Re: [PATCH v2] net: bcmgenet: fix throughtput regression

[Florian Fainelli <f.fainelli@gmail.com>]

On 27/02/15 14:10, David Miller wrote:
> From: Florian Fainelli <f.fainelli@gmail.com>
> Date: Fri, 27 Feb 2015 09:38:12 -0800
> 
>> On 27/02/15 06:27, Jaedon Shin wrote:
>>> This patch adds bcmgenet_tx_poll for all active tx_rings. It can reduce
>>> the interrupt load and send xmit in upper network stack on time.
>>>
>>> The bcmgenet_tx_reclaim of tx_ring[{0,1,2,3}] process only under 18
>>> descriptors is to late reclaiming transmitted skb. Therefore,
>>> performance degradation of xmit after 605ad7f ("tcp: refine TSO
>>> autosizing").
>>
>> This looks very similar to my previous attempts at using NAPI for TX
>> completion, thanks for doing this.
>>
>> One thing you are not mentioning in your commit message is that ring16
>> used to be reclaimed/cleaned as part of the shared RX/TX NAPI context
>> (bcmgenet_poll), while you are now dedicating one and using
>> bcmgenet_tx_poll() for reclaim. This is a big enough change in the
>> driver structure that deserves to be reflected in the commit message.
>>
>>>
>>> Signed-off-by: Jaedon Shin <jaedon.shin@gmail.com>
>>
>> Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
> 
> Besides needing an updated commit message, this overlaps with a
> cleanup patch in my queue from Petri Gynther that touches the same
> exact code.
> 
> 	http://patchwork.ozlabs.org/patch/443604/
> 
> How would you guys like me to sort this all out?  Drop Petri's
> change for now?

If you can take Petri's change now and Jaedon then resubmits on top of
that change, would that be acceptable?
-- 
Florian