Re: [RFC v3 4/8] wifi: mac80211: add support for DFS with multiple radios

[Felix Fietkau <nbd@nbd.name>]

On 07.06.24 10:54, Karthikeyan Periyasamy wrote:
> 
> 
> On 6/7/2024 1:46 PM, Felix Fietkau wrote:
>> On 07.06.24 08:45, Karthikeyan Periyasamy wrote:
>>>
>>>
>>> On 6/7/2024 10:33 AM, Felix Fietkau wrote:
>>>> On 07.06.24 06:54, Karthikeyan Periyasamy wrote:
>>>>>
>>>>>
>>>>> On 6/7/2024 10:05 AM, Felix Fietkau wrote:
>>>>>> On 07.06.24 06:25, Karthikeyan Periyasamy wrote:
>>>>>>>
>>>>>>>
>>>>>>> On 6/6/2024 11:37 PM, Felix Fietkau wrote:
>>>>>>>> DFS can be supported with multi-channel combinations, as long as 
>>>>>>>> each DFS
>>>>>>>> capable radio only supports one channel.
>>>>>>>>
>>>>>>>> Signed-off-by: Felix Fietkau <nbd@nbd.name>
>>>>>>>> ---
>>>>>>>>   net/mac80211/main.c | 32 ++++++++++++++++++++++++--------
>>>>>>>>   1 file changed, 24 insertions(+), 8 deletions(-)
>>>>>>>>
>>>>>>>> diff --git a/net/mac80211/main.c b/net/mac80211/main.c
>>>>>>>> index 40fbf397ce74..e9c4cf611e94 100644
>>>>>>>> --- a/net/mac80211/main.c
>>>>>>>> +++ b/net/mac80211/main.c
>>>>>>>
>>>>>>> ...
>>>>>>>
>>>>>>>>   int ieee80211_register_hw(struct ieee80211_hw *hw)
>>>>>>>>   {
>>>>>>>>       struct ieee80211_local *local = hw_to_local(hw);
>>>>>>>> @@ -1173,17 +1188,18 @@ int ieee80211_register_hw(struct 
>>>>>>>> ieee80211_hw *hw)
>>>>>>>>               if (comb->num_different_channels > 1)
>>>>>>>>                   return -EINVAL;
>>>>>>>>           }
>>>>>>>> -    } else {
>>>>>>>> -        /* DFS is not supported with multi-channel combinations 
>>>>>>>> yet */
>>>>>>>> -        for (i = 0; i < local->hw.wiphy->n_iface_combinations; 
>>>>>>>> i++) {
>>>>>>>> -            const struct ieee80211_iface_combination *comb;
>>>>>>>> -
>>>>>>>> -            comb = &local->hw.wiphy->iface_combinations[i];
>>>>>>>> +    } else if (hw->wiphy->n_radio) {
>>>>>>>> +        for (i = 0; i < hw->wiphy->n_radio; i++) {
>>>>>>>> +            const struct wiphy_radio *radio = &hw->wiphy->radio[i];
>>>>>>>> -            if (comb->radar_detect_widths &&
>>>>>>>> -                comb->num_different_channels > 1)
>>>>>>>> +            if 
>>>>>>>> (!ieee80211_ifcomb_check_radar(radio->iface_combinations,
>>>>>>>> +                              radio->n_iface_combinations))
>>>>>>>>                   return -EINVAL;
>>>>>>>>           }
>>>>>>>> +    } else {
>>>>>>>> +        if 
>>>>>>>> (!ieee80211_ifcomb_check_radar(hw->wiphy->iface_combinations,
>>>>>>>> +                          hw->wiphy->n_iface_combinations))
>>>>>>>> +            return -EINVAL;
>>>>>>>>       }
>>>>>>>>       /* Only HW csum features are currently compatible with 
>>>>>>>> mac80211 */
>>>>>>>
>>>>>>> Are we omitting the "wiphy->iface_combinations" if the radio specific
>>>>>>> iface combination advertised ?
>>>>>>>
>>>>>>> If so, it looks like unused "wiphy->iface_combinations" for radio
>>>>>>> specific combination advertised.
>>>>>>
>>>>>> This patch series assumes that you have both 
>>>>>> wiphy->iface_combinations and radio->iface_combinations set. 
>>>>>> wiphy->iface_combinations applies to the full wiphy, whereas 
>>>>>> radio->iface_combinations only applies to vifs assigned to the radio.
>>>>>
>>>>>
>>>>> If radio->iface_combinations is set then always vifs assigned to the
>>>>> radio. so wiphy->iface_combinations get avoid for all the use cases.
>>>>>
>>>>> Ultimately either of one combination only get used by this proposal.
>>>>>
>>>>> or I am missing something here ?
>>>>
>>>> The functions that perform interface combination checks are called 
>>>> both with -1 as radio_idx (meaning per-wiphy), as well as with the 
>>>> assigned radio id. That way, both kinds of combinations/limits are 
>>>> checked and enforced.
>>>>
>>> In the radio specific iface advertisement, global iface combination
>>> represent the union or intersection of all radio iface combination ?
>> 
>> The global interface combination should be a union of all radio 
>> interface combinations.
>> You can also use them to apply extra limits, e.g. if you have a limit on 
>> the per-wiphy number of interfaces (regardless of the assigned radio), 
>> you use the global interface combination to apply it.
>> 
> If the global combination follow union representation, the non-ML client
> takes wrong/invalid perception against the global advertisement.
> 
> Ex:
> 
> Global iface = 14 ( Radio iface: 2G = 4, 5G = 4, 6G = 6 ).
> 
> 2G non-client read the global configuration and understand it can able
> to create 14 interfaces. But in reality it allowed to create max 4
> interface only.
> 
> we have to follow intersection or minimal set, no ?

Maybe I misunderstood your question earlier. I meant that the driver can 
set the global limit as a sort of union of all per-radio limits.
For each individual radio, the intersection of the global limit (taking 
into account other radios as well) and the per-radio limit (considering 
only per-radio vifs) applies.

- Felix