Fix: handling case where RL is lower then expected (#1205)

1. Fix handling case where RL is lower then expected
2. Update ice driver to version 2.2.8

---------

Signed-off-by: Szumski, Tomasz<[email protected]>
Co-authored-by: Kasiewicz, Marek <[email protected]>

Author: tszumski

doc/e810.md

@@ -8,41 +8,41 @@ The Media Transport Library relies on certain rate limit patches that are curren
 
 > **Note:** Occasionally, after a system reboot, the operating system (Ubuntu) may automatically upgrade to a new kernel version. In such cases, it is important to remember to rebuild the driver to ensure compatibility with the new kernel version.
 
-### 1.1. Download Driver Version 1.16.3
+### 1.1. Download Driver Version 2.2.8
 
-You can download the CVL 1.16.3 driver source code directly with the command:
+You can download the CVL 2.2.8 driver source code directly with the command:
 
 ```bash
-wget https://downloadmirror.intel.com/843928/ice-1.16.3.tar.gz
+wget https://downloadmirror.intel.com/859252/ice-2.2.8.tar.gz
 ```
 
-(As an alternative, you can also visit <https://www.intel.com/content/www/us/en/download/19630/intel-network-adapter-driver-for-e810-series-devices-under-linux.html> and select `Download ice-1.16.3.tar.gz`.
+(As an alternative, you can also visit <https://www.intel.com/content/www/us/en/download/19630/intel-network-adapter-driver-for-e810-series-devices-under-linux.html> and select `Download ice-2.2.8.tar.gz`.
 
-Make sure to select the 1.16.3 version from the available version options, as there may be newer versions available. It is important to note that version 1.16.3 is the latest version we have verified. Typically, we revisit driver version upgrades on a quarterly schedule.)
+Make sure to select the 2.2.8 version from the available version options, as there may be newer versions available. It is important to note that version 2.2.8 is the latest version we have verified. Typically, we revisit driver version upgrades on a quarterly schedule.)
 
-The steps are based on downloading the file `ice-1.16.3.tar.gz`.
+The steps are based on downloading the file `ice-2.2.8.tar.gz`.
 
-### 1.2. Unzip 1.16.3 Driver and Enter into the Source Code Directory
+### 1.2. Unzip 2.2.8 Driver and Enter into the Source Code Directory
 
 ```bash
-tar xvzf ice-1.16.3.tar.gz
-cd ice-1.16.3
+tar xvzf ice-2.2.8.tar.gz
+cd ice-2.2.8
 ```
 
-### 1.3. Patch 1.16.3 Driver with Rate Limit Patches
+### 1.3. Patch 2.2.8 Driver with Rate Limit Patches
 
-Apply all the patches placed in [directory for ice_driver](../patches/ice_drv/1.16.3/).
+Apply all the patches placed in [directory for ice_driver](../patches/ice_drv/2.2.8/).
 
 ```bash
 git init
 git add .
-git commit -m "init version 1.16.3"
-git am $mtl_source_code/patches/ice_drv/1.16.3/*.patch
+git commit -m "init version 2.2.8"
+git am $mtl_source_code/patches/ice_drv/2.2.8/*.patch
 ```
 
 Note: The variable `$mtl_source_code` should be set to the top directory of the Media Transport Library source code. Please ensure that it is correctly configured. Additionally, when running the `git am` command, please verify that it executes without any errors.
 
-Use `git log` to check if the latest commit is `version: update to Kahawai_1.16.3_20250327`.
+Use `git log` to check if the latest commit is `version: update to Kahawai_2.2.8`.
 
