diff options
author | Ganesh Goudar <ganeshgr@chelsio.com> | 2018-01-10 15:45:26 +0300 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2018-01-11 18:58:56 +0300 |
commit | d0a1299c6bf7d80c8bb8e181f36a7c407a4cabca (patch) | |
tree | a44a62ea249c18c50c13ed8b49566ba620e58708 | |
parent | 846eac3fccec85d464211ecb7aa53b3a33c69c4c (diff) | |
download | linux-d0a1299c6bf7d80c8bb8e181f36a7c407a4cabca.tar.xz |
cxgb4: add support for vxlan segmentation offload
add changes to t4_eth_xmit to enable vxlan segmentation
offload support.
Original work by: Santosh Rastapur <santosh@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
-rw-r--r-- | drivers/net/ethernet/chelsio/cxgb4/cxgb4.h | 1 | ||||
-rw-r--r-- | drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c | 4 | ||||
-rw-r--r-- | drivers/net/ethernet/chelsio/cxgb4/sge.c | 218 |
3 files changed, 186 insertions, 37 deletions
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h index f56e893feace..baa67d362051 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h @@ -1314,6 +1314,7 @@ void t4_sge_start(struct adapter *adap); void t4_sge_stop(struct adapter *adap); void cxgb4_set_ethtool_ops(struct net_device *netdev); int cxgb4_write_rss(const struct port_info *pi, const u16 *queues); +enum cpl_tx_tnl_lso_type cxgb_encap_offload_supported(struct sk_buff *skb); extern int dbfifo_int_thresh; #define for_each_port(adapter, iter) \ diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c index 8097762c2bdd..d0e08f6a441d 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c @@ -5210,6 +5210,10 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent) NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_TC; + + if (CHELSIO_CHIP_VERSION(chip) > CHELSIO_T5) + netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL; + if (highdma) netdev->hw_features |= NETIF_F_HIGHDMA; netdev->features |= netdev->hw_features; diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c index 922f2f937789..eab781fab2a8 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/sge.c +++ b/drivers/net/ethernet/chelsio/cxgb4/sge.c @@ -770,12 +770,19 @@ static inline unsigned int flits_to_desc(unsigned int n) * Returns whether an Ethernet packet is small enough to fit as * immediate data. Return value corresponds to headroom required. */ -static inline int is_eth_imm(const struct sk_buff *skb) +static inline int is_eth_imm(const struct sk_buff *skb, unsigned int chip_ver) { - int hdrlen = skb_shinfo(skb)->gso_size ? - sizeof(struct cpl_tx_pkt_lso_core) : 0; + int hdrlen = 0; - hdrlen += sizeof(struct cpl_tx_pkt); + if (skb->encapsulation && skb_shinfo(skb)->gso_size && + chip_ver > CHELSIO_T5) { + hdrlen = sizeof(struct cpl_tx_tnl_lso); + hdrlen += sizeof(struct cpl_tx_pkt_core); + } else { + hdrlen = skb_shinfo(skb)->gso_size ? + sizeof(struct