issue: 3807308 Add Ring TSO statistics

TSO statistics are very important for tracking TX performance.
Two variables are added, TSO packets number and TSO bytes number.
With these two and with conjunction of total TX packets and bytes,
the needed information can be derrived.

Signed-off-by: Alexander Grissik <agrissik@nvidia.com>

src/core/dev/hw_queue_tx.cpp

@@ -700,7 +700,7 @@ inline int hw_queue_tx::fill_wqe(xlio_ibv_send_wr *pswr)
         } else {
             /* Support XLIO_IBV_WR_SEND_TSO operation
              */
-            wqe_size = fill_wqe_lso(pswr);
+            wqe_size = fill_wqe_lso(pswr, data_len);
             return wqe_size;
         }
     }
@@ -750,7 +750,7 @@ inline int hw_queue_tx::fill_wqe_send(xlio_ibv_send_wr *pswr)
 }
 
 //! Filling wqe for LSO
-inline int hw_queue_tx::fill_wqe_lso(xlio_ibv_send_wr *pswr)
+inline int hw_queue_tx::fill_wqe_lso(xlio_ibv_send_wr *pswr, int data_len)
 {
     struct mlx5_wqe_ctrl_seg *ctrl = nullptr;
     struct mlx5_wqe_eth_seg *eseg = nullptr;
@@ -771,6 +771,8 @@ inline int hw_queue_tx::fill_wqe_lso(xlio_ibv_send_wr *pswr)
     if (0 == pswr->tso.mss) {
         ctrl->opmod_idx_opcode =
             htonl(((m_sq_wqe_counter & 0xffff) << 8) | (get_mlx5_opcode(XLIO_IBV_WR_SEND) & 0xff));
+    } else {
+        m_p_ring->update_tso_stats(static_cast<uint64_t>(data_len));
     }
 
     eseg = (struct mlx5_wqe_eth_seg *)((uint8_t *)m_sq_wqe_hot + sizeof(*ctrl));

src/core/dev/hw_queue_tx.h

@@ -261,7 +261,7 @@ class hw_queue_tx : public xlio_ti_owner {
     inline void store_current_wqe_prop(mem_buf_desc_t *wr_id, unsigned credits, xlio_ti *ti);
     inline int fill_wqe(xlio_ibv_send_wr *p_send_wqe);
     inline int fill_wqe_send(xlio_ibv_send_wr *pswr);
-    inline int fill_wqe_lso(xlio_ibv_send_wr *pswr);
+    inline int fill_wqe_lso(xlio_ibv_send_wr *pswr, int data_len);
     inline int fill_inl_segment(sg_array &sga, uint8_t *cur_seg, uint8_t *data_addr,
                                 int max_inline_len, int inline_len);
     inline void ring_doorbell(int num_wqebb, bool skip_comp = false);

src/core/dev/ring_simple.h

@@ -131,6 +131,12 @@ class ring_simple : public ring_slave {
     struct ibv_comp_channel *get_tx_comp_event_channel() { return m_p_tx_comp_event_channel; }
     void modify_cq_moderation(uint32_t period, uint32_t count);
 
+    void update_tso_stats(uint64_t bytes)
+    {
+        ++m_p_ring_stat->simple.n_tx_tso_pkt_count;
+        m_p_ring_stat->simple.n_tx_tso_byte_count += bytes;
+    }
+
 #ifdef DEFINED_UTLS
     bool tls_tx_supported(void) override { return m_tls.tls_tx; }
     bool tls_rx_supported(void) override { return m_tls.tls_rx; }

src/core/util/xlio_stats.h

@@ -373,6 +373,8 @@ typedef struct {
     ring_type_t n_type;
     union {
         struct {
+            uint64_t n_tx_tso_pkt_count;
+            uint64_t n_tx_tso_byte_count;
             uint64_t n_rx_interrupt_requests;
             uint64_t n_rx_interrupt_received;
             uint32_t n_rx_cq_moderation_count;

src/stats/stats_reader.cpp

@@ -379,6 +379,14 @@ void update_delta_ring_stat(ring_stats_t *p_curr_ring_stats, ring_stats_t *p_pre
             p_prev_ring_stats->tap.n_vf_plugouts =
                 (p_curr_ring_stats->tap.n_vf_plugouts - p_prev_ring_stats->tap.n_vf_plugouts);
         } else {
+            p_prev_ring_stats->simple.n_tx_tso_pkt_count =
+                (p_curr_ring_stats->simple.n_tx_tso_pkt_count -
+                 p_prev_ring_stats->simple.n_tx_tso_pkt_count) /
+                delay;
+            p_prev_ring_stats->simple.n_tx_tso_byte_count =
+                (p_curr_ring_stats->simple.n_tx_tso_byte_count -
+                 p_prev_ring_stats->simple.n_tx_tso_byte_count) /
+                delay;
             p_prev_ring_stats->simple.n_rx_interrupt_received =
                 (p_curr_ring_stats->simple.n_rx_interrupt_received -
                  p_prev_ring_stats->simple.n_rx_interrupt_received) /
@@ -535,6 +543,12 @@ void print_ring_stats(ring_instance_block_t *p_ring_inst_arr)
                 printf(FORMAT_STATS_32bit, "Tap fd:", p_ring_stats->tap.n_tap_fd);
                 printf(FORMAT_RING_TAP_NAME, "Tap Device:", p_ring_stats->tap.s_tap_name);
             } else {
+                if (p_ring_stats->simple.n_tx_tso_pkt_count ||
+                    p_ring_stats->simple.n_tx_tso_byte_count) {
+                    printf(FORMAT_RING_PACKETS, "TSO Offload:",
+                           p_ring_stats->simple.n_tx_tso_byte_count / BYTES_TRAFFIC_UNIT,
+                           p_ring_stats->simple.n_tx_tso_pkt_count, post_fix);
+                }
                 if (p_ring_stats->simple.n_rx_interrupt_requests ||
                     p_ring_stats->simple.n_rx_interrupt_received) {
                     printf(FORMAT_RING_INTERRUPT,
@@ -1799,6 +1813,8 @@ void zero_ring_stats(ring_stats_t *p_ring_stats)
     if (p_ring_stats->n_type == RING_TAP) {
         p_ring_stats->tap.n_vf_plugouts = 0;
     } else {
+        p_ring_stats->simple.n_tx_tso_pkt_count = 0;
+        p_ring_stats->simple.n_tx_tso_byte_count = 0;
         p_ring_stats->simple.n_rx_interrupt_received = 0;
         p_ring_stats->simple.n_rx_interrupt_requests = 0;
         p_ring_stats->simple.n_tx_dropped_wqes = 0;