Skip to contentSkip to conversation

mac80211: report per-chain signal values through ethtool.

PATCH In review [email protected] 2 participants 5 messages Mar 29, 2022 – Jul 1, 2022 mac80211

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mac80211: report per-chain signal values through ethtool.

From: Ben Greear <[email protected]>

Combine them into a u64, each byte is one chain. Re-work the way that APs averaged stats to be more efficient.

---
 net/mac80211/ethtool.c | 103 +++++++++++++++++++++++++++++++++++++----
 1 file changed, 93 insertions(+), 10 deletions(-)
net/mac80211/ethtool.c +93 -10
index 028ffe1a4d2d..10a9a30bcbf3 100644
@@ -64,6 +64,7 @@ static const char ieee80211_gstrings_sta_stats[][ETH_GSTRING_LEN] = {
 	"tx_packets", "tx_bytes",
 	"tx_filtered", "tx_retry_failed", "tx_retries",
 	"sta_state", "txrate", "rxrate", "signal", "signal_beacon",
+	"signal_chains", "signal_chains_avg",
 	"channel", "noise", "ch_time", "ch_time_busy",
 	"ch_time_ext_busy", "ch_time_rx", "ch_time_tx"
 };
@@ -96,6 +97,7 @@ static void ieee80211_get_stats2(struct net_device *dev,
 	struct station_info sinfo;
 	struct survey_info survey;
 	int i, q;
+	int z;
 #define STA_STATS_SURVEY_LEN 7
 
 	memset(data, 0, sizeof(u64) * STA_STATS_LEN);
@@ -154,14 +156,49 @@ static void ieee80211_get_stats2(struct net_device *dev,
 		if (sinfo.filled & BIT_ULL(NL80211_STA_INFO_SIGNAL_AVG))
 			data[i] = (u8)sinfo.rx_beacon_signal_avg;
 		i++;
+
+		if (sinfo.filled & BIT_ULL(NL80211_STA_INFO_CHAIN_SIGNAL)) {
+			int mn = min_t(int, sizeof(u64), ARRAY_SIZE(sinfo.chain_signal));
+			u64 accum = (u8)sinfo.chain_signal[0];
+
+			mn = min_t(int, mn, sinfo.chains);
+			for (z = 1; z < mn; z++) {
+				u64 csz = sinfo.chain_signal[z] & 0xFF;
+				u64 cs = csz << (8 * z);
+
+				accum |= cs;
+			}
+			data[i] = accum;
+		}
+		i++;
+
+		if (sinfo.filled & BIT_ULL(NL80211_STA_INFO_CHAIN_SIGNAL_AVG)) {
+			int mn = min_t(int, sizeof(u64), ARRAY_SIZE(sinfo.chain_signal_avg));
+			u64 accum = (u8)sinfo.chain_signal_avg[0];
+
+			for (z = 1; z < mn; z++) {
+				u64 csz = sinfo.chain_signal_avg[z] & 0xFF;
+				u64 cs = csz << (8 * z);
+
+				accum |= cs;
+			}
+			data[i] = accum;
+		}
+		i++;
 	} else {
 		int amt_tx = 0;
 		int amt_rx = 0;
 		int amt_sig = 0;
+		s16 amt_accum_chain[8] = {0};
+		s16 amt_accum_chain_avg[8] = {0};
 		s64 tx_accum = 0;
 		s64 rx_accum = 0;
 		s64 sig_accum = 0;
 		s64 sig_accum_beacon = 0;
+		s64 sig_accum_chain[8] = {0};
+		s64 sig_accum_chain_avg[8] = {0};
+		int start_accum_idx = 0;
+
 		list_for_each_entry(sta, &local->sta_list, list) {
 			/* Make sure this station belongs to the proper dev */
 			if (sta->sdata->dev != dev)
@@ -173,35 +210,48 @@ static void ieee80211_get_stats2(struct net_device *dev,
 			ADD_STA_STATS(sta);
 
 			i++; /* skip sta state */
+			start_accum_idx = i;
 
 			if (sinfo.filled & BIT(NL80211_STA_INFO_TX_BITRATE)) {
 				tx_accum += 100000ULL *
 					cfg80211_calculate_bitrate(&sinfo.txrate);
 				amt_tx++;
 			}
-			if (amt_tx)
-				data[i] = mac_div(tx_accum, amt_tx);
-			i++;
 
