igb: add filters
authorJingjing Wu <jingjing.wu@intel.com>
Mon, 16 Jun 2014 07:31:44 +0000 (15:31 +0800)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Mon, 16 Jun 2014 17:55:04 +0000 (19:55 +0200)
This patch adds following igb NIC filters implement:
  syn filter, ethertype filter, 2tuple filter, flex filter for intel NIC 82580 and i350
  syn filter, ethertype filter, 5tuple filter for intel NIC 82576

Signed-off-by: jingjing.wu <jingjing.wu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Reviewed-by: Vladimir Medvedkin <medvedkinv@gmail.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
lib/librte_pmd_e1000/e1000_ethdev.h
lib/librte_pmd_e1000/igb_ethdev.c

index a287f40..71eb5fb 100644 (file)
 #define E1000_CTRL_EXT_EXTEND_VLAN  (1<<26)    /* EXTENDED VLAN */
 #define IGB_VFTA_SIZE 128
 
+#define IGB_MAX_RX_QUEUE_NUM           8
+#define IGB_MAX_RX_QUEUE_NUM_82576     16
+
+#define E1000_SYN_FILTER_ENABLE        0x00000001 /* syn filter enable field */
+#define E1000_SYN_FILTER_QUEUE         0x0000000E /* syn filter queue field */
+#define E1000_SYN_FILTER_QUEUE_SHIFT   1          /* syn filter queue field */
+#define E1000_RFCTL_SYNQFP             0x00080000 /* SYNQFP in RFCTL register */
+
+#define E1000_ETQF_ETHERTYPE           0x0000FFFF
+#define E1000_ETQF_QUEUE               0x00070000
+#define E1000_ETQF_QUEUE_SHIFT         16
+#define E1000_MAX_ETQF_FILTERS         8
+
+#define E1000_IMIR_DSTPORT             0x0000FFFF
+#define E1000_IMIR_PRIORITY            0xE0000000
+#define E1000_IMIR_EXT_SIZE_BP         0x00001000
+#define E1000_IMIR_EXT_CTRL_UGR        0x00002000
+#define E1000_IMIR_EXT_CTRL_ACK        0x00004000
+#define E1000_IMIR_EXT_CTRL_PSH        0x00008000
+#define E1000_IMIR_EXT_CTRL_RST        0x00010000
+#define E1000_IMIR_EXT_CTRL_SYN        0x00020000
+#define E1000_IMIR_EXT_CTRL_FIN        0x00040000
+#define E1000_IMIR_EXT_CTRL_BP         0x00080000
+#define E1000_MAX_TTQF_FILTERS         8
+#define E1000_2TUPLE_MAX_PRI           7
+
+#define E1000_MAX_FLEXIBLE_FILTERS       8
+#define E1000_MAX_FHFT                   4
+#define E1000_MAX_FHFT_EXT               4
+#define E1000_MAX_FLEX_FILTER_PRI        7
+#define E1000_MAX_FLEX_FILTER_LEN        128
+#define E1000_FHFT_QUEUEING_LEN          0x0000007F
+#define E1000_FHFT_QUEUEING_QUEUE        0x00000700
+#define E1000_FHFT_QUEUEING_PRIO         0x00070000
+#define E1000_FHFT_QUEUEING_OFFSET       0xFC
+#define E1000_FHFT_QUEUEING_QUEUE_SHIFT  8
+#define E1000_FHFT_QUEUEING_PRIO_SHIFT   16
+#define E1000_WUFC_FLEX_HQ               0x00004000
+
+#define E1000_SPQF_SRCPORT               0x0000FFFF
+
+#define E1000_MAX_FTQF_FILTERS           8
+#define E1000_FTQF_PROTOCOL_MASK         0x000000FF
+#define E1000_FTQF_5TUPLE_MASK_SHIFT     28
+#define E1000_FTQF_PROTOCOL_COMP_MASK    0x10000000
+#define E1000_FTQF_SOURCE_ADDR_MASK      0x20000000
+#define E1000_FTQF_DEST_ADDR_MASK        0x40000000
+#define E1000_FTQF_SOURCE_PORT_MASK      0x80000000
+#define E1000_FTQF_VF_MASK_EN            0x00008000
+#define E1000_FTQF_QUEUE_MASK            0x03ff0000
+#define E1000_FTQF_QUEUE_SHIFT           16
+#define E1000_FTQF_QUEUE_ENABLE          0x00000100
+
 /* structure for interrupt relative data */
 struct e1000_interrupt {
        uint32_t flags;
index 58ba5d3..e651e3d 100644 (file)
@@ -124,6 +124,44 @@ static int eth_igb_rss_reta_update(struct rte_eth_dev *dev,
 static int eth_igb_rss_reta_query(struct rte_eth_dev *dev,
                struct rte_eth_rss_reta *reta_conf);
 
+static int eth_igb_add_syn_filter(struct rte_eth_dev *dev,
+                       struct rte_syn_filter *filter, uint16_t rx_queue);
+static int eth_igb_remove_syn_filter(struct rte_eth_dev *dev);
