Re: [PATCHv2] mac80211: increment rx stats according to USES_RSS flag

[Johannes Berg <johannes@sipsolutions.net>]

On Wed, 2021-02-17 at 17:26 +0530, Thiraviyam Mariyappan wrote:
> Currently, rx_stats were updated regardless of USES_RSS flag is
> enabled/disabled. So, updating the rx_stats from percpu pointers
> according to the USES_RSS flag.

OK actually, I'm not going to fix the commit log, you'll probably have
to resend it anyway.


> @@ -425,7 +426,8 @@ static void mesh_sta_info_init(struct ieee80211_sub_if_data *sdata,
>  					&basic_rates);

>  	spin_lock_bh(&sta->mesh->plink_lock);
> -	sta->rx_stats.last_rx = jiffies;
> +	stats = ieee80211_get_rx_stats(&local->hw, sta);
> +	stats->last_rx = jiffies;

This doesn't really make much sense? Not sure why that is even updating
anything at all, it doesn't update anything else?

Or at least you didn't change anything else, maybe you should have?
> 
> @@ -1734,49 +1745,49 @@ ieee80211_rx_h_sta_process(struct ieee80211_rx_data *rx)
>  	 * something went wrong the first time.
>  	 */
>  	if (rx->sdata->vif.type == NL80211_IFTYPE_ADHOC) {
> -		u8 *bssid = ieee80211_get_bssid(hdr, rx->skb->len,
> +		u8 *bssid = ieee80211_get_bssid(hdr, skb->len,

That seems unrelated.

> @@ -3625,8 +3648,10 @@ ieee80211_rx_h_mgmt(struct ieee80211_rx_data *rx)
>  	/* queue up frame and kick off work to process it */
>  	skb_queue_tail(&sdata->skb_queue, rx->skb);
>  	ieee80211_queue_work(&rx->local->hw, &sdata->work);
> -	if (rx->sta)
> -		rx->sta->rx_stats.packets++;
> +	if (rx->sta) {
> +		stats = ieee80211_get_rx_stats(&rx->sdata->local->hw, rx->sta);
> +		stats->packets++;
> +	}
> 

Picking this for no particular reason - everything else in this patch is
unnecessary since we have rx_path_lock held afaict, so it doesn't
matter. The whole per-cpu status stuff only matters once you get into
fast-rx path.


I'd argue that had you written a proper commit log that actually says
why you need to change things, you'd probably even have noticed these
issues yourself.

johannes