}
+
+static const struct ipn3ke_flow_pattern ipn3ke_supported_patterns[] = {
+       [IPN3KE_HASH_KEY_VXLAN] = {
+               .items = FLOW_PATTERNS(RTE_FLOW_ITEM_TYPE_VXLAN,
+                                       RTE_FLOW_ITEM_TYPE_ETH),
+               .filter = ipn3ke_pattern_vxlan,
+       },
+
+       [IPN3KE_HASH_KEY_MAC] = {
+               .items = FLOW_PATTERNS(RTE_FLOW_ITEM_TYPE_ETH),
+               .filter = ipn3ke_pattern_mac,
+       },
+
+       [IPN3KE_HASH_KEY_QINQ] = {
+               .items = FLOW_PATTERNS(RTE_FLOW_ITEM_TYPE_VLAN,
+                                       RTE_FLOW_ITEM_TYPE_VLAN),
+               .filter = ipn3ke_pattern_qinq,
+       },
+
+       [IPN3KE_HASH_KEY_MPLS] = {
+               .items = FLOW_PATTERNS(RTE_FLOW_ITEM_TYPE_MPLS,
+                                       RTE_FLOW_ITEM_TYPE_MPLS),
+               .filter = ipn3ke_pattern_mpls,
+       },
+
+       [IPN3KE_HASH_KEY_IP_TCP] = {
+               .items = FLOW_PATTERNS(RTE_FLOW_ITEM_TYPE_IPV4,
+                                       RTE_FLOW_ITEM_TYPE_TCP),
+               .filter = ipn3ke_pattern_ip_tcp,
+       },
+
+       [IPN3KE_HASH_KEY_IP_UDP] = {
+               .items = FLOW_PATTERNS(RTE_FLOW_ITEM_TYPE_IPV4,
+                                       RTE_FLOW_ITEM_TYPE_UDP),
+               .filter = ipn3ke_pattern_ip_udp,
+       },
+
+       [IPN3KE_HASH_KEY_IP_NVGRE] = {
+               .items = FLOW_PATTERNS(RTE_FLOW_ITEM_TYPE_IPV4,
+                                       RTE_FLOW_ITEM_TYPE_UDP,
+                                       RTE_FLOW_ITEM_TYPE_NVGRE),
+               .filter = ipn3ke_pattern_ip_nvgre,
+       },
+
+       [IPN3KE_HASH_KEY_VXLAN_IP_UDP] = {
+               .items = FLOW_PATTERNS(RTE_FLOW_ITEM_TYPE_VXLAN,
+                                       RTE_FLOW_ITEM_TYPE_IPV4,
+                                       RTE_FLOW_ITEM_TYPE_UDP),
+               .filter = ipn3ke_pattern_vxlan_ip_udp,
+       },
+};
+
+static int
+ipn3ke_flow_convert_attributes(const struct rte_flow_attr *attr,
+                               struct rte_flow_error *error)
+{
+       if (!attr) {
+               rte_flow_error_set(error,
+                               EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ATTR,
+                               NULL,
+                               "NULL attribute.");
+               return -rte_errno;
+       }
+
+       if (attr->group) {
+               rte_flow_error_set(error,
+                               ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
+                               NULL,
+                               "groups are not supported");
+               return -rte_errno;
+       }
+
+       if (attr->egress) {
+               rte_flow_error_set(error,
+                               ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
+                               NULL,
+                               "egress is not supported");
+               return -rte_errno;
+       }
+
+       if (attr->transfer) {
+               rte_flow_error_set(error,
+                               ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
+                               NULL,
+                               "transfer is not supported");
+               return -rte_errno;
+       }
+
+       if (!attr->ingress) {
+               rte_flow_error_set(error,
+                               ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
+                               NULL,
+                               "only ingress is supported");
+               return -rte_errno;
+       }
+
+       return 0;
+}
+
+static int
+ipn3ke_flow_convert_actions(const struct rte_flow_action actions[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       const struct rte_flow_action_mark *mark = NULL;
+
+       if (!actions) {
+               rte_flow_error_set(error,
+                               EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ACTION_NUM,
+                               NULL,
+                               "NULL action.");
+               return -rte_errno;
+       }
+
+       for (; actions->type != RTE_FLOW_ACTION_TYPE_END; ++actions) {
+               switch (actions->type) {
+               case RTE_FLOW_ACTION_TYPE_VOID:
+                       break;
+
