On Wed, 2019-10-23 at 11:59 +0200, Toke Høiland-Jørgensen wrote:
>
> + if (info->tx_time_est) {
> + struct sta_info *sta = NULL, *s;
> + struct rhlist_head *tmp;
> +
> + rcu_read_lock();
> +
> + for_each_sta_info(local, hdr->addr1, s, tmp) {
> + /* skip wrong virtual interface */
> + if (!ether_addr_equal(hdr->addr2, s->sdata->vif.addr))
> + continue;
> +
> + sta = s;
> + break;
> + }
I guess that is better than looking up the sdata and then using
sta_info_get(), but I think I'd like to see this wrapped into a function
(even if it's an inline) in sta_info.{c,h}.
> + airtime = ieee80211_calc_expected_tx_airtime(hw, vif, txq->sta,
> + skb->len);
> + if (airtime) {
> + /* We only have 10 bits in tx_time_est, so store airtime
> + * in increments of 4us and clamp the maximum to 2**12-1
> + */
> + airtime = min_t(u32, airtime, 4095) & ~3U;
> + info->tx_time_est = airtime >> 2;
> + ieee80211_sta_update_pending_airtime(local, tx.sta,
> + txq->ac, airtime,
> + false);
I wonder if it'd be better to pass the shifted value to
ieee80211_sta_update_pending_airtime() to avoid all the shifting in all
places?
You could even store the shifted value in "aql_tx_pending" and
"aql_total_pending_airtime" etc., it's completely equivalent, and only
shift it out for people looking at debugfs.
johannes
