net/cnxk: disable default inner checksum for outbound inline
[dpdk.git] / drivers / net / cnxk / cn10k_tx.h
index 4ae6bbf..feaa41f 100644 (file)
@@ -246,6 +246,7 @@ cn10k_nix_prep_sec_vec(struct rte_mbuf *m, uint64x2_t *cmd0, uint64x2_t *cmd1,
 {
        struct cn10k_sec_sess_priv sess_priv;
        uint32_t pkt_len, dlen_adj, rlen;
+       uint8_t l3l4type, chksum;
        uint64x2_t cmd01, cmd23;
        uintptr_t dptr, nixtx;
        uint64_t ucode_cmd[4];
@@ -256,10 +257,23 @@ cn10k_nix_prep_sec_vec(struct rte_mbuf *m, uint64x2_t *cmd0, uint64x2_t *cmd1,
 
        sess_priv.u64 = *rte_security_dynfield(m);
 
-       if (flags & NIX_TX_NEED_SEND_HDR_W1)
+       if (flags & NIX_TX_NEED_SEND_HDR_W1) {
                l2_len = vgetq_lane_u8(*cmd0, 8);
-       else
+               /* Extract l3l4type either from il3il4type or ol3ol4type */
+               if (flags & NIX_TX_OFFLOAD_L3_L4_CSUM_F &&
+                   flags & NIX_TX_OFFLOAD_OL3_OL4_CSUM_F)
+                       l3l4type = vgetq_lane_u8(*cmd0, 13);
+               else
+                       l3l4type = vgetq_lane_u8(*cmd0, 12);
+
+               chksum = (l3l4type & 0x1) << 1 | !!(l3l4type & 0x30);
+               chksum = ~chksum;
+               sess_priv.chksum = sess_priv.chksum & chksum;
+               /* Clear SEND header flags */
+               *cmd0 = vsetq_lane_u16(0, *cmd0, 6);
+       } else {
                l2_len = m->l2_len;
+       }
 
        /* Retrieve DPTR */
        dptr = vgetq_lane_u64(*cmd1, 1);
@@ -291,8 +305,8 @@ cn10k_nix_prep_sec_vec(struct rte_mbuf *m, uint64x2_t *cmd0, uint64x2_t *cmd1,
        sa_base &= ~0xFFFFUL;
        sa = (uintptr_t)roc_nix_inl_ot_ipsec_outb_sa(sa_base, sess_priv.sa_idx);
        ucode_cmd[3] = (ROC_CPT_DFLT_ENG_GRP_SE_IE << 61 | 1UL << 60 | sa);
-       ucode_cmd[0] =
-               (ROC_IE_OT_MAJOR_OP_PROCESS_OUTBOUND_IPSEC << 48 | pkt_len);
+       ucode_cmd[0] = (ROC_IE_OT_MAJOR_OP_PROCESS_OUTBOUND_IPSEC << 48 |
+                       ((uint64_t)sess_priv.chksum) << 32 | pkt_len);
 
        /* CPT Word 0 and Word 1 */
        cmd01 = vdupq_n_u64((nixtx + 16) | (cn10k_nix_tx_ext_subs(flags) + 1));
@@ -343,6 +357,7 @@ cn10k_nix_prep_sec(struct rte_mbuf *m, uint64_t *cmd, uintptr_t *nixtx_addr,
        struct cn10k_sec_sess_priv sess_priv;
        uint32_t pkt_len, dlen_adj, rlen;
        struct nix_send_hdr_s *send_hdr;
+       uint8_t l3l4type, chksum;
        uint64x2_t cmd01, cmd23;
        union nix_send_sg_s *sg;
        uintptr_t dptr, nixtx;
@@ -360,10 +375,23 @@ cn10k_nix_prep_sec(struct rte_mbuf *m, uint64_t *cmd, uintptr_t *nixtx_addr,
        else
                sg = (union nix_send_sg_s *)&cmd[2];
 
