patch 'net/mlx5: fix counters map in bonding mode' has been queued to stable release 22.11.5

[luca.boccassi@gmail.com]

Hi,

FYI, your patch has been queued to stable release 22.11.5

Note it hasn't been pushed to http://dpdk.org/browse/dpdk-stable yet.
It will be pushed if I get no objections before 03/16/24. So please
shout if anyone has objections.

Also note that after the patch there's a diff of the upstream commit vs the
patch applied to the branch. This will indicate if there was any rebasing
needed to apply to the stable branch. If there were code changes for rebasing
(ie: not only metadata diffs), please double check that the rebase was
correctly done.

Queued patches are on a temporary branch at:
https://github.com/bluca/dpdk-stable

This queued commit can be viewed at:
https://github.com/bluca/dpdk-stable/commit/cc21b4a2d2d343307af178b1fd31eed1df0b1979

Thanks.

Luca Boccassi

---
From cc21b4a2d2d343307af178b1fd31eed1df0b1979 Mon Sep 17 00:00:00 2001
From: Bing Zhao <bingz@nvidia.com>
Date: Thu, 29 Feb 2024 11:34:56 +0200
Subject: [PATCH] net/mlx5: fix counters map in bonding mode

[ upstream commit a687c3e658c2d889052089af8340bc0b9299c856 ]

In the HW-LAG mode, there is only one mlx5 IB device with 2 ETH
interfaces. In theory, the settings on both ports should be the same.
But in the real life, some inconsistent settings may be done by the
user and the PMD is not aware of this.

In the previous implementation, the xstats map was generated from the
information fetched on the 1st port of a bonding interface. If the
2nd port had a different settings, the number and the order of the
counters may differ from that of the 1st one. The ioctl() call may
corrupt the user buffers (copy_to_user) and cause a crash.

The commit will change the map between the driver counters to the
PMD user counters.
  1. Switch the inner and outer loop to speed up the initialization
     time AMAP - since there will be >300 counters returned from the
     driver.
  2. Generate an unique map for both ports in LAG mode.
    a. Scan the 1st port and find the supported counters' strings,
       then add to the map.
    b. In bonding, scan the 2nd port and find the strings. If one is
       already in the map, use the index. Or append to the next free
       slot.
    c. Append the device counters that needs to be fetched via sysfs
       or Devx command. This kind of counter(s) is unique per IB
       device.

After querying the statistics from the driver, the value will be read
from the proper offset in the "struct ethtool_stats" and then added
into the output array based on the map information. In bonding mode,
the statistics from both ports will be accumulated if the counters
are valid on both ports.

Compared to the system call or Devx command, the overhead introduced
by the extra index comparison is light and should not cause a
significant degradation.

The application should ensure that the port settings should not be
changed out of the DPDK application dynamically in most cases. Or
else the change cannot be notified and the counters map might not
be valid when the number doesn't change but the counters set had
changed. A device restart will help to re-initialize the map from
scrath.

Fixes: 7ed15acdcd69 ("net/mlx5: improve xstats of bonding port")

Signed-off-by: Bing Zhao <bingz@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_ethdev_os.c   | 249 +++++++++++++++-------
 drivers/net/mlx5/mlx5.h                   |  15 +-
 drivers/net/mlx5/mlx5_stats.c             |  58 +++--
 drivers/net/mlx5/windows/mlx5_ethdev_os.c |  22 +-
 4 files changed, 242 insertions(+), 102 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_ethdev_os.c b/drivers/net/mlx5/linux/mlx5_ethdev_os.c
index dd5a0c546d..0ee8c58ba7 100644
--- a/drivers/net/mlx5/linux/mlx5_ethdev_os.c
+++ b/drivers/net/mlx5/linux/mlx5_ethdev_os.c
@@ -1286,13 +1286,16 @@ _mlx5_os_read_dev_counters(struct rte_eth_dev *dev, int pf, uint64_t *stats)
 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
 	unsigned int i;
 	struct ifreq ifr;
-	unsigned int stats_sz = xstats_ctrl->stats_n * sizeof(uint64_t);
+	unsigned int max_stats_n = RTE_MAX(xstats_ctrl->stats_n, xstats_ctrl->stats_n_2nd);
+	unsigned int stats_sz = max_stats_n * sizeof(uint64_t);
 	unsigned char et_stat_buf[sizeof(struct ethtool_stats) + stats_sz];
 	struct ethtool_stats *et_stats = (struct ethtool_stats *)et_stat_buf;
 	int ret;
+	uint16_t i_idx, o_idx;
 
