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

[Felix Fietkau <nbd@nbd.name>]

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.

- Felix