+static int eth_igb_get_syn_filter(struct rte_eth_dev *dev,
+                       struct rte_syn_filter *filter, uint16_t *rx_queue);
+static int eth_igb_add_ethertype_filter(struct rte_eth_dev *dev,
+                       uint16_t index,
+                       struct rte_ethertype_filter *filter, uint16_t rx_queue);
+static int eth_igb_remove_ethertype_filter(struct rte_eth_dev *dev,
+                       uint16_t index);
+static int eth_igb_get_ethertype_filter(struct rte_eth_dev *dev,
+                       uint16_t index,
+                       struct rte_ethertype_filter *filter, uint16_t *rx_queue);
+static int eth_igb_add_2tuple_filter(struct rte_eth_dev *dev,
+                       uint16_t index,
+                       struct rte_2tuple_filter *filter, uint16_t rx_queue);
+static int eth_igb_remove_2tuple_filter(struct rte_eth_dev *dev,
+                       uint16_t index);
+static int eth_igb_get_2tuple_filter(struct rte_eth_dev *dev,
+                       uint16_t index,
+                       struct rte_2tuple_filter *filter, uint16_t *rx_queue);
+static int eth_igb_add_flex_filter(struct rte_eth_dev *dev,
+                       uint16_t index,
+                       struct rte_flex_filter *filter, uint16_t rx_queue);
+static int eth_igb_remove_flex_filter(struct rte_eth_dev *dev,
+                       uint16_t index);
+static int eth_igb_get_flex_filter(struct rte_eth_dev *dev,
+                       uint16_t index,
+                       struct rte_flex_filter *filter, uint16_t *rx_queue);
+static int eth_igb_add_5tuple_filter(struct rte_eth_dev *dev,
+                       uint16_t index,
+                       struct rte_5tuple_filter *filter, uint16_t rx_queue);
+static int eth_igb_remove_5tuple_filter(struct rte_eth_dev *dev,
+                       uint16_t index);
+static int eth_igb_get_5tuple_filter(struct rte_eth_dev *dev,
+                       uint16_t index,
+                       struct rte_5tuple_filter *filter, uint16_t *rx_queue);
+
 /*
  * Define VF Stats MACRO for Non "cleared on read" register
  */
@@ -196,6 +234,21 @@ static struct eth_dev_ops eth_igb_ops = {
        .reta_query           = eth_igb_rss_reta_query,
        .rss_hash_update      = eth_igb_rss_hash_update,
        .rss_hash_conf_get    = eth_igb_rss_hash_conf_get,
+       .add_syn_filter          = eth_igb_add_syn_filter,
+       .remove_syn_filter       = eth_igb_remove_syn_filter,
+       .get_syn_filter          = eth_igb_get_syn_filter,
+       .add_ethertype_filter    = eth_igb_add_ethertype_filter,
+       .remove_ethertype_filter = eth_igb_remove_ethertype_filter,
+       .get_ethertype_filter    = eth_igb_get_ethertype_filter,
+       .add_2tuple_filter       = eth_igb_add_2tuple_filter,
+       .remove_2tuple_filter    = eth_igb_remove_2tuple_filter,
+       .get_2tuple_filter       = eth_igb_get_2tuple_filter,
+       .add_flex_filter         = eth_igb_add_flex_filter,
+       .remove_flex_filter      = eth_igb_remove_flex_filter,
+       .get_flex_filter         = eth_igb_get_flex_filter,
+       .add_5tuple_filter       = eth_igb_add_5tuple_filter,
+       .remove_5tuple_filter    = eth_igb_remove_5tuple_filter,
+       .get_5tuple_filter       = eth_igb_get_5tuple_filter,
 };
 
 /*
@@ -2186,6 +2239,720 @@ eth_igb_rss_reta_query(struct rte_eth_dev *dev,
        return 0;
 }
 
+#define MAC_TYPE_FILTER_SUP(type)    do {\
+       if ((type) != e1000_82580 && (type) != e1000_i350 &&\
+               (type) != e1000_82576)\
+               return -ENOSYS;\
+} while (0)
+
+/*
+ * add the syn filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * filter: ponter to the filter that will be added.