 	et_stats->cmd = ETHTOOL_GSTATS;
-	et_stats->n_stats = xstats_ctrl->stats_n;
+	/* Pass the maximum value, the driver may ignore this. */
+	et_stats->n_stats = max_stats_n;
 	ifr.ifr_data = (caddr_t)et_stats;
 	if (pf >= 0)
 		ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[pf].ifname,
@@ -1305,21 +1308,34 @@ _mlx5_os_read_dev_counters(struct rte_eth_dev *dev, int pf, uint64_t *stats)
 			dev->data->port_id);
 		return ret;
 	}
-	for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) {
-		if (xstats_ctrl->info[i].dev)
-			continue;
-		stats[i] += (uint64_t)
-			    et_stats->data[xstats_ctrl->dev_table_idx[i]];
+	if (pf <= 0) {
+		for (i = 0; i != xstats_ctrl->mlx5_stats_n; i++) {
+			i_idx = xstats_ctrl->dev_table_idx[i];
+			if (i_idx == UINT16_MAX || xstats_ctrl->info[i].dev)
+				continue;
+			o_idx = xstats_ctrl->xstats_o_idx[i];
+			stats[o_idx] += (uint64_t)et_stats->data[i_idx];
+		}
+	} else {
+		for (i = 0; i != xstats_ctrl->mlx5_stats_n; i++) {
+			i_idx = xstats_ctrl->dev_table_idx_2nd[i];
+			if (i_idx == UINT16_MAX)
+				continue;
+			o_idx = xstats_ctrl->xstats_o_idx_2nd[i];
+			stats[o_idx] += (uint64_t)et_stats->data[i_idx];
+		}
 	}
 	return 0;
 }
 
-/**
+/*
  * Read device counters.
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param[out] stats
+ * @param bond_master
+ *   Indicate if the device is a bond master.
+ * @param stats
  *   Counters table output buffer.
  *
  * @return
@@ -1327,7 +1343,7 @@ _mlx5_os_read_dev_counters(struct rte_eth_dev *dev, int pf, uint64_t *stats)
  *   rte_errno is set.
  */
 int
-mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
+mlx5_os_read_dev_counters(struct rte_eth_dev *dev, bool bond_master, uint64_t *stats)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
@@ -1335,7 +1351,7 @@ mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
 
 	memset(stats, 0, sizeof(*stats) * xstats_ctrl->mlx5_stats_n);
 	/* Read ifreq counters. */
-	if (priv->master && priv->pf_bond >= 0) {
+	if (bond_master) {
 		/* Sum xstats from bonding device member ports. */
 		for (i = 0; i < priv->sh->bond.n_port; i++) {
 			ret = _mlx5_os_read_dev_counters(dev, i, stats);
@@ -1347,13 +1363,17 @@ mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
 		if (ret)
 			return ret;
 	}
-	/* Read IB counters. */
-	for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) {
+	/*
+	 * Read IB counters.
+	 * The counters are unique per IB device but not per net IF.
+	 * In bonding mode, getting the stats name only from 1 port is enough.
+	 */
+	for (i = 0; i != xstats_ctrl->mlx5_stats_n; i++) {
 		if (!xstats_ctrl->info[i].dev)
 			continue;
 		/* return last xstats counter if fail to read. */
 		if (mlx5_os_read_dev_stat(priv, xstats_ctrl->info[i].ctr_name,
-			    &stats[i]) == 0)
+					  &stats[i]) == 0)
 			xstats_ctrl->xstats[i] = stats[i];
 		else
 			stats[i] = xstats_ctrl->xstats[i];
@@ -1361,18 +1381,24 @@ mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
 	return ret;
 }
 
