[PATCH net-2.6.25 3/3] Uninline the inet_twsk_put function

[PATCH net-2.6.25 3/3] Uninline the inet_twsk_put function

[Pavel Emelyanov]

This one is not that big, but is widely used: saves 1200 bytes 
from net/ipv4/built-in.o

add/remove: 1/0 grow/shrink: 1/12 up/down: 97/-1300 (-1203)
function                                     old     new   delta
inet_twsk_put                                  -      87     +87
__inet_lookup_listener                       274     284     +10
tcp_sacktag_write_queue                     2255    2254      -1
tcp_time_wait                                482     411     -71
__inet_check_established                     796     722     -74
tcp_v4_err                                   973     898     -75
__inet_twsk_kill                             230     154     -76
inet_twsk_deschedule                         180     103     -77
tcp_v4_do_rcv                                462     384     -78
inet_hash_connect                            686     607     -79
inet_twdr_do_twkill_work                     236     150     -86
inet_twdr_twcal_tick                         395     307     -88
tcp_v4_rcv                                  1744    1480    -264
tcp_timewait_state_process                   975     644    -331

Export it for ipv6 module.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>

---
 include/net/inet_timewait_sock.h |   14 +-------------
 net/ipv4/inet_timewait_sock.c    |   15 +++++++++++++++
 2 files changed, 16 insertions(+), 13 deletions(-)

diff --git a/include/net/inet_timewait_sock.h b/include/net/inet_timewait_sock.h
index abaff05..67e9250 100644
--- a/include/net/inet_timewait_sock.h
+++ b/include/net/inet_timewait_sock.h
@@ -193,19 +193,7 @@ static inline __be32 inet_rcv_saddr(const struct sock *sk)
 		inet_sk(sk)->rcv_saddr : inet_twsk(sk)->tw_rcv_saddr;
 }
 
-static inline void inet_twsk_put(struct inet_timewait_sock *tw)
-{
-	if (atomic_dec_and_test(&tw->tw_refcnt)) {
-		struct module *owner = tw->tw_prot->owner;
-		twsk_destructor((struct sock *)tw);
-#ifdef SOCK_REFCNT_DEBUG
-		printk(KERN_DEBUG "%s timewait_sock %p released\n",
-		       tw->tw_prot->name, tw);
-#endif
-		kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
-		module_put(owner);
-	}
-}
+extern void inet_twsk_put(struct inet_timewait_sock *tw);
 
 extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
 						  const int state);
diff --git a/net/ipv4/inet_timewait_sock.c b/net/ipv4/inet_timewait_sock.c
index d43e787..1b7db42 100644
--- a/net/ipv4/inet_timewait_sock.c
+++ b/net/ipv4/inet_timewait_sock.c
@@ -48,6 +48,21 @@ static void __inet_twsk_kill(struct inet_timewait_sock *tw,
 	inet_twsk_put(tw);
 }
 
+void inet_twsk_put(struct inet_timewait_sock *tw)
+{
+	if (atomic_dec_and_test(&tw->tw_refcnt)) {
+		struct module *owner = tw->tw_prot->owner;
+		twsk_destructor((struct sock *)tw);
+#ifdef SOCK_REFCNT_DEBUG
+		printk(KERN_DEBUG "%s timewait_sock %p released\n",
+		       tw->tw_prot->name, tw);
+#endif
+		kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
+		module_put(owner);
+	}
+}
+EXPORT_SYMBOL_GPL(inet_twsk_put);
+
 /*
  * Enter the time wait state. This is called with locally disabled BH.
  * Essentially we whip up a timewait bucket, copy the relevant info into it
-- 
1.5.3.4

Re: [PATCH net-2.6.25 3/3] Uninline the inet_twsk_put function

[Ingo Oeser]

Pavel Emelyanov schrieb:
> This one is not that big, but is widely used: saves 1200 bytes 
> from net/ipv4/built-in.o
> +void inet_twsk_put(struct inet_timewait_sock *tw)
> +{
> +	if (atomic_dec_and_test(&tw->tw_refcnt)) {
> +		struct module *owner = tw->tw_prot->owner;
> +		twsk_destructor((struct sock *)tw);
> +#ifdef SOCK_REFCNT_DEBUG
> +		printk(KERN_DEBUG "%s timewait_sock %p released\n",
> +		       tw->tw_prot->name, tw);
> +#endif
> +		kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
> +		module_put(owner);
> +	}
> +}
> +EXPORT_SYMBOL_GPL(inet_twsk_put);

More correct fix seems to be conversion to kref.

