Re: [PATCH wireless-next v2 7/7] wifi: rt2x00usb: use a threaded NAPI with GRO for RX

[Stanislaw Gruszka <stf_xl@wp.pl>]

On Sun, Nov 01, 2026 at 12:51:14AM +0800, Ho Jie Feng wrote:
> GRO makes sense now that RX aggregation is enabled, since multiple
> frames arrive at once.
> 
> This code is referenced from commit 5323d3e50c20
> ("wifi: mt76: mt76u: use a threaded NAPI for the RX path")
> 
> Test results:
>    before after
> TX 30.5   30.5
> RX 80.9   161.7

What is the explanation for this gain?

I tested on my devices and do not see RX performance gain.
I would expect some benefits, but I don't see any and wonder why.
Maybe on your setup CPU get saturated on single thread ?
Or maybe I'm not testing traffic that benefits from this?

> Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
> ---
>  drivers/net/wireless/ralink/rt2x00/rt2x00.h   |  7 ++-
>  .../net/wireless/ralink/rt2x00/rt2x00dev.c    |  6 ++-
>  .../net/wireless/ralink/rt2x00/rt2x00usb.c    | 49 ++++++++++++++++---
>  3 files changed, 52 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00.h b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
> index 4c83168e9c169..25262b96c6941 100644
> --- a/drivers/net/wireless/ralink/rt2x00/rt2x00.h
> +++ b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
> @@ -955,9 +955,14 @@ struct rt2x00_dev {
>  	/**
>  	 * Scheduled work for TX/RX done handling (USB devices)
>  	 */
> -	struct work_struct rxdone_work;
>  	struct work_struct txdone_work;
>  
> +	/*
> +	 * NAPI context for RX done handling (USB devices).
> +	 */
> +	struct net_device *napi_dev;
> +	struct napi_struct napi;
> +
>  	/*
>  	 * Powersaving work
>  	 */
> diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
> index 8f02b62cecdf6..6226bcbf24e5b 100644
> --- a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
> +++ b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
> @@ -847,7 +847,11 @@ static void rt2x00lib_rxdone_skb(struct queue_entry *entry)
>  	rx_status->bw = rxdesc.bw;
>  	rx_status->antenna = rt2x00dev->link.ant.active.rx;
>  
> -	ieee80211_rx_ni(rt2x00dev->hw, entry->skb);
> +	if (rt2x00dev->napi_dev)
> +		ieee80211_rx_napi(rt2x00dev->hw, NULL, entry->skb,
> +				  &rt2x00dev->napi);

The docs for ieee80211_rx_napi() states:

"Must not run concurrently with
ieee80211_tx_status_skb() or ieee80211_tx_status_ni()."

So we would need to either introduce lock similar like mt76 or
assure ieee80211_ts_status() and ieee80211_rx_napi() are done
in the same thread.

Regards
Stanislaw