+ * rx_queue: the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_add_syn_filter(struct rte_eth_dev *dev,
+                       struct rte_syn_filter *filter, uint16_t rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t synqf, rfctl;
+
+       MAC_TYPE_FILTER_SUP(hw->mac.type);
+
+       if (rx_queue >= IGB_MAX_RX_QUEUE_NUM)
+               return -EINVAL;
+
+       synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
+       if (synqf & E1000_SYN_FILTER_ENABLE)
+               return -EINVAL;
+
+       synqf = (uint32_t)(((rx_queue << E1000_SYN_FILTER_QUEUE_SHIFT) &
+               E1000_SYN_FILTER_QUEUE) | E1000_SYN_FILTER_ENABLE);
+
+       rfctl = E1000_READ_REG(hw, E1000_RFCTL);
+       if (filter->hig_pri)
+               rfctl |= E1000_RFCTL_SYNQFP;
+       else
+               rfctl &= ~E1000_RFCTL_SYNQFP;
+
+       E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
+       E1000_WRITE_REG(hw, E1000_RFCTL, rfctl);
+       return 0;
+}
+
+/*
+ * remove the syn filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_remove_syn_filter(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       MAC_TYPE_FILTER_SUP(hw->mac.type);
+
+       E1000_WRITE_REG(hw, E1000_SYNQF(0), 0);
+       return 0;
+}
+
+/*
+ * get the syn filter's info
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * filter: ponter to the filter that returns.
+ * *rx_queue: pointer to the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_get_syn_filter(struct rte_eth_dev *dev,
+                       struct rte_syn_filter *filter, uint16_t *rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t synqf, rfctl;
+
+       MAC_TYPE_FILTER_SUP(hw->mac.type);
+       synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
+       if (synqf & E1000_SYN_FILTER_ENABLE) {
+               rfctl = E1000_READ_REG(hw, E1000_RFCTL);
+               filter->hig_pri = (rfctl & E1000_RFCTL_SYNQFP) ? 1 : 0;
+               *rx_queue = (uint8_t)((synqf & E1000_SYN_FILTER_QUEUE) >>
+                               E1000_SYN_FILTER_QUEUE_SHIFT);
+               return 0;
+       }
+       return -ENOENT;
+}
+
+/*
+ * add an ethertype filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ * filter: ponter to the filter that will be added.
+ * rx_queue: the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_add_ethertype_filter(struct rte_eth_dev *dev, uint16_t index,
+                       struct rte_ethertype_filter *filter, uint16_t rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t etqf;
+
+       MAC_TYPE_FILTER_SUP(hw->mac.type);
+
+       if (index >= E1000_MAX_ETQF_FILTERS || rx_queue >= IGB_MAX_RX_QUEUE_NUM)
+               return -EINVAL;
+
+       etqf = E1000_READ_REG(hw, E1000_ETQF(index));
+       if (etqf & E1000_ETQF_FILTER_ENABLE)
+               return -EINVAL;  /* filter index is in use. */
+       else
+               etqf = 0;
+
+       etqf |= E1000_ETQF_FILTER_ENABLE | E1000_ETQF_QUEUE_ENABLE;
+       etqf |= (uint32_t)(filter->ethertype & E1000_ETQF_ETHERTYPE);
+       etqf |= rx_queue << E1000_ETQF_QUEUE_SHIFT;
+
+       if (filter->priority_en) {
+               PMD_INIT_LOG(ERR, "vlan and priority (%d) is not supported"
+                       " in E1000.", filter->priority);
+               return -EINVAL;
+       }
+
+       E1000_WRITE_REG(hw, E1000_ETQF(index), etqf);
+       return 0;
+}
+
+/*
+ * remove an ethertype filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_remove_ethertype_filter(struct rte_eth_dev *dev, uint16_t index)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       MAC_TYPE_FILTER_SUP(hw->mac.type);
+
+       if (index >= E1000_MAX_ETQF_FILTERS)
+               return -EINVAL;
+
+       E1000_WRITE_REG(hw, E1000_ETQF(index), 0);
+       return 0;
+}
+
+/*
+ * get an ethertype filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ * filter: ponter to the filter that will be gotten.