+               case RTE_FLOW_ACTION_TYPE_MARK:
+                       if (mark) {
+                               rte_flow_error_set(error,
+                                               ENOTSUP,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               actions,
+                                               "duplicated mark");
+                               return -rte_errno;
+                       }
+
+                       mark = actions->conf;
+                       if (!mark) {
+                               rte_flow_error_set(error,
+                                               EINVAL,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               actions,
+                                               "mark must be defined");
+                               return -rte_errno;
+                       } else if (mark->id > IPN3KE_FLOW_RESULT_UID_MAX) {
+                               rte_flow_error_set(error,
+                                               ENOTSUP,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               actions,
+                                               "mark id is out of range");
+                               return -rte_errno;
+                       }
+
+                       parser->mark = 1;
+                       parser->mark_id = mark->id;
+                       break;
+
+               case RTE_FLOW_ACTION_TYPE_DROP:
+                       parser->drop = 1;
+                       break;
+
+               default:
+                       rte_flow_error_set(error,
+                                       ENOTSUP,
+                                       RTE_FLOW_ERROR_TYPE_ACTION,
+                                       actions,
+                                       "invalid action");
+                       return -rte_errno;
+               }
+       }
+
+       if (!parser->drop && !parser->mark) {
+               rte_flow_error_set(error,
+                               EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ACTION,
+                               actions,
+                               "no valid actions");
+               return -rte_errno;
+       }
+
+       return 0;
+}
+
+static bool
+ipn3ke_match_pattern(const enum rte_flow_item_type *patterns,
+                               const struct rte_flow_item *input)
+{
+       const struct rte_flow_item *item = input;
+
+       while ((*patterns == item->type) &&
+               (*patterns != RTE_FLOW_ITEM_TYPE_END)) {
+               patterns++;
+               item++;
+       }
+
+       return (*patterns == RTE_FLOW_ITEM_TYPE_END &&
+               item->type == RTE_FLOW_ITEM_TYPE_END);
+}
+
+static pattern_filter_t
+ipn3ke_find_filter_func(const struct rte_flow_item *input,
+                               uint32_t *idx)
+{
+       pattern_filter_t filter = NULL;
+       uint32_t i;
+
+       for (i = 0; i < RTE_DIM(ipn3ke_supported_patterns); i++) {
+               if (ipn3ke_match_pattern(ipn3ke_supported_patterns[i].items,
+                                       input)) {
+                       filter = ipn3ke_supported_patterns[i].filter;
+                       *idx = i;
+                       break;
+               }
+       }
+
+       return filter;
+}
+
+static int
+ipn3ke_flow_convert_items(const struct rte_flow_item items[],
+       struct rte_flow_error *error, struct ipn3ke_flow_parse *parser)
+{
+       pattern_filter_t filter = NULL;
+       uint32_t idx;
+
+       if (!items) {
+               rte_flow_error_set(error,
+                               EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ITEM_NUM,
+                               NULL,
+                               "NULL pattern.");
+               return -rte_errno;
+       }
+
+       filter = ipn3ke_find_filter_func(items, &idx);
+
+       if (!filter) {
+               rte_flow_error_set(error,
+                               EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ITEM,
+                               items,
+                               "Unsupported pattern");
+               return -rte_errno;
+       }
+
+       parser->key_type = idx;
+
+       return filter(items, error, parser);
+}
+
+/* Put the least @nbits of @data into @offset of @dst bits stream, and
+ * the @offset starts from MSB to LSB in each byte.