 ### 1.4. Build and Install the Driver
 
@@ -85,7 +85,7 @@ sudo dmesg | grep "Intel(R) Ethernet Connection E800 Series Linux Driver"
 ```
 
 ```text
-ice: Intel(R) Ethernet Connection E800 Series Linux Driver - version Kahawai_1.16.3_20250327
+ice: Intel(R) Ethernet Connection E800 Series Linux Driver - version Kahawai_2.2.8
 ```
 
 Similar steps to confirm the DDP version.
@@ -150,7 +150,7 @@ A correct setup should have an output similar to the following:
 
 ```text
 driver: ice
-version: Kahawai_1.16.3_20250327
+version: Kahawai_2.2.8
 firmware-version: 4.40 0x8001c967 1.3534.0
 expansion-rom-version:
 bus-info: 0000:af:00.0

include/st20_api.h

@@ -62,9 +62,11 @@ extern "C" {
 #define ST20_TX_FLAG_ENABLE_VSYNC (MTL_BIT32(5))
 /**
  * Flag bit in flags of struct st20_tx_ops.
- * If disable the static RL pad interval profiling.
+ * If enable the static RL pad interval profiling.
+ * Static padding is trained only for e810, it is not recommended to use this flag
+ * for other NICs.
  */
-#define ST20_TX_FLAG_DISABLE_STATIC_PAD_P (MTL_BIT32(6))
+#define ST20_TX_FLAG_ENABLE_STATIC_PAD_P (MTL_BIT32(6))
 /**
  * Flag bit in flags of struct st20_tx_ops.
  * If enable the rtcp.

include/st_pipeline_api.h

@@ -446,9 +446,11 @@ enum st20p_tx_flag {
    */
   ST20P_TX_FLAG_ENABLE_VSYNC = (MTL_BIT32(5)),
   /**
-   * If disable the static RL pad interval profiling.
+   * If enable the static RL pad interval profiling.
+   * Static padding is trained only for e810, it is not recommended to use this flag
+   * for other NICs.
    */
-  ST20P_TX_FLAG_DISABLE_STATIC_PAD_P = (MTL_BIT32(6)),
+  ST20P_TX_FLAG_ENABLE_STATIC_PAD_P = (MTL_BIT32(6)),
   /**
    * If enable the rtcp.
    */

lib/src/mt_main.h

@@ -48,7 +48,7 @@
 #define MT_MAX_SCH_NUM (18) /* max 18 scheduler lcore */
 
 /* max RL items */
-#define MT_MAX_RL_ITEMS (64)
+#define MT_MAX_RL_ITEMS (128)
 
 #define MT_ARP_ENTRY_MAX (60)
 
@@ -616,13 +616,9 @@ struct mt_sch_mgr {
   enum mt_lcore_type local_lcores_type[RTE_MAX_LCORE];
 };
 
-struct mt_audio_pacing_train_result {
-  uint64_t input_bps;    /* input, byte per sec */
-  uint64_t profiled_bps; /* profiled result */
-};
-
 struct mt_pacing_train_result {
-  uint64_t rl_bps;           /* input, byte per sec */
+  uint64_t input_bps;        /* input, byte per sec */
+  uint64_t profiled_bps;     /* profiled result */
   float pacing_pad_interval; /* result */
 };
 
@@ -742,8 +738,6 @@ struct mt_interface {
   bool tx_rl_root_active;
   /* video rl pacing train result */
   struct mt_pacing_train_result pt_results[MT_MAX_RL_ITEMS];
-  /* audio rl pacing train result */
-  struct mt_audio_pacing_train_result audio_pt_results[MT_MAX_RL_ITEMS];
 
   /* function ops per interface(pf/vf) */
   uint64_t (*ptp_get_time_fn)(struct mtl_main_impl* impl, enum mtl_port port);

lib/src/mt_util.c

@@ -251,13 +251,13 @@ int mt_build_port_map(struct mtl_main_impl* impl, char** ports, enum mtl_port* m
   return 0;
 }
 
-int mt_pacing_train_result_add(struct mtl_main_impl* impl, enum mtl_port port,
-                               uint64_t rl_bps, float pad_interval) {
+int mt_pacing_train_pad_result_add(struct mtl_main_impl* impl, enum mtl_port port,
+                                   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++) {
-    if (ptr[i].rl_bps) continue;
-    ptr[i].rl_bps = rl_bps;
+    if (ptr[i].input_bps) continue;
+    ptr[i].input_bps = input_bps;
     ptr[i].pacing_pad_interval = pad_interval;
     return 0;
   }
@@ -266,12 +266,12 @@ int mt_pacing_train_result_add(struct mtl_main_impl* impl, enum mtl_port port,
   return -ENOMEM;
 }
 