-/**
+/*
  * Query the number of statistics provided by ETHTOOL.
  *
  * @param dev
  *   Pointer to Ethernet device.
+ * @param bond_master
+ *   Indicate if the device is a bond master.
+ * @param n_stats
+ *   Pointer to number of stats to store.
+ * @param n_stats_sec
+ *   Pointer to number of stats to store for the 2nd port of the bond.
  *
  * @return
- *   Number of statistics on success, negative errno value otherwise and
- *   rte_errno is set.
+ *   0 on success, negative errno value otherwise and rte_errno is set.
  */
 int
-mlx5_os_get_stats_n(struct rte_eth_dev *dev)
+mlx5_os_get_stats_n(struct rte_eth_dev *dev, bool bond_master,
+		    uint16_t *n_stats, uint16_t *n_stats_sec)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct ethtool_drvinfo drvinfo;
@@ -1381,18 +1407,34 @@ mlx5_os_get_stats_n(struct rte_eth_dev *dev)
 
 	drvinfo.cmd = ETHTOOL_GDRVINFO;
 	ifr.ifr_data = (caddr_t)&drvinfo;
-	if (priv->master && priv->pf_bond >= 0)
-		/* Bonding PF. */
+	/* Bonding PFs. */
+	if (bond_master) {
 		ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[0].ifname,
 					   SIOCETHTOOL, &ifr);
-	else
+		if (ret) {
+			DRV_LOG(WARNING, "bonding port %u unable to query number of"
+				" statistics for the 1st slave, %d", PORT_ID(priv), ret);
+			return ret;
+		}
+		*n_stats = drvinfo.n_stats;
+		ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[1].ifname,
+					   SIOCETHTOOL, &ifr);
+		if (ret) {
+			DRV_LOG(WARNING, "bonding port %u unable to query number of"
+				" statistics for the 2nd slave, %d", PORT_ID(priv), ret);
+			return ret;
+		}
+		*n_stats_sec = drvinfo.n_stats;
+	} else {
 		ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
-	if (ret) {
-		DRV_LOG(WARNING, "port %u unable to query number of statistics",
-			dev->data->port_id);
-		return ret;
+		if (ret) {
+			DRV_LOG(WARNING, "port %u unable to query number of statistics",
+				PORT_ID(priv));
+			return ret;
+		}
+		*n_stats = drvinfo.n_stats;
 	}
-	return drvinfo.n_stats;
+	return 0;
 }
 
 static const struct mlx5_counter_ctrl mlx5_counters_init[] = {
@@ -1578,6 +1620,101 @@ static const struct mlx5_counter_ctrl mlx5_counters_init[] = {
 
 static const unsigned int xstats_n = RTE_DIM(mlx5_counters_init);
 