+ * *rx_queue: the ponited of the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_get_ethertype_filter(struct rte_eth_dev *dev, uint16_t index,
+                       struct rte_ethertype_filter *filter, uint16_t *rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t etqf;
+
+       MAC_TYPE_FILTER_SUP(hw->mac.type);
+
+       if (index >= E1000_MAX_ETQF_FILTERS)
+               return -EINVAL;
+
+       etqf = E1000_READ_REG(hw, E1000_ETQF(index));
+       if (etqf & E1000_ETQF_FILTER_ENABLE) {
+               filter->ethertype = etqf & E1000_ETQF_ETHERTYPE;
+               filter->priority_en = 0;
+               *rx_queue = (etqf & E1000_ETQF_QUEUE) >> E1000_ETQF_QUEUE_SHIFT;
+               return 0;
+       }
+       return -ENOENT;
+}
+
+#define MAC_TYPE_FILTER_SUP_EXT(type)    do {\
+       if ((type) != e1000_82580 && (type) != e1000_i350)\
+               return -ENOSYS; \
+} while (0)
+
+/*
+ * add a 2tuple filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ * filter: ponter to the filter that will be added.
+ * rx_queue: the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_add_2tuple_filter(struct rte_eth_dev *dev, uint16_t index,
+                       struct rte_2tuple_filter *filter, uint16_t rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t ttqf, imir = 0;
+       uint32_t imir_ext = 0;
+
+       MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
+
+       if (index >= E1000_MAX_TTQF_FILTERS ||
+               rx_queue >= IGB_MAX_RX_QUEUE_NUM ||
+               filter->priority > E1000_2TUPLE_MAX_PRI)
+               return -EINVAL;  /* filter index is out of range. */
+       if  (filter->tcp_flags > TCP_FLAG_ALL)
+               return -EINVAL;  /* flags is invalid. */
+
+       ttqf = E1000_READ_REG(hw, E1000_TTQF(index));
+       if (ttqf & E1000_TTQF_QUEUE_ENABLE)
+               return -EINVAL;  /* filter index is in use. */
+
+       imir = (uint32_t)(filter->dst_port & E1000_IMIR_DSTPORT);
+       if (filter->dst_port_mask == 1) /* 1b means not compare. */
+               imir |= E1000_IMIR_PORT_BP;
+       else
+               imir &= ~E1000_IMIR_PORT_BP;
+
+       imir |= filter->priority << E1000_IMIR_PRIORITY_SHIFT;
+
+       ttqf = 0;
+       ttqf |= E1000_TTQF_QUEUE_ENABLE;
+       ttqf |= (uint32_t)(rx_queue << E1000_TTQF_QUEUE_SHIFT);
+       ttqf |= (uint32_t)(filter->protocol & E1000_TTQF_PROTOCOL_MASK);
+       if (filter->protocol_mask == 1)
+               ttqf |= E1000_TTQF_MASK_ENABLE;
+       else
+               ttqf &= ~E1000_TTQF_MASK_ENABLE;
+
+       imir_ext |= E1000_IMIR_EXT_SIZE_BP;
+       /* tcp flags bits setting. */
+       if (filter->tcp_flags & TCP_FLAG_ALL) {
+               if (filter->tcp_flags & TCP_UGR_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_UGR;
+               if (filter->tcp_flags & TCP_ACK_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_ACK;
+               if (filter->tcp_flags & TCP_PSH_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_PSH;
+               if (filter->tcp_flags & TCP_RST_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_RST;
+               if (filter->tcp_flags & TCP_SYN_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_SYN;
+               if (filter->tcp_flags & TCP_FIN_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_FIN;
+               imir_ext &= ~E1000_IMIR_EXT_CTRL_BP;
+       } else
+               imir_ext |= E1000_IMIR_EXT_CTRL_BP;
+       E1000_WRITE_REG(hw, E1000_IMIR(index), imir);
+       E1000_WRITE_REG(hw, E1000_TTQF(index), ttqf);
+       E1000_WRITE_REG(hw, E1000_IMIREXT(index), imir_ext);
+       return 0;
+}
+
+/*
+ * remove a 2tuple filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_remove_2tuple_filter(struct rte_eth_dev *dev,
+                       uint16_t index)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
+
+       if (index >= E1000_MAX_TTQF_FILTERS)
+               return -EINVAL;  /* filter index is out of range */
+
+       E1000_WRITE_REG(hw, E1000_TTQF(index), 0);
+       E1000_WRITE_REG(hw, E1000_IMIR(index), 0);
+       E1000_WRITE_REG(hw, E1000_IMIREXT(index), 0);
+       return 0;
+}
+
+/*
+ * get a 2tuple filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ * filter: ponter to the filter that returns.
