netdev - Re: [RFC][PATCH] Turn part of SNMP accounting macros into functions

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <48B68827.2060105@openvz.org>
Date:	Thu, 28 Aug 2008 15:12:39 +0400
From:	Pavel Emelyanov <xemul@...nvz.org>
To:	Eric Dumazet <dada1@...mosbay.com>
CC:	Linux Netdev List <netdev@...r.kernel.org>
Subject: Re: [RFC][PATCH] Turn part of SNMP accounting macros into functions

Eric Dumazet wrote:
> Pavel Emelyanov a écrit :
>> After turning IP_XXX_STATS, TCP_XXX_STATS and NET_XXX_STATS from
>> macros into functions the net/ipv4/built-in.o shrank significantly:
>>
>> add/remove: 14/0 grow/shrink: 0/67 up/down: 482/-2246 (-1764)
>>
>> Turning the CONFIG_NET_NS option on makes this shrink even larger:
>>
>> add/remove: 14/0 grow/shrink: 0/67 up/down: 478/-2646 (-2168)
>>
>> So the question is - what was the reason to keep those as macros?
>> I thought about the possible performance questions, but netperf
>> didn't show any (I admit I just cannot cook it properly).
>>
>> The sample patch is here, but it's not good (EXPORTs for ipv6
>> and a better place for functions rather than net/ipv4/af_inet.c
>> are required).
>>
>> Signed-off-by: Pavel Emelyanov <xemul@...nvz.org>
>>
> 
> 
> I dont know, but passing all those "struct net *net" to every 
> network function in the kernel sounds overkill, especially for
> !CONFIG_NET_NS users. This is pure bloat.
> 
> We could define two macros so that function prototypes dont include
> useless pointers, especially on arches where only first and second
> parameter is passed in eax and edx register ;)
> 
> #ifdef CONFIG_NET_NS
> # define VNETPTR ,struct net *net
> # define NETPTR  net
> #else
> # define VNETPTR
> # defint NETPTR  &init_net
> #endif
> 
> ...
> void TCP_DEC_STATS(int field VNETPTR);
> ...
> void TCP_DEC_STATS(int field VNETPTR)
> {
> 	SNMP_DEC_STATS((NETPTR)->mib.tcp_statistics, field);
> }
> ...

I agree with this, but I've checked whether compiling out the first argument
of XXX_STATS would help and found out that there's almost no difference (-9
bytes) in the net/ipv4/built-in.o for NET_NS=n case. 

After turning the macros into functions, compiling the first argument out gives
us 500 more bytes out. Good catch, but I have a question - the way you proposed
to implement the function itself is rather nice, but how would you implement
the *call* to that function in order to make it variable-arguments depending on
the config option?

I see the similar way:

+#ifdef CONFIG_NET_NS
+#define NET_ARG(net) net,
+#else
+#define NET_ARG(net)
+#endif
...
-	TCP_INC_STATS(sock_net(sk), field);
+	TCP_INC_STATS(NET_ARG(sock_net(sk) field);

but do these 500 bytes compensate for this ugly style?
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html