+ *
+ * MSB    LSB
+ *  +------+------+------+------+
+ *  |      |      |      |      |
+ *  +------+------+------+------+
+ *       ^                 ^
+ *       |<- data: nbits ->|
+ *       |
+ *     offset
+ */
+static void
+copy_data_bits(uint8_t *dst, uint64_t data,
+               uint32_t offset, uint8_t nbits)
+{
+       uint8_t set, *p = &dst[offset / BITS_PER_BYTE];
+       uint8_t bits_to_set = BITS_PER_BYTE - (offset % BITS_PER_BYTE);
+       uint8_t mask_to_set = 0xff >> (offset % BITS_PER_BYTE);
+       uint32_t size = offset + nbits;
+
+       if (nbits > (sizeof(data) * BITS_PER_BYTE)) {
+               IPN3KE_AFU_PMD_ERR("nbits is out of range");
+               return;
+       }
+
+       while (nbits - bits_to_set >= 0) {
+               set = data >> (nbits - bits_to_set);
+
+               *p &= ~mask_to_set;
+               *p |= (set & mask_to_set);
+
+               nbits -= bits_to_set;
+               bits_to_set = BITS_PER_BYTE;
+               mask_to_set = 0xff;
+               p++;
+       }
+
+       if (nbits) {
+               uint8_t shift = BITS_PER_BYTE - (size % BITS_PER_BYTE);
+
+               set = data << shift;
+               mask_to_set = 0xff << shift;
+
+               *p &= ~mask_to_set;
+               *p |= (set & mask_to_set);
+       }
+}
+
+static void
+ipn3ke_flow_key_generation(struct ipn3ke_flow_parse *parser,
+                               struct rte_flow *flow)
+{
+       uint32_t i, shift_bytes, len_in_bytes, offset;
+       uint64_t key;
+       uint8_t *dst;
+
+       dst = flow->rule.key;
+
+       copy_data_bits(dst,
+                       parser->key_type,
+                       IPN3KE_FLOW_KEY_ID_OFFSET,
+                       IPN3KE_FLOW_KEY_ID_BITS);
+
+       /* The MSb of key is filled to 0 when it is less than
+        * IPN3KE_FLOW_KEY_DATA_BITS bit. And the parsed key data is
+        * save as MSB byte first in the array, it needs to move
+        * the bits before formatting them.
+        */
+       key = 0;
+       shift_bytes = 0;
+       len_in_bytes = BITS_TO_BYTES(parser->key_len);
+       offset = (IPN3KE_FLOW_KEY_DATA_OFFSET +
+               IPN3KE_FLOW_KEY_DATA_BITS -
+               parser->key_len);
+
+       for (i = 0; i < len_in_bytes; i++) {
+               key = (key << 8) | parser->key[i];
+
+               if (++shift_bytes == sizeof(key)) {
+                       shift_bytes = 0;
+
+                       copy_data_bits(dst, key, offset,
+                                       sizeof(key) * BITS_PER_BYTE);
+                       offset += sizeof(key) * BITS_PER_BYTE;
+                       key = 0;
+               }
+       }
+
+       if (shift_bytes != 0) {
+               uint32_t rem_bits;
+
+               rem_bits = parser->key_len % (sizeof(key) * BITS_PER_BYTE);
+               key >>= (shift_bytes * 8 - rem_bits);
+               copy_data_bits(dst, key, offset, rem_bits);
+       }
+}
+
+static void
+ipn3ke_flow_result_generation(struct ipn3ke_flow_parse *parser,
+                               struct rte_flow *flow)
+{
+       uint8_t *dst;
+
+       if (parser->drop)
+               return;
+
+       dst = flow->rule.result;
+
+       copy_data_bits(dst,
+                       1,
+                       IPN3KE_FLOW_RESULT_ACL_OFFSET,
+                       IPN3KE_FLOW_RESULT_ACL_BITS);
+
+       copy_data_bits(dst,
+                       parser->mark_id,
+                       IPN3KE_FLOW_RESULT_UID_OFFSET,