-int mt_pacing_train_result_search(struct mtl_main_impl* impl, enum mtl_port port,
-                                  uint64_t rl_bps, float* pad_interval) {
+int mt_pacing_train_pad_result_search(struct mtl_main_impl* impl, enum mtl_port port,
+                                      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++) {
-    if (rl_bps == ptr[i].rl_bps) {
+    if (rl_bps == ptr[i].input_bps && ptr[i].pacing_pad_interval) {
       *pad_interval = ptr[i].pacing_pad_interval;
       return 0;
     }
@@ -281,9 +281,9 @@ int mt_pacing_train_result_search(struct mtl_main_impl* impl, enum mtl_port port
   return -EINVAL;
 }
 
-int mt_audio_pacing_train_result_add(struct mtl_main_impl* impl, enum mtl_port port,
-                                     uint64_t input_bps, uint64_t profiled_bps) {
-  struct mt_audio_pacing_train_result* ptr = &mt_if(impl, port)->audio_pt_results[0];
+int mt_pacing_train_bps_result_add(struct mtl_main_impl* impl, enum mtl_port port,
+                                   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++) {
     if (ptr[i].input_bps) continue;
@@ -296,12 +296,12 @@ int mt_audio_pacing_train_result_add(struct mtl_main_impl* impl, enum mtl_port p
   return -ENOMEM;
 }
 
-int mt_audio_pacing_train_result_search(struct mtl_main_impl* impl, enum mtl_port port,
-                                        uint64_t input_bps, uint64_t* profiled_bps) {
-  struct mt_audio_pacing_train_result* ptr = &mt_if(impl, port)->audio_pt_results[0];
+int mt_pacing_train_bps_result_search(struct mtl_main_impl* impl, enum mtl_port port,
+                                      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++) {
-    if (input_bps == ptr[i].input_bps) {
+    if (input_bps == ptr[i].input_bps && ptr[i].profiled_bps) {
       *profiled_bps = ptr[i].profiled_bps;
       return 0;
     }

lib/src/mt_util.h

@@ -56,15 +56,15 @@ int mt_ring_dequeue_clean(struct rte_ring* ring);
 
 void mt_mbuf_sanity_check(struct rte_mbuf** mbufs, uint16_t nb, char* tag);
 
-int mt_pacing_train_result_add(struct mtl_main_impl* impl, enum mtl_port port,
-                               uint64_t rl_bps, float pad_interval);
-int mt_pacing_train_result_search(struct mtl_main_impl* impl, enum mtl_port port,
-                                  uint64_t rl_bps, float* pad_interval);
-
-int mt_audio_pacing_train_result_add(struct mtl_main_impl* impl, enum mtl_port port,
-                                     uint64_t input_bps, uint64_t profiled_bps);
-int mt_audio_pacing_train_result_search(struct mtl_main_impl* impl, enum mtl_port port,
-                                        uint64_t input_bps, uint64_t* profiled_bps);
+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);
+
+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);
 
 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/pipeline/st20_pipeline_tx.c

@@ -299,8 +299,8 @@ static int tx_st20p_create_transport(struct mtl_main_impl* impl, struct st20p_tx
   if (ops->flags & ST20P_TX_FLAG_USER_TIMESTAMP)
     ops_tx.flags |= ST20_TX_FLAG_USER_TIMESTAMP;
   if (ops->flags & ST20P_TX_FLAG_ENABLE_VSYNC) ops_tx.flags |= ST20_TX_FLAG_ENABLE_VSYNC;
-  if (ops->flags & ST20P_TX_FLAG_DISABLE_STATIC_PAD_P)
-    ops_tx.flags |= ST20_TX_FLAG_DISABLE_STATIC_PAD_P;
+  if (ops->flags & ST20P_TX_FLAG_ENABLE_STATIC_PAD_P)
+    ops_tx.flags |= ST20_TX_FLAG_ENABLE_STATIC_PAD_P;
   if (ops->flags & ST20P_TX_FLAG_ENABLE_RTCP) {
     ops_tx.flags |= ST20_TX_FLAG_ENABLE_RTCP;
     ops_tx.rtcp = ops->rtcp;

lib/src/st2110/st_tx_audio_session.c

@@ -1257,8 +1257,8 @@ static int tx_audio_session_init_rl(struct mtl_main_impl* impl,
     port = mt_port_logic2phy(s->port_maps, i);
 
     uint64_t initial_bytes_per_sec = tx_audio_session_initial_rl_bps(s);
-    int profiled = mt_audio_pacing_train_result_search(impl, port, initial_bytes_per_sec,
-                                                       &profiled_per_sec);
+    int profiled = mt_pacing_train_bps_result_search(impl, port, initial_bytes_per_sec,
+                                                     &profiled_per_sec);
 
