[PATCH wireless-next 0/8] wifi: rt2x00: improve the throughput of rt2800usb

[PATCH wireless-next 0/8] wifi: rt2x00: improve the throughput of rt2800usb

[Ho Jie Feng]

This series improves the throughput of rt2800usb.

Implements
- Frequency offset calibration
- Better fallback rates
- Aggregation
- GRO

Tested on WNDA4100 RT3593, 5 GHz with perf3

  patch                          before        after
                                 TX    RX      TX    RX
  2 crystal calibration          27.7  31.6    28.5  69.7
  4 MCS16 to MCS8 TX status      27.0  70.0    29.4  70.0
  7 RX bulk aggregation          30.5  67.8    30.5  80.9
  8 threaded NAPI with GRO       30.5  80.9    30.5  161.7

RT3070 did not have any visible change for both TX and RX.

Ho Jie Feng (8):
  wifi: rt2x00: start the link tuner when an interface is added
  wifi: rt2800: calibrate the crystal of RT3593 with the beacons
  wifi: rt2800: use the 3SS fallback table of RT3883 on RT3593
  wifi: rt2800: report the MCS16 to MCS8 fallback of RT3593 in TX status
  wifi: rt2800: don't report an A-MPDU length to minstrel_ht
  wifi: rt2800: set the mac80211 fast-xmit and aggregation flags
  wifi: rt2800usb: enable RX bulk aggregation
  wifi: rt2x00usb: use a threaded NAPI with GRO for RX

 drivers/net/wireless/ralink/rt2x00/rt2800.h   |   1 +
 .../net/wireless/ralink/rt2x00/rt2800lib.c    | 115 +++++++++++++++++-
 .../net/wireless/ralink/rt2x00/rt2800lib.h    |   4 +
 .../net/wireless/ralink/rt2x00/rt2800usb.c    |  33 ++++-
 .../net/wireless/ralink/rt2x00/rt2800usb.h    |   6 +
 drivers/net/wireless/ralink/rt2x00/rt2x00.h   |  16 ++-
 .../net/wireless/ralink/rt2x00/rt2x00dev.c    | 105 ++++++++++++----
 .../net/wireless/ralink/rt2x00/rt2x00lib.h    |   2 +
 .../net/wireless/ralink/rt2x00/rt2x00link.c   |   9 ++
 .../net/wireless/ralink/rt2x00/rt2x00mac.c    |   6 +
 .../net/wireless/ralink/rt2x00/rt2x00mmio.c   |   2 +-
 .../net/wireless/ralink/rt2x00/rt2x00usb.c    |  52 ++++++--
 12 files changed, 307 insertions(+), 44 deletions(-)


base-commit: 734bef548a96e8774d6eef5dfa2026c6edc4e80b
-- 
2.53.0

[PATCH wireless-next 1/8] wifi: rt2x00: start the link tuner when an interface is added

[Ho Jie Feng]

rt2x00lib_start() enables the radio before mac80211 adds the first
interface, so rt2x00link_start_tuner(), called from
rt2x00lib_enable_radio(), finds no station or AP interface and returns.

Start the link tuner again when an interface is added.

Fixes: d96324703ffa ("rt2x00: merge agc and vco works with link tuner")
Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
---
 drivers/net/wireless/ralink/rt2x00/rt2x00mac.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c b/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
index 3bc0c1c906c91..d9e323ccc6d62 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
@@ -264,6 +264,12 @@ int rt2x00mac_add_interface(struct ieee80211_hw *hw,
 	 */
 	rt2x00dev->packet_filter = 0;
 
+	/*
+	 * Start the link tuner again if the radio is enabled before the
+	 * first interface is added.
+	 */
+	rt2x00link_start_tuner(rt2x00dev);
+
 	return 0;
 }
 EXPORT_SYMBOL_GPL(rt2x00mac_add_interface);
-- 
2.53.0

[PATCH wireless-next 2/8] wifi: rt2800: calibrate the crystal of RT3593 with the beacons

[Ho Jie Feng]

On a Netgear WNDA4100 (RT3593), as the adapter warms up, the
frequency offset drifts enough to affect the link quality.
In certain cases it can cause the link speed to drop
to 1Mbit/s or below.

Add a frequency calibration step function to RT3593 to use
the frequency offset obtained from the beacon frames of the AP
for tuning.

