net/enic: fix Tx IP and UDP/TCP checksum offload
authorJohn Daley <johndale@cisco.com>
Fri, 3 Jun 2016 00:22:57 +0000 (17:22 -0700)
committerBruce Richardson <bruce.richardson@intel.com>
Wed, 15 Jun 2016 15:13:56 +0000 (17:13 +0200)
Private/conflicting ol_flags where used to enable UDP/TCP Tx
offloads. Use the common flags in PKT_TX_L4_MASK to support them.
When updating flags, also do some minor code rearranging for
slightly better performane.

Fixes: fefed3d1e62c ("enic: new driver")
Signed-off-by: John Daley <johndale@cisco.com>
drivers/net/enic/enic.h
drivers/net/enic/enic_rxtx.c

index 4b76e6d..1e6914e 100644 (file)
@@ -62,7 +62,6 @@
 #define ENICPMD_SETTING(enic, f) ((enic->config.flags & VENETF_##f) ? 1 : 0)
 
 #define ENICPMD_BDF_LENGTH      13   /* 0000:00:00.0'\0' */
-#define PKT_TX_TCP_UDP_CKSUM    0x6000
 #define ENIC_CALC_IP_CKSUM      1
 #define ENIC_CALC_TCP_UDP_CKSUM 2
 #define ENIC_MAX_MTU            9000
index 08ab630..eb8fbdb 100644 (file)
@@ -385,10 +385,10 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        struct enic *enic = vnic_dev_priv(wq->vdev);
        unsigned short vlan_id;
        uint64_t ol_flags;
+       uint64_t ol_flags_mask;
        unsigned int wq_desc_avail;
        int head_idx;
        struct vnic_wq_buf *buf;
-       unsigned int hw_ip_cksum_enabled;
        unsigned int desc_count;
        struct wq_enet_desc *descs, *desc_p, desc_tmp;
        uint16_t mss;
@@ -400,10 +400,10 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        wq_desc_avail = vnic_wq_desc_avail(wq);
        head_idx = wq->head_idx;
        desc_count = wq->ring.desc_count;
+       ol_flags_mask = PKT_TX_VLAN_PKT | PKT_TX_IP_CKSUM | PKT_TX_L4_MASK;
 
        nb_pkts = RTE_MIN(nb_pkts, ENIC_TX_XMIT_MAX);
 
-       hw_ip_cksum_enabled = enic->hw_ip_checksum;
        for (index = 0; index < nb_pkts; index++) {
                tx_pkt = *tx_pkts++;
                nb_segs = tx_pkt->nb_segs;
@@ -415,10 +415,9 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 
                pkt_len = tx_pkt->pkt_len;
                data_len = tx_pkt->data_len;
-               vlan_id = tx_pkt->vlan_tci;
                ol_flags = tx_pkt->ol_flags;
-
                mss = 0;
+               vlan_id = 0;
                vlan_tag_insert = 0;
                bus_addr = (dma_addr_t)
                           (tx_pkt->buf_physaddr + tx_pkt->data_off);
@@ -428,14 +427,23 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 
                eop = (data_len == pkt_len);
 
-               if (ol_flags & PKT_TX_VLAN_PKT)
-                       vlan_tag_insert = 1;
+               if (ol_flags & ol_flags_mask) {
+                       if (ol_flags & PKT_TX_VLAN_PKT) {
+                               vlan_tag_insert = 1;
+                               vlan_id = tx_pkt->vlan_tci;
+                       }
 
-               if (hw_ip_cksum_enabled && (ol_flags & PKT_TX_IP_CKSUM))
-                       mss |= ENIC_CALC_IP_CKSUM;
+                       if (ol_flags & PKT_TX_IP_CKSUM)
+                               mss |= ENIC_CALC_IP_CKSUM;
 
-               if (hw_ip_cksum_enabled && (ol_flags & PKT_TX_TCP_UDP_CKSUM))
-                       mss |= ENIC_CALC_TCP_UDP_CKSUM;
+                       /* Nic uses just 1 bit for UDP and TCP */
+                       switch (ol_flags & PKT_TX_L4_MASK) {
+                       case PKT_TX_TCP_CKSUM:
+                       case PKT_TX_UDP_CKSUM:
+                               mss |= ENIC_CALC_TCP_UDP_CKSUM;
+                               break;
+                       }
+               }
 
                wq_enet_desc_enc(&desc_tmp, bus_addr, data_len, mss, 0, 0, eop,
                                 eop, 0, vlan_tag_insert, vlan_id, 0);