     /* pad pkt */
     rl_port->pad = mt_build_pad(impl, mt_sys_tx_mempool(impl, port), port,
@@ -1297,7 +1297,7 @@ static int tx_audio_session_init_rl(struct mtl_main_impl* impl,
           return -EIO;
         }
 
-        mt_audio_pacing_train_result_add(impl, port, initial_bytes_per_sec, trained);
+        mt_pacing_train_bps_result_add(impl, port, initial_bytes_per_sec, trained);
         info("%s(%d), trained bytes_per_sec %" PRIu64 "\n", __func__, idx, trained);
         int ret = mt_txq_set_tx_bps(rl_port->queue[j], trained);
         if (ret < 0) {

lib/src/st2110/st_tx_video_session.c

@@ -322,14 +322,16 @@ static int tv_train_pacing(struct mtl_main_impl* impl, struct st_tx_video_sessio
   float pad_interval;
   uint64_t rl_bps = tv_rl_bps(s);
   uint64_t train_start_time, train_end_time;
+  double measured_bps;
+  uint64_t bps_to_set;
 
   uint16_t resolved = s->ops.pad_interval;
   if (resolved) {
     s->pacing.pad_interval = resolved;
     info("%s(%d), user customized pad_interval %u\n", __func__, idx, resolved);
     return 0;
   }
-  if (!(s->ops.flags & ST20_TX_FLAG_DISABLE_STATIC_PAD_P)) {
+  if ((s->ops.flags & ST20_TX_FLAG_ENABLE_STATIC_PAD_P)) {
     resolved = st20_pacing_static_profiling(impl, s, s_port);
     if (resolved) {
       s->pacing.pad_interval = resolved;
@@ -338,7 +340,7 @@ static int tv_train_pacing(struct mtl_main_impl* impl, struct st_tx_video_sessio
     }
   }
 
-  ret = mt_pacing_train_result_search(impl, port, rl_bps, &pad_interval);
+  ret = mt_pacing_train_pad_result_search(impl, port, rl_bps, &pad_interval);
   if (ret >= 0) {
     s->pacing.pad_interval = pad_interval;
     info("%s(%d), use pre-train pad_interval %f\n", __func__, idx, pad_interval);
@@ -423,10 +425,30 @@ static int tv_train_pacing(struct mtl_main_impl* impl, struct st_tx_video_sessio
     reactive = (s->ops.height == 480) ? 487.0 / 525.0 : 576.0 / 625.0;
   }
   pkts_per_frame = pkts_per_frame * reactive;
-  if (pkts_per_frame < s->st20_total_pkts) {
-    err("%s(%d), error pkts_per_frame %f, st20_total_pkts %d\n", __func__, idx,
-        pkts_per_frame, s->st20_total_pkts);
-    return -EINVAL;
+  measured_bps = s->st20_pkt_size * pkts_per_sec * reactive;
+
+  /* If the measured speed is lower than expected. Set higher bps and retrain to add
+   * padding */
+  if (measured_bps < rl_bps) {
+    info("%s(%d), measured bps %" PRIu64 " is lower then set bps %" PRIu64 "\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;
+    }
+
+/* Slightly increase the target bitrate to compensate for measurement inaccuracies,
+ * rounding errors, and system overhead. This helps ensure the actual transmission bitrate
+ * meets or exceeds the required rate
+ */
+#define INCREASE_BPS_FACTOR 1.005
+    bps_to_set = INCREASE_BPS_FACTOR * (rl_bps * rl_bps) / measured_bps;
+    info("%s(%d), increase bps to %" PRIu64 "\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);
+    return ret;
   }
 
