Skip to contentSkip to conversation

netdevsim: implement RX statistics using NETDEV_PCPU_STAT_DSTATS

PATCH Applied Breno Leitao 2 participants 6 messages Jun 18, 2025 – Jun 19, 2025

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v4 cover netdevsim: implement RX statistics using NETDEV_PCPU_STAT_DSTATS

The netdevsim driver previously lacked RX statistics support, which prevented its use with the GenerateTraffic() test framework, as this framework verifies traffic flow by checking RX byte counts.

This patch migrates netdevsim from its custom statistics collection to the NETDEV_PCPU_STAT_DSTATS framework, as suggested by Jakub. This change not only standardizes the statistics handling but also adds the necessary RX statistics support required by the test framework.

---
Changes in v4:
- Protect dev_dstats_rx_dropped_add() by disabling BH (Jakub)
- Link to v3: https://lore.kernel.org/r/[email protected]

Changes in v3:
- Rely on netdev from caller instead of napi->dev in nsim_queue_free().
- Link to v2: https://lore.kernel.org/r/[email protected]

Changes in v2:
- Changed the RX collection place from nsim_napi_rx() to nsim_rcv (Joe Damato)
- Collect RX dropped packets statistic in nsim_queue_free() (Jakub)
- Added a helper in dstat to add values to RX dropped packets
- Link to v1: https://lore.kernel.org/r/[email protected]

---
Breno Leitao (4):
netdevsim: migrate to dstats stats collection
netdevsim: collect statistics at RX side
net: add dev_dstats_rx_dropped_add() helper
netdevsim: account dropped packet length in stats on queue free

 drivers/net/netdevsim/netdev.c    | 56 ++++++++++++++++-----------------------
 drivers/net/netdevsim/netdevsim.h |  5 ----
 include/linux/netdevice.h         | 10 +++++++
 3 files changed, 33 insertions(+), 38 deletions(-)

---
base-commit: 3b5b1c428260152e47c9584bc176f358b87ca82d
change-id: 20250610-netdevsim_stat-95995921e03e

Best regards,

Signature · 1 line
Breno Leitao <[email protected]>
#1 lore raw

PATCH v4 1/4 netdevsim: migrate to dstats stats collection

Replace custom statistics tracking with the kernel's dstats infrastructure to simplify code and improve consistency with other network drivers.

This change:
- Sets dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS for automatic automatic allocation and deallocation.
- Removes manual stats fields and their update
- Replaces custom nsim_get_stats64() with dev_get_stats()
- Uses dev_dstats_tx_add() and dev_dstats_tx_dropped() helpers
- Eliminates the need for manual synchronization primitives

The dstats framework provides the same functionality with less code.

---
 drivers/net/netdevsim/netdev.c    | 33 ++++++---------------------------
 drivers/net/netdevsim/netdevsim.h |  5 -----
 2 files changed, 6 insertions(+), 32 deletions(-)
drivers/net/netdevsim/netdev.c +6 -27
index fa5fbd97ad69e..5010d8eefc854 100644
@@ -93,19 +93,14 @@ static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
 		hrtimer_start(&rq->napi_timer, us_to_ktime(5), HRTIMER_MODE_REL);
 
 	rcu_read_unlock();
-	u64_stats_update_begin(&ns->syncp);
-	ns->tx_packets++;
-	ns->tx_bytes += len;
-	u64_stats_update_end(&ns->syncp);
+	dev_dstats_tx_add(dev, skb->len);
 	return NETDEV_TX_OK;
 
 out_drop_free:
 	dev_kfree_skb(skb);
 out_drop_cnt:
 	rcu_read_unlock();
-	u64_stats_update_begin(&ns->syncp);
-	ns->tx_dropped++;
-	u64_stats_update_end(&ns->syncp);
+	dev_dstats_tx_dropped(dev);
 	return NETDEV_TX_OK;
 }
 
@@ -126,20 +121,6 @@ static int nsim_change_mtu(struct net_device *dev, int new_mtu)
 	return 0;
 }
 
-static void
-nsim_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
-{
-	struct netdevsim *ns = netdev_priv(dev);
-	unsigned int start;
-
-	do {
-		start = u64_stats_fetch_begin(&ns->syncp);
-		stats->tx_bytes = ns->tx_bytes;
-		stats->tx_packets = ns->tx_packets;
-		stats->tx_dropped = ns->tx_dropped;
-	} while (u64_stats_fetch_retry(&ns->syncp, start));
-}
-
 static int
 nsim_setup_tc_block_cb(enum tc_setup_type type, void *type_data, void *cb_priv)
 {
@@ -556,7 +537,6 @@ static const struct net_device_ops nsim_netdev_ops = {
 	.ndo_set_mac_address	= eth_mac_addr,
 	.ndo_validate_addr	= eth_validate_addr,
 	.ndo_change_mtu		= nsim_change_mtu,
-	.ndo_get_stats64	= nsim_get_stats64,
 	.ndo_set_vf_mac		= nsim_set_vf_mac,
 	.ndo_set_vf_vlan	= nsim_set_vf_vlan,
 	.ndo_set_vf_rate	= nsim_set_vf_rate,
@@ -580,7 +560,6 @@ static const struct net_device_ops nsim_vf_netdev_ops = {
 	.ndo_set_mac_address	= eth_mac_addr,
 	.ndo_validate_addr	= eth_validate_addr,
 	.ndo_change_mtu		= nsim_change_mtu,
-	.ndo_get_stats64	= nsim_get_stats64,
 	.ndo_setup_tc		= nsim_setup_tc,
 	.ndo_set_features	= nsim_set_features,
 };
@@ -594,7 +573,7 @@ static void nsim_get_queue_stats_rx(struct net_device *dev, int idx,
 	struct rtnl_link_stats64 rtstats = {};
 
