[dpdk-dev] [PATCH 1/1] net/pcap: imissed stats support

Ferruh Yigit ferruh.yigit at intel.com
Mon Feb 1 10:25:39 CET 2021


On 2/1/2021 8:53 AM, Ido Goshen wrote:
> 
> 
>> -----Original Message-----
>> From: Ferruh Yigit <ferruh.yigit at intel.com>
>> Sent: Thursday, 28 January 2021 20:21
>> To: Ido Goshen <Ido at cgstowernetworks.com>
>> Cc: dev at dpdk.org
>> Subject: Re: [PATCH 1/1] net/pcap: imissed stats support
>>
>> On 1/25/2021 5:58 PM, Ido Goshen wrote:
>>> Signed-off-by: Ido Goshen <ido at cgstowernetworks.com>
>>
>> <...>
>>
>>> @@ -695,6 +708,10 @@ eth_stats_get(struct rte_eth_dev *dev, struct
>> rte_eth_stats *stats)
>>>    		stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
>>>    		rx_packets_total += stats->q_ipackets[i];
>>>    		rx_bytes_total += stats->q_ibytes[i];
>>> +		unsigned long rx_missed = eth_stats_get_pcap_missed(dev, i);
>>> +		if (rx_missed)
>>> +			rx_missed_total = rx_missed -
>>> +				internal->rx_queue[i].rx_stat.missed_reset;
>>
>> 'ps_drop' seems u_32 type, do you know how it behaves on overflow? Do you
>> think do we need a check here for overflow?
> 
> Right, it may overflow after few hours.
> I don't see a way to fully solve it w/o periodic sampling which is quite an overhead

Agree

> To compensate and avoid getting weird high ("negative") values
> I can check if the last retrieved value is higher than the current, then either
> zero it (restart) which will reflect rollover, or
> add UINT_MAX hoping there was only one rollover since last sample
> Please advice
> 

I would go with single rollover assumption, but comment this in the code.


More information about the dev mailing list