event/octeontx2: add Rx adapter fastpath ops
authorPavan Nikhilesh <pbhagavatula@marvell.com>
Thu, 4 Jul 2019 02:19:37 +0000 (07:49 +0530)
committerJerin Jacob <jerinj@marvell.com>
Sun, 7 Jul 2019 13:16:33 +0000 (15:16 +0200)
Add support for event eth Rx adapter fastpath operations.

Signed-off-by: Jerin Jacob <jerinj@marvell.com>
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Signed-off-by: Nithin Dabilpuram <ndabilpuram@marvell.com>
drivers/event/octeontx2/otx2_evdev.c
drivers/event/octeontx2/otx2_evdev.h
drivers/event/octeontx2/otx2_evdev_adptr.c
drivers/event/octeontx2/otx2_worker.c
drivers/event/octeontx2/otx2_worker.h
drivers/event/octeontx2/otx2_worker_dual.c
drivers/event/octeontx2/otx2_worker_dual.h

index 2956a57..f45fc00 100644 (file)
@@ -43,17 +43,199 @@ void
 sso_fastpath_fns_set(struct rte_eventdev *event_dev)
 {
        struct otx2_sso_evdev *dev = sso_pmd_priv(event_dev);
+       /* Single WS modes */
+       const event_dequeue_t ssogws_deq[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_deq_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_burst_t ssogws_deq_burst[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_deq_burst_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_t ssogws_deq_timeout[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_deq_timeout_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_burst_t
+               ssogws_deq_timeout_burst[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] =                              \
+                       otx2_ssogws_deq_timeout_burst_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_t ssogws_deq_seg[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_deq_seg_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_burst_t ssogws_deq_seg_burst[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_deq_seg_burst_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_t ssogws_deq_seg_timeout[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_deq_seg_timeout_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_burst_t
+               ssogws_deq_seg_timeout_burst[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] =                              \
+                               otx2_ssogws_deq_seg_timeout_burst_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+
+       /* Dual WS modes */
+       const event_dequeue_t ssogws_dual_deq[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_dual_deq_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_burst_t ssogws_dual_deq_burst[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_dual_deq_burst_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_t ssogws_dual_deq_timeout[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_dual_deq_timeout_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_burst_t
+               ssogws_dual_deq_timeout_burst[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+       [f5][f4][f3][f2][f1][f0] = otx2_ssogws_dual_deq_timeout_burst_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_t ssogws_dual_deq_seg[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] = otx2_ssogws_dual_deq_seg_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_burst_t
+               ssogws_dual_deq_seg_burst[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] =                              \
+                               otx2_ssogws_dual_deq_seg_burst_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_t ssogws_dual_deq_seg_timeout[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+               [f5][f4][f3][f2][f1][f0] =                              \
+                               otx2_ssogws_dual_deq_seg_timeout_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
+
+       const event_dequeue_burst_t
+               ssogws_dual_deq_seg_timeout_burst[2][2][2][2][2][2] = {
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+       [f5][f4][f3][f2][f1][f0] =                                      \
+               otx2_ssogws_dual_deq_seg_timeout_burst_ ##name,
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
+       };
 
        event_dev->enqueue                      = otx2_ssogws_enq;
        event_dev->enqueue_burst                = otx2_ssogws_enq_burst;
        event_dev->enqueue_new_burst            = otx2_ssogws_enq_new_burst;
        event_dev->enqueue_forward_burst        = otx2_ssogws_enq_fwd_burst;
-
-       event_dev->dequeue                      = otx2_ssogws_deq;
-       event_dev->dequeue_burst                = otx2_ssogws_deq_burst;
-       if (dev->is_timeout_deq) {
-               event_dev->dequeue              = otx2_ssogws_deq_timeout;
-               event_dev->dequeue_burst        = otx2_ssogws_deq_timeout_burst;
+       if (dev->rx_offloads & NIX_RX_MULTI_SEG_F) {
+               event_dev->dequeue              = ssogws_deq_seg
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_CHECKSUM_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+               event_dev->dequeue_burst        = ssogws_deq_seg_burst
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_CHECKSUM_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+               if (dev->is_timeout_deq) {
+                       event_dev->dequeue      = ssogws_deq_seg_timeout
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_CHECKSUM_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+                       event_dev->dequeue_burst        =
+                               ssogws_deq_seg_timeout_burst
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_CHECKSUM_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+               }
+       } else {
+               event_dev->dequeue                      = ssogws_deq
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_CHECKSUM_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+               event_dev->dequeue_burst                = ssogws_deq_burst
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_CHECKSUM_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+               if (dev->is_timeout_deq) {
+                       event_dev->dequeue              = ssogws_deq_timeout
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_CHECKSUM_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+                       event_dev->dequeue_burst        =
+                               ssogws_deq_timeout_burst
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_CHECKSUM_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                       [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+               }
        }
 