 			if (sinfo.filled & BIT(NL80211_STA_INFO_RX_BITRATE)) {
 				rx_accum += 100000ULL *
 					cfg80211_calculate_bitrate(&sinfo.rxrate);
 				amt_rx++;
 			}
-			if (amt_rx)
-				data[i] = mac_div(rx_accum, amt_rx);
-			i++;
 
 			if (sinfo.filled & BIT(NL80211_STA_INFO_SIGNAL_AVG)) {
 				sig_accum += sinfo.signal_avg;
 				sig_accum_beacon += sinfo.rx_beacon_signal_avg;
 				amt_sig++;
 			}
-			if (amt_sig) {
-				data[i] = (mac_div(sig_accum, amt_sig) & 0xFF);
-				data[i+1] = (mac_div(sig_accum_beacon, amt_sig) & 0xFF);
+
+			if (sinfo.filled & BIT_ULL(NL80211_STA_INFO_CHAIN_SIGNAL)) {
+				int mn = min_t(int, sizeof(u64), ARRAY_SIZE(sinfo.chain_signal));
+
+				mn = min_t(int, mn, sinfo.chains);
+				for (z = 0; z < mn; z++) {
+					sig_accum_chain[z] += sinfo.chain_signal[z];
+					amt_accum_chain[z]++;
+				}
 			}
-			i += 2;
+			i++;
+
+			if (sinfo.filled & BIT_ULL(NL80211_STA_INFO_CHAIN_SIGNAL_AVG)) {
+				int mn;
+
+				mn = min_t(int, sizeof(u64), ARRAY_SIZE(sinfo.chain_signal_avg));
+				mn = min_t(int, mn, sinfo.chains);
+				for (z = 0; z < mn; z++) {
+					sig_accum_chain_avg[z] += sinfo.chain_signal_avg[z];
+					amt_accum_chain_avg[z]++;
+				}
+			}
+			i++;
 
 			/*pr_err("sta: %p (%s) sig_accum: %ld  amt-sig: %d filled: 0x%x:%08x rpt-sig-avg: %d  i: %d  div: %d sinfo.signal_avg: %d\n",
 			       sta, sta->sdata->name, (long)(sig_accum), amt_sig, (u32)(sinfo.filled >> 32),
@@ -209,6 +259,39 @@ static void ieee80211_get_stats2(struct net_device *dev,
 			       sinfo.signal_avg);*/
 
 		}
+
+		/* Do averaging */
+		i = start_accum_idx;
+
+		if (amt_tx)
+			data[i] = mac_div(tx_accum, amt_tx);
+		i++;
+
+		if (amt_rx)
+			data[i] = mac_div(rx_accum, amt_rx);
+		i++;
+
+		if (amt_sig) {
+			data[i] = (mac_div(sig_accum, amt_sig) & 0xFF);
+			data[i + 1] = (mac_div(sig_accum_beacon, amt_sig) & 0xFF);
+		}
+		i += 2;
+
+		for (z = 0; z < sizeof(u64); z++) {
+			if (amt_accum_chain[z]) {
+				u64 val = mac_div(sig_accum_chain[z], amt_accum_chain[z]);
+
+				val |= 0xFF;
+				data[i] |= (val << (z * 8));
+			}
+			if (amt_accum_chain_avg[z]) {
+				u64 val = mac_div(sig_accum_chain_avg[z], amt_accum_chain_avg[z]);
+
+				val |= 0xFF;
+				data[i + 1] |= (val << (z * 8));
+			}
+		}
+		i += 2;
 	}
 
 do_survey:
Signature · 1 line
2.20.1
maintainer
#1 lore raw
Quoted 3 lines · On Tue, 2022-03-29 at 14:02 -0700, [email protected] wrote:

From: Ben Greear <[email protected]>

Combine them into a u64, each byte is one chain.

This only works up to 4 chains, but the specs at least support 8. I don't think we have any drivers for that, but ...

And it's also rather ugly, IMHO.

We're reporting these through nl80211 anyway though, no? Why should we prefer ethtool, which fundamentally limits to a single value for the AP rather than giving the full per-station view.

Quoted 2 lines

Re-work the way that APs averaged stats to be more efficient.

Isn't that completely unrelated?

johannes

author
#2 lore raw
Quoted 7 lines · Johannes Berg
Quoted 3 lines · On Tue, 2022-03-29 at 14:02 -0700, [email protected] wrote:

From: Ben Greear <[email protected]>

Combine them into a u64, each byte is one chain.

