Fix: seting rl rate on VF

Signed-off-by: Kasiewicz, Marek <marek.kasiewicz@intel.com>

Author: Sakoram

lib/src/dev/mt_dev.c

@@ -683,13 +683,8 @@ static int dev_init_ratelimit_all(struct mt_interface* inf) {
          shaper->shaper_profile_id);
   }
 
-  ret = rte_tm_hierarchy_commit(port_id, 1, &error);
-  if (ret < 0)
-    err("%s(%d), commit error (%d)%s\n", __func__, port, ret,
-        mt_string_safe(error.message));
-
   dbg("%s(%d), succ\n", __func__, port);
-  return ret;
+  return 0;
 }
 
 static int dev_tx_queue_set_rl_rate(struct mt_interface* inf, uint16_t queue,
@@ -712,13 +707,22 @@ static int dev_tx_queue_set_rl_rate(struct mt_interface* inf, uint16_t queue,
   /* not changed */
   if (bps == tx_queue->bps) return 0;
 
+  rte_atomic32_set(&inf->resetting, true);
+  mt_pthread_mutex_lock(&inf->vf_cmd_mutex);
+
+  ret = rte_eth_dev_stop(port_id);
+  if (ret) {
+    err("%s(%d), stop port %d fail %d\n", __func__, port, port_id, ret);
+    goto exit;
+  }
+
   /* delete old queue node */
   if (tx_queue->rl_shapers_mapping >= 0) {
     ret = rte_tm_node_delete(port_id, queue, &error);
     if (ret < 0) {
       err("%s(%d), node %d delete fail %d(%s)\n", __func__, port, queue, ret,
           mt_string_safe(error.message));
-      return ret;
+      goto exit;
     }
     tx_queue->rl_shapers_mapping = -1;
   }
@@ -727,7 +731,8 @@ static int dev_tx_queue_set_rl_rate(struct mt_interface* inf, uint16_t queue,
     shaper = dev_rl_shaper_get(inf, bps);
     if (!shaper) {
       err("%s(%d), rl shaper get fail for q %d\n", __func__, port, queue);
-      return -EIO;
+      ret = -EIO;
+      goto exit;
     }
     memset(&qp, 0, sizeof(qp));
     qp.shaper_profile_id = shaper->shaper_profile_id;
@@ -740,29 +745,64 @@ static int dev_tx_queue_set_rl_rate(struct mt_interface* inf, uint16_t queue,
       ret = rte_tm_node_add(port_id, queue, ST_TM_LAST_NONLEAF_NODE_ID_PF, 0, 1,
                             ST_TM_NONLEAF_NODES_NUM_PF, &qp, &error);
     }
-    if (ret < 0) {
+    if (ret) {
       err("%s(%d), q %d add fail %d(%s)\n", __func__, port, queue, ret,
           mt_string_safe(error.message));
-      return ret;
+      goto exit;
     }
+
     tx_queue->rl_shapers_mapping = shaper->idx;
     info("%s(%d), q %d link to shaper id %d(%" PRIu64 ")\n", __func__, port, queue,
          shaper->shaper_profile_id, shaper->rl_bps);
   }
-  rte_atomic32_set(&inf->resetting, true);
-  mt_pthread_mutex_lock(&inf->vf_cmd_mutex);
-  ret = rte_tm_hierarchy_commit(port_id, 1, &error);
-  mt_pthread_mutex_unlock(&inf->vf_cmd_mutex);
-  rte_atomic32_set(&inf->resetting, false);
-  if (ret < 0) {
-    err("%s(%d), commit error (%d)%s\n", __func__, port, ret,
-        mt_string_safe(error.message));
-    return ret;
+
+  if (inf->drv_info.drv_type == MT_DRV_IAVF) {
+    /* Note: IAVF behavior in DPDK 25.03:
+     * 1. If the hierarchy is committed on a started device, nodes cannot be removed after
+     * the commit.
+     * 2. After committing the hierarchy, the device must be restarted for changes to take
+     * effect.
+     */
+    ret = rte_eth_dev_start(port_id);
+    if (ret) {
+      err("%s(%d), start port %d fail %d\n", __func__, port, port_id, ret);
+      goto exit;
+    }
+
+    ret = rte_tm_hierarchy_commit(port_id, 1, &error);
+    if (ret) {
+      err("%s(%d), commit error (%d)%s\n", __func__, port, ret,
+          mt_string_safe(error.message));
+      goto exit;
+    }
+
+    ret = rte_eth_dev_stop(port_id);
+    if (ret) {
+      err("%s(%d), stop port %d fail %d\n", __func__, port, port_id, ret);
+      goto exit;
+    }
+
+  } else {
+    ret = rte_tm_hierarchy_commit(port_id, 1, &error);
+    if (ret) {
+      err("%s(%d), commit error (%d)%s\n", __func__, port, ret,
+          mt_string_safe(error.message));
+      goto exit;
+    }
+  }
+
+  ret = rte_eth_dev_start(port_id);
+  if (ret) {
+    err("%s(%d), start port %d fail %d\n", __func__, port, port_id, ret);
+    goto exit;
   }
 