        if (dev->dual_ws) {
@@ -63,12 +245,112 @@ sso_fastpath_fns_set(struct rte_eventdev *event_dev)
                                        otx2_ssogws_dual_enq_new_burst;
                event_dev->enqueue_forward_burst =
                                        otx2_ssogws_dual_enq_fwd_burst;
-               event_dev->dequeue              = otx2_ssogws_dual_deq;
-               event_dev->dequeue_burst        = otx2_ssogws_dual_deq_burst;
-               if (dev->is_timeout_deq) {
-                       event_dev->dequeue      = otx2_ssogws_dual_deq_timeout;
-                       event_dev->dequeue_burst =
-                                       otx2_ssogws_dual_deq_timeout_burst;
+
+               if (dev->rx_offloads & NIX_RX_MULTI_SEG_F) {
+                       event_dev->dequeue      = ssogws_dual_deq_seg
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_TSTAMP_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_CHECKSUM_F)]
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+                       event_dev->dequeue_burst = ssogws_dual_deq_seg_burst
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_TSTAMP_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_CHECKSUM_F)]
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+                       if (dev->is_timeout_deq) {
+                               event_dev->dequeue      =
+                                       ssogws_dual_deq_seg_timeout
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_TSTAMP_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_CHECKSUM_F)]
+                                       [!!(dev->rx_offloads &
+                                                       NIX_RX_OFFLOAD_PTYPE_F)]
+                                       [!!(dev->rx_offloads &
+                                                       NIX_RX_OFFLOAD_RSS_F)];
+                               event_dev->dequeue_burst =
+                                       ssogws_dual_deq_seg_timeout_burst
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_TSTAMP_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_CHECKSUM_F)]
+                                       [!!(dev->rx_offloads &
+                                                       NIX_RX_OFFLOAD_PTYPE_F)]
+                                       [!!(dev->rx_offloads &
+                                                       NIX_RX_OFFLOAD_RSS_F)];
+                       }
+               } else {
+                       event_dev->dequeue              = ssogws_dual_deq
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_TSTAMP_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_CHECKSUM_F)]
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+                       event_dev->dequeue_burst        = ssogws_dual_deq_burst
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_TSTAMP_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                               [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_CHECKSUM_F)]
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_PTYPE_F)]
+                               [!!(dev->rx_offloads & NIX_RX_OFFLOAD_RSS_F)];
+                       if (dev->is_timeout_deq) {
+                               event_dev->dequeue      =
+                                       ssogws_dual_deq_timeout
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_TSTAMP_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_CHECKSUM_F)]
+                                       [!!(dev->rx_offloads &
+                                                       NIX_RX_OFFLOAD_PTYPE_F)]
+                                       [!!(dev->rx_offloads &
+                                                       NIX_RX_OFFLOAD_RSS_F)];
+                               event_dev->dequeue_burst =
+                                       ssogws_dual_deq_timeout_burst
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_TSTAMP_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_MARK_UPDATE_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_VLAN_STRIP_F)]
+                                       [!!(dev->rx_offloads &
+                                               NIX_RX_OFFLOAD_CHECKSUM_F)]
+                                       [!!(dev->rx_offloads &
+                                                       NIX_RX_OFFLOAD_PTYPE_F)]
+                                       [!!(dev->rx_offloads &
+                                                       NIX_RX_OFFLOAD_RSS_F)];
+                       }
                }
        }
        rte_mb();
@@ -1126,6 +1408,8 @@ static struct rte_eventdev_ops otx2_sso_ops = {
        .eth_rx_adapter_caps_get  = otx2_sso_rx_adapter_caps_get,
        .eth_rx_adapter_queue_add = otx2_sso_rx_adapter_queue_add,
        .eth_rx_adapter_queue_del = otx2_sso_rx_adapter_queue_del,
+       .eth_rx_adapter_start = otx2_sso_rx_adapter_start,
+       .eth_rx_adapter_stop = otx2_sso_rx_adapter_stop,
 
        .timer_adapter_caps_get = otx2_tim_caps_get,
 
index 107aa79..a81a8be 100644 (file)
@@ -132,6 +132,7 @@ struct otx2_sso_evdev {
        uint64_t nb_xaq_cfg;
        rte_iova_t fc_iova;
        struct rte_mempool *xaq_pool;
+       uint64_t rx_offloads;
        uint16_t rx_adptr_pool_cnt;
        uint32_t adptr_xae_cnt;
        uint64_t *rx_adptr_pools;
@@ -166,6 +167,7 @@ struct otx2_ssogws {
        /* Get Work Fastpath data */
        OTX2_SSOGWS_OPS;
        uint8_t swtag_req;
+       void *lookup_mem;
        uint8_t port;
        /* Add Work Fastpath data */
        uint64_t xaq_lmt __rte_cache_aligned;
@@ -182,6 +184,7 @@ struct otx2_ssogws_dual {
        struct otx2_ssogws_state ws_state[2]; /* Ping and Pong */
        uint8_t swtag_req;
        uint8_t vws; /* Ping pong bit */
+       void *lookup_mem;
        uint8_t port;
        /* Add Work Fastpath data */
        uint64_t xaq_lmt __rte_cache_aligned;
@@ -195,6 +198,28 @@ sso_pmd_priv(const struct rte_eventdev *event_dev)
        return event_dev->data->dev_private;
 }
 
+static const union mbuf_initializer mbuf_init = {
+       .fields = {
+               .data_off = RTE_PKTMBUF_HEADROOM,
+               .refcnt = 1,
+               .nb_segs = 1,
+               .port = 0
+       }
+};
+
+static __rte_always_inline void
+otx2_wqe_to_mbuf(uint64_t get_work1, const uint64_t mbuf, uint8_t port_id,
+                const uint32_t tag, const uint32_t flags,
+                const void * const lookup_mem)
+{
+       struct nix_wqe_hdr_s *wqe = (struct nix_wqe_hdr_s *)get_work1;
+
+       otx2_nix_cqe_to_mbuf((struct nix_cqe_hdr_s *)wqe, tag,
+                            (struct rte_mbuf *)mbuf, lookup_mem,
+                            mbuf_init.value | (uint64_t)port_id << 48, flags);
+
+}
+
 static inline int
 parse_kvargs_flag(const char *key, const char *value, void *opaque)
 {
@@ -213,6 +238,9 @@ parse_kvargs_value(const char *key, const char *value, void *opaque)
        return 0;
 }
 