+static int
+mlx5_os_get_stats_strings(struct rte_eth_dev *dev, bool bond_master,
+			  struct ethtool_gstrings *strings,
+			  uint32_t stats_n, uint32_t stats_n_2nd)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
+	struct ifreq ifr;
+	int ret;
+	uint32_t i, j, idx;
+
+	/* Ensure no out of bounds access before. */
+	MLX5_ASSERT(xstats_n <= MLX5_MAX_XSTATS);
+	strings->cmd = ETHTOOL_GSTRINGS;
+	strings->string_set = ETH_SS_STATS;
+	strings->len = stats_n;
+	ifr.ifr_data = (caddr_t)strings;
+	if (bond_master)
+		ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[0].ifname,
+					   SIOCETHTOOL, &ifr);
+	else
+		ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
+	if (ret) {
+		DRV_LOG(WARNING, "port %u unable to get statistic names with %d",
+			PORT_ID(priv), ret);
+		return ret;
+	}
+	/* Reorganize the orders to reduce the iterations. */
+	for (j = 0; j < xstats_n; j++) {
+		xstats_ctrl->dev_table_idx[j] = UINT16_MAX;
+		for (i = 0; i < stats_n; i++) {
+			const char *curr_string =
+				(const char *)&strings->data[i * ETH_GSTRING_LEN];
+
+			if (!strcmp(mlx5_counters_init[j].ctr_name, curr_string)) {
+				idx = xstats_ctrl->mlx5_stats_n++;
+				xstats_ctrl->dev_table_idx[j] = i;
+				xstats_ctrl->xstats_o_idx[j] = idx;
+				xstats_ctrl->info[idx] = mlx5_counters_init[j];
+			}
+		}
+	}
+	if (!bond_master) {
+		/* Add dev counters, unique per IB device. */
+		for (j = 0; j != xstats_n; j++) {
+			if (mlx5_counters_init[j].dev) {
+				idx = xstats_ctrl->mlx5_stats_n++;
+				xstats_ctrl->info[idx] = mlx5_counters_init[j];
+				xstats_ctrl->hw_stats[idx] = 0;
+			}
+		}
+		return 0;
+	}
+
+	strings->len = stats_n_2nd;
+	ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[1].ifname,
+				   SIOCETHTOOL, &ifr);
+	if (ret) {
+		DRV_LOG(WARNING, "port %u unable to get statistic names for 2nd slave with %d",
+			PORT_ID(priv), ret);
+		return ret;
+	}
+	/* The 2nd slave port may have a different strings set, based on the configuration. */
+	for (j = 0; j != xstats_n; j++) {
+		xstats_ctrl->dev_table_idx_2nd[j] = UINT16_MAX;
+		for (i = 0; i != stats_n_2nd; i++) {
+			const char *curr_string =
+				(const char *)&strings->data[i * ETH_GSTRING_LEN];
+
+			if (!strcmp(mlx5_counters_init[j].ctr_name, curr_string)) {
+				xstats_ctrl->dev_table_idx_2nd[j] = i;
+				if (xstats_ctrl->dev_table_idx[j] != UINT16_MAX) {
+					/* Already mapped in the 1st slave port. */
+					idx = xstats_ctrl->xstats_o_idx[j];
+					xstats_ctrl->xstats_o_idx_2nd[j] = idx;
+				} else {
+					/* Append the new items to the end of the map. */
+					idx = xstats_ctrl->mlx5_stats_n++;
+					xstats_ctrl->xstats_o_idx_2nd[j] = idx;
+					xstats_ctrl->info[idx] = mlx5_counters_init[j];
+				}
+			}
+		}
+	}
+	/* Dev counters are always at the last now. */
+	for (j = 0; j != xstats_n; j++) {
+		if (mlx5_counters_init[j].dev) {
+			idx = xstats_ctrl->mlx5_stats_n++;
+			xstats_ctrl->info[idx] = mlx5_counters_init[j];
+			xstats_ctrl->hw_stats[idx] = 0;
+		}
+	}
+	return 0;
+}
+
 /**
  * Init the structures to read device counters.
  *
@@ -1590,76 +1727,44 @@ mlx5_os_stats_init(struct rte_eth_dev *dev)
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
 	struct mlx5_stats_ctrl *stats_ctrl = &priv->stats_ctrl;
-	unsigned int i;
-	unsigned int j;
-	struct ifreq ifr;
 	struct ethtool_gstrings *strings = NULL;
-	unsigned int dev_stats_n;
+	uint16_t dev_stats_n = 0;
+	uint16_t dev_stats_n_2nd = 0;
+	unsigned int max_stats_n;
 	unsigned int str_sz;
 	int ret;
+	bool bond_master = (priv->master && priv->pf_bond >= 0);
 