This only works up to 4 chains, but the specs at least support 8. I don't think we have any drivers for that, but ...

u64 gives 8 bytes, so the ethtool part can support 8 chains. The mac80211 part only supports up to 4 chains currently though.

Quoted 6 lines

And it's also rather ugly, IMHO.

We're reporting these through nl80211 anyway though, no? Why should we prefer ethtool, which fundamentally limits to a single value for the AP rather than giving the full per-station view.

I already gather ethtool stats for STA vdevs, so adding another stat is basically free as far as performance is concerned. That is important to me. I do not currently program much against netlink API (just scrape existing tools output).

I understand if you don't want it upstream though.

Quoted 5 lines
Quoted 3 lines

Re-work the way that APs averaged stats to be more efficient.

Isn't that completely unrelated?

At least somewhat unrelated.

Thanks,
Ben

Quoted 3 lines

johannes

Signature · 2 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com
maintainer
#3 lore raw
Quoted 11 lines · Ben Greear
Quoted 7 lines · Johannes Berg
Quoted 3 lines · On Tue, 2022-03-29 at 14:02 -0700, [email protected] wrote:

From: Ben Greear <[email protected]>

Combine them into a u64, each byte is one chain.

This only works up to 4 chains, but the specs at least support 8. I don't think we have any drivers for that, but ...

u64 gives 8 bytes, so the ethtool part can support 8 chains. The mac80211 part only supports up to 4 chains currently though.

Oops, right, sorry.

Still, I'm not sure I like munging it all up into one value - the value itself then doesn't mean anything, and the normal ethtool APIs userspace tools would be pretty much useless for it?

Quoted 8 lines
Quoted 3 lines

We're reporting these through nl80211 anyway though, no? Why should we prefer ethtool, which fundamentally limits to a single value for the AP rather than giving the full per-station view.

I already gather ethtool stats for STA vdevs, so adding another stat is basically free as far as performance is concerned. That is important to me. I do not currently program much against netlink API (just scrape existing tools output).

Well I guess then I think you probably should program against the netlink API :)

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

Re-work the way that APs averaged stats to be more efficient.

Isn't that completely unrelated?

At least somewhat unrelated.

Fair enough. Maybe send that separately? I guess that's something I'd understand a bit more and improving the existing code is an easier sell than adding a whole new thing there :)

johannes

author
#4 lore raw
Quoted 10 lines · Johannes Berg
Quoted 9 lines · Ben Greear
Quoted 7 lines · Johannes Berg
Quoted 3 lines · On Tue, 2022-03-29 at 14:02 -0700, [email protected] wrote:

From: Ben Greear <[email protected]>

Combine them into a u64, each byte is one chain.

This only works up to 4 chains, but the specs at least support 8. I don't think we have any drivers for that, but ...

Quoted 8 lines
Quoted 2 lines

u64 gives 8 bytes, so the ethtool part can support 8 chains. The mac80211 part only supports up to 4 chains currently though.

Oops, right, sorry.

Still, I'm not sure I like munging it all up into one value - the value itself then doesn't mean anything, and the normal ethtool APIs userspace tools would be pretty much useless for it?

A human looking at the value would be confused, but a program easily deals with it. In a lot of ways, ethtool is way more straight forward to program against than netlink, and it works well with non-wifi drivers, so one chunk of stats-gathering code for all network devices.

Quoted 5 lines
Quoted 5 lines
Quoted 4 lines

We're reporting these through nl80211 anyway though, no? Why should we prefer ethtool, which fundamentally limits to a single value for the AP rather than giving the full per-station view.

Quoted 7 lines
Quoted 4 lines

I already gather ethtool stats for STA vdevs, so adding another stat is basically free as far as performance is concerned. That is important to me. I do not currently program much against netlink API (just scrape existing tools output).

Well I guess then I think you probably should program against the netlink API :)

I've managed to mostly dodge it for 20 years....there is hope to make it another 20!

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines

Re-work the way that APs averaged stats to be more efficient.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

Isn't that completely unrelated?

At least somewhat unrelated.

Fair enough. Maybe send that separately? I guess that's something I'd understand a bit more and improving the existing code is an easier sell than adding a whole new thing there :)

Ok, I'll add that to my list and will plan to do it next time I rebase on a newer upstream kernel.

Thanks,
Ben

Quoted 3 lines

johannes

Signature · 2 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com