Skip to contentSkip to conversation

mac80211: get the rates masks from the txrc in rate_control_get_rate

PATCH Superseded Karl Beldan 3 participants 16 messages Mar 4, 2013 – Mar 16, 2013 mac80211

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/2 mac80211: get the rates masks from the txrc in rate_control_get_rate

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

---
 net/mac80211/rate.c |   17 ++++++++---------
 1 file changed, 8 insertions(+), 9 deletions(-)
net/mac80211/rate.c +8 -9
index dd88381..c1e5f25 100644
@@ -435,8 +435,8 @@ void rate_control_get_rate(struct ieee80211_sub_if_data *sdata,
 	struct ieee80211_sta *ista = NULL;
 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(txrc->skb);
 	int i;
-	u32 mask;
-	u8 mcs_mask[IEEE80211_HT_MCS_MASK_LEN];
+	u32 *mask;
+	u8 *mcs_mask;
 
 	if (sta && test_sta_flag(sta, WLAN_STA_RATE_CONTROL)) {
 		ista = &sta->sta;
@@ -459,14 +459,13 @@ void rate_control_get_rate(struct ieee80211_sub_if_data *sdata,
 	 * default mask (allow all rates) is used to save some processing for
 	 * the common case.
 	 */
-	mask = sdata->rc_rateidx_mask[info->band];
-	memcpy(mcs_mask, sdata->rc_rateidx_mcs_mask[info->band],
-	       sizeof(mcs_mask));
-	if (mask != (1 << txrc->sband->n_bitrates) - 1) {
+	mask = &txrc->rate_idx_mask;
+	mcs_mask = txrc->rate_idx_mcs_mask;
+	if (*mask != (1 << txrc->sband->n_bitrates) - 1) {
 		if (sta) {
 			/* Filter out rates that the STA does not support */
-			mask &= sta->sta.supp_rates[info->band];
-			for (i = 0; i < sizeof(mcs_mask); i++)
+			*mask &= sta->sta.supp_rates[info->band];
+			for (i = 0; i < sizeof(txrc->rate_idx_mcs_mask); i++)
 				mcs_mask[i] &= sta->sta.ht_cap.mcs.rx_mask[i];
 		}
 		/*
@@ -479,7 +478,7 @@ void rate_control_get_rate(struct ieee80211_sub_if_data *sdata,
 			if (info->control.rates[i].idx < 0)
 				break;
 			rate_idx_match_mask(&info->control.rates[i], txrc,
-					    mask, mcs_mask);
+					    *mask, mcs_mask);
 		}
 	}
 
Signature · 1 line
1.7.10.GIT
#1 lore raw

PATCH v2 2/2 mac80211: fix the check for mcs rates masking

From: Karl Beldan <[email protected]>

Currently the mcs bitrates mask rate_idx_mcs_mask is only applied when the pre-ht bitrates mask rate_idx_mask of the same band differs from the default mask.
Fix it by comparing the rate_idx_mcs_mask with the driver ht caps.

---
 net/mac80211/rate.c |   10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
net/mac80211/rate.c +6 -4
index c1e5f25..28f1936 100644
@@ -455,13 +455,15 @@ void rate_control_get_rate(struct ieee80211_sub_if_data *sdata,
 	ref->ops->get_rate(ref->priv, ista, priv_sta, txrc);
 
 	/*
-	 * Try to enforce the rateidx mask the user wanted. skip this if the
-	 * default mask (allow all rates) is used to save some processing for
-	 * the common case.
+	 * Try to enforce the rateidx masks the user wanted. Skip this if mask
+	 * is the default mask (allow all rates) and mcs_mask does not differ
+	 * from the driver ht caps to save some processing for the common case.
 	 */
 	mask = &txrc->rate_idx_mask;
 	mcs_mask = txrc->rate_idx_mcs_mask;
-	if (*mask != (1 << txrc->sband->n_bitrates) - 1) {
+	if (*mask != (1 << txrc->sband->n_bitrates) - 1 ||
+		memcmp(txrc->sband->ht_cap.mcs.rx_mask, txrc->rate_idx_mcs_mask,
+		       sizeof(txrc->rate_idx_mcs_mask))) {
 		if (sta) {
 			/* Filter out rates that the STA does not support */
 			*mask &= sta->sta.supp_rates[info->band];
Signature · 1 line
1.7.10.GIT
maintainer
#2 lore raw