+                       IPN3KE_FLOW_RESULT_UID_BITS);
+}
+
+#define MHL_COMMAND_TIME_COUNT        0xFFFF
+#define MHL_COMMAND_TIME_INTERVAL_US  10
+
+static int
+ipn3ke_flow_hw_update(struct ipn3ke_hw *hw,
+                       struct rte_flow *flow, uint32_t is_add)
+{
+       uint32_t *pdata = NULL;
+       uint32_t data;
+       uint32_t time_out = MHL_COMMAND_TIME_COUNT;
+       uint32_t i;
+
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE flow dump start\n");
+
+       pdata = (uint32_t *)flow->rule.key;
+       IPN3KE_AFU_PMD_DEBUG(" - key   :");
+
+       for (i = 0; i < RTE_DIM(flow->rule.key); i++)
+               IPN3KE_AFU_PMD_DEBUG(" %02x", flow->rule.key[i]);
+
+       for (i = 0; i < 4; i++)
+               IPN3KE_AFU_PMD_DEBUG(" %02x", ipn3ke_swap32(pdata[3 - i]));
+       IPN3KE_AFU_PMD_DEBUG("\n");
+
+       pdata = (uint32_t *)flow->rule.result;
+       IPN3KE_AFU_PMD_DEBUG(" - result:");
+
+       for (i = 0; i < RTE_DIM(flow->rule.result); i++)
+               IPN3KE_AFU_PMD_DEBUG(" %02x", flow->rule.result[i]);
+
+       for (i = 0; i < 1; i++)
+               IPN3KE_AFU_PMD_DEBUG(" %02x", pdata[i]);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE flow dump end\n");
+
+       pdata = (uint32_t *)flow->rule.key;
+
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_MHL_KEY_0,
+                       0,
+                       ipn3ke_swap32(pdata[3]),
+                       IPN3KE_CLF_MHL_KEY_MASK);
+
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_MHL_KEY_1,
+                       0,
+                       ipn3ke_swap32(pdata[2]),
+                       IPN3KE_CLF_MHL_KEY_MASK);
+
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_MHL_KEY_2,
+                       0,
+                       ipn3ke_swap32(pdata[1]),
+                       IPN3KE_CLF_MHL_KEY_MASK);
+
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_MHL_KEY_3,
+                       0,
+                       ipn3ke_swap32(pdata[0]),
+                       IPN3KE_CLF_MHL_KEY_MASK);
+
+       pdata = (uint32_t *)flow->rule.result;
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_MHL_RES,
+                       0,
+                       ipn3ke_swap32(pdata[0]),
+                       IPN3KE_CLF_MHL_RES_MASK);
+
+       /* insert/delete the key and result */
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_MHL_MGMT_CTRL,
+                               0,
+                               0x80000000);
+       time_out = MHL_COMMAND_TIME_COUNT;
+       while (IPN3KE_BIT_ISSET(data, IPN3KE_CLF_MHL_MGMT_CTRL_BIT_BUSY) &&
+               (time_out > 0)) {
+               data = IPN3KE_MASK_READ_REG(hw,
+                                       IPN3KE_CLF_MHL_MGMT_CTRL,
+                                       0,
+                                       0x80000000);
+               time_out--;
+               rte_delay_us(MHL_COMMAND_TIME_INTERVAL_US);
+       }
+       if (!time_out)
+               return -1;
+       if (is_add)
+               IPN3KE_MASK_WRITE_REG(hw,
+                               IPN3KE_CLF_MHL_MGMT_CTRL,
+                               0,
+                               IPN3KE_CLF_MHL_MGMT_CTRL_INSERT,
+                               0x3);
+       else
+               IPN3KE_MASK_WRITE_REG(hw,
+                               IPN3KE_CLF_MHL_MGMT_CTRL,
+                               0,
+                               IPN3KE_CLF_MHL_MGMT_CTRL_DELETE,
+                               0x3);
+