 	/* So that it won't aggregate for each init. */
 	xstats_ctrl->mlx5_stats_n = 0;
-	ret = mlx5_os_get_stats_n(dev);
+	ret = mlx5_os_get_stats_n(dev, bond_master, &dev_stats_n, &dev_stats_n_2nd);
 	if (ret < 0) {
 		DRV_LOG(WARNING, "port %u no extended statistics available",
 			dev->data->port_id);
 		return;
 	}
-	dev_stats_n = ret;
+	max_stats_n = RTE_MAX(dev_stats_n, dev_stats_n_2nd);
 	/* Allocate memory to grab stat names and values. */
-	str_sz = dev_stats_n * ETH_GSTRING_LEN;
+	str_sz = max_stats_n * ETH_GSTRING_LEN;
 	strings = (struct ethtool_gstrings *)
 		  mlx5_malloc(0, str_sz + sizeof(struct ethtool_gstrings), 0,
 			      SOCKET_ID_ANY);
 	if (!strings) {
 		DRV_LOG(WARNING, "port %u unable to allocate memory for xstats",
-		     dev->data->port_id);
+			dev->data->port_id);
 		return;
 	}
-	strings->cmd = ETHTOOL_GSTRINGS;
-	strings->string_set = ETH_SS_STATS;
-	strings->len = dev_stats_n;
-	ifr.ifr_data = (caddr_t)strings;
-	if (priv->master && priv->pf_bond >= 0)
-		/* Bonding master. */
-		ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[0].ifname,
-					   SIOCETHTOOL, &ifr);
-	else
-		ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
-	if (ret) {
-		DRV_LOG(WARNING, "port %u unable to get statistic names",
+	ret = mlx5_os_get_stats_strings(dev, bond_master, strings,
+					dev_stats_n, dev_stats_n_2nd);
+	if (ret < 0) {
+		DRV_LOG(WARNING, "port %u failed to get the stats strings",
 			dev->data->port_id);
 		goto free;
 	}
-	for (i = 0; i != dev_stats_n; ++i) {
-		const char *curr_string = (const char *)
-			&strings->data[i * ETH_GSTRING_LEN];
-
-		for (j = 0; j != xstats_n; ++j) {
-			if (!strcmp(mlx5_counters_init[j].ctr_name,
-				    curr_string)) {
-				unsigned int idx = xstats_ctrl->mlx5_stats_n++;
-
-				xstats_ctrl->dev_table_idx[idx] = i;
-				xstats_ctrl->info[idx] = mlx5_counters_init[j];
-				break;
-			}
-		}
-	}
-	/* Add dev counters. */
-	MLX5_ASSERT(xstats_ctrl->mlx5_stats_n <= MLX5_MAX_XSTATS);
-	for (i = 0; i != xstats_n; ++i) {
-		if (mlx5_counters_init[i].dev) {
-			unsigned int idx = xstats_ctrl->mlx5_stats_n++;
-
-			xstats_ctrl->info[idx] = mlx5_counters_init[i];
-			xstats_ctrl->hw_stats[idx] = 0;
-		}
-	}
 	xstats_ctrl->stats_n = dev_stats_n;
+	xstats_ctrl->stats_n_2nd = dev_stats_n_2nd;
 	/* Copy to base at first time. */
-	ret = mlx5_os_read_dev_counters(dev, xstats_ctrl->base);
+	ret = mlx5_os_read_dev_counters(dev, bond_master, xstats_ctrl->base);
 	if (ret)
 		DRV_LOG(ERR, "port %u cannot read device counters: %s",
 			dev->data->port_id, strerror(rte_errno));
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 96a269ccd0..9832b6df52 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -234,14 +234,22 @@ struct mlx5_counter_ctrl {
 struct mlx5_xstats_ctrl {
 	/* Number of device stats. */
 	uint16_t stats_n;
+	/* Number of device stats, for the 2nd port in bond. */
+	uint16_t stats_n_2nd;
 	/* Number of device stats identified by PMD. */
-	uint16_t  mlx5_stats_n;
+	uint16_t mlx5_stats_n;
 	/* Index in the device counters table. */
 	uint16_t dev_table_idx[MLX5_MAX_XSTATS];
+	/* Index in the output table. */
+	uint16_t xstats_o_idx[MLX5_MAX_XSTATS];
 	uint64_t base[MLX5_MAX_XSTATS];
 	uint64_t xstats[MLX5_MAX_XSTATS];
 	uint64_t hw_stats[MLX5_MAX_XSTATS];
 	struct mlx5_counter_ctrl info[MLX5_MAX_XSTATS];
+	/* Index in the device counters table, for the 2nd port in bond. */
+	uint16_t dev_table_idx_2nd[MLX5_MAX_XSTATS];
+	/* Index in the output table, for the 2nd port in bond. */
+	uint16_t xstats_o_idx_2nd[MLX5_MAX_XSTATS];
 };
 
 struct mlx5_stats_ctrl {
@@ -1974,8 +1982,9 @@ int mlx5_get_module_eeprom(struct rte_eth_dev *dev,
 			   struct rte_dev_eeprom_info *info);
 int mlx5_os_read_dev_stat(struct mlx5_priv *priv,
 			  const char *ctr_name, uint64_t *stat);
-int mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats);
-int mlx5_os_get_stats_n(struct rte_eth_dev *dev);
+int mlx5_os_read_dev_counters(struct rte_eth_dev *dev, bool bond_master, uint64_t *stats);
+int mlx5_os_get_stats_n(struct rte_eth_dev *dev, bool bond_master,
+			uint16_t *n_stats, uint16_t *n_stats_sec);
 void mlx5_os_stats_init(struct rte_eth_dev *dev);
 int mlx5_get_flag_dropless_rq(struct rte_eth_dev *dev);
 
