Stop reading (and copying) from ieee80211_tx_rate to ath_tx_info.rates
after list end tag (count == 0, idx < 0), prevents copying of garbage
to card registers.
Note: no need to write to the remaining ath_tx_info.rates entries
as the complete ath_tx_info struct is already initialized to zero from
both call sites.
Signed-off-by: Peter Seiderer <[email protected]>
---
drivers/net/wireless/ath/ath9k/xmit.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/wireless/ath/ath9k/xmit.c b/drivers/net/wireless/ath/ath9k/xmit.c
index d0caf1de2bde..ec9bad2d9510 100644
--- a/drivers/net/wireless/ath/ath9k/xmit.c
+++ b/drivers/net/wireless/ath/ath9k/xmit.c
@@ -1271,7 +1271,7 @@ static void ath_buf_set_rate(struct ath_softc *sc, struct ath_buf *bf,
int phy;
if (!rates[i].count || (rates[i].idx < 0))
- continue;
+ break;
rix = rates[i].idx;
info->rates[i].Tries = rates[i].count;
--
2.35.1
Fill all requested rates (in case of ath9k 4 rate slots are
available, so fill all 4 instead of only 3), improves throughput in
noisy environment.
Signed-off-by: Peter Seiderer <[email protected]>
---
net/mac80211/rc80211_minstrel_ht.c | 14 +++++++-------
1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/net/mac80211/rc80211_minstrel_ht.c b/net/mac80211/rc80211_minstrel_ht.c
index 9c6ace858107..cd6a0f153688 100644
--- a/net/mac80211/rc80211_minstrel_ht.c
+++ b/net/mac80211/rc80211_minstrel_ht.c
@@ -1436,17 +1436,17 @@ minstrel_ht_update_rates(struct minstrel_priv *mp, struct minstrel_ht_sta *mi)
/* Start with max_tp_rate[0] */
minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_tp_rate[0]);
- if (mp->hw->max_rates >= 3) {
- /* At least 3 tx rates supported, use max_tp_rate[1] next */
- minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_tp_rate[1]);
- }
+ /* Fill up remaining, keep one entry for max_probe_rate */
+ for (; i < (mp->hw->max_rates - 1); i++)
+ minstrel_ht_set_rate(mp, mi, rates, i, mi->max_tp_rate[i]);
- if (mp->hw->max_rates >= 2) {
+ if (i < mp->hw->max_rates)
minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_prob_rate);
- }
+
+ if (i < IEEE80211_TX_RATE_TABLE_SIZE)
+ rates->rate[i].idx = -1;
mi->sta->max_rc_amsdu_len = minstrel_ht_get_max_amsdu_len(mi);
- rates->rate[i].idx = -1;
rate_control_set_rates(mp->hw, mi->sta, rates);
}
--
2.35.1
+Felix
Peter Seiderer <[email protected]> writes:
> Fill all requested rates (in case of ath9k 4 rate slots are
> available, so fill all 4 instead of only 3), improves throughput in
> noisy environment.
How did you test this? Could you quantify the gains in throughput you saw?