+       return 0;
+}
+
+static int
+ipn3ke_flow_hw_flush(struct ipn3ke_hw *hw)
+{
+       uint32_t data;
+       uint32_t time_out = MHL_COMMAND_TIME_COUNT;
+
+       /* flush the MHL lookup table */
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_MHL_MGMT_CTRL,
+                               0,
+                               0x80000000);
+       time_out = MHL_COMMAND_TIME_COUNT;
+       while (IPN3KE_BIT_ISSET(data, IPN3KE_CLF_MHL_MGMT_CTRL_BIT_BUSY) &&
+               (time_out > 0)) {
+               data = IPN3KE_MASK_READ_REG(hw,
+                                       IPN3KE_CLF_MHL_MGMT_CTRL,
+                                       0,
+                                       0x80000000);
+               time_out--;
+               rte_delay_us(MHL_COMMAND_TIME_INTERVAL_US);
+       }
+       if (!time_out)
+               return -1;
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_MHL_MGMT_CTRL,
+                       0,
+                       IPN3KE_CLF_MHL_MGMT_CTRL_FLUSH,
+                       0x3);
+
+       return 0;
+}
+
+static void
+ipn3ke_flow_convert_finalise(struct ipn3ke_hw *hw,
+       struct ipn3ke_flow_parse *parser, struct rte_flow *flow)
+{
+       ipn3ke_flow_key_generation(parser, flow);
+       ipn3ke_flow_result_generation(parser, flow);
+       ipn3ke_flow_hw_update(hw, flow, 1);
+}
+
+static int
+ipn3ke_flow_convert(const struct rte_flow_attr *attr,
+       const struct rte_flow_item items[],
+       const struct rte_flow_action actions[], struct rte_flow_error *error,
+       struct ipn3ke_flow_parse *parser)
+{
+       int ret;
+
+       ret = ipn3ke_flow_convert_attributes(attr, error);
+       if (ret)
+               return ret;
+
+       ret = ipn3ke_flow_convert_actions(actions, error, parser);
+       if (ret)
+               return ret;
+
+       ret = ipn3ke_flow_convert_items(items, error, parser);
+       if (ret)
+               return ret;
+
+       return 0;
+}
+
+static int
+ipn3ke_flow_validate(__rte_unused struct rte_eth_dev *dev,
+       const struct rte_flow_attr *attr, const struct rte_flow_item pattern[],
+       const struct rte_flow_action actions[], struct rte_flow_error *error)
+{
+       struct ipn3ke_flow_parse parser = {0};
+       return ipn3ke_flow_convert(attr, pattern, actions, error, &parser);
+}
+
+static struct rte_flow *
+ipn3ke_flow_create(struct rte_eth_dev *dev,
+       const struct rte_flow_attr *attr, const struct rte_flow_item pattern[],
+       const struct rte_flow_action actions[], struct rte_flow_error *error)
+{
+       struct ipn3ke_hw *hw = IPN3KE_DEV_PRIVATE_TO_HW(dev);
+       struct ipn3ke_flow_parse parser = {0};
+       struct rte_flow *flow;
+       int ret;
+
+       if (hw->flow_num_entries == hw->flow_max_entries) {
+               rte_flow_error_set(error,
+                               ENOBUFS,
+                               RTE_FLOW_ERROR_TYPE_HANDLE,
+                               NULL,
+                               "The flow table is full.");
+               return NULL;
+       }
+
+       ret = ipn3ke_flow_convert(attr, pattern, actions, error, &parser);
+       if (ret < 0) {
+               rte_flow_error_set(error,
+                               -ret,
+                               RTE_FLOW_ERROR_TYPE_HANDLE,
+                               NULL,
+                               "Failed to create flow.");
+               return NULL;
+       }
+
+       flow = rte_zmalloc("ipn3ke_flow", sizeof(struct rte_flow), 0);