diff --git a/drivers/net/mlx5/mlx5_stats.c b/drivers/net/mlx5/mlx5_stats.c
index 615e1d073d..f4ac58e2f9 100644
--- a/drivers/net/mlx5/mlx5_stats.c
+++ b/drivers/net/mlx5/mlx5_stats.c
@@ -39,24 +39,36 @@ mlx5_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *stats,
 		unsigned int n)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	unsigned int i;
-	uint64_t counters[n];
+	uint64_t counters[MLX5_MAX_XSTATS];
 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
+	unsigned int i;
+	uint16_t stats_n = 0;
+	uint16_t stats_n_2nd = 0;
 	uint16_t mlx5_stats_n = xstats_ctrl->mlx5_stats_n;
+	bool bond_master = (priv->master && priv->pf_bond >= 0);
 
 	if (n >= mlx5_stats_n && stats) {
-		int stats_n;
 		int ret;
 
-		stats_n = mlx5_os_get_stats_n(dev);
-		if (stats_n < 0)
-			return stats_n;
-		if (xstats_ctrl->stats_n != stats_n)
+		ret = mlx5_os_get_stats_n(dev, bond_master, &stats_n, &stats_n_2nd);
+		if (ret < 0)
+			return ret;
+		/*
+		 * The number of statistics fetched via "ETH_SS_STATS" may vary because
+		 * of the port configuration each time. This is also true between 2
+		 * ports. There might be a case that the numbers are the same even if
+		 * configurations are different.
+		 * It is not recommended to change the configuration without using
+		 * RTE API. The port(traffic) restart may trigger another initialization
+		 * to make sure the map are correct.
+		 */
+		if (xstats_ctrl->stats_n != stats_n ||
+		    (bond_master && xstats_ctrl->stats_n_2nd != stats_n_2nd))
 			mlx5_os_stats_init(dev);
-		ret = mlx5_os_read_dev_counters(dev, counters);
-		if (ret)
+		ret = mlx5_os_read_dev_counters(dev, bond_master, counters);
+		if (ret < 0)
 			return ret;
-		for (i = 0; i != mlx5_stats_n; ++i) {
+		for (i = 0; i != mlx5_stats_n; i++) {
 			stats[i].id = i;
 			if (xstats_ctrl->info[i].dev) {
 				uint64_t wrap_n;
@@ -225,30 +237,32 @@ mlx5_xstats_reset(struct rte_eth_dev *dev)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
-	int stats_n;
 	unsigned int i;
 	uint64_t *counters;
 	int ret;
+	uint16_t stats_n = 0;
+	uint16_t stats_n_2nd = 0;
+	bool bond_master = (priv->master && priv->pf_bond >= 0);
 
-	stats_n = mlx5_os_get_stats_n(dev);
-	if (stats_n < 0) {
+	ret = mlx5_os_get_stats_n(dev, bond_master, &stats_n, &stats_n_2nd);
+	if (ret < 0) {
 		DRV_LOG(ERR, "port %u cannot get stats: %s", dev->data->port_id,
-			strerror(-stats_n));
-		return stats_n;
+			strerror(-ret));
+		return ret;
 	}
-	if (xstats_ctrl->stats_n != stats_n)
+	if (xstats_ctrl->stats_n != stats_n ||
+	    (bond_master && xstats_ctrl->stats_n_2nd != stats_n_2nd))
 		mlx5_os_stats_init(dev);
-	counters =  mlx5_malloc(MLX5_MEM_SYS, sizeof(*counters) *
-			xstats_ctrl->mlx5_stats_n, 0,
-			SOCKET_ID_ANY);
+	/* Considering to use stack directly. */
+	counters = mlx5_malloc(MLX5_MEM_SYS, sizeof(*counters) * xstats_ctrl->mlx5_stats_n,
+			       0, SOCKET_ID_ANY);
 	if (!counters) {
-		DRV_LOG(WARNING, "port %u unable to allocate memory for xstats "
-				"counters",
+		DRV_LOG(WARNING, "port %u unable to allocate memory for xstats counters",
 		     dev->data->port_id);
 		rte_errno = ENOMEM;
 		return -rte_errno;
 	}
-	ret = mlx5_os_read_dev_counters(dev, counters);
+	ret = mlx5_os_read_dev_counters(dev, bond_master, counters);
 	if (ret) {
 		DRV_LOG(ERR, "port %u cannot read device counters: %s",
 			dev->data->port_id, strerror(rte_errno));
diff --git a/drivers/net/mlx5/windows/mlx5_ethdev_os.c b/drivers/net/mlx5/windows/mlx5_ethdev_os.c
index a31e1b5494..49f750be68 100644
--- a/drivers/net/mlx5/windows/mlx5_ethdev_os.c
+++ b/drivers/net/mlx5/windows/mlx5_ethdev_os.c
@@ -178,20 +178,29 @@ mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 	return -ENOTSUP;
 }
 
