On 8/23/19 9:32 AM, Vladimir Oltean wrote:
> Hi Florian,
>
> On Fri, 23 Aug 2019 at 19:23, Florian Fainelli <f.faine...@gmail.com> wrote:
>>
>> On 8/22/19 4:44 PM, Vladimir Oltean wrote:
>>> On Fri, 23 Aug 2019 at 02:43, Vivien Didelot <vivien.dide...@gmail.com>
>>> wrote:
>>>>
>>>> Hi Vladimir,
>>>>
>>>> On Fri, 23 Aug 2019 01:06:58 +0300, Vladimir Oltean <olte...@gmail.com>
>>>> wrote:
>>>>> Hi Vivien,
>>>>>
>>>>> On 8/22/19 11:13 PM, Vivien Didelot wrote:
>>>>>> Currently dsa_port_vid_add returns 0 if the switch returns -EOPNOTSUPP.
>>>>>>
>>>>>> This function is used in the tag_8021q.c code to offload the PVID of
>>>>>> ports, which would simply not work if .port_vlan_add is not supported
>>>>>> by the underlying switch.
>>>>>>
>>>>>> Do not skip -EOPNOTSUPP in dsa_port_vid_add but only when necessary,
>>>>>> that is to say in dsa_slave_vlan_rx_add_vid.
>>>>>>
>>>>>
>>>>> Do you know why Florian suppressed -EOPNOTSUPP in 061f6a505ac3 ("net:
>>>>> dsa: Add ndo_vlan_rx_{add, kill}_vid implementation")?
>>>>> I forced a return value of -EOPNOTSUPP here and when I create a VLAN
>>>>> sub-interface nothing breaks, it just says:
>>>>> RTNETLINK answers: Operation not supported
>>>>> which IMO is expected.
>>>>
>>>> I do not know what you mean. This patch does not change the behavior of
>>>> dsa_slave_vlan_rx_add_vid, which returns 0 if -EOPNOTSUPP is caught.
>>>>
>>>
>>> Yes, but what's wrong with just forwarding -EOPNOTSUPP?
>>
>> It makes us fail adding the VLAN to the slave network device, which
>> sounds silly, if we can't offload it in HW, that's fine, we can still do
>> a SW VLAN instead, see net/8021q/vlan_core.c::vlan_add_rx_filter_info().
>>
>> Maybe a more correct solution is to set the NETIF_F_HW_VLAN_CTAG_FILTER
>> feature bit only if we have the ability to offload, now that I think
>> about it. Would you want me to cook a patch doing that?
>
> sja1105 doesn't support offloading NETIF_F_HW_VLAN_CTAG_FILTER even
> though it does support programming VLANs.
The additional of the ndo_vlan_rx_{add,kill}_vid() is such that
standalone DSA ports continue to work while there is a bridge with
vlan_filtering=1 spanning other ports. In order for that ndo operation
to be called, we need to advertise the NETIF_F_HW_VLAN_CTAG_FILTER feature.
> Adding an offloaded VLAN sub-interface on a standalone switch port
> (vlan_filtering=0, uses dsa_8021q) would make the driver insert a VLAN
> entry whilst the TPID is ETH_P_DSA_8021Q.
> Maybe just let the driver set the netdev features, similar to how it
> does for the number of TX queues?
Why should we bend the framework because sja1105 and dsa_8021q are
special? Let me counter the argument: if the tagging is DSA_8021Q, why
not clear the feature then?
--
Florian