+       if (!flow) {
+               rte_flow_error_set(error,
+                               ENOMEM,
+                               RTE_FLOW_ERROR_TYPE_HANDLE,
+                               NULL,
+                               "Failed to allocate memory");
+               return flow;
+       }
+
+       ipn3ke_flow_convert_finalise(hw, &parser, flow);
+
+       TAILQ_INSERT_TAIL(&hw->flow_list, flow, next);
+
+       return flow;
+}
+
+static int
+ipn3ke_flow_destroy(struct rte_eth_dev *dev,
+       struct rte_flow *flow, struct rte_flow_error *error)
+{
+       struct ipn3ke_hw *hw = IPN3KE_DEV_PRIVATE_TO_HW(dev);
+       int ret = 0;
+
+       ret = ipn3ke_flow_hw_update(hw, flow, 0);
+       if (!ret) {
+               TAILQ_REMOVE(&hw->flow_list, flow, next);
+               rte_free(flow);
+       } else {
+               rte_flow_error_set(error,
+                               -ret,
+                               RTE_FLOW_ERROR_TYPE_HANDLE,
+                               NULL,
+                               "Failed to destroy flow.");
+       }
+
+       return ret;
+}
+
+static int
+ipn3ke_flow_flush(struct rte_eth_dev *dev,
+               __rte_unused struct rte_flow_error *error)
+{
+       struct ipn3ke_hw *hw = IPN3KE_DEV_PRIVATE_TO_HW(dev);
+       struct rte_flow *flow, *temp;
+
+       TAILQ_FOREACH_SAFE(flow, &hw->flow_list, next, temp) {
+               TAILQ_REMOVE(&hw->flow_list, flow, next);
+               rte_free(flow);
+       }
+
+       return ipn3ke_flow_hw_flush(hw);
+}
+
+int ipn3ke_flow_init(void *dev)
+{
+       struct ipn3ke_hw *hw = (struct ipn3ke_hw *)dev;
+       uint32_t data;
+
+       /* disable rx classifier bypass */
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_RX_TEST,
+                       0, 0, 0x1);
+
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_RX_TEST,
+                               0,
+                               0x1);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE_CLF_RX_TEST: %x\n", data);
+
+       /* configure base mac address */
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_BASE_DST_MAC_ADDR_HI,
+                       0,
+                       0x2457,
+                       0xFFFF);
+
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_BASE_DST_MAC_ADDR_HI,
+                               0,
+                               0xFFFF);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE_CLF_BASE_DST_MAC_ADDR_HI: %x\n", data);
+
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_BASE_DST_MAC_ADDR_LOW,
+                       0,
+                       0x9bdf1000,
+                       0xFFFFFFFF);
+
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_BASE_DST_MAC_ADDR_LOW,
+                               0,
+                               0xFFFFFFFF);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE_CLF_BASE_DST_MAC_ADDR_LOW: %x\n", data);
+
+
+       /* configure hash lookup rules enable */
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_LKUP_ENABLE,
+                       0,
+                       0xFD,
+                       0xFF);
+
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_LKUP_ENABLE,
+                               0,
+                               0xFF);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE_CLF_LKUP_ENABLE: %x\n", data);
+
+
+       /* configure rx parse config, settings associatied with VxLAN */
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_RX_PARSE_CFG,