-/**
+/*
  * Query the number of statistics provided by ETHTOOL.
  *
  * @param dev
  *   Pointer to Ethernet device.
+ * @param bond_master
+ *   Indicate if the device is a bond master.
+ * @param n_stats
+ *   Pointer to number of stats to store.
+ * @param n_stats_sec
+ *   Pointer to number of stats to store for the 2nd port of the bond.
  *
  * @return
- *   Number of statistics on success, negative errno value otherwise and
- *   rte_errno is set.
+ *   0 on success, negative errno value otherwise and rte_errno is set.
  */
 int
-mlx5_os_get_stats_n(struct rte_eth_dev *dev)
+mlx5_os_get_stats_n(struct rte_eth_dev *dev, bool bond_master,
+		    uint16_t *n_stats, uint16_t *n_stats_sec)
 {
 	RTE_SET_USED(dev);
+	RTE_SET_USED(bond_master);
+	RTE_SET_USED(n_stats);
+	RTE_SET_USED(n_stats_sec);
 	return -ENOTSUP;
 }
 
@@ -221,6 +230,8 @@ mlx5_os_stats_init(struct rte_eth_dev *dev)
  *
  * @param dev
  *   Pointer to Ethernet device.
+ * @param bond_master
+ *   Indicate if the device is a bond master.
  * @param[out] stats
  *   Counters table output buffer.
  *
@@ -229,9 +240,10 @@ mlx5_os_stats_init(struct rte_eth_dev *dev)
  *   rte_errno is set.
  */
 int
-mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
+mlx5_os_read_dev_counters(struct rte_eth_dev *dev, bool bond_master, uint64_t *stats)
 {
 	RTE_SET_USED(dev);
+	RTE_SET_USED(bond_master);
 	RTE_SET_USED(stats);
 	return -ENOTSUP;
 }
-- 
2.39.2

---
  Diff of the applied patch vs upstream commit (please double-check if non-empty:
---
--- -	2024-03-14 00:09:22.961587116 +0000
+++ 0049-net-mlx5-fix-counters-map-in-bonding-mode.patch	2024-03-14 00:09:20.697616888 +0000
@@ -1 +1 @@
-From a687c3e658c2d889052089af8340bc0b9299c856 Mon Sep 17 00:00:00 2001
+From cc21b4a2d2d343307af178b1fd31eed1df0b1979 Mon Sep 17 00:00:00 2001
@@ -5,0 +6,2 @@
+[ upstream commit a687c3e658c2d889052089af8340bc0b9299c856 ]
+
@@ -50 +51,0 @@
-Cc: stable@dpdk.org
@@ -62 +63 @@
-index 92c47a3b3d..eb47c284ec 100644
+index dd5a0c546d..0ee8c58ba7 100644
@@ -237 +238 @@
-@@ -1615,6 +1657,101 @@ static const struct mlx5_counter_ctrl mlx5_counters_init[] = {
+@@ -1578,6 +1620,101 @@ static const struct mlx5_counter_ctrl mlx5_counters_init[] = {
@@ -339 +340 @@
-@@ -1627,76 +1764,44 @@ mlx5_os_stats_init(struct rte_eth_dev *dev)
+@@ -1590,76 +1727,44 @@ mlx5_os_stats_init(struct rte_eth_dev *dev)
@@ -431 +432 @@
-index f11a0181b8..fb3df76cac 100644
+index 96a269ccd0..9832b6df52 100644
@@ -434 +435 @@
-@@ -263,14 +263,22 @@ struct mlx5_counter_ctrl {
+@@ -234,14 +234,22 @@ struct mlx5_counter_ctrl {
@@ -458 +459 @@
-@@ -2182,8 +2190,9 @@ int mlx5_get_module_eeprom(struct rte_eth_dev *dev,
+@@ -1974,8 +1982,9 @@ int mlx5_get_module_eeprom(struct rte_eth_dev *dev,