+ * *rx_queue: pointer of the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_get_2tuple_filter(struct rte_eth_dev *dev, uint16_t index,
+                       struct rte_2tuple_filter *filter, uint16_t *rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t imir, ttqf, imir_ext;
+
+       MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
+
+       if (index >= E1000_MAX_TTQF_FILTERS)
+               return -EINVAL;  /* filter index is out of range. */
+
+       ttqf = E1000_READ_REG(hw, E1000_TTQF(index));
+       if (ttqf & E1000_TTQF_QUEUE_ENABLE) {
+               imir = E1000_READ_REG(hw, E1000_IMIR(index));
+               filter->protocol = ttqf & E1000_TTQF_PROTOCOL_MASK;
+               filter->protocol_mask = (ttqf & E1000_TTQF_MASK_ENABLE) ? 1 : 0;
+               *rx_queue = (ttqf & E1000_TTQF_RX_QUEUE_MASK) >>
+                               E1000_TTQF_QUEUE_SHIFT;
+               filter->dst_port = (uint16_t)(imir & E1000_IMIR_DSTPORT);
+               filter->dst_port_mask = (imir & E1000_IMIR_PORT_BP) ? 1 : 0;
+               filter->priority = (imir & E1000_IMIR_PRIORITY) >>
+                       E1000_IMIR_PRIORITY_SHIFT;
+
+               imir_ext = E1000_READ_REG(hw, E1000_IMIREXT(index));
+               if (!(imir_ext & E1000_IMIR_EXT_CTRL_BP)) {
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_UGR)
+                               filter->tcp_flags |= TCP_UGR_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_ACK)
+                               filter->tcp_flags |= TCP_ACK_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_PSH)
+                               filter->tcp_flags |= TCP_PSH_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_RST)
+                               filter->tcp_flags |= TCP_RST_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_SYN)
+                               filter->tcp_flags |= TCP_SYN_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_FIN)
+                               filter->tcp_flags |= TCP_FIN_FLAG;
+               } else
+                       filter->tcp_flags = 0;
+               return 0;
+       }
+       return -ENOENT;
+}
+
+/*
+ * add a flex filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ * filter: ponter to the filter that will be added.
+ * rx_queue: the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *     - On failure, a negative value.