+#define SSO_RX_ADPTR_ENQ_FASTPATH_FUNC NIX_RX_FASTPATH_MODES
+#define SSO_TX_ADPTR_ENQ_FASTPATH_FUNC NIX_TX_FASTPATH_MODES
+
 /* Single WS API's */
 uint16_t otx2_ssogws_enq(void *port, const struct rte_event *ev);
 uint16_t otx2_ssogws_enq_burst(void *port, const struct rte_event ev[],
@@ -222,15 +250,6 @@ uint16_t otx2_ssogws_enq_new_burst(void *port, const struct rte_event ev[],
 uint16_t otx2_ssogws_enq_fwd_burst(void *port, const struct rte_event ev[],
                                   uint16_t nb_events);
 
-uint16_t otx2_ssogws_deq(void *port, struct rte_event *ev,
-                        uint64_t timeout_ticks);
-uint16_t otx2_ssogws_deq_burst(void *port, struct rte_event ev[],
-                              uint16_t nb_events, uint64_t timeout_ticks);
-uint16_t otx2_ssogws_deq_timeout(void *port, struct rte_event *ev,
-                                uint64_t timeout_ticks);
-uint16_t otx2_ssogws_deq_timeout_burst(void *port, struct rte_event ev[],
-                                      uint16_t nb_events,
-                                      uint64_t timeout_ticks);
 /* Dual WS API's */
 uint16_t otx2_ssogws_dual_enq(void *port, const struct rte_event *ev);
 uint16_t otx2_ssogws_dual_enq_burst(void *port, const struct rte_event ev[],
@@ -240,15 +259,63 @@ uint16_t otx2_ssogws_dual_enq_new_burst(void *port, const struct rte_event ev[],
 uint16_t otx2_ssogws_dual_enq_fwd_burst(void *port, const struct rte_event ev[],
                                        uint16_t nb_events);
 
-uint16_t otx2_ssogws_dual_deq(void *port, struct rte_event *ev,
-                             uint64_t timeout_ticks);
-uint16_t otx2_ssogws_dual_deq_burst(void *port, struct rte_event ev[],
-                                   uint16_t nb_events, uint64_t timeout_ticks);
-uint16_t otx2_ssogws_dual_deq_timeout(void *port, struct rte_event *ev,
-                                     uint64_t timeout_ticks);
-uint16_t otx2_ssogws_dual_deq_timeout_burst(void *port, struct rte_event ev[],
-                                           uint16_t nb_events,
-                                           uint64_t timeout_ticks);
+/* Auto generated API's */
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                                \
+uint16_t otx2_ssogws_deq_ ##name(void *port, struct rte_event *ev,            \
+                                uint64_t timeout_ticks);                      \
+uint16_t otx2_ssogws_deq_burst_ ##name(void *port, struct rte_event ev[],      \
+                                      uint16_t nb_events,                     \
+                                      uint64_t timeout_ticks);                \
+uint16_t otx2_ssogws_deq_timeout_ ##name(void *port,                          \
+                                        struct rte_event *ev,                 \
+                                        uint64_t timeout_ticks);              \
+uint16_t otx2_ssogws_deq_timeout_burst_ ##name(void *port,                    \
+                                              struct rte_event ev[],          \
+                                              uint16_t nb_events,             \
+                                              uint64_t timeout_ticks);        \
+uint16_t otx2_ssogws_deq_seg_ ##name(void *port, struct rte_event *ev,        \
+                                    uint64_t timeout_ticks);                  \
+uint16_t otx2_ssogws_deq_seg_burst_ ##name(void *port,                        \
+                                          struct rte_event ev[],              \
+                                          uint16_t nb_events,                 \
+                                          uint64_t timeout_ticks);            \
+uint16_t otx2_ssogws_deq_seg_timeout_ ##name(void *port,                      \
+                                            struct rte_event *ev,             \
+                                            uint64_t timeout_ticks);          \
+uint16_t otx2_ssogws_deq_seg_timeout_burst_ ##name(void *port,                \
+                                                  struct rte_event ev[],      \
+                                                  uint16_t nb_events,         \
+                                                  uint64_t timeout_ticks);    \
+                                                                              \
+uint16_t otx2_ssogws_dual_deq_ ##name(void *port, struct rte_event *ev,               \
+                                     uint64_t timeout_ticks);                 \
+uint16_t otx2_ssogws_dual_deq_burst_ ##name(void *port,                               \
+                                           struct rte_event ev[],             \
+                                           uint16_t nb_events,                \
+                                           uint64_t timeout_ticks);           \
+uint16_t otx2_ssogws_dual_deq_timeout_ ##name(void *port,                     \
+                                             struct rte_event *ev,            \
+                                             uint64_t timeout_ticks);         \
+uint16_t otx2_ssogws_dual_deq_timeout_burst_ ##name(void *port,                       \
+                                                   struct rte_event ev[],     \
+                                                   uint16_t nb_events,        \
+                                                   uint64_t timeout_ticks);   \
+uint16_t otx2_ssogws_dual_deq_seg_ ##name(void *port, struct rte_event *ev,    \
+                                         uint64_t timeout_ticks);             \
+uint16_t otx2_ssogws_dual_deq_seg_burst_ ##name(void *port,                   \
+                                               struct rte_event ev[],         \
+                                               uint16_t nb_events,            \
+                                               uint64_t timeout_ticks);       \
+uint16_t otx2_ssogws_dual_deq_seg_timeout_ ##name(void *port,                 \
+                                                 struct rte_event *ev,        \
+                                                 uint64_t timeout_ticks);     \
+uint16_t otx2_ssogws_dual_deq_seg_timeout_burst_ ##name(void *port,           \
+                                                       struct rte_event ev[], \
+                                                       uint16_t nb_events,    \
+                                                      uint64_t timeout_ticks);\
+
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
 
 void sso_updt_xae_cnt(struct otx2_sso_evdev *dev, void *data,
                      uint32_t event_type);
@@ -265,7 +332,10 @@ int otx2_sso_rx_adapter_queue_add(const struct rte_eventdev *event_dev,
 int otx2_sso_rx_adapter_queue_del(const struct rte_eventdev *event_dev,
                                  const struct rte_eth_dev *eth_dev,
                                  int32_t rx_queue_id);
-
+int otx2_sso_rx_adapter_start(const struct rte_eventdev *event_dev,
+                             const struct rte_eth_dev *eth_dev);
+int otx2_sso_rx_adapter_stop(const struct rte_eventdev *event_dev,
+                            const struct rte_eth_dev *eth_dev);
 /* Clean up API's */
 typedef void (*otx2_handle_event_t)(void *arg, struct rte_event ev);
 void ssogws_flush_events(struct otx2_ssogws *ws, uint8_t queue_id,
index 12469fa..e605fd1 100644 (file)
@@ -232,6 +232,25 @@ sso_updt_xae_cnt(struct otx2_sso_evdev *dev, void *data, uint32_t event_type)
        }
 }
 