+                       0,
+                       0x212b5,
+                       0x3FFFF);
+
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_RX_PARSE_CFG,
+                               0,
+                               0x3FFFF);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE_CLF_RX_PARSE_CFG: %x\n", data);
+
+
+       /* configure QinQ S-Tag */
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_QINQ_STAG,
+                       0,
+                       0x88a8,
+                       0xFFFF);
+
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_QINQ_STAG,
+                               0,
+                               0xFFFF);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE_CLF_QINQ_STAG: %x\n", data);
+
+
+       /* configure gen ctrl */
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_MHL_GEN_CTRL,
+                       0,
+                       0x3,
+                       0x3);
+
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_MHL_GEN_CTRL,
+                               0,
+                               0x1F);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE_CLF_MHL_GEN_CTRL: %x\n", data);
+
+
+       /* clear monitoring register */
+       IPN3KE_MASK_WRITE_REG(hw,
+                       IPN3KE_CLF_MHL_MON_0,
+                       0,
+                       0xFFFFFFFF,
+                       0xFFFFFFFF);
+
+       data = 0;
+       data = IPN3KE_MASK_READ_REG(hw,
+                               IPN3KE_CLF_MHL_MON_0,
+                               0,
+                               0xFFFFFFFF);
+       IPN3KE_AFU_PMD_DEBUG("IPN3KE_CLF_MHL_MON_0: %x\n", data);
+
+
+       ipn3ke_flow_hw_flush(hw);
+
+       TAILQ_INIT(&hw->flow_list);
+       hw->flow_max_entries = IPN3KE_MASK_READ_REG(hw,
+                                               IPN3KE_CLF_EM_NUM,
+                                               0,
+                                               0xFFFFFFFF);
+       hw->flow_num_entries = 0;
+
+       return 0;
+}
+
+const struct rte_flow_ops ipn3ke_flow_ops = {
+       .validate = ipn3ke_flow_validate,
+       .create = ipn3ke_flow_create,
+       .destroy = ipn3ke_flow_destroy,
+       .flush = ipn3ke_flow_flush,
+};
+
diff --git a/drivers/net/ipn3ke/ipn3ke_flow.h b/drivers/net/ipn3ke/ipn3ke_flow.h
new file mode 100644 (file)
index 0000000..ef1a61f
--- /dev/null
@@ -0,0 +1,106 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2019 Intel Corporation
+ */
+
+#ifndef _IPN3KE_FLOW_H_
+#define _IPN3KE_FLOW_H_
+
+/**
+ * Expand the length to DWORD alignment with 'Unused' field.
+ *
+ * FLOW KEY:
+ *  | Unused |Ruler id (id)  | Key1 Key2 … (data) |
+ *  |--------+---------------+--------------------|
+ *  | 17bits |    3 bits     |   Total 108 bits   |
+ * MSB                 --->                      LSB
+ *
+ * Note: And the MSb of key data is filled to 0 when it is less
+ *       than 108 bit.
+ */
+#define IPN3KE_FLOW_KEY_UNUSED_BITS  17
+#define IPN3KE_FLOW_KEY_ID_BITS      3
+#define IPN3KE_FLOW_KEY_DATA_BITS    108
+
+#define IPN3KE_FLOW_KEY_TOTAL_BITS \
+               (IPN3KE_FLOW_KEY_UNUSED_BITS + \
+               IPN3KE_FLOW_KEY_ID_BITS + \
+               IPN3KE_FLOW_KEY_DATA_BITS)
+
+#define IPN3KE_FLOW_KEY_ID_OFFSET \
+               (IPN3KE_FLOW_KEY_UNUSED_BITS)
+
+#define IPN3KE_FLOW_KEY_DATA_OFFSET \
+               (IPN3KE_FLOW_KEY_ID_OFFSET + IPN3KE_FLOW_KEY_ID_BITS)
+
+/**
+ * Expand the length to DWORD alignment with 'Unused' field.