   pad_interval = (float)s->st20_total_pkts / (pkts_per_frame - s->st20_total_pkts);
@@ -437,7 +459,7 @@ static int tv_train_pacing(struct mtl_main_impl* impl, struct st_tx_video_sessio
   }
 
   s->pacing.pad_interval = pad_interval;
-  mt_pacing_train_result_add(impl, port, rl_bps, pad_interval);
+  mt_pacing_train_pad_result_add(impl, port, rl_bps, pad_interval);
   train_end_time = mt_get_tsc(impl);
   info("%s(%d,%d), trained pad_interval %f pkts_per_frame %f with time %fs\n", __func__,
        idx, s_port, pad_interval, pkts_per_frame,
@@ -2526,6 +2548,7 @@ static int tv_init_hw(struct mtl_main_impl* impl, struct st_tx_video_sessions_mg
     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, i, flow.bytes_per_sec, &flow.bytes_per_sec);
     mtl_memcpy(&flow.dip_addr, &s->ops.dip_addr[i], MTL_IP_ADDR_LEN);
     flow.dst_port = s->ops.udp_port[i];
     if (ST21_TX_PACING_WAY_TSN == s->pacing_way[i])
@@ -3858,6 +3881,8 @@ 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);
   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);

patches/ice_drv/2.2.8/0001-vf-support-kahawai-runtime-rl-queue.patch

@@ -0,0 +1,55 @@
+From cca447de21727b653dd5251ceb960aa0a8d9a447 Mon Sep 17 00:00:00 2001
+From: Szumski Tomasz <[email protected]>
+Date: Fri, 11 Jul 2025 13:32:58 +0200
+Subject: [PATCH 1/4] vf: support kahawai runtime rl queue
+
+---
+ src/ice_virtchnl.c | 12 +++++++++---
+ 1 file changed, 9 insertions(+), 3 deletions(-)
+
+diff --git a/src/ice_virtchnl.c b/src/ice_virtchnl.c
+index 0eb4250..040a1b4 100644
+--- a/src/ice_virtchnl.c
++++ b/src/ice_virtchnl.c
+@@ -3657,6 +3657,11 @@ static int ice_vc_cfg_q_bw(struct ice_vf *vf, u8 *msg)
+ 
+ 	memcpy(vf->qs_bw, qs_bw, len);
+ 
++	/* for kahawai runtime rl */
++	if (vf->qs_bw)
++		if (ice_vf_cfg_qs_bw(vf, qbw->num_queues))
++			v_ret = VIRTCHNL_STATUS_ERR_PARAM;
++
+ err_bw:
+ 	kfree(qs_bw);
+ 
+@@ -3835,12 +3840,12 @@ static int ice_vc_cfg_qs_msg(struct ice_vf *vf, u8 *msg)
+ 		if (qpi->txq.ring_len > 0) {
+ 			vsi->tx_rings[q_idx]->dma = qpi->txq.dma_ring_addr;
+ 			vsi->tx_rings[q_idx]->count = qpi->txq.ring_len;
+-
++#if 0 /* for kahawai runtime rl */
+ 			/* Disable any existing queue first */
+ 			if (ice_vf_vsi_dis_single_txq(vf, vsi, q_idx,
+ 						      qpi->txq.queue_id))
+ 				goto error_param;
+-
++#endif
+ 			/* Configure a queue with the requested settings */
+ 			if (ice_vsi_cfg_single_txq(vsi, vsi->tx_rings, q_idx)) {
+ 				dev_warn(ice_pf_to_dev(pf), "VF-%d failed to configure TX queue %d\n",
+@@ -3929,9 +3934,10 @@ static int ice_vc_cfg_qs_msg(struct ice_vf *vf, u8 *msg)
+ 			}
+ 		}
+ 	}
+-
++#if 0 /* for kahawai runtime rl */
+ 	if (ice_vf_cfg_qs_bw(vf, qci->num_queue_pairs))
+ 		goto error_param;
++#endif
+ #ifdef HAVE_NETDEV_UPPER_INFO
+ 
+ 	if (lag && lag->bonded && lag->primary &&
+-- 
+2.34.1
+