Just create out of line inet_twsk_release() containing
sth. similiar to the code inside these braces and modify 
inet_twsk_put() to sth. like this:

static inline inet_twsk_put(struct inet_timewait_sock *tw)
{
	kref_put(&tw->kref, inet_twsk_release);
}

David, can you see any reason (e.g. some crazy lock stuff) NOT to do this?


Best Regards

Ingo Oeser

Re: [PATCH net-2.6.25 3/3] Uninline the inet_twsk_put function

[David Miller]

From: Pavel Emelyanov <xemul@openvz.org>
Date: Wed, 19 Dec 2007 13:56:13 +0300

> This one is not that big, but is widely used: saves 1200 bytes 
> from net/ipv4/built-in.o
> 
> add/remove: 1/0 grow/shrink: 1/12 up/down: 97/-1300 (-1203)
> function                                     old     new   delta
> inet_twsk_put                                  -      87     +87
> __inet_lookup_listener                       274     284     +10
> tcp_sacktag_write_queue                     2255    2254      -1
> tcp_time_wait                                482     411     -71
> __inet_check_established                     796     722     -74
> tcp_v4_err                                   973     898     -75
> __inet_twsk_kill                             230     154     -76
> inet_twsk_deschedule                         180     103     -77
> tcp_v4_do_rcv                                462     384     -78
> inet_hash_connect                            686     607     -79
> inet_twdr_do_twkill_work                     236     150     -86
> inet_twdr_twcal_tick                         395     307     -88
> tcp_v4_rcv                                  1744    1480    -264
> tcp_timewait_state_process                   975     644    -331
> 
> Export it for ipv6 module.
> 
> Signed-off-by: Pavel Emelyanov <xemul@openvz.org>

Applied, thanks.

Re: [PATCH net-2.6.25 3/3] Uninline the inet_twsk_put function

[David Miller]

From: Ingo Oeser <netdev@axxeo.de>
Date: Thu, 20 Dec 2007 19:32:45 +0100

> static inline inet_twsk_put(struct inet_timewait_sock *tw)
> {
> 	kref_put(&tw->kref, inet_twsk_release);
> }
> 
> David, can you see any reason (e.g. some crazy lock stuff) NOT to do this?

Look at how this datastructure actually works before making
such suggestions, don't just look at the context provided
purely by a patch.

"inet_timewait_sock" begins with a "struct sock_common"
which is where the atomic_t is, and:

#define tw_refcnt		__tw_common.skc_refcnt

So you would have to change struct sock_common over to kref, and thus
the entire networking, in order to make such a change.

I see zero value in this.  There are millions of more useful things to
invest that kind of time on.

But you would have seen this instantly if you had spent 5 seconds
looking at how these datastructures are defined.  Instead you choose
to make me do it and explain it to you instead.

Re: [PATCH net-2.6.25 3/3] Uninline the inet_twsk_put function

[Ingo Oeser]

Hi David,

David Miller schrieb:
> "inet_timewait_sock" begins with a "struct sock_common"
> which is where the atomic_t is, and:
> 
> #define tw_refcnt		__tw_common.skc_refcnt
> 
> So you would have to change struct sock_common over to kref, and thus
> the entire networking, in order to make such a change.

Ok, that sounds too much. Many thanks for following up and taking the time
to explain it.
 
> But you would have seen this instantly if you had spent 5 seconds
> looking at how these datastructures are defined.  Instead you choose
> to make me do it and explain it to you instead.

Sorry, just matched the wrong pattern here :-)


Best Regards

Ingo Oeser