+static inline void
+sso_updt_lookup_mem(const struct rte_eventdev *event_dev, void *lookup_mem)
+{
+       struct otx2_sso_evdev *dev = sso_pmd_priv(event_dev);
+       int i;
+
+       for (i = 0; i < dev->nb_event_ports; i++) {
+               if (dev->dual_ws) {
+                       struct otx2_ssogws_dual *ws = event_dev->data->ports[i];
+
+                       ws->lookup_mem = lookup_mem;
+               } else {
+                       struct otx2_ssogws *ws = event_dev->data->ports[i];
+
+                       ws->lookup_mem = lookup_mem;
+               }
+       }
+}
+
 int
 otx2_sso_rx_adapter_queue_add(const struct rte_eventdev *event_dev,
                              const struct rte_eth_dev *eth_dev,
@@ -258,6 +277,8 @@ otx2_sso_rx_adapter_queue_add(const struct rte_eventdev *event_dev,
                                             queue_conf->ev.sched_type,
                                             queue_conf->ev.queue_id, port);
                }
+               rxq = eth_dev->data->rx_queues[0];
+               sso_updt_lookup_mem(event_dev, rxq->lookup_mem);
        } else {
                rxq = eth_dev->data->rx_queues[rx_queue_id];
                sso_updt_xae_cnt(dev, rxq, RTE_EVENT_TYPE_ETHDEV);
@@ -266,6 +287,7 @@ otx2_sso_rx_adapter_queue_add(const struct rte_eventdev *event_dev,
                rc |= sso_rxq_enable(otx2_eth_dev, (uint16_t)rx_queue_id,
                                     queue_conf->ev.sched_type,
                                     queue_conf->ev.queue_id, port);
+               sso_updt_lookup_mem(event_dev, rxq->lookup_mem);
        }
 
        if (rc < 0) {
@@ -274,6 +296,9 @@ otx2_sso_rx_adapter_queue_add(const struct rte_eventdev *event_dev,
                return rc;
        }
 
+       dev->rx_offloads |= otx2_eth_dev->rx_offload_flags;
+       sso_fastpath_fns_set((struct rte_eventdev *)(uintptr_t)event_dev);
+
        return 0;
 }
 
@@ -303,3 +328,23 @@ otx2_sso_rx_adapter_queue_del(const struct rte_eventdev *event_dev,
 
        return rc;
 }
+
+int
+otx2_sso_rx_adapter_start(const struct rte_eventdev *event_dev,
+                         const struct rte_eth_dev *eth_dev)
+{
+       RTE_SET_USED(event_dev);
+       RTE_SET_USED(eth_dev);
+
+       return 0;
+}
+
+int
+otx2_sso_rx_adapter_stop(const struct rte_eventdev *event_dev,
+                        const struct rte_eth_dev *eth_dev)
+{
+       RTE_SET_USED(event_dev);
+       RTE_SET_USED(eth_dev);
+
+       return 0;
+}
index 7a6d4ca..ea2d0b5 100644 (file)
@@ -81,60 +81,132 @@ otx2_ssogws_release_event(struct otx2_ssogws *ws)
        otx2_ssogws_swtag_flush(ws);
 }
 
