Re: [PATCH v3 2/2] wifi: rtw88: support channel switch in AP mode

[Luka Gejak]

On Mon, 05 Oct 2026, Mehmet Fide wrote:

> @@ -626,7 +652,8 @@ static int rtw_ops_set_key(struct ieee80211_hw *hw, enum set_key_cmd cmd,
> [...]
> -	if (rtw_get_lps_deep_mode(rtwdev) == LPS_DEEP_MODE_PG)
> +	if (rtw_get_lps_deep_mode(rtwdev) == LPS_DEEP_MODE_PG &&
> +	    !rtw_fw_csa_active(rtwdev))
>  		rtw_fw_download_rsvd_page(rtwdev);

The assoc change downloads the same page, and it is not guarded:

	if (changed & BSS_CHANGED_ASSOC) {
		rtw_vif_assoc_changed(rtwvif, conf);
		if (vif->cfg.assoc) {
			rtw_coex_connect_notify(rtwdev, COEX_ASSOCIATE_FINISH);

			rtw_fw_download_rsvd_page(rtwdev);

The page holds the beacon of the AP vif, and building it fetches that
beacon:

	case RSVD_BEACON:
		skb_new = ieee80211_beacon_get_tim(hw, vif, &tim_offset, NULL, 0);

mac80211 steps the countdown on that fetch, not once per transmitted
beacon:

	if (beacon->cntdwn_counter_offsets[0]) {
		if (!is_template)
			ieee80211_beacon_update_cntdwn(vif, link->link_id);

so an association during the countdown moves the count a second time in
the same interval. Can this download skip the page while a switch is
announced, the way set_key does?

> @@ -900,6 +945,13 @@ static int rtw_ops_hw_scan(struct ieee80211_hw *hw, struct ieee80211_vif *vif,
>  	mutex_lock(&rtwdev->mutex);
> +
> +	rtw_iterate_vifs(rtwdev, rtw_csa_active_iter, &csa_active);
> +	if (csa_active) {
> +		mutex_unlock(&rtwdev->mutex);
> +		return -EBUSY;
> +	}

The hw scan check sits below the offload test, so with firmware without
scan offload the op returns before it:

	if (!rtw_fw_feature_check(&rtwdev->fw, FW_FEATURE_SCAN_OFFLOAD))
		return 1;

A return of one tells mac80211 to run the scan in software:

	if (hw_scan && rc == 1) {
		/*
		 * we can't fall back to software for P2P-GO
		 * as it must update NoA etc.
		 */
		if (ieee80211_vif_type_p2p(&sdata->vif) ==
				NL80211_IFTYPE_P2P_GO)
			return -EOPNOTSUPP;
		hw_scan = false;
		goto again;
	}

so the AP leaves the channel during the countdown anyway. Can the check
move above the feature test?

Best regards,
Luka Gejak