+ */
+static int
+eth_igb_add_flex_filter(struct rte_eth_dev *dev, uint16_t index,
+                       struct rte_flex_filter *filter, uint16_t rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t wufc, en_bits = 0;
+       uint32_t queueing = 0;
+       uint32_t reg_off = 0;
+       uint8_t i, j = 0;
+
+       MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
+
+       if (index >= E1000_MAX_FLEXIBLE_FILTERS)
+               return -EINVAL;  /* filter index is out of range. */
+
+       if (filter->len == 0 || filter->len > E1000_MAX_FLEX_FILTER_LEN ||
+               filter->len % 8 != 0 ||
+               filter->priority > E1000_MAX_FLEX_FILTER_PRI)
+               return -EINVAL;
+
+       wufc = E1000_READ_REG(hw, E1000_WUFC);
+       en_bits = E1000_WUFC_FLEX_HQ | (E1000_WUFC_FLX0 << index);
+       if ((wufc & en_bits) == en_bits)
+               return -EINVAL;  /* the filter is in use. */
+
+       E1000_WRITE_REG(hw, E1000_WUFC,
+               wufc | E1000_WUFC_FLEX_HQ | (E1000_WUFC_FLX0 << index));
+
+       j = 0;
+       if (index < E1000_MAX_FHFT)
+               reg_off = E1000_FHFT(index);
+       else
+               reg_off = E1000_FHFT_EXT(index - E1000_MAX_FHFT);
+
+       for (i = 0; i < 16; i++) {
+               E1000_WRITE_REG(hw, reg_off + i*4*4, filter->dwords[j]);
+               E1000_WRITE_REG(hw, reg_off + (i*4+1)*4, filter->dwords[++j]);
+               E1000_WRITE_REG(hw, reg_off + (i*4+2)*4,
+                               (uint32_t)filter->mask[i]);
+               ++j;
+       }
+       queueing |= filter->len |
+               (rx_queue << E1000_FHFT_QUEUEING_QUEUE_SHIFT) |
+               (filter->priority << E1000_FHFT_QUEUEING_PRIO_SHIFT);
+       E1000_WRITE_REG(hw, reg_off + E1000_FHFT_QUEUEING_OFFSET, queueing);
+       return 0;
+}
+
+/*
+ * remove a flex filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_remove_flex_filter(struct rte_eth_dev *dev,
+                               uint16_t index)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t wufc, reg_off = 0;
+       uint8_t i;
+
+       MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
+
+       if (index >= E1000_MAX_FLEXIBLE_FILTERS)
+               return -EINVAL;  /* filter index is out of range. */
+
+       wufc = E1000_READ_REG(hw, E1000_WUFC);
+       E1000_WRITE_REG(hw, E1000_WUFC, wufc & (~(E1000_WUFC_FLX0 << index)));
+
+       if (index < E1000_MAX_FHFT)
+               reg_off = E1000_FHFT(index);
+       else
+               reg_off = E1000_FHFT_EXT(index - E1000_MAX_FHFT);
+
+       for (i = 0; i < 64; i++)
+               E1000_WRITE_REG(hw, reg_off + i*4, 0);
+       return 0;
+}
+
+/*
+ * get a flex filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ * filter: ponter to the filter that returns.
+ * *rx_queue: the pointer of the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_get_flex_filter(struct rte_eth_dev *dev, uint16_t index,
+                       struct rte_flex_filter *filter, uint16_t *rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t wufc, queueing, wufc_en = 0;
+       uint8_t i, j;
+
+       MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
+
+       if (index >= E1000_MAX_FLEXIBLE_FILTERS)
+               return -EINVAL;  /* filter index is out of range. */
+
+       wufc = E1000_READ_REG(hw, E1000_WUFC);
+       wufc_en = E1000_WUFC_FLEX_HQ | (E1000_WUFC_FLX0 << index);
+
+       if ((wufc & wufc_en) == wufc_en) {
+               uint32_t reg_off = 0;
+               j = 0;
+               if (index < E1000_MAX_FHFT)
+                       reg_off = E1000_FHFT(index);
+               else
+                       reg_off = E1000_FHFT_EXT(index - E1000_MAX_FHFT);
+
+               for (i = 0; i < 16; i++, j = i * 2) {
+                       filter->dwords[j] =
+                               E1000_READ_REG(hw, reg_off + i*4*4);
+                       filter->dwords[j+1] =
+                               E1000_READ_REG(hw, reg_off + (i*4+1)*4);
+                       filter->mask[i] =
+                               E1000_READ_REG(hw, reg_off + (i*4+2)*4);
+               }
+               queueing = E1000_READ_REG(hw,
+                               reg_off + E1000_FHFT_QUEUEING_OFFSET);
+               filter->len = queueing & E1000_FHFT_QUEUEING_LEN;
+               filter->priority = (queueing & E1000_FHFT_QUEUEING_PRIO) >>
+                       E1000_FHFT_QUEUEING_PRIO_SHIFT;
+               *rx_queue = (queueing & E1000_FHFT_QUEUEING_QUEUE) >>
+                       E1000_FHFT_QUEUEING_QUEUE_SHIFT;
+               return 0;
+       }
+       return -ENOENT;
+}
+
+/*
+ * add a 5tuple filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates.
