Patchwork cxgb3: Replace LRO with GRO

login
register
mail settings
Submitter Herbert Xu
Date Feb. 16, 2009, 3:36 a.m.
Message ID <20090216033644.GA14431@gondor.apana.org.au>
Download mbox | patch
Permalink /patch/23199/
State RFC
Delegated to: David Miller
Headers show

Comments

Herbert Xu - Feb. 16, 2009, 3:36 a.m.
On Thu, Jan 22, 2009 at 01:42:22AM -0800, Divy Le Ray wrote:
>
> I'm now getting about an average 5.7Gbs, oprofile typically showing:
> 
>       37.294500  copy_user_generic_unrolled          vmlinux
>        8.733800  process_responses                  cxgb3.ko
>        6.123700  refill_fl                          cxgb3.ko
>        5.006900  put_page                            vmlinux
>        3.400700  napi_fraginfo_skb                   vmlinux
>        2.484600  tcp_gro_receive                     vmlinux
>        2.308900  inet_gro_receive                    vmlinux
>        2.196000  free_hot_cold_page                  vmlinux
>        2.032900  skb_gro_header                      vmlinux
>        1.970100  get_page_from_freelist              vmlinux
>        1.869700  skb_copy_datagram_iovec             vmlinux
>        1.380300  dev_gro_receive                     vmlinux
>        1.242300  tcp_recvmsg                         vmlinux
>        1.079200  irq_entries_start                   vmlinux
>        1.003900  get_pageblock_flags_group           vmlinux
>        0.991300  skb_gro_receive                     vmlinux
>        0.878400  _raw_spin_lock                      vmlinux
>        0.878400  memcpy                              vmlinux

When you can get a chance can you see if this patch makes any

Thanks,
Divy Le Ray - Feb. 16, 2009, 3:47 a.m.
Herbert Xu wrote:

> When you can get a chance can you see if this patch makes any
> difference at all?

Hi Herbert,

Thanks for the patch.
I should be able to test within the next few days.

cheers,
Divy
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Divy Le Ray - March 13, 2009, 7:28 a.m.
> When you can get a chance can you see if this patch makes any
> difference at all?

Hi Herbert,

Sorry for the delay.
I had to switch to different development platforms.
I've not seen much perf change with this patch, it looks good though.

Cheers,
Divy

> 
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index d7efaf9..6a542fa 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -2586,8 +2586,10 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
>  {
>  	struct sk_buff *p = *head;
>  	struct sk_buff *nskb;
> +	skb_frag_t *frag;
>  	unsigned int headroom;
>  	unsigned int len = skb_gro_len(skb);
> +	int i;
>  
>  	if (p->len + len >= 65536)
>  		return -E2BIG;
> @@ -2604,9 +2606,9 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
>  		skb_shinfo(skb)->frags[0].size -=
>  			skb_gro_offset(skb) - skb_headlen(skb);
>  
> -		memcpy(skb_shinfo(p)->frags + skb_shinfo(p)->nr_frags,
> -		       skb_shinfo(skb)->frags,
> -		       skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
> +		frag = skb_shinfo(p)->frags + skb_shinfo(p)->nr_frags;
> +		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
> +			*frag++ = skb_shinfo(skb)->frags[i];
>  
>  		skb_shinfo(p)->nr_frags += skb_shinfo(skb)->nr_frags;
>  		skb_shinfo(skb)->nr_frags = 0;
> 
> Thanks,

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

difference at all?

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index d7efaf9..6a542fa 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -2586,8 +2586,10 @@  int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 {
 	struct sk_buff *p = *head;
 	struct sk_buff *nskb;
+	skb_frag_t *frag;
 	unsigned int headroom;
 	unsigned int len = skb_gro_len(skb);
+	int i;
 
 	if (p->len + len >= 65536)
 		return -E2BIG;
@@ -2604,9 +2606,9 @@  int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 		skb_shinfo(skb)->frags[0].size -=
 			skb_gro_offset(skb) - skb_headlen(skb);
 
-		memcpy(skb_shinfo(p)->frags + skb_shinfo(p)->nr_frags,
-		       skb_shinfo(skb)->frags,
-		       skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
+		frag = skb_shinfo(p)->frags + skb_shinfo(p)->nr_frags;
+		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
+			*frag++ = skb_shinfo(skb)->frags[i];
 
 		skb_shinfo(p)->nr_frags += skb_shinfo(skb)->nr_frags;
 		skb_shinfo(skb)->nr_frags = 0;