cpl_tx_pkt_lso_core) : 0; + hdrlen += sizeof(struct cpl_tx_pkt); + } if (skb->len <= MAX_IMM_TX_PKT_LEN - hdrlen) return hdrlen; return 0; @@ -788,10 +795,11 @@ static inline int is_eth_imm(const struct sk_buff *skb) * Returns the number of flits needed for a Tx WR for the given Ethernet * packet, including the needed WR and CPL headers. */ -static inline unsigned int calc_tx_flits(const struct sk_buff *skb) +static inline unsigned int calc_tx_flits(const struct sk_buff *skb, + unsigned int chip_ver) { unsigned int flits; - int hdrlen = is_eth_imm(skb); + int hdrlen = is_eth_imm(skb, chip_ver); /* If the skb is small enough, we can pump it out as a work request * with only immediate data. In that case we just have to have the @@ -810,13 +818,20 @@ static inline unsigned int calc_tx_flits(const struct sk_buff *skb) * with an embedded TX Packet Write CPL message. */ flits = sgl_len(skb_shinfo(skb)->nr_frags + 1); - if (skb_shinfo(skb)->gso_size) - flits += (sizeof(struct fw_eth_tx_pkt_wr) + - sizeof(struct cpl_tx_pkt_lso_core) + - sizeof(struct cpl_tx_pkt_core)) / sizeof(__be64); - else + if (skb_shinfo(skb)->gso_size) { + if (skb->encapsulation && chip_ver > CHELSIO_T5) + hdrlen = sizeof(struct fw_eth_tx_pkt_wr) + + sizeof(struct cpl_tx_tnl_lso); + else + hdrlen = sizeof(struct fw_eth_tx_pkt_wr) + + sizeof(struct cpl_tx_pkt_lso_core); + + hdrlen += sizeof(struct cpl_tx_pkt_core); + flits += (hdrlen / sizeof(__be64)); + } else { flits += (sizeof(struct fw_eth_tx_pkt_wr) + sizeof(struct cpl_tx_pkt_core)) / sizeof(__be64); + } return flits; } @@ -827,9 +842,10 @@ static inline unsigned int calc_tx_flits(const struct sk_buff *skb) * Returns the number of Tx descriptors needed for the given Ethernet * packet, including the needed WR and CPL headers. */ -static inline unsigned int calc_tx_descs(const struct sk_buff *skb) +static inline unsigned int calc_tx_descs(const struct sk_buff *skb, + unsigned int chip_ver) { - return flits_to_desc(calc_tx_flits(skb)); + return flits_to_desc(calc_tx_flits(skb, chip_ver)); } /** @@ -1154,6 +1170,102 @@ cxgb_fcoe_offload(struct sk_buff *skb, struct adapter *adap, } #endif /* CONFIG_CHELSIO_T4_FCOE */ +/* Returns tunnel type if hardware supports offloading of the same. + * It is called only for T5 and onwards. + */ +enum cpl_tx_tnl_lso_type cxgb_encap_offload_supported(struct sk_buff *skb) +{ + u8 l4_hdr = 0; + enum cpl_tx_tnl_lso_type tnl_type = TX_TNL_TYPE_OPAQUE; + struct port_info *pi = netdev_priv(skb->dev); + struct adapter *adapter = pi->adapter; + + if (skb->inner_protocol_type != ENCAP_TYPE_ETHER || + skb->inner_protocol != htons(ETH_P_TEB)) + return tnl_type; + + switch (vlan_get_protocol(skb)) { + case htons(ETH_P_IP): + l4_hdr = ip_hdr(skb)->protocol; + break; + case htons(ETH_P_IPV6): + l4_hdr = ipv6_hdr(skb)->nexthdr; + break; + default: + return tnl_type; + } + + switch (l4_hdr) { + case IPPROTO_UDP: + if (adapter->vxlan_port == udp_hdr(skb)->dest) + tnl_type = TX_TNL_TYPE_VXLAN; + break; + default: + return tnl_type; + } + + return tnl_type; +} + +static inline void t6_fill_tnl_lso(struct sk_buff *skb, + struct cpl_tx_tnl_lso *tnl_lso, + enum cpl_tx_tnl_lso_type tnl_type) +{ + u32 val; + int in_eth_xtra_len; + int l3hdr_len = skb_network_header_len(skb); + int eth_xtra_len = skb_network_offset(skb) - ETH_HLEN; + const struct skb_shared_info *ssi = skb_shinfo(skb); + bool v6 = (ip_hdr(skb)->version == 6); + + val = CPL_TX_TNL_LSO_OPCODE_V(CPL_TX_TNL_LSO) | + CPL_TX_TNL_LSO_FIRST_F | + CPL_TX_TNL_LSO_LAST_F | + (v6 ? CPL_TX_TNL_LSO_IPV6OUT_F : 0) | + CPL_TX_TNL_LSO_ETHHDRLENOUT_V(eth_xtra_len / 4) | + CPL_TX_TNL_LSO_IPHDRLENOUT_V(l3hdr_len / 4) | + (v6 ? 0 : CPL_TX_TNL_LSO_IPHDRCHKOUT_F) | + CPL_TX_TNL_LSO_IPLENSETOUT_F | + (v6 ? 0 : CPL_TX_TNL_LSO_IPIDINCOUT_F); + tnl_lso->op_to_IpIdSplitOut = htonl(val); + + tnl_lso->IpIdOffsetOut = 0; + + /* Get the tunnel header length */ + val = skb_inner_mac_header(skb) - skb_mac_header(skb); + in_eth_xtra_len = skb_inner_network_header(skb) - + skb_inner_mac_header(skb) - ETH_HLEN; + + switch (tnl_type) { + case TX_TNL_TYPE_VXLAN: + tnl_lso->UdpLenSetOut_to_TnlHdrLen = + htons(CPL_TX_TNL_LSO_UDPCHKCLROUT_F | + CPL_TX_TNL_LSO_UDPLENSETOUT_F); + break; + default: + tnl_lso->UdpLenSetOut_to_TnlHdrLen = 0; + break; + } + + tnl_lso->UdpLenSetOut_to_TnlHdrLen |= + htons(CPL_TX_TNL_LSO_TNLHDRLEN_V(val) | + CPL_TX_TNL_LSO_TNLTYPE_V(tnl_type)); + + tnl_lso->r1 = 0; + + val = CPL_TX_TNL_LSO_ETHHDRLEN_V(in_eth_xtra_len / 4) | + CPL_TX_TNL_LSO_IPV6_V(inner_ip_hdr(skb)->version == 6) | + CPL_TX_TNL_LSO_IPHDRLEN_V(skb_inner_network_header_len(skb) / 4) | + CPL_TX_TNL_LSO_TCPHDRLEN_V(inner_tcp_hdrlen(skb) / 4); + tnl_lso->Flow_to_TcpHdrLen = htonl(val); + + tnl_lso->IpIdOffset = htons(0); + + tnl_lso->IpIdSplit_to_Mss = htons(CPL_TX_TNL_LSO_MSS_V(ssi->gso_size)); + tnl_lso->TCPSeqOffset = htonl(0); + tnl_lso->EthLenOffset_Size = htonl(CPL_TX_TNL_LSO_SIZE_V(skb->len)); +} + /** * t4_eth_xmit - add a packet to an Ethernet Tx queue * @skb: the packet @@ -1177,6 +1289,9 @@ netdev_tx_t t4_eth_xmit(struct sk_buff *skb, struct net_device *dev) bool immediate = false; int len, max_pkt_len; bool ptp_enabled = is_ptp_enabled(skb, dev); + unsigned int chip_ver; + enum cpl_tx_tnl_lso_type tnl_type = TX_TNL_TYPE_OPAQUE; + #ifdef CONFIG_CHELSIO_T4_FCOE int err; #endif /* CONFIG_CHELSIO_T4_FCOE */ @@ -1227,7 +1342,8 @@ out_free: dev_kfree_skb_any(skb); } #endif /* CONFIG_CHELSIO_T4_FCOE */ - flits = calc_tx_flits(skb); + chip_ver = CHELSIO_CHIP_VERSION(adap->params.chip); + flits = calc_tx_flits(skb, chip_ver); ndesc = flits_to_desc(flits); credits = txq_avail(&q->q) - ndesc; @@ -1241,9 +1357,12 @@ out_free: dev_kfree_skb_any(skb); return NETDEV_TX_BUSY; } - if (is_eth_imm(skb)) + if (is_eth_imm(skb, chip_ver)) immediate = true; + if (skb->encapsulation && chip_ver > CHELSIO_T5) + tnl_type = cxgb_encap_offload_supported(skb); + if (!immediate && unlikely(map_skb(adap->pdev_dev, skb, addr) < 0)) { q->mapping_err++; @@ -1270,33 +1389,58 @@ out_free: dev_kfree_skb_any(skb); bool v6 = (ssi->gso_type & SKB_GSO_TCPV6) != 0; int l3hdr_len = skb_network_header_len(skb); int eth_xtra_len = skb_network_offset(skb) - ETH_HLEN; + struct cpl_tx_tnl_lso *tnl_lso = (void *)(wr + 1); + + if (tnl_type) + len += sizeof(*tnl_lso); + else + len += sizeof(*lso); - len += sizeof(*lso); wr->op_immdlen = htonl(FW_WR_OP_V(FW_ETH_TX_PKT_WR) | FW_WR_IMMDLEN_V(len)); - lso->c.lso_ctrl = htonl(LSO_OPCODE_V(CPL_TX_PKT_LSO) | - LSO_FIRST_SLICE_F | LSO_LAST_SLICE_F | - LSO_IPV6_V(v6) | - LSO_ETHHDR_LEN_V(eth_xtra_len / 4) | - LSO_IPHDR_LEN_V(l3hdr_len / 4) | - LSO_TCPHDR_LEN_V(tcp_hdr(skb)->doff)); - lso->c.ipid_ofst = htons(0); - lso->c.mss = htons(ssi->gso_size); - lso->c.seqno_offset = htonl(0); - if (is_t4(adap->params.chip)) - lso->c.len = htonl(skb->len); - else - lso->c.len = htonl(LSO_T5_XFER_SIZE_V(skb->len)); - cpl = (void *)(lso + 1); + if (tnl_type) { + struct iphdr *iph = ip_hdr(skb); - if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5) - cntrl = TXPKT_ETHHDR_LEN_V(eth_xtra_len); - else - cntrl = T6_TXPKT_ETHHDR_LEN_V(eth_xtra_len); + t6_fill_tnl_lso(skb, tnl_lso, tnl_type); + cpl = (void *)(tnl_lso + 1); + /* Driver is expected to compute partial checksum that + * does not include the IP Total Length. + */ + if (iph->version == 4) { + iph->check = 0; + iph->tot_len = 0; + iph->check = (u16)(~ip_fast_csum((u8 *)iph, + iph->ihl)); + } + if (skb->ip_summed == CHECKSUM_PARTIAL) + cntrl = hwcsum(adap->params.chip, skb); + } else { + lso->c.lso_ctrl = htonl(LSO_OPCODE_V(CPL_TX_PKT_LSO) | + LSO_FIRST_SLICE_F | LSO_LAST_SLICE_F | + LSO_IPV6_V(v6) | + LSO_ETHHDR_LEN_V(eth_xtra_len / 4) | + LSO_IPHDR_LEN_V(l3hdr_len / 4) | + LSO_TCPHDR_LEN_V(tcp_hdr(skb)->doff)); + lso->c.ipid_ofst = htons(0); + lso->c.mss = htons(ssi->gso_size); + lso->c.seqno_offset = htonl(0); + if (is_t4(adap->params.chip)) + lso->c.len = htonl(skb->len); + else + lso->c.len = + htonl(LSO_T5_XFER_SIZE_V(skb->len)); + cpl = (void *)(lso + 1); - cntrl |= TXPKT_CSUM_TYPE_V(v6 ? - TX_CSUM_TCPIP6 : TX_CSUM_TCPIP) | - TXPKT_IPHDR_LEN_V(l3hdr_len); + if (CHELSIO_CHIP_VERSION(adap->params.chip) + <= CHELSIO_T5) + cntrl = TXPKT_ETHHDR_LEN_V(eth_xtra_len); + else + cntrl = T6_TXPKT_ETHHDR_LEN_V(eth_xtra_len); + + cntrl |= TXPKT_CSUM_TYPE_V(v6 ? + TX_CSUM_TCPIP6 : TX_CSUM_TCPIP) | + TXPKT_IPHDR_LEN_V(l3hdr_len); + } q->tso++; q->tx_cso += ssi->gso_segs; } else { |