-uint16_t __hot
-otx2_ssogws_deq(void *port, struct rte_event *ev, uint64_t timeout_ticks)
-{
-       struct otx2_ssogws *ws = port;
-
-       RTE_SET_USED(timeout_ticks);
-
-       if (ws->swtag_req) {
-               ws->swtag_req = 0;
-               otx2_ssogws_swtag_wait(ws);
-               return 1;
-       }
-
-       return otx2_ssogws_get_work(ws, ev);
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+uint16_t __hot                                                         \
+otx2_ssogws_deq_ ##name(void *port, struct rte_event *ev,              \
+                       uint64_t timeout_ticks)                         \
+{                                                                      \
+       struct otx2_ssogws *ws = port;                                  \
+                                                                       \
+       RTE_SET_USED(timeout_ticks);                                    \
+                                                                       \
+       if (ws->swtag_req) {                                            \
+               ws->swtag_req = 0;                                      \
+               otx2_ssogws_swtag_wait(ws);                             \
+               return 1;                                               \
+       }                                                               \
+                                                                       \
+       return otx2_ssogws_get_work(ws, ev, flags, ws->lookup_mem);     \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_deq_burst_ ##name(void *port, struct rte_event ev[],       \
+                             uint16_t nb_events,                       \
+                             uint64_t timeout_ticks)                   \
+{                                                                      \
+       RTE_SET_USED(nb_events);                                        \
+                                                                       \
+       return otx2_ssogws_deq_ ##name(port, ev, timeout_ticks);        \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_deq_timeout_ ##name(void *port, struct rte_event *ev,      \
+                               uint64_t timeout_ticks)                 \
+{                                                                      \
+       struct otx2_ssogws *ws = port;                                  \
+       uint16_t ret = 1;                                               \
+       uint64_t iter;                                                  \
+                                                                       \
+       if (ws->swtag_req) {                                            \
+               ws->swtag_req = 0;                                      \
+               otx2_ssogws_swtag_wait(ws);                             \
+               return ret;                                             \
+       }                                                               \
+                                                                       \
+       ret = otx2_ssogws_get_work(ws, ev, flags, ws->lookup_mem);      \
+       for (iter = 1; iter < timeout_ticks && (ret == 0); iter++)      \
+               ret = otx2_ssogws_get_work(ws, ev, flags,               \
+                                          ws->lookup_mem);             \
+                                                                       \
+       return ret;                                                     \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_deq_timeout_burst_ ##name(void *port, struct rte_event ev[],\
+                                     uint16_t nb_events,               \
+                                     uint64_t timeout_ticks)           \
+{                                                                      \
+       RTE_SET_USED(nb_events);                                        \
+                                                                       \
+       return otx2_ssogws_deq_timeout_ ##name(port, ev, timeout_ticks);\
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_deq_seg_ ##name(void *port, struct rte_event *ev,          \
+                           uint64_t timeout_ticks)                     \
+{                                                                      \
+       struct otx2_ssogws *ws = port;                                  \
+                                                                       \
+       RTE_SET_USED(timeout_ticks);                                    \
+                                                                       \
+       if (ws->swtag_req) {                                            \
+               ws->swtag_req = 0;                                      \
+               otx2_ssogws_swtag_wait(ws);                             \
+               return 1;                                               \
+       }                                                               \
+                                                                       \
+       return otx2_ssogws_get_work(ws, ev, flags | NIX_RX_MULTI_SEG_F, \
+                                   ws->lookup_mem);                    \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_deq_seg_burst_ ##name(void *port, struct rte_event ev[],   \
+                                 uint16_t nb_events,                   \
+                                 uint64_t timeout_ticks)               \
+{                                                                      \
+       RTE_SET_USED(nb_events);                                        \
+                                                                       \
+       return otx2_ssogws_deq_seg_ ##name(port, ev, timeout_ticks);    \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_deq_seg_timeout_ ##name(void *port, struct rte_event *ev,  \
+                                   uint64_t timeout_ticks)             \
+{                                                                      \
+       struct otx2_ssogws *ws = port;                                  \
+       uint16_t ret = 1;                                               \
+       uint64_t iter;                                                  \
+                                                                       \
+       if (ws->swtag_req) {                                            \
+               ws->swtag_req = 0;                                      \
+               otx2_ssogws_swtag_wait(ws);                             \
+               return ret;                                             \
+       }                                                               \
+                                                                       \
+       ret = otx2_ssogws_get_work(ws, ev, flags | NIX_RX_MULTI_SEG_F,  \
+                                  ws->lookup_mem);                     \
+       for (iter = 1; iter < timeout_ticks && (ret == 0); iter++)      \
+               ret = otx2_ssogws_get_work(ws, ev,                      \
+                                          flags | NIX_RX_MULTI_SEG_F,  \
+                                          ws->lookup_mem);             \
+                                                                       \
+       return ret;                                                     \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_deq_seg_timeout_burst_ ##name(void *port,                  \
+                                         struct rte_event ev[],        \
+                                         uint16_t nb_events,           \
+                                         uint64_t timeout_ticks)       \
+{                                                                      \
+       RTE_SET_USED(nb_events);                                        \
+                                                                       \
+       return otx2_ssogws_deq_seg_timeout_ ##name(port, ev,            \
+                                                  timeout_ticks);      \
 }
 
-uint16_t __hot
-otx2_ssogws_deq_burst(void *port, struct rte_event ev[], uint16_t nb_events,
-                     uint64_t timeout_ticks)
-{
-       RTE_SET_USED(nb_events);
-
-       return otx2_ssogws_deq(port, ev, timeout_ticks);
-}
-
-uint16_t __hot
-otx2_ssogws_deq_timeout(void *port, struct rte_event *ev,
-                       uint64_t timeout_ticks)
-{
-       struct otx2_ssogws *ws = port;
-       uint16_t ret = 1;
-       uint64_t iter;
-
-       if (ws->swtag_req) {
-               ws->swtag_req = 0;
-               otx2_ssogws_swtag_wait(ws);
-               return ret;
-       }
-
-       ret = otx2_ssogws_get_work(ws, ev);
-       for (iter = 1; iter < timeout_ticks && (ret == 0); iter++)
-               ret = otx2_ssogws_get_work(ws, ev);
-
-       return ret;
-}
-
-uint16_t __hot
-otx2_ssogws_deq_timeout_burst(void *port, struct rte_event ev[],
-                             uint16_t nb_events, uint64_t timeout_ticks)
-{
-       RTE_SET_USED(nb_events);
-
-       return otx2_ssogws_deq_timeout(port, ev, timeout_ticks);
-}
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
 
 uint16_t __hot
 otx2_ssogws_enq(void *port, const struct rte_event *ev)
@@ -221,7 +293,7 @@ ssogws_flush_events(struct otx2_ssogws *ws, uint8_t queue_id, uintptr_t base,
 
        while (aq_cnt || cq_ds_cnt || ds_cnt) {
                otx2_write64(val, ws->getwrk_op);
-               otx2_ssogws_get_work_empty(ws, &ev);
+               otx2_ssogws_get_work_empty(ws, &ev, 0);
                if (fn != NULL && ev.u64 != 0)
                        fn(arg, ev);
                if (ev.sched_type != SSO_TT_EMPTY)
index f06ff06..accf7f9 100644 (file)
 /* SSO Operations */
 
 static __rte_always_inline uint16_t
-otx2_ssogws_get_work(struct otx2_ssogws *ws, struct rte_event *ev)
+otx2_ssogws_get_work(struct otx2_ssogws *ws, struct rte_event *ev,
+                    const uint32_t flags, const void * const lookup_mem)
 {
        union otx2_sso_event event;
        uint64_t get_work1;
+       uint64_t mbuf;
 
        otx2_write64(BIT_ULL(16) | /* wait for work. */
                     1, /* Use Mask set 0. */
                     ws->getwrk_op);
 
