Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1653269 > unrolled thread

[PATCH net-next 6/7] net: dsa: remove useless copy of tagger xmit

Started byVivien Didelot <vivien.didelot@savoirfairelinux.com>
First post2017-05-30 16:30 +0200
Last post2017-05-30 17:50 +0200
Articles 3 — 2 participants

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.


Contents

  [PATCH net-next 6/7] net: dsa: remove useless copy of tagger xmit Vivien Didelot <vivien.didelot@savoirfairelinux.com> - 2017-05-30 16:30 +0200
    Re: [PATCH net-next 6/7] net: dsa: remove useless copy of tagger xmit Andrew Lunn <andrew@lunn.ch> - 2017-05-30 17:30 +0200
      Re: [PATCH net-next 6/7] net: dsa: remove useless copy of tagger xmit Vivien Didelot <vivien.didelot@savoirfairelinux.com> - 2017-05-30 17:50 +0200

#1653269 — [PATCH net-next 6/7] net: dsa: remove useless copy of tagger xmit

FromVivien Didelot <vivien.didelot@savoirfairelinux.com>
Date2017-05-30 16:30 +0200
Subject[PATCH net-next 6/7] net: dsa: remove useless copy of tagger xmit
Message-ID<tMQjL-8vK-1@gated-at.bofh.it>
The dsa_slave_priv structure holds a copy of the dsa_device_ops xmit
function. It is always assigned to the switch tree xmit function.

Remove this useless copy.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
---
 net/dsa/dsa_priv.h | 3 ---
 net/dsa/slave.c    | 4 +---
 2 files changed, 1 insertion(+), 6 deletions(-)

diff --git a/net/dsa/dsa_priv.h b/net/dsa/dsa_priv.h
index 0fdd2a8a4ad8..282a55639285 100644
--- a/net/dsa/dsa_priv.h
+++ b/net/dsa/dsa_priv.h
@@ -71,9 +71,6 @@ struct dsa_device_ops {
 };
 
 struct dsa_slave_priv {
-	struct sk_buff *	(*xmit)(struct sk_buff *skb,
-					struct net_device *dev);
-
 	/* DSA port data, such as switch, port index, etc. */
 	struct dsa_port		*dp;
 
diff --git a/net/dsa/slave.c b/net/dsa/slave.c
index 887e26695519..ebad2af4d3a8 100644
--- a/net/dsa/slave.c
+++ b/net/dsa/slave.c
@@ -363,7 +363,7 @@ static netdev_tx_t dsa_slave_xmit(struct sk_buff *skb, struct net_device *dev)
 	dev->stats.tx_bytes += skb->len;
 
 	/* Transmit function may have to reallocate the original SKB */
-	nskb = p->xmit(skb, dev);
+	nskb = p->dp->ds->dst->tag_ops->xmit(skb, dev);
 	if (!nskb)
 		return NETDEV_TX_OK;
 
@@ -1136,7 +1136,6 @@ int dsa_slave_resume(struct net_device *slave_dev)
 int dsa_slave_create(struct dsa_switch *ds, struct device *parent,
 		     int port, const char *name)
 {
-	struct dsa_switch_tree *dst = ds->dst;
 	struct net_device *master;
 	struct net_device *slave_dev;
 	struct dsa_slave_priv *p;
@@ -1172,7 +1171,6 @@ int dsa_slave_create(struct dsa_switch *ds, struct device *parent,
 	p = netdev_priv(slave_dev);
 	p->dp = &ds->ports[port];
 	INIT_LIST_HEAD(&p->mall_tc_list);
-	p->xmit = dst->tag_ops->xmit;
 
 	p->old_pause = -1;
 	p->old_link = -1;
-- 
2.13.0

[toc] | [next] | [standalone]


#1653321

FromAndrew Lunn <andrew@lunn.ch>
Date2017-05-30 17:30 +0200
Message-ID<tMRfQ-DG-17@gated-at.bofh.it>
In reply to#1653269
On Tue, May 30, 2017 at 10:21:30AM -0400, Vivien Didelot wrote:
> The dsa_slave_priv structure holds a copy of the dsa_device_ops xmit
> function. It is always assigned to the switch tree xmit function.
> 
> Remove this useless copy.
> 
> Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
> ---
>  net/dsa/dsa_priv.h | 3 ---
>  net/dsa/slave.c    | 4 +---
>  2 files changed, 1 insertion(+), 6 deletions(-)
> 
> diff --git a/net/dsa/dsa_priv.h b/net/dsa/dsa_priv.h
> index 0fdd2a8a4ad8..282a55639285 100644
> --- a/net/dsa/dsa_priv.h
> +++ b/net/dsa/dsa_priv.h
> @@ -71,9 +71,6 @@ struct dsa_device_ops {
>  };
>  
>  struct dsa_slave_priv {
> -	struct sk_buff *	(*xmit)(struct sk_buff *skb,
> -					struct net_device *dev);
> -
>  	/* DSA port data, such as switch, port index, etc. */
>  	struct dsa_port		*dp;
>  
> diff --git a/net/dsa/slave.c b/net/dsa/slave.c
> index 887e26695519..ebad2af4d3a8 100644
> --- a/net/dsa/slave.c
> +++ b/net/dsa/slave.c
> @@ -363,7 +363,7 @@ static netdev_tx_t dsa_slave_xmit(struct sk_buff *skb, struct net_device *dev)
>  	dev->stats.tx_bytes += skb->len;
>  
>  	/* Transmit function may have to reallocate the original SKB */
> -	nskb = p->xmit(skb, dev);
> +	nskb = p->dp->ds->dst->tag_ops->xmit(skb, dev);

This is also the hot path for DSA transmit. Do we really want to do 4
extra pointer dereferences a million times per second, compared to one
copy during setup?

     Andrew

[toc] | [prev] | [next] | [standalone]


#1653328

FromVivien Didelot <vivien.didelot@savoirfairelinux.com>
Date2017-05-30 17:50 +0200
Message-ID<tMRzc-KJ-3@gated-at.bofh.it>
In reply to#1653321
Hi Andrew,

Andrew Lunn <andrew@lunn.ch> writes:

>>  	/* Transmit function may have to reallocate the original SKB */
>> -	nskb = p->xmit(skb, dev);
>> +	nskb = p->dp->ds->dst->tag_ops->xmit(skb, dev);
>
> This is also the hot path for DSA transmit. Do we really want to do 4
> extra pointer dereferences a million times per second, compared to one
> copy during setup?

Yep I get the idea. It felt weird to copy structure members like this
and not at least reusing the dsa_device_ops structure.

Thanks,

        Vivien

[toc] | [prev] | [standalone]


Back to top | Article view | linux.kernel


csiph-web