Re: [PATCH] net: fix softnet_stat

[Changli Gao <xiaosuo@gmail.com>]

On Thu, Apr 15, 2010 at 11:28 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
>> ----
>>  net/core/dev.c |    5 ++---
>>  1 file changed, 2 insertions(+), 3 deletions(-)
>> diff --git a/net/core/dev.c b/net/core/dev.c
>> index a10a216..b38a991 100644
>> --- a/net/core/dev.c
>> +++ b/net/core/dev.c
>> @@ -2336,7 +2336,6 @@ static int enqueue_to_backlog(struct sk_buff *skb, int cpu)
>>       queue = &per_cpu(softnet_data, cpu);
>>
>>       local_irq_save(flags);
>> -     __get_cpu_var(netdev_rx_stat).total++;
>
> I think this was nice, because we can see number of frames received by
> this cpu, before giving them to another cpu.
>

:( Packets will be counted again in __net_receive_skb(). Now the
meaning of total is confusing.

> Maybe we want a new counter ?
>
>        __get_cpu_var(netdev_rx_stat).inpackets++;
>

How about replacing total with received?

>>
>>       rps_lock(queue);
>>       if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
>> @@ -2366,9 +2365,9 @@ enqueue:
>>               goto enqueue;
>>       }
>>
>> +     per_cpu(netdev_rx_stat, cpu).dropped++;
>
> This is a bit expensive, and could be a queue->rx_stat.dropped++
> if netdev_rx_stat is moved inside queue structure.
>

It will make softnet_data larger.?

>
>>       rps_unlock(queue);
>>
>> -     __get_cpu_var(netdev_rx_stat).dropped++;
>>       local_irq_restore(flags);
>>
>>       kfree_skb(skb);
>> @@ -3565,7 +3564,7 @@ static int softnet_seq_show(struct seq_file *seq, void *v)
>>       struct netif_rx_stats *s = v;
>>
>>       seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
>> -                s->total, s->dropped, s->time_squeeze, 0,
>> +                s->total + s->dropped, s->dropped, s->time_squeeze, 0,
>
> This is wrong I believe... I prefer to add a new column, s->inpackets ?

If I rename total to received, is it still confusing?



-- 
Regards，
Changli Gao(xiaosuo@gmail.com)