mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: "Jérôme Pouiller" <jerome.pouiller@silabs.com>
To: linux-wireless@vger.kernel.org, Kalle Valo <kvalo@kernel.org>
Cc: linux-kernel@vger.kernel.org, "Felix Fietkau" <nbd@nbd.name>,
	"Felipe Negrelli Wolter" <felipe.negrelliwolter@silabs.com>,
	"Olivier Souloumiac" <olivier.souloumiac@silabs.com>,
	"Alexandr Suslenko" <suslenko.o@ajax.systems>,
	"Jérôme Pouiller" <jerome.pouiller@silabs.com>
Subject: [PATCH] wifi: wfx: fix case where rates are out of order
Date: Thu, 28 Sep 2023 12:10:48 +0200	[thread overview]
Message-ID: <20230928101048.5543-1-jerome.pouiller@silabs.com> (raw)

From: Felipe Negrelli Wolter <felipe.negrelliwolter@silabs.com>

When frames are sent over the air, the device always applies the data
rates in descending order. The driver assumed Minstrel also provided
rate in descending order.

However, in some cases, Minstrel can a choose a fallback rate greater
than the primary rate. In this case, the two rates was inverted, the
device try highest rate first and we get many retries.

Since the device always apply rates in descending order. The workaround
is to drop rate when it higher than its predecessor in the rate list.
Thus [ 4, 5, 3 ] become [ 4, 3 ].

This patch has been tested in isolated room with a series of
attenuators. Here are the Minstrel statistics with 80dBm of attenuation:

  Without the fix:

                  best    ____________rate__________    ____statistics___    _____last____    ______sum-of________
    mode guard #  rate   [name   idx airtime  max_tp]  [avg(tp) avg(prob)]  [retry|suc|att]  [#success | #attempts]
    HT20  LGI  1       S  MCS0     0    1477     5.6       5.2      82.7       3     0 0             3   4
    HT20  LGI  1          MCS1     1     738    10.6       0.0       0.0       0     0 0             0   1
    HT20  LGI  1     D    MCS2     2     492    14.9      13.5      81.5       5     0 0             5   9
    HT20  LGI  1    C     MCS3     3     369    18.8      17.6      84.3       5     0 0            76   96
    HT20  LGI  1  A   P   MCS4     4     246    25.4      22.4      79.5       5     0 0         11268   14026
    HT20  LGI  1   B   S  MCS5     5     185    30.7      19.7      57.7       5     8 9          3918   9793
    HT20  LGI  1          MCS6     6     164    33.0       0.0       0.0       5     0 0             6   102
    HT20  LGI  1          MCS7     7     148    35.1       0.0       0.0       0     0 0             0   44

  With the fix:

                  best    ____________rate__________    ____statistics___    _____last____    ______sum-of________
    mode guard #  rate   [name   idx airtime  max_tp]  [avg(tp) avg(prob)]  [retry|suc|att]  [#success | #attempts]
    HT20  LGI  1       S  MCS0     0    1477     5.6       1.8      28.6       1     0 0             1   5
    HT20  LGI  1     DP   MCS1     1     738    10.6       9.7      82.6       4     0 0            14   34
    HT20  LGI  1          MCS2     2     492    14.9       9.2      55.4       5     0 0            52   77
    HT20  LGI  1   B   S  MCS3     3     369    18.8      15.6      74.9       5     1 1           417   554
    HT20  LGI  1  A       MCS4     4     246    25.4      16.7      59.2       5     1 1         13812   17951
    HT20  LGI  1    C  S  MCS5     5     185    30.7      14.0      41.0       5     1 5            57   640
    HT20  LGI  1          MCS6     6     164    33.0       0.0       0.0       0     0 1             0   48
    HT20  LGI  1       S  MCS7     7     148    35.1       0.0       0.0       0     0 0             0   36

We can notice the device try now to send with lower rates (and high
success rates). At the end, we measured 20-25% better throughput with
this patch.

Fixes: 9bca45f3d6924 "staging: wfx: allow to send 802.11 frames"
Tested-by: Olivier Souloumiac <olivier.souloumiac@silabs.com>
Tested-by: Alexandr Suslenko <suslenko.o@ajax.systems>
Reported-by: Alexandr Suslenko <suslenko.o@ajax.systems>
Co-developed-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
---
 drivers/net/wireless/silabs/wfx/data_tx.c | 71 +++++++++--------------
 1 file changed, 29 insertions(+), 42 deletions(-)

diff --git a/drivers/net/wireless/silabs/wfx/data_tx.c b/drivers/net/wireless/silabs/wfx/data_tx.c
index 6a5e52a96d183..caa22226b01bc 100644
--- a/drivers/net/wireless/silabs/wfx/data_tx.c
+++ b/drivers/net/wireless/silabs/wfx/data_tx.c
@@ -226,53 +226,40 @@ static u8 wfx_tx_get_link_id(struct wfx_vif *wvif, struct ieee80211_sta *sta,
 
 static void wfx_tx_fixup_rates(struct ieee80211_tx_rate *rates)
 {
-	int i;
-	bool finished;
+	bool has_rate0 = false;
+	int i, j;
 