-       if (flags & NIX_TX_NEED_SEND_HDR_W1)
+       if (flags & NIX_TX_NEED_SEND_HDR_W1) {
                l2_len = cmd[1] & 0xFF;
-       else
+               /* Extract l3l4type either from il3il4type or ol3ol4type */
+               if (flags & NIX_TX_OFFLOAD_L3_L4_CSUM_F &&
+                   flags & NIX_TX_OFFLOAD_OL3_OL4_CSUM_F)
+                       l3l4type = (cmd[1] >> 40) & 0xFF;
+               else
+                       l3l4type = (cmd[1] >> 32) & 0xFF;
+
+               chksum = (l3l4type & 0x1) << 1 | !!(l3l4type & 0x30);
+               chksum = ~chksum;
+               sess_priv.chksum = sess_priv.chksum & chksum;
+               /* Clear SEND header flags */
+               cmd[1] &= ~(0xFFFFUL << 32);
+       } else {
                l2_len = m->l2_len;
+       }
 
        /* Retrieve DPTR */
        dptr = *(uint64_t *)(sg + 1);
@@ -395,8 +423,8 @@ cn10k_nix_prep_sec(struct rte_mbuf *m, uint64_t *cmd, uintptr_t *nixtx_addr,
        sa_base &= ~0xFFFFUL;
        sa = (uintptr_t)roc_nix_inl_ot_ipsec_outb_sa(sa_base, sess_priv.sa_idx);
        ucode_cmd[3] = (ROC_CPT_DFLT_ENG_GRP_SE_IE << 61 | 1UL << 60 | sa);
-       ucode_cmd[0] =
-               (ROC_IE_OT_MAJOR_OP_PROCESS_OUTBOUND_IPSEC << 48 | pkt_len);
+       ucode_cmd[0] = (ROC_IE_OT_MAJOR_OP_PROCESS_OUTBOUND_IPSEC << 48 |
+                       ((uint64_t)sess_priv.chksum) << 32 | pkt_len);
 
        /* CPT Word 0 and Word 1. Assume no multi-seg support */
        cmd01 = vdupq_n_u64((nixtx + 16) | (cn10k_nix_tx_ext_subs(flags) + 1));
@@ -511,13 +539,16 @@ cn10k_nix_xmit_prepare_tso(struct rte_mbuf *m, const uint64_t flags)
 
 static __rte_always_inline void
 cn10k_nix_xmit_prepare(struct rte_mbuf *m, uint64_t *cmd, const uint16_t flags,
-                      const uint64_t lso_tun_fmt, bool *sec)
+                      const uint64_t lso_tun_fmt, bool *sec, uint8_t mark_flag,
+                      uint64_t mark_fmt)
 {
+       uint8_t mark_off = 0, mark_vlan = 0, markptr = 0;
        struct nix_send_ext_s *send_hdr_ext;
        struct nix_send_hdr_s *send_hdr;
        uint64_t ol_flags = 0, mask;
        union nix_send_hdr_w1_u w1;
        union nix_send_sg_s *sg;
+       uint16_t mark_form = 0;
 
        send_hdr = (struct nix_send_hdr_s *)cmd;
        if (flags & NIX_TX_NEED_EXT_HDR) {
@@ -525,7 +556,9 @@ cn10k_nix_xmit_prepare(struct rte_mbuf *m, uint64_t *cmd, const uint16_t flags,
                sg = (union nix_send_sg_s *)(cmd + 4);
                /* Clear previous markings */
                send_hdr_ext->w0.lso = 0;
+               send_hdr_ext->w0.mark_en = 0;
                send_hdr_ext->w1.u = 0;
+               ol_flags = m->ol_flags;
        } else {
                sg = (union nix_send_sg_s *)(cmd + 2);
        }
@@ -621,6 +654,10 @@ cn10k_nix_xmit_prepare(struct rte_mbuf *m, uint64_t *cmd, const uint16_t flags,
        }
 