-Toke
> Signed-off-by: Peter Seiderer <[email protected]>
> ---
> net/mac80211/rc80211_minstrel_ht.c | 14 +++++++-------
> 1 file changed, 7 insertions(+), 7 deletions(-)
>
> diff --git a/net/mac80211/rc80211_minstrel_ht.c b/net/mac80211/rc80211_minstrel_ht.c
> index 9c6ace858107..cd6a0f153688 100644
> --- a/net/mac80211/rc80211_minstrel_ht.c
> +++ b/net/mac80211/rc80211_minstrel_ht.c
> @@ -1436,17 +1436,17 @@ minstrel_ht_update_rates(struct minstrel_priv *mp, struct minstrel_ht_sta *mi)
> /* Start with max_tp_rate[0] */
> minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_tp_rate[0]);
>
> - if (mp->hw->max_rates >= 3) {
> - /* At least 3 tx rates supported, use max_tp_rate[1] next */
> - minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_tp_rate[1]);
> - }
> + /* Fill up remaining, keep one entry for max_probe_rate */
> + for (; i < (mp->hw->max_rates - 1); i++)
> + minstrel_ht_set_rate(mp, mi, rates, i, mi->max_tp_rate[i]);
>
> - if (mp->hw->max_rates >= 2) {
> + if (i < mp->hw->max_rates)
> minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_prob_rate);
> - }
> +
> + if (i < IEEE80211_TX_RATE_TABLE_SIZE)
> + rates->rate[i].idx = -1;
>
> mi->sta->max_rc_amsdu_len = minstrel_ht_get_max_amsdu_len(mi);
> - rates->rate[i].idx = -1;
> rate_control_set_rates(mp->hw, mi->sta, rates);
> }
>
> --
> 2.35.1
Hello Toke, Felix,
On Mon, 04 Apr 2022 20:21:24 +0200, Toke Høiland-Jørgensen <[email protected]> wrote:
> +Felix
>
> Peter Seiderer <[email protected]> writes:
>
> > Fill all requested rates (in case of ath9k 4 rate slots are
> > available, so fill all 4 instead of only 3), improves throughput in
> > noisy environment.
>
> How did you test this? Could you quantify the gains in throughput you saw?
Investigating some performance degradation of a wifi system original with ath5k
cards using a legacy kernel and madwifi driver compared against the
performance using a ath9k card running IBSS mode over long distance/with
additional amplifier and using iperf for measurement...
With the ath5k cards under bad conditions the throughput is going down below
10 Mbits/s but with stable throughput..., with the ath9k card and the exact same
setup there are short periods with good throughput values and periods up to 10-15
seconds with 0 Mbits/s...
The actual in-door/laboratory test setup is a wired connection between the
two wifi systems (with fixed attenuators in between) and an adjustable noise/
disturb signal induced by a signal generator...
Without this patch the 0 Mbits/s periods from the field test are reproducible,
with this patch applied we see a more or less stable throughput of 1-5 Mbits/s...
Regards,
Peter
>
> -Toke
>
> > Signed-off-by: Peter Seiderer <[email protected]>
> > ---
> > net/mac80211/rc80211_minstrel_ht.c | 14 +++++++-------
> > 1 file changed, 7 insertions(+), 7 deletions(-)
> >
> > diff --git a/net/mac80211/rc80211_minstrel_ht.c b/net/mac80211/rc80211_minstrel_ht.c
> > index 9c6ace858107..cd6a0f153688 100644
> > --- a/net/mac80211/rc80211_minstrel_ht.c
> > +++ b/net/mac80211/rc80211_minstrel_ht.c
> > @@ -1436,17 +1436,17 @@ minstrel_ht_update_rates(struct minstrel_priv *mp, struct minstrel_ht_sta *mi)
> > /* Start with max_tp_rate[0] */
> > minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_tp_rate[0]);
> >
> > - if (mp->hw->max_rates >= 3) {
> > - /* At least 3 tx rates supported, use max_tp_rate[1] next */
> > - minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_tp_rate[1]);
> > - }
> > + /* Fill up remaining, keep one entry for max_probe_rate */
> > + for (; i < (mp->hw->max_rates - 1); i++)
> > + minstrel_ht_set_rate(mp, mi, rates, i, mi->max_tp_rate[i]);
> >
> > - if (mp->hw->max_rates >= 2) {
> > + if (i < mp->hw->max_rates)
> > minstrel_ht_set_rate(mp, mi, rates, i++, mi->max_prob_rate);
> > - }
> > +
> > + if (i < IEEE80211_TX_RATE_TABLE_SIZE)
> > + rates->rate[i].idx = -1;
> >
> > mi->sta->max_rc_amsdu_len = minstrel_ht_get_max_amsdu_len(mi);
> > - rates->rate[i].idx = -1;
> > rate_control_set_rates(mp->hw, mi->sta, rates);
> > }
> >
> > --
> > 2.35.1
Peter Seiderer <[email protected]> writes:
> Stop reading (and copying) from ieee80211_tx_rate to ath_tx_info.rates
> after list end tag (count == 0, idx < 0), prevents copying of garbage
> to card registers.