Re: [PATCH v2 2/2] mac80211: fix the check for mcs rates masking

Quoted 23 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently the mcs bitrates mask rate_idx_mcs_mask is only applied when the pre-ht bitrates mask rate_idx_mask of the same band differs from the default mask.
Fix it by comparing the rate_idx_mcs_mask with the driver ht caps.

---
 net/mac80211/rate.c |   10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
net/mac80211/rate.c +0 -3
index c1e5f25..28f1936 100644
@@ -455,13 +455,15 @@ void rate_control_get_rate(struct ieee80211_sub_if_data *sdata,
 	ref->ops->get_rate(ref->priv, ista, priv_sta, txrc);
 
 	/*
-	 * Try to enforce the rateidx mask the user wanted. skip this if the
-	 * default mask (allow all rates) is used to save some processing for
-	 * the common case.

net/mac80211/rate.c void rate_control_get_rate(struct ieee80211_sub_if_data *sdata, · line 466

Quoted 10 lines
net/mac80211/rate.c +6 -1
+	 * Try to enforce the rateidx masks the user wanted. Skip this if mask
+	 * is the default mask (allow all rates) and mcs_mask does not differ
+	 * from the driver ht caps to save some processing for the common case.
 	 */
 	mask = &txrc->rate_idx_mask;
 	mcs_mask = txrc->rate_idx_mcs_mask;
-	if (*mask != (1 << txrc->sband->n_bitrates) - 1) {
+	if (*mask != (1 << txrc->sband->n_bitrates) - 1 ||
+		memcmp(txrc->sband->ht_cap.mcs.rx_mask, txrc->rate_idx_mcs_mask,
+		       sizeof(txrc->rate_idx_mcs_mask))) {

Please fix the indentation, the memcmp should be indented to just after the opening parenthesis of "if ("

johannes

author
#3 lore raw

Re: [PATCH v2 2/2] mac80211: fix the check for mcs rates masking

Quoted 29 lines · Johannes Berg
Quoted 28 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently the mcs bitrates mask rate_idx_mcs_mask is only applied when the pre-ht bitrates mask rate_idx_mask of the same band differs from the default mask.
Fix it by comparing the rate_idx_mcs_mask with the driver ht caps.

---
 net/mac80211/rate.c |   10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
net/mac80211/rate.c +3 -3
index c1e5f25..28f1936 100644
@@ -455,13 +455,15 @@ void rate_control_get_rate(struct ieee80211_sub_if_data *sdata,
 	ref->ops->get_rate(ref->priv, ista, priv_sta, txrc);
 
 	/*
-	 * Try to enforce the rateidx mask the user wanted. skip this if the
-	 * default mask (allow all rates) is used to save some processing for
-	 * the common case.
+	 * Try to enforce the rateidx masks the user wanted. Skip this if mask
+	 * is the default mask (allow all rates) and mcs_mask does not differ
+	 * from the driver ht caps to save some processing for the common case.
 	 */
 	mask = &txrc->rate_idx_mask;
Quoted 9 lines

net/mac80211/rate.c void rate_control_get_rate(struct ieee80211_sub_if_data *sdata, · line 466

Quoted 5 lines
net/mac80211/rate.c +3 -1
 	mcs_mask = txrc->rate_idx_mcs_mask;
-	if (*mask != (1 << txrc->sband->n_bitrates) - 1) {
+	if (*mask != (1 << txrc->sband->n_bitrates) - 1 ||
+		memcmp(txrc->sband->ht_cap.mcs.rx_mask, txrc->rate_idx_mcs_mask,
+		       sizeof(txrc->rate_idx_mcs_mask))) {

Please fix the indentation, the memcmp should be indented to just after the opening parenthesis of "if ("

Yes, sorry.