-	/* Firmware is not able to mix rates with different flags */
-	for (i = 0; i < IEEE80211_TX_MAX_RATES; i++) {
-		if (rates[0].flags & IEEE80211_TX_RC_SHORT_GI)
-			rates[i].flags |= IEEE80211_TX_RC_SHORT_GI;
-		if (!(rates[0].flags & IEEE80211_TX_RC_SHORT_GI))
+	for (i = 1, j = 1; j < IEEE80211_TX_MAX_RATES; j++) {
+		if (rates[j].idx == -1)
+			break;
+		/* The device use the rates in descending order, whatever the request from minstrel.
+		 * We have to trade off here. Most important is to respect the primary rate
+		 * requested by minstrel. So, we drops the entries with rate higher than the
+		 * previous.
+		 */
+		if (rates[j].idx >= rates[i - 1].idx) {
+			rates[i - 1].count += rates[j].count;
+			rates[i - 1].count = min_t(u16, 15, rates[i - 1].count);
+		} else {
+			memcpy(rates + i, rates + j, sizeof(rates[i]));
+			if (rates[i].idx == 0)
+				has_rate0 = true;
+			/* The device apply Short GI only on the first rate */
 			rates[i].flags &= ~IEEE80211_TX_RC_SHORT_GI;
-		if (!(rates[0].flags & IEEE80211_TX_RC_USE_RTS_CTS))
-			rates[i].flags &= ~IEEE80211_TX_RC_USE_RTS_CTS;
-	}
-
-	/* Sort rates and remove duplicates */
-	do {
-		finished = true;
-		for (i = 0; i < IEEE80211_TX_MAX_RATES - 1; i++) {
-			if (rates[i + 1].idx == rates[i].idx &&
-			    rates[i].idx != -1) {
-				rates[i].count += rates[i + 1].count;
-				if (rates[i].count > 15)
-					rates[i].count = 15;
-				rates[i + 1].idx = -1;
-				rates[i + 1].count = 0;
-
-				finished = false;
-			}
-			if (rates[i + 1].idx > rates[i].idx) {
-				swap(rates[i + 1], rates[i]);
-				finished = false;
-			}
+			i++;
 		}
-	} while (!finished);
+	}
 	/* Ensure that MCS0 or 1Mbps is present at the end of the retry list */
-	for (i = 0; i < IEEE80211_TX_MAX_RATES; i++) {
-		if (rates[i].idx == 0)
-			break;
-		if (rates[i].idx == -1) {
-			rates[i].idx = 0;
-			rates[i].count = 8; /* == hw->max_rate_tries */
-			rates[i].flags = rates[i - 1].flags & IEEE80211_TX_RC_MCS;
-			break;
-		}
+	if (!has_rate0 && i < IEEE80211_TX_MAX_RATES) {
+		rates[i].idx = 0;
+		rates[i].count = 8; /* == hw->max_rate_tries */
+		rates[i].flags = rates[0].flags & IEEE80211_TX_RC_MCS;
+		i++;
+	}
+	for (; i < IEEE80211_TX_MAX_RATES; i++) {
+		memset(rates + i, 0, sizeof(rates[i]));
+		rates[i].idx = -1;
 	}
-	/* All retries use long GI */
-	for (i = 1; i < IEEE80211_TX_MAX_RATES; i++)
-		rates[i].flags &= ~IEEE80211_TX_RC_SHORT_GI;
 }
 
 static u8 wfx_tx_get_retry_policy_id(struct wfx_vif *wvif, struct ieee80211_tx_info *tx_info)
-- 
2.39.2


             reply	other threads:[~2023-09-28 10:11 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-09-28 10:10 Jérôme Pouiller [this message]
2023-10-04 10:29 ` Kalle Valo

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20230928101048.5543-1-jerome.pouiller@silabs.com \
    --to=jerome.pouiller@silabs.com \
    --cc=felipe.negrelliwolter@silabs.com \
    --cc=kvalo@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-wireless@vger.kernel.org \
    --cc=nbd@nbd.name \
    --cc=olivier.souloumiac@silabs.com \
    --cc=suslenko.o@ajax.systems \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®