[Make-wifi-fast] wifi 7 for mediatek lands... fq_codel? atf? can't tell...

[Make-wifi-fast] wifi 7 for mediatek lands... fq_codel? atf? can't tell...

[Dave Taht]

To my eye, this is offloading everything, but I could be wrong.

https://patchwork.kernel.org/project/linux-mediatek/list/?series=690553


-- 
This song goes out to all the folk that thought Stadia would work:
https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity-6981366665607352320-FXtz
Dave Täht CEO, TekLibre, LLC

Re: [Make-wifi-fast] wifi 7 for mediatek lands... fq_codel? atf? can't tell...

[Felix Fietkau]

On 01.11.22 02:49, Dave Taht wrote:
> To my eye, this is offloading everything, but I could be wrong.
> 
> https://patchwork.kernel.org/project/linux-mediatek/list/?series=690553
ATF and AQL are enabled in the generic mt76 code, not the individual 
drivers. itxq (used for fq_codel) is also pretty much mandatory for mt76 
drivers using the generic framework.
The driver also has code to read the airtime stats, so all of those 
features should work.

- Felix

Re: [Make-wifi-fast] wifi 7 for mediatek lands... fq_codel? atf? can't tell...

[Dave Taht]

On Tue, Nov 1, 2022 at 2:03 AM Felix Fietkau <nbd@nbd.name> wrote:
>
> On 01.11.22 02:49, Dave Taht wrote:
> > To my eye, this is offloading everything, but I could be wrong.
> >
> > https://patchwork.kernel.org/project/linux-mediatek/list/?series=690553
> ATF and AQL are enabled in the generic mt76 code, not the individual
> drivers. itxq (used for fq_codel) is also pretty much mandatory for mt76
> drivers using the generic framework.
> The driver also has code to read the airtime stats, so all of those
> features should work.

Thank you for the update! However, in order to take advantage of
certain wifi7 features, like RU,
a gang scheduler is needed, in order to feed each station
independently in the same txop, and
also there are other features like probes to set that up. Didn't see
any of that...

>
> - Felix



-- 
This song goes out to all the folk that thought Stadia would work:
https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity-6981366665607352320-FXtz
Dave Täht CEO, TekLibre, LLC

Re: [Make-wifi-fast] wifi 7 for mediatek lands... fq_codel? atf? can't tell...

[Felix Fietkau]

On 01.11.22 10:17, Dave Taht wrote:
> On Tue, Nov 1, 2022 at 2:03 AM Felix Fietkau <nbd@nbd.name> wrote:
>>
>> On 01.11.22 02:49, Dave Taht wrote:
>> > To my eye, this is offloading everything, but I could be wrong.
>> >
>> > https://patchwork.kernel.org/project/linux-mediatek/list/?series=690553
>> ATF and AQL are enabled in the generic mt76 code, not the individual
>> drivers. itxq (used for fq_codel) is also pretty much mandatory for mt76
>> drivers using the generic framework.
>> The driver also has code to read the airtime stats, so all of those
>> features should work.
> 
> Thank you for the update! However, in order to take advantage of
> certain wifi7 features, like RU,
> a gang scheduler is needed, in order to feed each station
> independently in the same txop, and
> also there are other features like probes to set that up. Didn't see
> any of that...
All of that scheduling happens inside the firmware. The driver/stack 
only ensures that the hw queueing per station is limited to the 
available AQL budget (and ATF constraints).

- Felix

Re: [Make-wifi-fast] wifi 7 for mediatek lands... fq_codel? atf? can't tell...

[Dave Taht]

On Tue, Nov 1, 2022 at 2:20 AM Felix Fietkau <nbd@nbd.name> wrote:
>
> On 01.11.22 10:17, Dave Taht wrote:
> > On Tue, Nov 1, 2022 at 2:03 AM Felix Fietkau <nbd@nbd.name> wrote:
> >>
> >> On 01.11.22 02:49, Dave Taht wrote:
> >> > To my eye, this is offloading everything, but I could be wrong.
> >> >
> >> > https://patchwork.kernel.org/project/linux-mediatek/list/?series=690553
> >> ATF and AQL are enabled in the generic mt76 code, not the individual
> >> drivers. itxq (used for fq_codel) is also pretty much mandatory for mt76
> >> drivers using the generic framework.
> >> The driver also has code to read the airtime stats, so all of those
> >> features should work.
> >
> > Thank you for the update! However, in order to take advantage of
> > certain wifi7 features, like RU,
> > a gang scheduler is needed, in order to feed each station
> > independently in the same txop, and
> > also there are other features like probes to set that up. Didn't see
> > any of that...
> All of that scheduling happens inside the firmware. The driver/stack
> only ensures that the hw queueing per station is limited to the
> available AQL budget (and ATF constraints).

I'm sorry to hear that. Tuning AQL has proven tricky, and the rightest
answer was for the firmware to expose per station queues and the
interface statistics so that a smarter host could shave another 10ms
of or more of possible latency off the connection, maybe even get
below 4ms consistently for service to multiple stations at once, which
would be ideal for multi-player gaming.

I should have given more presentations than just the one at IEEE
802.11 and battlemesh, huh?

https://www.youtube.com/watch?v=Rb-UnHDw02o&t=1560

AQL on the ath10k was a hack. The ath9k still runs circles around it
and that too could have been improved with a tx-is-almost-done
interrupt.

>
> - Felix
>


-- 
This song goes out to all the folk that thought Stadia would work:
https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity-6981366665607352320-FXtz
Dave Täht CEO, TekLibre, LLC