+ * filter: ponter to the filter that will be added.
+ * rx_queue: the queue id the filter assigned to.
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_add_5tuple_filter(struct rte_eth_dev *dev, uint16_t index,
+                       struct rte_5tuple_filter *filter, uint16_t rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t ftqf, spqf = 0;
+       uint32_t imir = 0;
+       uint32_t imir_ext = 0;
+
+       if (hw->mac.type != e1000_82576)
+               return -ENOSYS;
+
+       if (index >= E1000_MAX_FTQF_FILTERS ||
+               rx_queue >= IGB_MAX_RX_QUEUE_NUM_82576)
+               return -EINVAL;  /* filter index is out of range. */
+
+       ftqf = E1000_READ_REG(hw, E1000_FTQF(index));
+       if (ftqf & E1000_FTQF_QUEUE_ENABLE)
+               return -EINVAL;  /* filter index is in use. */
+
+       ftqf = 0;
+       ftqf |= filter->protocol & E1000_FTQF_PROTOCOL_MASK;
+       if (filter->src_ip_mask == 1) /* 1b means not compare. */
+               ftqf |= E1000_FTQF_SOURCE_ADDR_MASK;
+       if (filter->dst_ip_mask == 1)
+               ftqf |= E1000_FTQF_DEST_ADDR_MASK;
+       if (filter->src_port_mask == 1)
+               ftqf |= E1000_FTQF_SOURCE_PORT_MASK;
+       if (filter->protocol_mask == 1)
+               ftqf |= E1000_FTQF_PROTOCOL_COMP_MASK;
+       ftqf |= (rx_queue << E1000_FTQF_QUEUE_SHIFT) & E1000_FTQF_QUEUE_MASK;
+       ftqf |= E1000_FTQF_VF_MASK_EN;
+       ftqf |= E1000_FTQF_QUEUE_ENABLE;
+       E1000_WRITE_REG(hw, E1000_FTQF(index), ftqf);
+       E1000_WRITE_REG(hw, E1000_DAQF(index), filter->dst_ip);
+       E1000_WRITE_REG(hw, E1000_SAQF(index), filter->src_ip);
+
+       spqf |= filter->src_port & E1000_SPQF_SRCPORT;
+       E1000_WRITE_REG(hw, E1000_SPQF(index), spqf);
+
+       imir |= (uint32_t)(filter->dst_port & E1000_IMIR_DSTPORT);
+       if (filter->dst_port_mask == 1) /* 1b means not compare. */
+               imir |= E1000_IMIR_PORT_BP;
+       else
+               imir &= ~E1000_IMIR_PORT_BP;
+       imir |= filter->priority << E1000_IMIR_PRIORITY_SHIFT;
+
+       imir_ext |= E1000_IMIR_EXT_SIZE_BP;
+       /* tcp flags bits setting. */
+       if (filter->tcp_flags & TCP_FLAG_ALL) {
+               if (filter->tcp_flags & TCP_UGR_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_UGR;
+               if (filter->tcp_flags & TCP_ACK_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_ACK;
+               if (filter->tcp_flags & TCP_PSH_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_PSH;
+               if (filter->tcp_flags & TCP_RST_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_RST;
+               if (filter->tcp_flags & TCP_SYN_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_SYN;
+               if (filter->tcp_flags & TCP_FIN_FLAG)
+                       imir_ext |= E1000_IMIR_EXT_CTRL_FIN;
+       } else
+               imir_ext |= E1000_IMIR_EXT_CTRL_BP;
+       E1000_WRITE_REG(hw, E1000_IMIR(index), imir);
+       E1000_WRITE_REG(hw, E1000_IMIREXT(index), imir_ext);
+       return 0;
+}
+
+/*
+ * remove a 5tuple filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_remove_5tuple_filter(struct rte_eth_dev *dev,
+                               uint16_t index)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       if (hw->mac.type != e1000_82576)
+               return -ENOSYS;
+
+       if (index >= E1000_MAX_FTQF_FILTERS)
+               return -EINVAL;  /* filter index is out of range. */
+
+       E1000_WRITE_REG(hw, E1000_FTQF(index), 0);
+       E1000_WRITE_REG(hw, E1000_DAQF(index), 0);
+       E1000_WRITE_REG(hw, E1000_SAQF(index), 0);
+       E1000_WRITE_REG(hw, E1000_SPQF(index), 0);
+       E1000_WRITE_REG(hw, E1000_IMIR(index), 0);
+       E1000_WRITE_REG(hw, E1000_IMIREXT(index), 0);
+       return 0;
+}
+
+/*
+ * get a 5tuple filter
+ *
+ * @param
+ * dev: Pointer to struct rte_eth_dev.