+       if (flags & NIX_RX_OFFLOAD_PTYPE_F)
+               rte_prefetch_non_temporal(lookup_mem);
 #ifdef RTE_ARCH_ARM64
        asm volatile(
                        "               ldr %[tag], [%[tag_loc]]        \n"
@@ -34,9 +38,12 @@ otx2_ssogws_get_work(struct otx2_ssogws *ws, struct rte_event *ev)
                        "               ldr %[wqp], [%[wqp_loc]]        \n"
                        "               tbnz %[tag], 63, rty%=          \n"
                        "done%=:        dmb ld                          \n"
-                       "               prfm pldl1keep, [%[wqp]]        \n"
+                       "               prfm pldl1keep, [%[wqp], #8]    \n"
+                       "               sub %[mbuf], %[wqp], #0x80      \n"
+                       "               prfm pldl1keep, [%[mbuf]]       \n"
                        : [tag] "=&r" (event.get_work0),
-                         [wqp] "=&r" (get_work1)
+                         [wqp] "=&r" (get_work1),
+                         [mbuf] "=&r" (mbuf)
                        : [tag_loc] "r" (ws->tag_op),
                          [wqp_loc] "r" (ws->wqp_op)
                        );
@@ -47,6 +54,8 @@ otx2_ssogws_get_work(struct otx2_ssogws *ws, struct rte_event *ev)
 
        get_work1 = otx2_read64(ws->wqp_op);
        rte_prefetch0((const void *)get_work1);
+       mbuf = (uint64_t)((char *)get_work1 - sizeof(struct rte_mbuf));
+       rte_prefetch0((const void *)mbuf);
 #endif
 
        event.get_work0 = (event.get_work0 & (0x3ull << 32)) << 6 |
@@ -55,6 +64,12 @@ otx2_ssogws_get_work(struct otx2_ssogws *ws, struct rte_event *ev)
        ws->cur_tt = event.sched_type;
        ws->cur_grp = event.queue_id;
 
+       if (event.sched_type != SSO_TT_EMPTY &&
+           event.event_type == RTE_EVENT_TYPE_ETHDEV) {
+               otx2_wqe_to_mbuf(get_work1, mbuf, event.sub_event_type,
+                                (uint32_t) event.get_work0, flags, lookup_mem);
+               get_work1 = mbuf;
+       }
 
        ev->event = event.get_work0;
        ev->u64 = get_work1;
@@ -64,10 +79,12 @@ otx2_ssogws_get_work(struct otx2_ssogws *ws, struct rte_event *ev)
 
 /* Used in cleaning up workslot. */
 static __rte_always_inline uint16_t
-otx2_ssogws_get_work_empty(struct otx2_ssogws *ws, struct rte_event *ev)
+otx2_ssogws_get_work_empty(struct otx2_ssogws *ws, struct rte_event *ev,
+                          const uint32_t flags)
 {
        union otx2_sso_event event;
        uint64_t get_work1;
+       uint64_t mbuf;
 
 #ifdef RTE_ARCH_ARM64
        asm volatile(
@@ -80,9 +97,12 @@ otx2_ssogws_get_work_empty(struct otx2_ssogws *ws, struct rte_event *ev)
                        "               ldr %[wqp], [%[wqp_loc]]        \n"
                        "               tbnz %[tag], 63, rty%=          \n"
                        "done%=:        dmb ld                          \n"
-                       "               prfm pldl1keep, [%[wqp]]        \n"
+                       "               prfm pldl1keep, [%[wqp], #8]    \n"
+                       "               sub %[mbuf], %[wqp], #0x80      \n"
+                       "               prfm pldl1keep, [%[mbuf]]       \n"
                        : [tag] "=&r" (event.get_work0),
-                         [wqp] "=&r" (get_work1)
+                         [wqp] "=&r" (get_work1),
+                         [mbuf] "=&r" (mbuf)
                        : [tag_loc] "r" (ws->tag_op),
                          [wqp_loc] "r" (ws->wqp_op)
                        );
@@ -92,7 +112,9 @@ otx2_ssogws_get_work_empty(struct otx2_ssogws *ws, struct rte_event *ev)
                event.get_work0 = otx2_read64(ws->tag_op);
 
        get_work1 = otx2_read64(ws->wqp_op);
-       rte_prefetch0((const void *)get_work1);
+       rte_prefetch_non_temporal((const void *)get_work1);
+       mbuf = (uint64_t)((char *)get_work1 - sizeof(struct rte_mbuf));
+       rte_prefetch_non_temporal((const void *)mbuf);
 #endif
 
        event.get_work0 = (event.get_work0 & (0x3ull << 32)) << 6 |
@@ -101,6 +123,13 @@ otx2_ssogws_get_work_empty(struct otx2_ssogws *ws, struct rte_event *ev)
        ws->cur_tt = event.sched_type;
        ws->cur_grp = event.queue_id;
 
+       if (event.sched_type != SSO_TT_EMPTY &&
+           event.event_type == RTE_EVENT_TYPE_ETHDEV) {
+               otx2_wqe_to_mbuf(get_work1, mbuf, event.sub_event_type,
+                                (uint32_t) event.get_work0, flags, NULL);
+               get_work1 = mbuf;
+       }
+
        ev->event = event.get_work0;
        ev->u64 = get_work1;
 
index 58fd588..b5cf9ac 100644 (file)
@@ -140,68 +140,162 @@ otx2_ssogws_dual_enq_fwd_burst(void *port, const struct rte_event ev[],
        return 1;
 }
 
