Re: [RFC PATCH 01/13] net: dsa: add plumbing for custom netdev statistics

[Heiner Kallweit <hkallweit1@gmail.com>]

On 17.10.2020 23:35, Vladimir Oltean wrote:
> DSA needs to push a header onto every packet on TX, and this might cause
> reallocation under certain scenarios, which might affect, for example,
> performance.
> 
> But reallocated packets are not standardized in struct pcpu_sw_netstats,
> struct net_device_stats or anywhere else, it seems, so we need to roll
> our own extra netdevice statistics and expose them to ethtool.
> 
> Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
> ---
>  net/dsa/dsa_priv.h |  9 +++++++++
>  net/dsa/slave.c    | 25 ++++++++++++++++++++++---
>  2 files changed, 31 insertions(+), 3 deletions(-)
> 
> diff --git a/net/dsa/dsa_priv.h b/net/dsa/dsa_priv.h
> index 12998bf04e55..d39db7500cdd 100644
> --- a/net/dsa/dsa_priv.h
> +++ b/net/dsa/dsa_priv.h
> @@ -73,12 +73,21 @@ struct dsa_notifier_mtu_info {
>  	int mtu;
>  };
>  
> +/* Driver statistics, other than those in struct rtnl_link_stats64.
> + * These are collected per-CPU and aggregated by ethtool.
> + */
> +struct dsa_slave_stats {
> +	__u64			tx_reallocs;
> +	struct u64_stats_sync	syncp;
> +} __aligned(1 * sizeof(u64));
> +

Wouldn't a simple unsigned long (like in struct net_device_stats) be
sufficient here? This would make handling the counter much simpler.
And as far as I understand we talk about a packet counter that is
touched in certain scenarios only.

>  struct dsa_slave_priv {
>  	/* Copy of CPU port xmit for faster access in slave transmit hot path */
>  	struct sk_buff *	(*xmit)(struct sk_buff *skb,
>  					struct net_device *dev);
>  
>  	struct pcpu_sw_netstats	__percpu *stats64;
> +	struct dsa_slave_stats	__percpu *extra_stats;
>  
>  	struct gro_cells	gcells;
>  
> diff --git a/net/dsa/slave.c b/net/dsa/slave.c
> index 3bc5ca40c9fb..d4326940233c 100644
> --- a/net/dsa/slave.c
> +++ b/net/dsa/slave.c
> @@ -668,9 +668,10 @@ static void dsa_slave_get_strings(struct net_device *dev,
>  		strncpy(data + len, "tx_bytes", len);
>  		strncpy(data + 2 * len, "rx_packets", len);
>  		strncpy(data + 3 * len, "rx_bytes", len);
> +		strncpy(data + 4 * len, "tx_reallocs", len);
>  		if (ds->ops->get_strings)
>  			ds->ops->get_strings(ds, dp->index, stringset,
> -					     data + 4 * len);
> +					     data + 5 * len);
>  	}
>  }
>  
> @@ -682,11 +683,13 @@ static void dsa_slave_get_ethtool_stats(struct net_device *dev,
>  	struct dsa_slave_priv *p = netdev_priv(dev);
>  	struct dsa_switch *ds = dp->ds;
>  	struct pcpu_sw_netstats *s;
> +	struct dsa_slave_stats *e;
>  	unsigned int start;
>  	int i;
>  
>  	for_each_possible_cpu(i) {
>  		u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
> +		u64 tx_reallocs;
>  
>  		s = per_cpu_ptr(p->stats64, i);
>  		do {
> @@ -696,13 +699,21 @@ static void dsa_slave_get_ethtool_stats(struct net_device *dev,
>  			rx_packets = s->rx_packets;
>  			rx_bytes = s->rx_bytes;
>  		} while (u64_stats_fetch_retry_irq(&s->syncp, start));
> +
> +		e = per_cpu_ptr(p->extra_stats, i);
> +		do {
> +			start = u64_stats_fetch_begin_irq(&e->syncp);
> +			tx_reallocs	= e->tx_reallocs;
> +		} while (u64_stats_fetch_retry_irq(&e->syncp, start));
> +
>  		data[0] += tx_packets;
>  		data[1] += tx_bytes;
>  		data[2] += rx_packets;
>  		data[3] += rx_bytes;
> +		data[4] += tx_reallocs;
>  	}
>  	if (ds->ops->get_ethtool_stats)
> -		ds->ops->get_ethtool_stats(ds, dp->index, data + 4);
> +		ds->ops->get_ethtool_stats(ds, dp->index, data + 5);
>  }
>  
>  static int dsa_slave_get_sset_count(struct net_device *dev, int sset)
> @@ -713,7 +724,7 @@ static int dsa_slave_get_sset_count(struct net_device *dev, int sset)
>  	if (sset == ETH_SS_STATS) {
>  		int count;
>  
> -		count = 4;
> +		count = 5;
>  		if (ds->ops->get_sset_count)
>  			count += ds->ops->get_sset_count(ds, dp->index, sset);
>  
> @@ -1806,6 +1817,12 @@ int dsa_slave_create(struct dsa_port *port)
>  		free_netdev(slave_dev);
>  		return -ENOMEM;
>  	}
> +	p->extra_stats = netdev_alloc_pcpu_stats(struct dsa_slave_stats);
> +	if (!p->extra_stats) {
> +		free_percpu(p->stats64);
> +		free_netdev(slave_dev);
> +		return -ENOMEM;
> +	}
>  
>  	ret = gro_cells_init(&p->gcells, slave_dev);
>  	if (ret)
> @@ -1864,6 +1881,7 @@ int dsa_slave_create(struct dsa_port *port)
>  out_gcells:
>  	gro_cells_destroy(&p->gcells);
>  out_free:
> +	free_percpu(p->extra_stats);
>  	free_percpu(p->stats64);
>  	free_netdev(slave_dev);
>  	port->slave = NULL;
> @@ -1886,6 +1904,7 @@ void dsa_slave_destroy(struct net_device *slave_dev)
>  	dsa_slave_notify(slave_dev, DSA_PORT_UNREGISTER);
>  	phylink_destroy(dp->pl);
>  	gro_cells_destroy(&p->gcells);
> +	free_percpu(p->extra_stats);
>  	free_percpu(p->stats64);
>  	free_netdev(slave_dev);
>  }
>