Re: [PATCH net-next 2/3] net: qualcomm: rmnet: add tx packets aggregation

["Subash Abhinov Kasiviswanathan (KS)" <quic_subashab@quicinc.com>]

On 11/21/2022 12:00 AM, Daniele Palmas wrote:
> Il giorno dom 20 nov 2022 alle ore 18:48 Subash Abhinov
> Kasiviswanathan (KS) <quic_subashab@quicinc.com> ha scritto:
>>
>> On 11/20/2022 2:52 AM, Daniele Palmas wrote:
>>> Il giorno dom 20 nov 2022 alle ore 10:39 Bjørn Mork <bjorn@mork.no> ha scritto:
>>>>
>>>> Daniele Palmas <dnlplm@gmail.com> writes:
>>>>
>>>>> Ok, so rmnet would only take care of qmap rx packets deaggregation and
>>>>> qmi_wwan of the tx aggregation.
>>>>>
>>>>> At a conceptual level, implementing tx aggregation in qmi_wwan for
>>>>> passthrough mode could make sense, since the tx aggregation parameters
>>>>> belong to the physical device and are shared among the virtual rmnet
>>>>> netdevices, which can't have different aggr configurations if they
>>>>> belong to the same physical device.
>>>>>
>>>>> Bjørn, would this approach be ok for you?
>>>>
>>>> Sounds good to me, if this can be done within the userspace API
>>>> restrictions we've been through.
>>>>
>>>> I assume it's possible to make this Just Work(tm) in qmi_wwan
>>>> passthrough mode?  I do not think we want any solution where the user
>>>> will have to configure both qmi_wwan and rmnet to make things work
>>>> properly.
>>>>
>>>
>>> Yes, I think so: the ethtool configurations would apply to the
>>> qmi_wwan netdevice so that nothing should be done on the rmnet side.
>>>
>>> Regards,
>>> Daniele
>>
>> My only concern against this option is that we would now need to end up
>> implementing the same tx aggregation logic in the other physical device
>> drivers - mhi_netdev & ipa. Keeping this tx aggregation logic in rmnet
>> allows you to leverage it across all these various physical devices.
> 
> Yes, that's true.
> 
> But according to this discussion, the need for tx aggregation seems
> relevant just to USB devices (and maybe also a minority of them, since
> so far no one complained about it lacking), isn't it?
> 
> Regards,
> Daniele

Ah, it's more to do with it being future proof (in case some one does 
run into a similar throughput issue on the other platforms).

It also consolidates the functionality within rmnet driver for both tx & 
rx path. As you already know, the deaggregation, checksum offload & 
demuxing logic are already present in rmnet path in rx. The 
complementary muxing & checksum offload functionality are also present 
in the rmnet tx path, so I wanted to see if the aggregation 
functionality could be added in rmnet.