Wen Gong <wgong@xxxxxxxxxxxxxx> writes:
> On 2020-04-09 22:53, Kalle Valo wrote:
>> Wen Gong <wgong@xxxxxxxxxxxxxx> writes:
>>
>>> + ar->workqueue_tx_complete =
>>> + create_singlethread_workqueue("ath10k_tx_complete_wq");
>>> + if (!ar->workqueue_tx_complete)
>>> + goto err_free_aux_wq;
>>
>> We already have three threads:
>>
>> ath/ath10k/core.c: ar->workqueue =
>> create_singlethread_workqueue("ath10k_wq");
>> ath/ath10k/core.c: ar->workqueue_aux =
>> create_singlethread_workqueue("ath10k_aux_wq");
>> ath/ath10k/sdio.c: ar_sdio->workqueue =
>> create_singlethread_workqueue("ath10k_sdio_wq");
>>
>> Do we really need a fourth one? For example, why can't we use
>> ar->workqueue_aux?
>
> For tcp test, it has 4 thread work meanwhile:
> tx_bundle_skbs(ar->workqueue),
> rx_indication(ar->workqueue_aux),
> sdio_async_tx_request(ar_sdio->workqueue),
> tx_bundle_complete(ar->workqueue_tx_complete)
>
> It has 4+ cpu/core in system, if reduced to 3 threads, then tcp
> throughput will drop. only when it only has 1/2/3 cpu/core in system,
> then reduced to 3 threads will not drop.
How much does it drop? Please add the justification (with numbers) for
the new thread to the commit log, so that the reason is properly
documented.
--
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