        if (flags & NIX_TX_NEED_EXT_HDR && flags & NIX_TX_OFFLOAD_VLAN_QINQ_F) {
+               const uint8_t ipv6 = !!(ol_flags & RTE_MBUF_F_TX_IPV6);
+               const uint8_t ip = !!(ol_flags & (RTE_MBUF_F_TX_IPV4 |
+                                                 RTE_MBUF_F_TX_IPV6));
+
                send_hdr_ext->w1.vlan1_ins_ena = !!(ol_flags & RTE_MBUF_F_TX_VLAN);
                /* HW will update ptr after vlan0 update */
                send_hdr_ext->w1.vlan1_ins_ptr = 12;
@@ -630,6 +667,22 @@ cn10k_nix_xmit_prepare(struct rte_mbuf *m, uint64_t *cmd, const uint16_t flags,
                /* 2B before end of l2 header */
                send_hdr_ext->w1.vlan0_ins_ptr = 12;
                send_hdr_ext->w1.vlan0_ins_tci = m->vlan_tci_outer;
+               /* Fill for VLAN marking only when VLAN insertion enabled */
+               mark_vlan = ((mark_flag & CNXK_TM_MARK_VLAN_DEI) &
+                            (send_hdr_ext->w1.vlan1_ins_ena ||
+                             send_hdr_ext->w1.vlan0_ins_ena));
+
+               /* Mask requested flags with packet data information */
+               mark_off = mark_flag & ((ip << 2) | (ip << 1) | mark_vlan);
+               mark_off = ffs(mark_off & CNXK_TM_MARK_MASK);
+
+               mark_form = (mark_fmt >> ((mark_off - !!mark_off) << 4));
+               mark_form = (mark_form >> (ipv6 << 3)) & 0xFF;
+               markptr = m->l2_len + (mark_form >> 7) - (mark_vlan << 2);
+
+               send_hdr_ext->w0.mark_en = !!mark_off;
+               send_hdr_ext->w0.markform = mark_form & 0x7F;
+               send_hdr_ext->w0.markptr = markptr;
        }
 
        if (flags & NIX_TX_OFFLOAD_TSO_F && (ol_flags & RTE_MBUF_F_TX_TCP_SEG)) {
@@ -841,8 +894,10 @@ cn10k_nix_xmit_pkts(void *tx_queue, uint64_t *ws, struct rte_mbuf **tx_pkts,
        uintptr_t pa, lbase = txq->lmt_base;
        uint16_t lmt_id, burst, left, i;
        uintptr_t c_lbase = lbase;
+       uint64_t lso_tun_fmt = 0;
+       uint64_t mark_fmt = 0;
+       uint8_t mark_flag = 0;
        rte_iova_t c_io_addr;
-       uint64_t lso_tun_fmt;
        uint16_t c_lmt_id;
        uint64_t sa_base;
        uintptr_t laddr;
@@ -860,6 +915,11 @@ cn10k_nix_xmit_pkts(void *tx_queue, uint64_t *ws, struct rte_mbuf **tx_pkts,
        if (flags & NIX_TX_OFFLOAD_TSO_F)
                lso_tun_fmt = txq->lso_tun_fmt;
 
+       if (flags & NIX_TX_OFFLOAD_VLAN_QINQ_F) {
+               mark_fmt = txq->mark_fmt;
+               mark_flag = txq->mark_flag;
+       }
+
        /* Get LMT base address and LMT ID as lcore id */
        ROC_LMT_BASE_ID_GET(lbase, lmt_id);
        if (flags & NIX_TX_OFFLOAD_SECURITY_F) {
@@ -887,7 +947,7 @@ again:
                        cn10k_nix_xmit_prepare_tso(tx_pkts[i], flags);
 
                cn10k_nix_xmit_prepare(tx_pkts[i], cmd, flags, lso_tun_fmt,
-                                      &sec);
+                                      &sec, mark_flag, mark_fmt);
 
                laddr = (uintptr_t)LMT_OFF(lbase, lnum, 0);
 
@@ -905,8 +965,8 @@ again:
                        lnum++;
        }
 
-       if (flags & NIX_TX_VWQE_F)
-               roc_sso_hws_head_wait(ws[0]);
+       if ((flags & NIX_TX_VWQE_F) && !(ws[1] & BIT_ULL(35)))
+               ws[1] = roc_sso_hws_head_wait(ws[0]);
 
        left -= burst;
        tx_pkts += burst;
@@ -967,9 +1027,11 @@ cn10k_nix_xmit_pkts_mseg(void *tx_queue, uint64_t *ws,
        uint16_t segdw, lmt_id, burst, left, i;
        uint8_t lnum, c_lnum, c_loff;
        uintptr_t c_lbase = lbase;
+       uint64_t lso_tun_fmt = 0;
+       uint64_t mark_fmt = 0;
+       uint8_t mark_flag = 0;
        uint64_t data0, data1;
        rte_iova_t c_io_addr;
-       uint64_t lso_tun_fmt;
        uint8_t shft, c_shft;
        __uint128_t data128;
        uint16_t c_lmt_id;
@@ -988,6 +1050,11 @@ cn10k_nix_xmit_pkts_mseg(void *tx_queue, uint64_t *ws,
        if (flags & NIX_TX_OFFLOAD_TSO_F)
                lso_tun_fmt = txq->lso_tun_fmt;
 