   tx_queue->bps = bps;
 
-  return 0;
+exit:
+  mt_pthread_mutex_unlock(&inf->vf_cmd_mutex);
+  rte_atomic32_set(&inf->resetting, false);
+  return ret;
 }
 
 static int dev_stop_port(struct mt_interface* inf) {
@@ -1420,13 +1460,11 @@ static int dev_if_init_pacing(struct mt_interface* inf) {
       return ret;
     }
     /* IAVF require all q config with RL */
-    if (inf->drv_info.drv_type == MT_DRV_IAVF) {
+    if (inf->drv_info.drv_type == MT_DRV_IAVF)
       ret = dev_init_ratelimit_all(inf);
-    } else {
+    else
       ret = dev_tx_queue_set_rl_rate(inf, 0, ST_DEFAULT_RL_BPS);
-      if (ret >= 0) dev_tx_queue_set_rl_rate(inf, 0, 0);
-    }
-    if (ret < 0) { /* fallback to tsc if no rl */
+    if (ret) { /* fallback to tsc if no rl */
       if (auto_detect) {
         warn("%s(%d), fallback to tsc as rl init fail\n", __func__, port);
         inf->tx_pacing_way = ST21_TX_PACING_WAY_TSC;

lib/src/mt_main.h

@@ -617,7 +617,7 @@ struct mt_sch_mgr {
 };
 
 struct mt_pacing_train_result {
-  uint64_t input_bps;           /* input, byte per sec */
+  uint64_t input_bps;        /* input, byte per sec */
   uint64_t profiled_bps;     /* profiled result */
   float pacing_pad_interval; /* result */
 };

lib/src/mt_util.c

@@ -252,7 +252,7 @@ int mt_build_port_map(struct mtl_main_impl* impl, char** ports, enum mtl_port* m
 }
 
 int mt_pacing_train_pad_result_add(struct mtl_main_impl* impl, enum mtl_port port,
-                               uint64_t input_bps, float pad_interval) {
+                                   uint64_t input_bps, float pad_interval) {
   struct mt_pacing_train_result* ptr = &mt_if(impl, port)->pt_results[0];
 
   for (int i = 0; i < MT_MAX_RL_ITEMS; i++) {
@@ -267,7 +267,7 @@ int mt_pacing_train_pad_result_add(struct mtl_main_impl* impl, enum mtl_port por
 }
 
 int mt_pacing_train_pad_result_search(struct mtl_main_impl* impl, enum mtl_port port,
-                                  uint64_t rl_bps, float* pad_interval) {
+                                      uint64_t rl_bps, float* pad_interval) {
   struct mt_pacing_train_result* ptr = &mt_if(impl, port)->pt_results[0];
 
   for (int i = 0; i < MT_MAX_RL_ITEMS; i++) {
@@ -282,7 +282,7 @@ int mt_pacing_train_pad_result_search(struct mtl_main_impl* impl, enum mtl_port
 }
 
 int mt_pacing_train_bps_result_add(struct mtl_main_impl* impl, enum mtl_port port,
-                                     uint64_t input_bps, uint64_t profiled_bps) {
+                                   uint64_t input_bps, uint64_t profiled_bps) {
   struct mt_pacing_train_result* ptr = &mt_if(impl, port)->pt_results[0];
 
   for (int i = 0; i < MT_MAX_RL_ITEMS; i++) {
@@ -297,7 +297,7 @@ int mt_pacing_train_bps_result_add(struct mtl_main_impl* impl, enum mtl_port por
 }
 
