[RFC] SNMP_ADD_STATS

[RFC] SNMP_ADD_STATS

[Arnaldo Carvalho de Melo]

Hi,

	I've been cleaning up the network stacks and one thing that made me
have a use for the barf bags I have handy is this type of construct:

    ip_statistics[smp_processor_id()*2 + !in_softirq()].IpFragCreates += nfrags;

I've noticed that there is:

#define SNMP_INC_STATS(mib, field) ((mib)[2*smp_processor_id()+!in_softirq()].field++)

and

#define IP_INC_STATS(field) SNMP_INC_STATS(ip_statistics, field)

So I propose that we have

#define SNMP_ADD_STATS(mib, field, incr) ((mib)[2*smp_processor_id()+!in_softirq()].field += incr)

and

#define IP_ADD_STATS(field,incr) SNMP_ADD_STATS(ip_statistics, field, incr)

Any objections?

- Arnaldo

Re: [RFC] SNMP_ADD_STATS

[kuznet]

Hello!

> Any objections?

No objections, of course. In any case, access to per-cpu stats is to be
macroized to work with preemption.

Alexey

Re: [RFC] SNMP_ADD_STATS

[David S. Miller]

   From: Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   Date: Wed, 12 Jun 2002 15:23:48 -0300
   
   So I propose that we have
   
   #define SNMP_ADD_STATS(mib, field, incr) ((mib)[2*smp_processor_id()+!in_softirq()].field += incr)
   
   and
   
   #define IP_ADD_STATS(field,incr) SNMP_ADD_STATS(ip_statistics, field, incr)
   
   Any objections?

No problem.

Re: [RFC] SNMP_ADD_STATS

[Andi Kleen]

> #define IP_ADD_STATS(field,incr) SNMP_ADD_STATS(ip_statistics, field, incr)
> 
> Any objections?

Better use per_cpu data in 2.5 (<asm/percpu.h>) for the MIBs.

That's much more efficient on some architectures, uses less memory on all
and is also cleaner.

-Andi