Vasanthakumar Thiagarajan <vthiagar@xxxxxxxxxxxxxx> writes:
> On 2019-08-20 22:21, Toke Høiland-Jørgensen wrote:
>> [... snip ... ]
>>
>>> +static const struct ieee80211_ops ath11k_ops = {
>>> + .tx = ath11k_mac_op_tx,
>>
>> No wake_tx_queue? :(
>
> Yes, packet queueing is handled in firmware. This makes sense
> especially when we enable 802.11 encap offload support where most of
> the data path processing in mac80211 will be skipped and packet is
> given to driver/firmware in 802.3 format itself. Then firmware would
> take care of all the classification, queueing and encapsulation
> operations.
Well, so does ath10k, and yet we still saw a significant improvement by
moving queueing back into the host where it can be handled by the
FQ-CoDel-enabled queueing structure.
So, *how* does the firmware handle the queueing? Does it have
per-stations queues? Per-flow queues? What's the latency under load
figures for the firmware queues?
-Toke