Karl

maintainer
#4 lore raw
In reply to Karl Beldan
Quoted 4 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

Is there any need to update them?

The change for "mask" seems to make it less efficient since it could otherwise be put into a register.

johannes

author
#5 lore raw
Quoted 11 lines · Johannes Berg
Quoted 4 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

Is there any need to update them?

The change for "mask" seems to make it less efficient since it could otherwise be put into a register.

Totally, this commit spares the 10bytes copy of mcs_mask but adds a less efficient indirection to mask.
I thought of it but kept the symmetry with mcs_mask. Apparently you wouldn't mind the dissymmetry so I will re-send using mask by value, plus I wrote "updates .." where it is more like "lets the ad-hoc masks get overwritten".

Karl

#6 lore raw
Quoted 12 lines · john
Quoted 11 lines · Johannes Berg
Quoted 4 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

Is there any need to update them?

The change for "mask" seems to make it less efficient since it could otherwise be put into a register.

Quoted 6 lines

Totally, this commit spares the 10bytes copy of mcs_mask but adds a less efficient indirection to mask.
I thought of it but kept the symmetry with mcs_mask. Apparently you wouldn't mind the dissymmetry so I will re-send using mask by value, plus I wrote "updates .." where it is more like "lets the ad-hoc masks get overwritten".

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

- Felix

#7 lore raw
Quoted 19 lines · Felix Fietkau
Quoted 12 lines · john
Quoted 11 lines · Johannes Berg
Quoted 4 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

Is there any need to update them?

The change for "mask" seems to make it less efficient since it could otherwise be put into a register.

Quoted 6 lines

Totally, this commit spares the 10bytes copy of mcs_mask but adds a less efficient indirection to mask.
I thought of it but kept the symmetry with mcs_mask. Apparently you wouldn't mind the dissymmetry so I will re-send using mask by value, plus I wrote "updates .." where it is more like "lets the ad-hoc masks get overwritten".

Quoted 7 lines

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

Yes, some things look a bit overkill in the masks logic.

Karl

maintainer
#8 lore raw
Quoted 8 lines · Karl Beldan
Quoted 7 lines

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

Yes, some things look a bit overkill in the masks logic.

Are you planning to send new patches to improve this?

johannes

author
#9 lore raw
Quoted 9 lines · Johannes Berg
Quoted 7 lines · Karl Beldan
Quoted 7 lines

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

Quoted 4 lines
Quoted 1 line

Yes, some things look a bit overkill in the masks logic.

Are you planning to send new patches to improve this?

I'll see what I can come up with.

Karl

#10 lore raw
Quoted 10 lines · john
Quoted 9 lines · Johannes Berg
Quoted 7 lines · Karl Beldan
Quoted 7 lines

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

Yes, some things look a bit overkill in the masks logic.

Are you planning to send new patches to improve this?

I'll see what I can come up with.

Now, FWIW, I was looking at how the masks are applied - the code tries to be thorough wrt the various RC flags - 2 things at least are missing:
handle basic rates with multicast, and protection when downgrading to pre-ht rates.

Karl

#11 lore raw
In reply to Felix Fietkau
Quoted 19 lines · Felix Fietkau
Quoted 12 lines · john
Quoted 11 lines · Johannes Berg
Quoted 4 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

Is there any need to update them?

The change for "mask" seems to make it less efficient since it could otherwise be put into a register.

Quoted 6 lines

Totally, this commit spares the 10bytes copy of mcs_mask but adds a less efficient indirection to mask.
I thought of it but kept the symmetry with mcs_mask. Apparently you wouldn't mind the dissymmetry so I will re-send using mask by value, plus I wrote "updates .." where it is more like "lets the ad-hoc masks get overwritten".

Quoted 7 lines

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

Filtering is necessary to lookup alternative downgrade/upgrade rates.

Karl

#12 lore raw
Quoted 20 lines · Karl Beldan
Quoted 19 lines · Felix Fietkau
Quoted 18 lines · john
Quoted 11 lines · Johannes Berg
Quoted 4 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

Is there any need to update them?