-uint16_t __hot
-otx2_ssogws_dual_deq(void *port, struct rte_event *ev, uint64_t timeout_ticks)
-{
-       struct otx2_ssogws_dual *ws = port;
-       uint8_t gw;
-
-       RTE_SET_USED(timeout_ticks);
-       if (ws->swtag_req) {
-               otx2_ssogws_swtag_wait((struct otx2_ssogws *)
-                                      &ws->ws_state[!ws->vws]);
-               ws->swtag_req = 0;
-               return 1;
-       }
-
-       gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],
-                                      &ws->ws_state[!ws->vws], ev);
-       ws->vws = !ws->vws;
-
-       return gw;
-}
-
-uint16_t __hot
-otx2_ssogws_dual_deq_burst(void *port, struct rte_event ev[],
-                          uint16_t nb_events, uint64_t timeout_ticks)
-{
-       RTE_SET_USED(nb_events);
-
-       return otx2_ssogws_dual_deq(port, ev, timeout_ticks);
-}
-
-uint16_t __hot
-otx2_ssogws_dual_deq_timeout(void *port, struct rte_event *ev,
-                            uint64_t timeout_ticks)
-{
-       struct otx2_ssogws_dual *ws = port;
-       uint64_t iter;
-       uint8_t gw;
-
-       if (ws->swtag_req) {
-               otx2_ssogws_swtag_wait((struct otx2_ssogws *)
-                                      &ws->ws_state[!ws->vws]);
-               ws->swtag_req = 0;
-               return 1;
-       }
-
-       gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],
-                                      &ws->ws_state[!ws->vws], ev);
-       ws->vws = !ws->vws;
-       for (iter = 1; iter < timeout_ticks && (gw == 0); iter++) {
-               gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],
-                                              &ws->ws_state[!ws->vws], ev);
-               ws->vws = !ws->vws;
-       }
-
-       return gw;
+#define R(name, f5, f4, f3, f2, f1, f0, flags)                         \
+uint16_t __hot                                                         \
+otx2_ssogws_dual_deq_ ##name(void *port, struct rte_event *ev,         \
+                            uint64_t timeout_ticks)                    \
+{                                                                      \
+       struct otx2_ssogws_dual *ws = port;                             \
+       uint8_t gw;                                                     \
+                                                                       \
+       RTE_SET_USED(timeout_ticks);                                    \
+       if (ws->swtag_req) {                                            \
+               otx2_ssogws_swtag_wait((struct otx2_ssogws *)           \
+                                      &ws->ws_state[!ws->vws]);        \
+               ws->swtag_req = 0;                                      \
+               return 1;                                               \
+       }                                                               \
+                                                                       \
+       gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],          \
+                                      &ws->ws_state[!ws->vws], ev,     \
+                                      flags, ws->lookup_mem);          \
+       ws->vws = !ws->vws;                                             \
+                                                                       \
+       return gw;                                                      \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_dual_deq_burst_ ##name(void *port, struct rte_event ev[],  \
+                                  uint16_t nb_events,                  \
+                                  uint64_t timeout_ticks)              \
+{                                                                      \
+       RTE_SET_USED(nb_events);                                        \
+                                                                       \
+       return otx2_ssogws_dual_deq_ ##name(port, ev, timeout_ticks);   \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_dual_deq_timeout_ ##name(void *port, struct rte_event *ev, \
+                                    uint64_t timeout_ticks)            \
+{                                                                      \
+       struct otx2_ssogws_dual *ws = port;                             \
+       uint64_t iter;                                                  \
+       uint8_t gw;                                                     \
+                                                                       \
+       if (ws->swtag_req) {                                            \
+               otx2_ssogws_swtag_wait((struct otx2_ssogws *)           \
+                                      &ws->ws_state[!ws->vws]);        \
+               ws->swtag_req = 0;                                      \
+               return 1;                                               \
+       }                                                               \
+                                                                       \
+       gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],          \
+                                      &ws->ws_state[!ws->vws], ev,     \
+                                      flags, ws->lookup_mem);          \
+       ws->vws = !ws->vws;                                             \
+       for (iter = 1; iter < timeout_ticks && (gw == 0); iter++) {     \
+               gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],  \
+                                              &ws->ws_state[!ws->vws], \
+                                              ev, flags,               \
+                                              ws->lookup_mem);         \
+               ws->vws = !ws->vws;                                     \
+       }                                                               \
+                                                                       \
+       return gw;                                                      \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_dual_deq_timeout_burst_ ##name(void *port,                 \
+                                          struct rte_event ev[],       \
+                                          uint16_t nb_events,          \
+                                          uint64_t timeout_ticks)      \
+{                                                                      \
+       RTE_SET_USED(nb_events);                                        \
+                                                                       \
+       return otx2_ssogws_dual_deq_timeout_ ##name(port, ev,           \
+                                                   timeout_ticks);     \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_dual_deq_seg_ ##name(void *port, struct rte_event *ev,     \
+                                uint64_t timeout_ticks)                \
+{                                                                      \
+       struct otx2_ssogws_dual *ws = port;                             \
+       uint8_t gw;                                                     \
+                                                                       \
+       RTE_SET_USED(timeout_ticks);                                    \
+       if (ws->swtag_req) {                                            \
+               otx2_ssogws_swtag_wait((struct otx2_ssogws *)           \
+                                      &ws->ws_state[!ws->vws]);        \
+               ws->swtag_req = 0;                                      \
+               return 1;                                               \
+       }                                                               \
+                                                                       \
+       gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],          \
+                                      &ws->ws_state[!ws->vws], ev,     \
+                                      flags | NIX_RX_MULTI_SEG_F,      \
+                                      ws->lookup_mem);                 \
+       ws->vws = !ws->vws;                                             \
+                                                                       \
+       return gw;                                                      \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_dual_deq_seg_burst_ ##name(void *port,                     \
+                                      struct rte_event ev[],           \
+                                      uint16_t nb_events,              \
+                                      uint64_t timeout_ticks)          \
+{                                                                      \
+       RTE_SET_USED(nb_events);                                        \
+                                                                       \
+       return otx2_ssogws_dual_deq_seg_ ##name(port, ev,               \
+                                               timeout_ticks);         \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_dual_deq_seg_timeout_ ##name(void *port,                   \
+                                        struct rte_event *ev,          \
+                                        uint64_t timeout_ticks)        \
+{                                                                      \
+       struct otx2_ssogws_dual *ws = port;                             \
+       uint64_t iter;                                                  \
+       uint8_t gw;                                                     \
+                                                                       \
+       if (ws->swtag_req) {                                            \
+               otx2_ssogws_swtag_wait((struct otx2_ssogws *)           \
+                                      &ws->ws_state[!ws->vws]);        \
+               ws->swtag_req = 0;                                      \
+               return 1;                                               \
+       }                                                               \
+                                                                       \
+       gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],          \
+                                      &ws->ws_state[!ws->vws], ev,     \
+                                      flags | NIX_RX_MULTI_SEG_F,      \
+                                      ws->lookup_mem);                 \
+       ws->vws = !ws->vws;                                             \
+       for (iter = 1; iter < timeout_ticks && (gw == 0); iter++) {     \
+               gw = otx2_ssogws_dual_get_work(&ws->ws_state[ws->vws],  \
+                                              &ws->ws_state[!ws->vws], \
+                                              ev, flags |              \
+                                              NIX_RX_MULTI_SEG_F,      \
+                                              ws->lookup_mem);         \
+               ws->vws = !ws->vws;                                     \
+       }                                                               \
+                                                                       \
+       return gw;                                                      \
+}                                                                      \
+                                                                       \
+uint16_t __hot                                                         \
+otx2_ssogws_dual_deq_seg_timeout_burst_ ##name(void *port,             \
+                                              struct rte_event ev[],   \
+                                              uint16_t nb_events,      \
+                                              uint64_t timeout_ticks)  \
+{                                                                      \
+       RTE_SET_USED(nb_events);                                        \
+                                                                       \
+       return otx2_ssogws_dual_deq_seg_timeout_ ##name(port, ev,       \
+                                                       timeout_ticks); \
 }
 