+ *
+ * FLOW RESULT:
+ *  |  Unused | enable (acl) |    uid       |
+ *  |---------+--------------+--------------|
+ *  | 15 bits |    1 bit     |   16 bits    |
+ * MSB              --->                   LSB
+ */
+
+#define IPN3KE_FLOW_RESULT_UNUSED_BITS 15
+#define IPN3KE_FLOW_RESULT_ACL_BITS    1
+#define IPN3KE_FLOW_RESULT_UID_BITS    16
+
+#define IPN3KE_FLOW_RESULT_TOTAL_BITS \
+               (IPN3KE_FLOW_RESULT_UNUSED_BITS + \
+               IPN3KE_FLOW_RESULT_ACL_BITS + \
+               IPN3KE_FLOW_RESULT_UID_BITS)
+
+#define IPN3KE_FLOW_RESULT_ACL_OFFSET \
+               (IPN3KE_FLOW_RESULT_UNUSED_BITS)
+
+#define IPN3KE_FLOW_RESULT_UID_OFFSET \
+               (IPN3KE_FLOW_RESULT_ACL_OFFSET + IPN3KE_FLOW_RESULT_ACL_BITS)
+
+#define IPN3KE_FLOW_RESULT_UID_MAX \
+               ((1UL << IPN3KE_FLOW_RESULT_UID_BITS) - 1)
+
+#ifndef BITS_PER_BYTE
+#define BITS_PER_BYTE    8
+#endif
+#define BITS_TO_BYTES(bits) \
+       (((bits) + BITS_PER_BYTE - 1) / BITS_PER_BYTE)
+
+struct ipn3ke_flow_rule {
+       uint8_t key[BITS_TO_BYTES(IPN3KE_FLOW_KEY_TOTAL_BITS)];
+       uint8_t result[BITS_TO_BYTES(IPN3KE_FLOW_RESULT_TOTAL_BITS)];
+};
+
+struct rte_flow {
+       TAILQ_ENTRY(rte_flow) next; /**< Pointer to the next flow structure. */
+
+       struct ipn3ke_flow_rule rule;
+};
+
+TAILQ_HEAD(ipn3ke_flow_list, rte_flow);
+
+static inline uint16_t ipn3ke_swap16(uint16_t x)
+{
+       return ((x & 0xff) << 8) | ((x >> 8) & 0xff);
+}
+
+static inline uint32_t ipn3ke_swap32(uint32_t x)
+{
+       uint32_t high, low;
+       uint32_t high1, low1;
+
+       high = (x >> 16) & 0xffff;
+       low = x & 0xffff;
+       high1 = ipn3ke_swap16(low);
+       high1 = high1 << 16;
+       low1 = ipn3ke_swap16(high);
+       low1 = low1 & 0xffff;
+
+       return high1 | low1;
+}
+
+extern const struct rte_flow_ops ipn3ke_flow_ops;
+
+int ipn3ke_flow_init(void *dev);
+
+#endif /* _IPN3KE_FLOW_H_ */
index 63098bf..cf3b214 100644 (file)
@@ -21,6 +21,7 @@
 #include <ifpga_logs.h>
 
 #include "ipn3ke_rawdev_api.h"
+#include "ipn3ke_flow.h"
 #include "ipn3ke_logs.h"
 #include "ipn3ke_ethdev.h"
 
@@ -746,7 +747,7 @@ ipn3ke_afu_filter_ctrl(struct rte_eth_dev *ethdev,
                case RTE_ETH_FILTER_GENERIC:
                        if (filter_op != RTE_ETH_FILTER_GET)
                                return -EINVAL;
-                       *(const void **)arg = NULL;
+                       *(const void **)arg = &ipn3ke_flow_ops;
                        break;
                default:
                        IPN3KE_AFU_PMD_WARN("Filter type (%d) not supported",
index 8baa2fb..4bcf3aa 100644 (file)
@@ -23,6 +23,7 @@
 #include <ifpga_logs.h>
 
 #include "ipn3ke_rawdev_api.h"
+#include "ipn3ke_flow.h"
 #include "ipn3ke_logs.h"
 #include "ipn3ke_ethdev.h"
 
index 3a95efc..74b4d7c 100644 (file)
@@ -12,5 +12,6 @@ allow_experimental_apis = true
 
 sources += files('ipn3ke_ethdev.c',
        'ipn3ke_representor.c',
-       'ipn3ke_tm.c')
+       'ipn3ke_tm.c',
+       'ipn3ke_flow.c')
 deps += ['bus_ifpga', 'sched']