The change for "mask" seems to make it less efficient since it could otherwise be put into a register.

Totally, this commit spares the 10bytes copy of mcs_mask but adds a less efficient indirection to mask.
I thought of it but kept the symmetry with mcs_mask. Apparently you wouldn't mind the dissymmetry so I will re-send using mask by value, plus I wrote "updates .." where it is more like "lets the ad-hoc masks get overwritten".

Quoted 8 lines
Quoted 7 lines

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

Filtering is necessary to lookup alternative downgrade/upgrade rates.

Right, but the code could be changed to only do the filtering if mac80211 needs to look up an alternative downgrade/upgrade rate.

- Felix

#13 lore raw
Quoted 28 lines · Felix Fietkau
Quoted 27 lines · Karl Beldan
Quoted 26 lines · Felix Fietkau
Quoted 12 lines · john
Quoted 11 lines · Johannes Berg
Quoted 4 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

Is there any need to update them?

The change for "mask" seems to make it less efficient since it could otherwise be put into a register.

Quoted 6 lines

Totally, this commit spares the 10bytes copy of mcs_mask but adds a less efficient indirection to mask.
I thought of it but kept the symmetry with mcs_mask. Apparently you wouldn't mind the dissymmetry so I will re-send using mask by value, plus I wrote "updates .." where it is more like "lets the ad-hoc masks get overwritten".

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

Quoted 4 lines
Quoted 1 line

Filtering is necessary to lookup alternative downgrade/upgrade rates.

Right, but the code could be changed to only do the filtering if mac80211 needs to look up an alternative downgrade/upgrade rate.

With this I agree.
Do you have strong opinions wrt basic rates ? The current code might tx mc/bc with non-basic rates.

Karl

#14 lore raw
Quoted 29 lines · Karl Beldan
Quoted 28 lines · Felix Fietkau
Quoted 27 lines · Karl Beldan
Quoted 26 lines · Felix Fietkau
Quoted 12 lines · john
Quoted 11 lines · Johannes Berg
Quoted 4 lines · Karl Beldan

From: Karl Beldan <[email protected]>

Currently it gets it from the sdata. This uses and updates the ad-hoc masks of the ieee80211_tx_rate_control instead of copying them.

Is there any need to update them?

The change for "mask" seems to make it less efficient since it could otherwise be put into a register.

Quoted 6 lines

Totally, this commit spares the 10bytes copy of mcs_mask but adds a less efficient indirection to mask.
I thought of it but kept the symmetry with mcs_mask. Apparently you wouldn't mind the dissymmetry so I will re-send using mask by value, plus I wrote "updates .." where it is more like "lets the ad-hoc masks get overwritten".

It seems to me that all of this could be made more efficient by default if a mcs mask pointer is only passed to rate control if the user actually configured a MCS mask. Also, filtering out rates from the mask that the sta does not support seems a bit unnecessary, since the rate control usually looks at the HT capabilities and the sta's mcs rx mask anyway.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Filtering is necessary to lookup alternative downgrade/upgrade rates.

Right, but the code could be changed to only do the filtering if mac80211 needs to look up an alternative downgrade/upgrade rate.

With this I agree.
Do you have strong opinions wrt basic rates ? The current code might tx mc/bc with non-basic rates.

The only strong opinion I have about the rate masking code is that it shouldn't waste precious CPU cycles in a hot path ;)

- Felix

maintainer
#15 lore raw
In reply to Karl Beldan
Quoted 7 lines · Karl Beldan
Quoted 4 lines
Quoted 1 line

Filtering is necessary to lookup alternative downgrade/upgrade rates.

Right, but the code could be changed to only do the filtering if mac80211 needs to look up an alternative downgrade/upgrade rate.

With this I agree.
Do you have strong opinions wrt basic rates ? The current code might tx mc/bc with non-basic rates.

If the users want to shoot themselves in the foot I'd rather give them enough rope to hang themselves with ;-)

johannes