-uint16_t __hot
-otx2_ssogws_dual_deq_timeout_burst(void *port, struct rte_event ev[],
-                                  uint16_t nb_events, uint64_t timeout_ticks)
-{
-       RTE_SET_USED(nb_events);
-
-       return otx2_ssogws_dual_deq_timeout(port, ev, timeout_ticks);
-}
+SSO_RX_ADPTR_ENQ_FASTPATH_FUNC
+#undef R
index d8453d1..32fe61b 100644 (file)
 static __rte_always_inline uint16_t
 otx2_ssogws_dual_get_work(struct otx2_ssogws_state *ws,
                          struct otx2_ssogws_state *ws_pair,
-                         struct rte_event *ev)
+                         struct rte_event *ev, const uint32_t flags,
+                         const void * const lookup_mem)
 {
        const uint64_t set_gw = BIT_ULL(16) | 1;
        union otx2_sso_event event;
        uint64_t get_work1;
+       uint64_t mbuf;
 
+       if (flags & NIX_RX_OFFLOAD_PTYPE_F)
+               rte_prefetch_non_temporal(lookup_mem);
 #ifdef RTE_ARCH_ARM64
        asm volatile(
                        "        ldr %[tag], [%[tag_loc]]    \n"
@@ -33,9 +37,12 @@ otx2_ssogws_dual_get_work(struct otx2_ssogws_state *ws,
                        "        tbnz %[tag], 63, rty%=      \n"
                        "done%=: str %[gw], [%[pong]]        \n"
                        "        dmb ld                      \n"
-                       "        prfm pldl1keep, [%[wqp]]    \n"
+                       "        prfm pldl1keep, [%[wqp], #8]\n"
+                       "        sub %[mbuf], %[wqp], #0x80  \n"
+                       "        prfm pldl1keep, [%[mbuf]]   \n"
                        : [tag] "=&r" (event.get_work0),
-                         [wqp] "=&r" (get_work1)
+                         [wqp] "=&r" (get_work1),
+                         [mbuf] "=&r" (mbuf)
                        : [tag_loc] "r" (ws->tag_op),
                          [wqp_loc] "r" (ws->wqp_op),
                          [gw] "r" (set_gw),
@@ -49,6 +56,8 @@ otx2_ssogws_dual_get_work(struct otx2_ssogws_state *ws,
        otx2_write64(set_gw, ws_pair->getwrk_op);
 
        rte_prefetch0((const void *)get_work1);
+       mbuf = (uint64_t)((char *)get_work1 - sizeof(struct rte_mbuf));
+       rte_prefetch0((const void *)mbuf);
 #endif
        event.get_work0 = (event.get_work0 & (0x3ull << 32)) << 6 |
                (event.get_work0 & (0x3FFull << 36)) << 4 |
@@ -56,6 +65,13 @@ otx2_ssogws_dual_get_work(struct otx2_ssogws_state *ws,
        ws->cur_tt = event.sched_type;
        ws->cur_grp = event.queue_id;
 
+       if (event.sched_type != SSO_TT_EMPTY &&
+           event.event_type == RTE_EVENT_TYPE_ETHDEV) {
+               otx2_wqe_to_mbuf(get_work1, mbuf, event.sub_event_type,
+                                (uint32_t) event.get_work0, flags, lookup_mem);
+               get_work1 = mbuf;
+       }
+
        ev->event = event.get_work0;
        ev->u64 = get_work1;