[Make-wifi-fast] Carrying the CoDel timestamp into the driver

[Make-wifi-fast] Carrying the CoDel timestamp into the driver

[Toke Høiland-Jørgensen]

Hi Michal

In your mac80211 FQ-CoDel patch, you put the CoDel timestamp into a
union with the vif pointer in struct ieee80211_tx_info. This means the
timestamp is not available in the driver. I'm experimenting with some
changes where having the enqueue time available would be useful.

Do you have any good ideas as to where else we could store the enqueue
time somewhere the driver can retrieve it?

-Toke

Re: [Make-wifi-fast] Carrying the CoDel timestamp into the driver

[Michal Kazior]

On 10 August 2016 at 15:00, Toke Høiland-Jørgensen <toke@toke.dk> wrote:
> Hi Michal
>
> In your mac80211 FQ-CoDel patch, you put the CoDel timestamp into a
> union with the vif pointer in struct ieee80211_tx_info. This means the
> timestamp is not available in the driver. I'm experimenting with some
> changes where having the enqueue time available would be useful.
>
> Do you have any good ideas as to where else we could store the enqueue
> time somewhere the driver can retrieve it?

I did explore this idea for different purposes though - to maintain
per-packet expected_duration. I think it should be in the
linux-wireless archives. It compacted band, ack_frame_id I think but
you get limited number of *bits*.

.. or you could remove the rate control stuff from tx_info and convert
all drivers to use mac80211 API to fetch it per-station on-demand only
I guess..


Michał

Re: [Make-wifi-fast] Carrying the CoDel timestamp into the driver

[Toke Høiland-Jørgensen]

On 10 August 2016 17:27:16 CEST, Michal Kazior <michal.kazior@tieto.com> wrote:
>On 10 August 2016 at 15:00, Toke Høiland-Jørgensen <toke@toke.dk>
>wrote:
>> Hi Michal
>>
>> In your mac80211 FQ-CoDel patch, you put the CoDel timestamp into a
>> union with the vif pointer in struct ieee80211_tx_info. This means
>the
>> timestamp is not available in the driver. I'm experimenting with some
>> changes where having the enqueue time available would be useful.
>>
>> Do you have any good ideas as to where else we could store the
>enqueue
>> time somewhere the driver can retrieve it?
>
>I did explore this idea for different purposes though - to maintain
>per-packet expected_duration. I think it should be in the
>linux-wireless archives. It compacted band, ack_frame_id I think but
>you get limited number of *bits*.
>
>.. or you could remove the rate control stuff from tx_info and convert
>all drivers to use mac80211 API to fetch it per-station on-demand only
>I guess..

Right. Guess I'll put that on the "things to look into" list.

Exactly why is it we can't just grow tx_info by a couple of bytes?

-Toke

Re: [Make-wifi-fast] Carrying the CoDel timestamp into the driver

[Michal Kazior]

On 10 August 2016 at 17:41, Toke Høiland-Jørgensen <toke@toke.dk> wrote:
> Exactly why is it we can't just grow tx_info by a couple of bytes?

Since it's stored in the inline buffer sk_buff->cb[] it's limited to its size.


Michał

Re: [Make-wifi-fast] Carrying the CoDel timestamp into the driver

[Toke Høiland-Jørgensen]

On 10 August 2016 18:01:51 CEST, Michal Kazior <michal.kazior@tieto.com> wrote:
>On 10 August 2016 at 17:41, Toke Høiland-Jørgensen <toke@toke.dk>
>wrote:
>> Exactly why is it we can't just grow tx_info by a couple of bytes?
>
>Since it's stored in the inline buffer sk_buff->cb[] it's limited to
>its size.

Right, noted. Thanks! :)

-Toke
>
>
>Michał

Re: [Make-wifi-fast] Carrying the CoDel timestamp into the driver

[Dave Taht]

There is (and has always been) the underutilized skb->tstamp pointer
in the skb itself.

http://lxr.free-electrons.com/source/include/linux/skbuff.h#L416



On Wed, Aug 10, 2016 at 6:06 PM, Toke Høiland-Jørgensen <toke@toke.dk> wrote:
>
>
> On 10 August 2016 18:01:51 CEST, Michal Kazior <michal.kazior@tieto.com> wrote:
>>On 10 August 2016 at 17:41, Toke Høiland-Jørgensen <toke@toke.dk>
>>wrote:
>>> Exactly why is it we can't just grow tx_info by a couple of bytes?
>>
>>Since it's stored in the inline buffer sk_buff->cb[] it's limited to
>>its size.
>
> Right, noted. Thanks! :)
>
> -Toke
>>
>>
>>Michał
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast



-- 
Dave Täht
Let's go make home routers and wifi faster! With better software!
http://blog.cerowrt.org

Re: [Make-wifi-fast] Carrying the CoDel timestamp into the driver

[Dave Taht]

On Wed, Aug 10, 2016 at 7:06 PM, Dave Taht <dave.taht@gmail.com> wrote:
> There is (and has always been) the underutilized skb->tstamp pointer
> in the skb itself.
>
> http://lxr.free-electrons.com/source/include/linux/skbuff.h#L416

Also another field that could possibly be munged on at this point in
the mac80211 layer is priority?

>
>
>
> On Wed, Aug 10, 2016 at 6:06 PM, Toke Høiland-Jørgensen <toke@toke.dk> wrote:
>>
>>
>> On 10 August 2016 18:01:51 CEST, Michal Kazior <michal.kazior@tieto.com> wrote:
>>>On 10 August 2016 at 17:41, Toke Høiland-Jørgensen <toke@toke.dk>
>>>wrote:
>>>> Exactly why is it we can't just grow tx_info by a couple of bytes?
>>>
>>>Since it's stored in the inline buffer sk_buff->cb[] it's limited to
>>>its size.
>>
>> Right, noted. Thanks! :)
>>
>> -Toke
>>>
>>>
>>>Michał
>> _______________________________________________
>> Make-wifi-fast mailing list
>> Make-wifi-fast@lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>
>
>
> --
> Dave Täht
> Let's go make home routers and wifi faster! With better software!
> http://blog.cerowrt.org



-- 
Dave Täht
Let's go make home routers and wifi faster! With better software!
http://blog.cerowrt.org