Test results:
   before after
TX 27.7   28.5
RX 31.6   69.7

Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
---
 drivers/net/wireless/ralink/rt2x00/rt2800.h   |   1 +
 .../net/wireless/ralink/rt2x00/rt2800lib.c    | 101 ++++++++++++++++++
 .../net/wireless/ralink/rt2x00/rt2800lib.h    |   4 +
 .../net/wireless/ralink/rt2x00/rt2800usb.c    |   1 +
 drivers/net/wireless/ralink/rt2x00/rt2x00.h   |   6 ++
 .../net/wireless/ralink/rt2x00/rt2x00lib.h    |   2 +
 .../net/wireless/ralink/rt2x00/rt2x00link.c   |   9 ++
 7 files changed, 124 insertions(+)

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800.h b/drivers/net/wireless/ralink/rt2x00/rt2800.h
index 8930589b4967d..b9f6de3a462e3 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800.h
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800.h
@@ -3163,6 +3163,7 @@ enum rt2800_eeprom_word {
  */
 #define RXWI_W3_SNR0			FIELD32(0x000000ff)
 #define RXWI_W3_SNR1			FIELD32(0x0000ff00)
+#define RXWI_W3_FOFFSET			FIELD32(0xff000000)
 
 /*
  * Macros for converting txpower from EEPROM to mac80211 value
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
index 2e2e382b62475..e7b6edf485099 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
@@ -897,6 +897,28 @@ static int rt2800_agc_to_rssi(struct rt2x00_dev *rt2x00dev, u32 rxwi_w2)
 	return (int)max(rssi0, rssi2);
 }
 
+static void rt2800_rx_freq_offset(struct queue_entry *entry,
+				  struct rxdone_entry_desc *rxdesc)
+{
+	struct rt2800_drv_data *drv_data = entry->queue->rt2x00dev->drv_data;
+	__le32 *rxwi = (__le32 *) entry->skb->data;
+	struct ieee80211_hdr *hdr =
+	    (void *)(entry->skb->data + entry->queue->winfo_size);
+	u32 word;
+
+	if (!(rxdesc->dev_flags & RXDONE_MY_BSS) ||
+	    (rxdesc->flags & RX_FLAG_FAILED_FCS_CRC) ||
+	    !ieee80211_is_beacon(hdr->frame_control))
+		return;
+
+	word = rt2x00_desc_read(rxwi, 3);
+
+	spin_lock_bh(&drv_data->bcn_lock);
+	drv_data->bcn_freq_offset = rt2x00_get_field32(word, RXWI_W3_FOFFSET);
+	drv_data->bcn_rate_mode = rxdesc->rate_mode;
+	spin_unlock_bh(&drv_data->bcn_lock);
+}
+
 void rt2800_process_rxwi(struct queue_entry *entry,
 			 struct rxdone_entry_desc *rxdesc)
 {
@@ -935,6 +957,14 @@ void rt2800_process_rxwi(struct queue_entry *entry,
 	 * Convert descriptor AGC value to RSSI value.
 	 */
 	rxdesc->rssi = rt2800_agc_to_rssi(entry->queue->rt2x00dev, word);
+
+	/*
+	 * The frequency offset of the beacons is used for the frequency
+	 * calibration.
+	 */
+	if (rt2x00_rt(entry->queue->rt2x00dev, RT3593))
+		rt2800_rx_freq_offset(entry, rxdesc);
+
 	/*
 	 * Remove RXWI descriptor from start of the buffer.
 	 */
@@ -2443,6 +2473,7 @@ static inline bool rt2800_clk_is_20mhz(struct rt2x00_dev *rt2x00dev)
 }
 
 #define FREQ_OFFSET_BOUND	0x5f
+#define FREQ_OFFSET_INVALID	-128
 
 static void rt2800_freq_cal_mode1(struct rt2x00_dev *rt2x00dev)
 {
@@ -5633,6 +5664,73 @@ void rt2800_vco_calibration(struct rt2x00_dev *rt2x00dev)
 }
 EXPORT_SYMBOL_GPL(rt2800_vco_calibration);
 