 	if (!idx)
-		nsim_get_stats64(dev, &rtstats);
+		dev_get_stats(dev, &rtstats);
 
 	stats->packets = rtstats.rx_packets - !!rtstats.rx_packets;
 	stats->bytes = rtstats.rx_bytes;
@@ -606,7 +585,7 @@ static void nsim_get_queue_stats_tx(struct net_device *dev, int idx,
 	struct rtnl_link_stats64 rtstats = {};
 
 	if (!idx)
-		nsim_get_stats64(dev, &rtstats);
+		dev_get_stats(dev, &rtstats);
 
 	stats->packets = rtstats.tx_packets - !!rtstats.tx_packets;
 	stats->bytes = rtstats.tx_bytes;
@@ -618,7 +597,7 @@ static void nsim_get_base_stats(struct net_device *dev,
 {
 	struct rtnl_link_stats64 rtstats = {};
 
-	nsim_get_stats64(dev, &rtstats);
+	dev_get_stats(dev, &rtstats);
 
 	rx->packets = !!rtstats.rx_packets;
 	rx->bytes = 0;
@@ -890,6 +869,7 @@ static void nsim_setup(struct net_device *dev)
 			    NETIF_F_HW_CSUM |
 			    NETIF_F_LRO |
 			    NETIF_F_TSO;
+	dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS;
 	dev->max_mtu = ETH_MAX_MTU;
 	dev->xdp_features = NETDEV_XDP_ACT_HW_OFFLOAD;
 }
@@ -1022,7 +1002,6 @@ nsim_create(struct nsim_dev *nsim_dev, struct nsim_dev_port *nsim_dev_port)
 	dev_net_set(dev, nsim_dev_net(nsim_dev));
 	ns = netdev_priv(dev);
 	ns->netdev = dev;
-	u64_stats_init(&ns->syncp);
 	ns->nsim_dev = nsim_dev;
 	ns->nsim_dev_port = nsim_dev_port;
 	ns->nsim_bus_dev = nsim_dev->nsim_bus_dev;
drivers/net/netdevsim/netdevsim.h +0 -5
index d04401f0bdf79..343b8f19dbed6 100644
@@ -108,11 +108,6 @@ struct netdevsim {
 
 	int rq_reset_mode;
 
-	u64 tx_packets;
-	u64 tx_bytes;
-	u64 tx_dropped;
-	struct u64_stats_sync syncp;
-
 	struct nsim_bus_dev *nsim_bus_dev;
 
 	struct bpf_prog	*bpf_offloaded;
Signature · 1 line
2.47.1
#2 lore raw

PATCH v4 2/4 netdevsim: collect statistics at RX side

In reply to Breno Leitao

When the RX side of netdevsim was added, the RX statistics were missing, making the driver unusable for GenerateTraffic() test framework.

This patch adds proper statistics tracking on RX side, complementing the TX path.

---
 drivers/net/netdevsim/netdev.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
drivers/net/netdevsim/netdev.c +10 -2
index 5010d8eefc854..de309ff69e43e 100644
@@ -331,16 +331,24 @@ static int nsim_get_iflink(const struct net_device *dev)
 
 static int nsim_rcv(struct nsim_rq *rq, int budget)
 {
+	struct net_device *dev = rq->napi.dev;
 	struct sk_buff *skb;
-	int i;
+	unsigned int skblen;
+	int i, ret;
 
 	for (i = 0; i < budget; i++) {
 		if (skb_queue_empty(&rq->skb_queue))
 			break;
 
 		skb = skb_dequeue(&rq->skb_queue);
+		/* skb might be discard at netif_receive_skb, save the len */
+		skblen =  skb->len;
 		skb_mark_napi_id(skb, &rq->napi);
-		netif_receive_skb(skb);
+		ret = netif_receive_skb(skb);
+		if (ret == NET_RX_SUCCESS)
+			dev_dstats_rx_add(dev, skblen);
+		else
+			dev_dstats_rx_dropped(dev);
 	}
 
 	return i;
Signature · 1 line
2.47.1
#3 lore raw

PATCH v4 3/4 net: add dev_dstats_rx_dropped_add() helper

In reply to Breno Leitao

Introduce the dev_dstats_rx_dropped_add() helper to allow incrementing the rx_drops per-CPU statistic by an arbitrary value, rather than just one. This is useful for drivers or code paths that need to account for multiple dropped packets at once, such as when dropping entire queues.