 int mt_pacing_train_bps_result_search(struct mtl_main_impl* impl, enum mtl_port port,
-                                        uint64_t input_bps, uint64_t* profiled_bps) {
+                                      uint64_t input_bps, uint64_t* profiled_bps) {
   struct mt_pacing_train_result* ptr = &mt_if(impl, port)->pt_results[0];
 
   for (int i = 0; i < MT_MAX_RL_ITEMS; i++) {

lib/src/mt_util.h

@@ -59,12 +59,12 @@ void mt_mbuf_sanity_check(struct rte_mbuf** mbufs, uint16_t nb, char* tag);
 int mt_pacing_train_pad_result_add(struct mtl_main_impl* impl, enum mtl_port port,
                                    uint64_t input_bps, float pad_interval);
 int mt_pacing_train_pad_result_search(struct mtl_main_impl* impl, enum mtl_port port,
-                                  uint64_t input_bps, float* pad_interval);
+                                      uint64_t input_bps, float* pad_interval);
 
 int mt_pacing_train_bps_result_add(struct mtl_main_impl* impl, enum mtl_port port,
                                    uint64_t input_bps, uint64_t profiled_bps);
 int mt_pacing_train_bps_result_search(struct mtl_main_impl* impl, enum mtl_port port,
-                                  uint64_t input_bps, uint64_t* profiled_bps);
+                                      uint64_t input_bps, uint64_t* profiled_bps);
 
 enum mtl_port mt_port_by_name(struct mtl_main_impl* impl, const char* name);
 int mt_build_port_map(struct mtl_main_impl* impl, char** ports, enum mtl_port* maps,

lib/src/st2110/st_tx_audio_session.c

@@ -1252,7 +1252,7 @@ static int tx_audio_session_init_rl(struct mtl_main_impl* impl,
 
     uint64_t initial_bytes_per_sec = tx_audio_session_initial_rl_bps(s);
     int profiled = mt_pacing_train_bps_result_search(impl, port, initial_bytes_per_sec,
-                                                       &profiled_per_sec);
+                                                     &profiled_per_sec);
 
     /* pad pkt */
     rl_port->pad = mt_build_pad(impl, mt_sys_tx_mempool(impl, port), port,

lib/src/st2110/st_tx_video_session.c

@@ -427,17 +427,18 @@ static int tv_train_pacing(struct mtl_main_impl* impl, struct st_tx_video_sessio
   pkts_per_frame = pkts_per_frame * reactive;
   measured_bps = s->st20_pkt_size * pkts_per_sec * reactive;
 
+  /* If the measured speed is lower then expected. Set higher bps and retrain to add
+   * padding */
   if (measured_bps < rl_bps) {
-    /* The measured speed is lower then expected. Set higher bps and retrain to add padding */
-    info("%s(%d), measured bps %ld, set bps %ld, increasing rl bps\n", __func__, idx,
-      (uint64_t)measured_bps, rl_bps);
-
+    info("%s(%d), measured bps %ld is lower then set bps %ld\n", __func__, idx,
+         (uint64_t)measured_bps, rl_bps);
     if (!mt_pacing_train_bps_result_search(impl, port, rl_bps, &bps_to_set)) {
       err("%s(%d), measured speed is too low on already trained bps\n", __func__, idx);
       return -EINVAL;
     }
 
     bps_to_set = rl_bps / measured_bps * rl_bps;
+    info("%s(%d), increase bps to %ld\n", __func__, idx, bps_to_set);
     mt_pacing_train_bps_result_add(impl, port, rl_bps, bps_to_set);
     mt_txq_set_tx_bps(queue, bps_to_set);
     ret = tv_train_pacing(impl, s, s_port);
@@ -3874,7 +3875,8 @@ int st20_tx_queue_fatal_error(struct mtl_main_impl* impl,
   struct mt_txq_flow flow;
   memset(&flow, 0, sizeof(flow));
   flow.bytes_per_sec = tv_rl_bps(s);
-  mt_pacing_train_bps_result_search(impl, s_port, flow.bytes_per_sec, &flow.bytes_per_sec);
+  mt_pacing_train_bps_result_search(impl, s_port, flow.bytes_per_sec,
+                                    &flow.bytes_per_sec);
   mtl_memcpy(&flow.dip_addr, &s->ops.dip_addr[s_port], MTL_IP_ADDR_LEN);
   flow.dst_port = s->ops.udp_port[s_port];
   s->queue[s_port] = mt_txq_get(impl, port, &flow);