Bryan O'Donoghue <bryan.odonoghue@xxxxxxxxxx> writes:
> On 20/03/2022 13:21, Bryan O'Donoghue wrote:
>> On 18/03/2022 19:58, Edmond Gagnon wrote:
>>> + INIT_DELAYED_WORK(&wcn->get_stats_work, wcn36xx_get_stats_work);
>>
>> Instead of forking a worker and polling we could add the relevant
>> SMD command to
>>
>> static int wcn36xx_smd_tx_compl_ind(struct wcn36xx *wcn, void *buf,
>> size_t len)
>> {
>> wcn36xx_smd_get_stats(wcn, 0xSomeMask);
>> }
>>
>> That way we only ever ask for and report a new TX data rate when we
>> know a TX event - and hence a potential TX data-rate update - has
>> taken place.
>>
>> ---
>> bod
>>
>
> Thinking a bit more
>
> - Do the SMD get_stats in the tx completion
> This might be a problem initiating another SMD transaction inside
> of an SMD callback. But is the most straight forward way to
> get the data while avoiding alot of needless polling.
>
> - Schedule your worker from the TX completion
> Again you should only care about gathering the data when you know
> something has happened which necessitates gathering that data
> like TX completion
>
> - Schedule your worker from the RX indication routine
> Seems not as logical as the first two but it might be easier
> to schedule the worker in the RX data handler
>
> Either way, I do think you should only gather this data on an event,
> not as a continuous poll.
I agree, a continuous poll is not a good idea as it affects power
consumption. What about struct ieee80211_ops::sta_statistics? AFAIK
that's called only when user space is requestings stats so the overhead
should be minimal.
--
https://patchwork.kernel.org/project/linux-wireless/list/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
