[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 4 Apr 2023 11:18:24 +0800
From: Yunsheng Lin <linyunsheng@...wei.com>
To: Jakub Kicinski <kuba@...nel.org>
CC: <davem@...emloft.net>, <netdev@...r.kernel.org>,
<edumazet@...gle.com>, <pabeni@...hat.com>, <hawk@...nel.org>,
<ilias.apalodimas@...aro.org>
Subject: Re: [RFC net-next 1/2] page_pool: allow caching from safely localized
NAPI
On 2023/4/4 9:45, Jakub Kicinski wrote:
> On Tue, 4 Apr 2023 08:53:36 +0800 Yunsheng Lin wrote:
>> I wonder if we can make this more generic by adding the skb to per napi
>> list instead of sd->defer_list, so that we can always use NAPI kicking to
>> flush skb as net_tx_action() done for sd->completion_queue instead of
>> softirq kicking?
>>
>> And it seems we know which napi binds to a specific socket through
>> busypoll mechanism, we can reuse that to release a skb to the napi
>> bound to that socket?
>
> Seems doable. My thinking was to first see how well the simpler scheme
> performs with production workloads because it should have no downsides.
Look forwording to some performs data with production workloads:)
> Tracking real NAPI pointers per socket and extra RCU sync to manage
> per-NAPI defer queues may have perf cost.
I suppose the extra RCU sync only happen on the napi add/del process,
not in the data path?
> .
>
Powered by blists - more mailing lists