>
> Note: no need to write to the remaining ath_tx_info.rates entries
> as the complete ath_tx_info struct is already initialized to zero from
> both call sites.
>
> Signed-off-by: Peter Seiderer <[email protected]>
Acked-by: Toke Høiland-Jørgensen <[email protected]>
Peter Seiderer <[email protected]> writes:
> Hello Toke,
>
> On Mon, 04 Apr 2022 20:19:39 +0200, Toke Høiland-Jørgensen <[email protected]> wrote:
>
>> Peter Seiderer <[email protected]> writes:
>>
>> > Stop reading (and copying) from ieee80211_tx_rate to ath_tx_info.rates
>> > after list end tag (count == 0, idx < 0), prevents copying of garbage
>> > to card registers.
>>
>> In the normal case I don't think this patch does anything, since any
>> invalid rate entries will already be skipped (just one at a time instead
>> of all at once). So this comment is a bit misleading.
>
> Save some (minimal) compute time? Found it something misleading while
> debugging to see random values written out to the card and found this
> comment in net/mac80211/rate.c:
>
> 648 /*
> 649 * make sure there's no valid rate following
> 650 * an invalid one, just in case drivers don't
> 651 * take the API seriously to stop at -1.
> 652 */
>
> and multiple places doing the same check (count == 0, idx < 0) for validation
> e.g.:
>
> 723 if (i < ARRAY_SIZE(info->control.rates) &&
> 724 info->control.rates[i].idx >= 0 &&
> 725 info->control.rates[i].count) {
>
> or
>
> 742 if (rates[i].idx < 0 || !rates[i].count)
> 743 break;
>
>>
>> Also, Minstrel could in principle produce a rate sequence where the
>> indexes are all positive, but there's one in the middle with a count of
>> 0, couldn't it? With this patch, the last entries of such a sequence
>> would now be skipped...
>
> According to net/mac80211/rc80211_minstrel_ht.c:
>
> 1128 static bool
> 1129 minstrel_ht_txstat_valid(struct minstrel_priv *mp, struct minstrel_ht_sta * mi,
> 1130 struct ieee80211_tx_rate *rate)
> 1131 {
> 1132 int i;
> 1133
> 1134 if (rate->idx < 0)
> 1135 return false;
> 1136
> 1137 if (!rate->count)
> 1138 return false;
> 1139
>
> minstrel although evaluates a rate count of zero as invalid...
So my concern was mostly that the documentation (in mac80211.h) says
that an idx of -1 indicates the end, but says nothing about the count.
Which implies that in principle you could have a rate table of { idx,
count } like { 1, 1 }, { 2, 0 }, { 3, 1 } which would mean all three
rates was valid but the second one would just be "skipped" due to a
count of zero.
But it seems that the code populating the rate table that you linked
above (lines 742/743) actually do abort on either condition, so I guess
it's safe to do so in the driver as well...
-Toke
Peter Seiderer <[email protected]> wrote:
> Stop reading (and copying) from ieee80211_tx_rate to ath_tx_info.rates
> after list end tag (count == 0, idx < 0), prevents copying of garbage
> to card registers.
>
> Note: no need to write to the remaining ath_tx_info.rates entries
> as the complete ath_tx_info struct is already initialized to zero from
> both call sites.
>
> Signed-off-by: Peter Seiderer <[email protected]>
> Acked-by: Toke Høiland-Jørgensen <[email protected]>
> Signed-off-by: Kalle Valo <[email protected]>
Patch applied to ath-next branch of ath.git, thanks.
24584d4f0afc ath9k: fix ath_get_rate_txpower() to respect the rate list end tag
--
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches