rx-dropped increases if the can module is not loaded (yet)

rx-dropped increases if the can module is not loaded (yet)

[Jeroen Hofstee]

Hi,

Since it took me a while to figure it out, it might be worthwhile to share.
If a CAN-bus network interface is first set to up and only later used,
the messages received in the meantime are counted by [1] as dropped.

If the the can module is modprobed before the interface is upped, this
is no longer the case.

Perhaps it saves someone some time.

Jeroen

[1] https://github.com/torvalds/linux/blob/v5.9/net/core/dev.c#L5257

Re: rx-dropped increases if the can module is not loaded (yet)

[Oliver Hartkopp]

On 27.11.20 17:09, Jeroen Hofstee wrote:

> Since it took me a while to figure it out, it might be worthwhile to share.
> If a CAN-bus network interface is first set to up and only later used,
> the messages received in the meantime are counted by [1] as dropped.

Interesting! But makes sense and is consistent: When the CAN frame is 
not processed by the network layer is is just dropped in the rx path.

> If the the can module is modprobed before the interface is upped, this
> is no longer the case.
> 
> Perhaps it saves someone some time.

So far we have not documented these details for the existing netdev 
statistics. Do you think this is needed when the behaviour, which leads 
to rx-dropped, is common for other netdevs also?

Regards,
Oliver

> 
> [1] https://github.com/torvalds/linux/blob/v5.9/net/core/dev.c#L5257
> 

Re: rx-dropped increases if the can module is not loaded (yet)

[Jeroen Hofstee]

Hello Oliver,

On 11/28/20 6:44 PM, Oliver Hartkopp wrote:
> On 27.11.20 17:09, Jeroen Hofstee wrote:
>
>> Since it took me a while to figure it out, it might be worthwhile to 
>> share.
>> If a CAN-bus network interface is first set to up and only later used,
>> the messages received in the meantime are counted by [1] as dropped.
>
> Interesting! But makes sense and is consistent: When the CAN frame is 
> not processed by the network layer is is just dropped in the rx path.
>
>> If the the can module is modprobed before the interface is upped, this
>> is no longer the case.
>>
>> Perhaps it saves someone some time.
>
> So far we have not documented these details for the existing netdev 
> statistics. Do you think this is needed when the behaviour, which 
> leads to rx-dropped, is common for other netdevs also?
>
>

Don't know... Perhaps we should simply wait and see how many
people show up wondering where the rx-dropped frames are
coming from?

Regards,

Jeroen

Re: rx-dropped increases if the can module is not loaded (yet)

[Oliver Hartkopp]

On 02.12.20 16:47, Jeroen Hofstee wrote:
> Hello Oliver,
> 
> On 11/28/20 6:44 PM, Oliver Hartkopp wrote:
>> On 27.11.20 17:09, Jeroen Hofstee wrote:
>>
>>> Since it took me a while to figure it out, it might be worthwhile to 
>>> share.
>>> If a CAN-bus network interface is first set to up and only later used,
>>> the messages received in the meantime are counted by [1] as dropped.
>>
>> Interesting! But makes sense and is consistent: When the CAN frame is 
>> not processed by the network layer is is just dropped in the rx path.
>>
>>> If the the can module is modprobed before the interface is upped, this
>>> is no longer the case.
>>>
>>> Perhaps it saves someone some time.
>>
>> So far we have not documented these details for the existing netdev 
>> statistics. Do you think this is needed when the behaviour, which 
>> leads to rx-dropped, is common for other netdevs also?
>>
>>
> 
> Don't know... Perhaps we should simply wait and see how many
> people show up wondering where the rx-dropped frames are
> coming from?
> 

ACK ;-)

Regards,
Oliver