+ * index: the index the filter allocates
+ * filter: ponter to the filter that returns
+ * *rx_queue: pointer of the queue id the filter assigned to
+ *
+ * @return
+ *    - On success, zero.
+ *    - On failure, a negative value.
+ */
+static int
+eth_igb_get_5tuple_filter(struct rte_eth_dev *dev, uint16_t index,
+                       struct rte_5tuple_filter *filter, uint16_t *rx_queue)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t spqf, ftqf, imir, imir_ext;
+
+       if (hw->mac.type != e1000_82576)
+               return -ENOSYS;
+
+       if (index >= E1000_MAX_FTQF_FILTERS)
+               return -EINVAL;  /* filter index is out of range. */
+
+       ftqf = E1000_READ_REG(hw, E1000_FTQF(index));
+       if (ftqf & E1000_FTQF_QUEUE_ENABLE) {
+               filter->src_ip_mask =
+                       (ftqf & E1000_FTQF_SOURCE_ADDR_MASK) ? 1 : 0;
+               filter->dst_ip_mask =
+                       (ftqf & E1000_FTQF_DEST_ADDR_MASK) ? 1 : 0;
+               filter->src_port_mask =
+                       (ftqf & E1000_FTQF_SOURCE_PORT_MASK) ? 1 : 0;
+               filter->protocol_mask =
+                       (ftqf & E1000_FTQF_PROTOCOL_COMP_MASK) ? 1 : 0;
+               filter->protocol =
+                       (uint8_t)ftqf & E1000_FTQF_PROTOCOL_MASK;
+               *rx_queue = (uint16_t)((ftqf & E1000_FTQF_QUEUE_MASK) >>
+                               E1000_FTQF_QUEUE_SHIFT);
+
+               spqf = E1000_READ_REG(hw, E1000_SPQF(index));
+               filter->src_port = spqf & E1000_SPQF_SRCPORT;
+
+               filter->dst_ip = E1000_READ_REG(hw, E1000_DAQF(index));
+               filter->src_ip = E1000_READ_REG(hw, E1000_SAQF(index));
+
+               imir = E1000_READ_REG(hw, E1000_IMIR(index));
+               filter->dst_port_mask = (imir & E1000_IMIR_PORT_BP) ? 1 : 0;
+               filter->dst_port = (uint16_t)(imir & E1000_IMIR_DSTPORT);
+               filter->priority = (imir & E1000_IMIR_PRIORITY) >>
+                       E1000_IMIR_PRIORITY_SHIFT;
+
+               imir_ext = E1000_READ_REG(hw, E1000_IMIREXT(index));
+               if (!(imir_ext & E1000_IMIR_EXT_CTRL_BP)) {
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_UGR)
+                               filter->tcp_flags |= TCP_UGR_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_ACK)
+                               filter->tcp_flags |= TCP_ACK_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_PSH)
+                               filter->tcp_flags |= TCP_PSH_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_RST)
+                               filter->tcp_flags |= TCP_RST_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_SYN)
+                               filter->tcp_flags |= TCP_SYN_FLAG;
+                       if (imir_ext & E1000_IMIR_EXT_CTRL_FIN)
+                               filter->tcp_flags |= TCP_FIN_FLAG;
+               } else
+                       filter->tcp_flags = 0;
+               return 0;
+       }
+       return -ENOENT;
+}
+
 static struct rte_driver pmd_igb_drv = {
        .type = PMD_PDEV,
        .init = rte_igb_pmd_init,