+       if (flags & NIX_TX_OFFLOAD_VLAN_QINQ_F) {
+               mark_fmt = txq->mark_fmt;
+               mark_flag = txq->mark_flag;
+       }
+
        /* Get LMT base address and LMT ID as lcore id */
        ROC_LMT_BASE_ID_GET(lbase, lmt_id);
        if (flags & NIX_TX_OFFLOAD_SECURITY_F) {
@@ -1017,7 +1084,7 @@ again:
                        cn10k_nix_xmit_prepare_tso(tx_pkts[i], flags);
 
                cn10k_nix_xmit_prepare(tx_pkts[i], cmd, flags, lso_tun_fmt,
-                                      &sec);
+                                      &sec, mark_flag, mark_fmt);
 
                laddr = (uintptr_t)LMT_OFF(lbase, lnum, 0);
 
@@ -1041,8 +1108,8 @@ again:
                }
        }
 
-       if (flags & NIX_TX_VWQE_F)
-               roc_sso_hws_head_wait(ws[0]);
+       if ((flags & NIX_TX_VWQE_F) && !(ws[1] & BIT_ULL(35)))
+               ws[1] = roc_sso_hws_head_wait(ws[0]);
 
        left -= burst;
        tx_pkts += burst;
@@ -2569,6 +2636,13 @@ again:
                        lnum += 1;
                }
 
+               if (flags & NIX_TX_MULTI_SEG_F) {
+                       tx_pkts[0]->next = NULL;
+                       tx_pkts[1]->next = NULL;
+                       tx_pkts[2]->next = NULL;
+                       tx_pkts[3]->next = NULL;
+               }
+
                tx_pkts = tx_pkts + NIX_DESCS_PER_LOOP;
        }
 
@@ -2582,8 +2656,8 @@ again:
        if (flags & (NIX_TX_MULTI_SEG_F | NIX_TX_OFFLOAD_SECURITY_F))
                wd.data[0] >>= 16;
 
-       if (flags & NIX_TX_VWQE_F)
-               roc_sso_hws_head_wait(ws[0]);
+       if ((flags & NIX_TX_VWQE_F) && !(ws[1] & BIT_ULL(35)))
+               ws[1] = roc_sso_hws_head_wait(ws[0]);
 
        left -= burst;
 