---
 include/linux/netdevice.h | 10 ++++++++++
 1 file changed, 10 insertions(+)
include/linux/netdevice.h +10 -0
index 9cbc4e54b7e4a..03c26bb0fbbef 100644
@@ -3016,6 +3016,16 @@ static inline void dev_dstats_rx_dropped(struct net_device *dev)
 	u64_stats_update_end(&dstats->syncp);
 }
 
+static inline void dev_dstats_rx_dropped_add(struct net_device *dev,
+					     unsigned int packets)
+{
+	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
+
+	u64_stats_update_begin(&dstats->syncp);
+	u64_stats_add(&dstats->rx_drops, packets);
+	u64_stats_update_end(&dstats->syncp);
+}
+
 static inline void dev_dstats_tx_add(struct net_device *dev,
 				     unsigned int len)
 {
Signature · 1 line
2.47.1
#4 lore raw

PATCH v4 4/4 netdevsim: account dropped packet length in stats on queue free

In reply to Breno Leitao

Add a call to dev_dstats_rx_dropped_add() in nsim_queue_free() to account for the number of packets dropped when purging the skb queue. This improves the accuracy of RX drop statistics reported by netdevsim.

local_bh_{disable, enable}() protection is used to disable preemption, which is necessary given that dev_dstats_rx_dropped_add() access this_cpu_ptr(). See discussion in [1].

---
 drivers/net/netdevsim/netdev.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
drivers/net/netdevsim/netdev.c +7 -4
index de309ff69e43e..07171cf8b07ee 100644
@@ -632,9 +632,12 @@ static struct nsim_rq *nsim_queue_alloc(void)
 	return rq;
 }
 
-static void nsim_queue_free(struct nsim_rq *rq)
+static void nsim_queue_free(struct net_device *dev, struct nsim_rq *rq)
 {
 	hrtimer_cancel(&rq->napi_timer);
+	local_bh_disable();
+	dev_dstats_rx_dropped_add(dev, rq->skb_queue.qlen);
+	local_bh_enable();
 	skb_queue_purge_reason(&rq->skb_queue, SKB_DROP_REASON_QUEUE_PURGE);
 	kfree(rq);
 }
@@ -681,7 +684,7 @@ nsim_queue_mem_alloc(struct net_device *dev, void *per_queue_mem, int idx)
 	return 0;
 
 err_free:
-	nsim_queue_free(qmem->rq);
+	nsim_queue_free(dev, qmem->rq);
 	return err;
 }
 
@@ -695,7 +698,7 @@ static void nsim_queue_mem_free(struct net_device *dev, void *per_queue_mem)
 		if (!ns->rq_reset_mode)
 			netif_napi_del_locked(&qmem->rq->napi);
 		page_pool_destroy(qmem->rq->page_pool);
-		nsim_queue_free(qmem->rq);
+		nsim_queue_free(dev, qmem->rq);
 	}
 }
 
@@ -913,7 +916,7 @@ static void nsim_queue_uninit(struct netdevsim *ns)
 	int i;
 
 	for (i = 0; i < dev->num_rx_queues; i++)
-		nsim_queue_free(ns->rq[i]);
+		nsim_queue_free(dev, ns->rq[i]);
 
 	kfree(ns->rq);
 	ns->rq = NULL;
Signature · 1 line
2.47.1
#5 lore raw
In reply to Breno Leitao
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:

Hello:

This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:

Quoted 10 lines · On Wed, 18 Jun 2025 01:32:41 -0700 you wrote:

The netdevsim driver previously lacked RX statistics support, which prevented its use with the GenerateTraffic() test framework, as this framework verifies traffic flow by checking RX byte counts.

This patch migrates netdevsim from its custom statistics collection to the NETDEV_PCPU_STAT_DSTATS framework, as suggested by Jakub. This change not only standardizes the statistics handling but also adds the necessary RX statistics support required by the test framework.

[...]

Here is the summary with links:
- [net-next,v4,1/4] netdevsim: migrate to dstats stats collection

    https://git.kernel.org/netdev/net-next/c/f9e2511d80c2
  - [net-next,v4,2/4] netdevsim: collect statistics at RX side
    https://git.kernel.org/netdev/net-next/c/788eb4de608b
  - [net-next,v4,3/4] net: add dev_dstats_rx_dropped_add() helper
    https://git.kernel.org/netdev/net-next/c/27480a7c8f02
  - [net-next,v4,4/4] netdevsim: account dropped packet length in stats on queue free
    https://git.kernel.org/netdev/net-next/c/2a68a22304f9

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html