On hardware with static fallback tables (e.g. mt76x2), rate probing attempts can be very expensive. On such devices, avoid sampling rates slower than the per-group max throughput rate, based on the assumption that the fallback table will take care of probing lower rates within that group if the higher rates fail. To make this work, this also fixes a wrong initialization in the previously unused per-group sorted rate array. To further reduce unnecessary probing attempts, skip duplicate attempts on rates slower than the max throughput rate. Signed-off-by: Felix Fietkau <nbd@xxxxxxxx> --- net/mac80211/rc80211_minstrel_ht.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/net/mac80211/rc80211_minstrel_ht.c b/net/mac80211/rc80211_minstrel_ht.c index ba230b037257..ad5da9a71da0 100644 --- a/net/mac80211/rc80211_minstrel_ht.c +++ b/net/mac80211/rc80211_minstrel_ht.c @@ -1059,6 +1059,21 @@ minstrel_get_sample_rate(struct minstrel_priv *mp, struct minstrel_ht_sta *mi) minstrel_get_duration(mi->max_prob_rate) * 3 < sample_dur) return -1; + + /* + * For devices with no configurable multi-rate retry, skip sampling + * below the per-group max throughput rate, and only use one sampling + * attempt per rate + */ + if (mp->hw->max_rates == 1 && + (minstrel_get_duration(mg->max_group_tp_rate[0]) < sample_dur || + mrs->attempts)) + return -1; + + /* Skip already sampled slow rates */ + if (sample_dur >= minstrel_get_duration(tp_rate1) && mrs->attempts) + return -1; + /* * Make sure that lower rates get sampled only occasionally, * if the link is working perfectly. -- 2.17.0