@@ -2682,18 +2756,18 @@ cn10k_nix_xmit_pkts_vector(void *tx_queue, uint64_t *ws,
 
 /* [T_SEC_F] [TSP] [TSO] [NOFF] [VLAN] [OL3OL4CSUM] [L3L4CSUM] */
 #define NIX_TX_FASTPATH_MODES_0_15                                             \
-       T(no_offload, 4, NIX_TX_OFFLOAD_NONE)                                  \
-       T(l3l4csum, 4, L3L4CSUM_F)                                             \
-       T(ol3ol4csum, 4, OL3OL4CSUM_F)                                         \
-       T(ol3ol4csum_l3l4csum, 4, OL3OL4CSUM_F | L3L4CSUM_F)                   \
+       T(no_offload, 6, NIX_TX_OFFLOAD_NONE)                                  \
+       T(l3l4csum, 6, L3L4CSUM_F)                                             \
+       T(ol3ol4csum, 6, OL3OL4CSUM_F)                                         \
+       T(ol3ol4csum_l3l4csum, 6, OL3OL4CSUM_F | L3L4CSUM_F)                   \
        T(vlan, 6, VLAN_F)                                                     \
        T(vlan_l3l4csum, 6, VLAN_F | L3L4CSUM_F)                               \
        T(vlan_ol3ol4csum, 6, VLAN_F | OL3OL4CSUM_F)                           \
        T(vlan_ol3ol4csum_l3l4csum, 6, VLAN_F | OL3OL4CSUM_F | L3L4CSUM_F)     \
-       T(noff, 4, NOFF_F)                                                     \
-       T(noff_l3l4csum, 4, NOFF_F | L3L4CSUM_F)                               \
-       T(noff_ol3ol4csum, 4, NOFF_F | OL3OL4CSUM_F)                           \
-       T(noff_ol3ol4csum_l3l4csum, 4, NOFF_F | OL3OL4CSUM_F | L3L4CSUM_F)     \
+       T(noff, 6, NOFF_F)                                                     \
+       T(noff_l3l4csum, 6, NOFF_F | L3L4CSUM_F)                               \
+       T(noff_ol3ol4csum, 6, NOFF_F | OL3OL4CSUM_F)                           \
+       T(noff_ol3ol4csum_l3l4csum, 6, NOFF_F | OL3OL4CSUM_F | L3L4CSUM_F)     \
        T(noff_vlan, 6, NOFF_F | VLAN_F)                                       \
        T(noff_vlan_l3l4csum, 6, NOFF_F | VLAN_F | L3L4CSUM_F)                 \
        T(noff_vlan_ol3ol4csum, 6, NOFF_F | VLAN_F | OL3OL4CSUM_F)             \
@@ -2767,19 +2841,19 @@ cn10k_nix_xmit_pkts_vector(void *tx_queue, uint64_t *ws,
          TSP_F | TSO_F | NOFF_F | VLAN_F | OL3OL4CSUM_F | L3L4CSUM_F)
 
 #define NIX_TX_FASTPATH_MODES_64_79                                            \
-       T(sec, 4, T_SEC_F)                                                     \
-       T(sec_l3l4csum, 4, T_SEC_F | L3L4CSUM_F)                               \
-       T(sec_ol3ol4csum, 4, T_SEC_F | OL3OL4CSUM_F)                           \
-       T(sec_ol3ol4csum_l3l4csum, 4, T_SEC_F | OL3OL4CSUM_F | L3L4CSUM_F)     \
+       T(sec, 6, T_SEC_F)                                                     \
+       T(sec_l3l4csum, 6, T_SEC_F | L3L4CSUM_F)                               \
+       T(sec_ol3ol4csum, 6, T_SEC_F | OL3OL4CSUM_F)                           \
+       T(sec_ol3ol4csum_l3l4csum, 6, T_SEC_F | OL3OL4CSUM_F | L3L4CSUM_F)     \
        T(sec_vlan, 6, T_SEC_F | VLAN_F)                                       \
        T(sec_vlan_l3l4csum, 6, T_SEC_F | VLAN_F | L3L4CSUM_F)                 \
        T(sec_vlan_ol3ol4csum, 6, T_SEC_F | VLAN_F | OL3OL4CSUM_F)             \
        T(sec_vlan_ol3ol4csum_l3l4csum, 6,                                     \
          T_SEC_F | VLAN_F | OL3OL4CSUM_F | L3L4CSUM_F)                        \
-       T(sec_noff, 4, T_SEC_F | NOFF_F)                                       \
-       T(sec_noff_l3l4csum, 4, T_SEC_F | NOFF_F | L3L4CSUM_F)                 \
-       T(sec_noff_ol3ol4csum, 4, T_SEC_F | NOFF_F | OL3OL4CSUM_F)             \
-       T(sec_noff_ol3ol4csum_l3l4csum, 4,                                     \
+       T(sec_noff, 6, T_SEC_F | NOFF_F)                                       \
+       T(sec_noff_l3l4csum, 6, T_SEC_F | NOFF_F | L3L4CSUM_F)                 \
+       T(sec_noff_ol3ol4csum, 6, T_SEC_F | NOFF_F | OL3OL4CSUM_F)             \
+       T(sec_noff_ol3ol4csum_l3l4csum, 6,                                     \
          T_SEC_F | NOFF_F | OL3OL4CSUM_F | L3L4CSUM_F)                        \
        T(sec_noff_vlan, 6, T_SEC_F | NOFF_F | VLAN_F)                         \
        T(sec_noff_vlan_l3l4csum, 6, T_SEC_F | NOFF_F | VLAN_F | L3L4CSUM_F)   \