+void rt2800_freq_calibration(struct rt2x00_dev *rt2x00dev)
+{
+	struct rt2800_drv_data *drv_data = rt2x00dev->drv_data;
+	struct link *link = &rt2x00dev->link;
+	u8 activate_threshold, deactivate_threshold;
+	u8 freq_offset, rate_mode;
+	u16 eeprom;
+	s8 offset;
+
+	if (!rt2x00_rt(rt2x00dev, RT3593))
+		return;
+
+	if (!rt2x00dev->intf_associated) {
+		eeprom = rt2800_eeprom_read(rt2x00dev, EEPROM_FREQ);
+		rt2x00dev->freq_offset = rt2x00_get_field16(eeprom,
+							    EEPROM_FREQ_OFFSET);
+		link->freq_cal_adjusting = false;
+		return;
+	}
+
+	spin_lock_bh(&drv_data->bcn_lock);
+	offset = drv_data->bcn_freq_offset;
+	rate_mode = drv_data->bcn_rate_mode;
+	drv_data->bcn_freq_offset = FREQ_OFFSET_INVALID;
+	spin_unlock_bh(&drv_data->bcn_lock);
+
+	/* No beacons received */
+	if (offset == FREQ_OFFSET_INVALID)
+		return;
+
+	if (rate_mode == RATE_MODE_CCK) {
+		activate_threshold = 4;
+		deactivate_threshold = 2;
+	} else {
+		activate_threshold = 20;
+		deactivate_threshold = 10;
+	}
+
+	/* Hysteresis */
+	if (abs(offset) >= activate_threshold)
+		link->freq_cal_adjusting = true;
+	else if (abs(offset) <= deactivate_threshold)
+		link->freq_cal_adjusting = false;
+
+	if (!link->freq_cal_adjusting)
+		return;
+
+	freq_offset = rt2x00_get_field8(rt2x00dev->freq_offset, RFCSR17_CODE);
+	freq_offset = min_t(u8, freq_offset, FREQ_OFFSET_BOUND);
+
+	if (offset > 0) {
+		if (freq_offset > 0)
+			freq_offset--;
+		else
+			link->freq_cal_adjusting = false;
+	} else {
+		if (freq_offset < FREQ_OFFSET_BOUND)
+			freq_offset++;
+		else
+			link->freq_cal_adjusting = false;
+	}
+
+	rt2x00dev->freq_offset = freq_offset;
+	rt2800_freq_cal_mode1(rt2x00dev);
+}
+EXPORT_SYMBOL_GPL(rt2800_freq_calibration);
+
 static void rt2800_config_retry_limit(struct rt2x00_dev *rt2x00dev,
 				      struct rt2x00lib_conf *libconf)
 {
@@ -12032,6 +12130,7 @@ static int rt2800_probe_rt(struct rt2x00_dev *rt2x00dev)
 
 int rt2800_probe_hw(struct rt2x00_dev *rt2x00dev)
 {
+	struct rt2800_drv_data *drv_data = rt2x00dev->drv_data;
 	int retval;
 	u32 reg;
 
@@ -12105,6 +12204,8 @@ int rt2800_probe_hw(struct rt2x00_dev *rt2x00dev)
 	 */
 	rt2x00dev->rssi_offset = DEFAULT_RSSI_OFFSET;
 
+	spin_lock_init(&drv_data->bcn_lock);
+
 	return 0;
 }
 EXPORT_SYMBOL_GPL(rt2800_probe_hw);
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.h b/drivers/net/wireless/ralink/rt2x00/rt2800lib.h
index a3c3a751f57e3..9a04a0fba113f 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.h
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.h
@@ -45,6 +45,9 @@ struct rt2800_drv_data {
 	unsigned int ampdu_factor_cnt[4];
 	DECLARE_BITMAP(sta_ids, STA_IDS_SIZE);
 	struct ieee80211_sta *wcid_to_sta[STA_IDS_SIZE];
+	spinlock_t bcn_lock; /* protects bcn_freq_offset, bcn_rate_mode */
+	s8 bcn_freq_offset;
+	u8 bcn_rate_mode;
 };
 
 struct rt2800_ops {
@@ -241,6 +244,7 @@ void rt2800_link_tuner(struct rt2x00_dev *rt2x00dev, struct link_qual *qual,
 		       const u32 count);
 void rt2800_gain_calibration(struct rt2x00_dev *rt2x00dev);
 void rt2800_vco_calibration(struct rt2x00_dev *rt2x00dev);
+void rt2800_freq_calibration(struct rt2x00_dev *rt2x00dev);
 
 int rt2800_enable_radio(struct rt2x00_dev *rt2x00dev);
 void rt2800_disable_radio(struct rt2x00_dev *rt2x00dev);
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800usb.c b/drivers/net/wireless/ralink/rt2x00/rt2800usb.c
index 15de3db124298..387c07e2251b7 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800usb.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800usb.c
@@ -693,6 +693,7 @@ static const struct rt2x00lib_ops rt2800usb_rt2x00_ops = {
 	.link_tuner		= rt2800_link_tuner,
 	.gain_calibration	= rt2800_gain_calibration,
 	.vco_calibration	= rt2800_vco_calibration,
+	.freq_calibration	= rt2800_freq_calibration,
 	.watchdog		= rt2800_watchdog,
 	.start_queue		= rt2800usb_start_queue,
 	.kick_queue		= rt2x00usb_kick_queue,
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00.h b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
index 7d313e86d3f20..a80508db32d9c 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00.h
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
@@ -319,6 +319,11 @@ struct link {
 	 */
 	struct ewma_rssi avg_rssi;
 
+	/*
+	 * Controls interval of frequency calibration
+	 */
+	bool freq_cal_adjusting;
+
 	/*
 	 * Work structure for scheduling periodic link tuning.
 	 */
@@ -559,6 +564,7 @@ struct rt2x00lib_ops {
 			    struct link_qual *qual, const u32 count);
 	void (*gain_calibration) (struct rt2x00_dev *rt2x00dev);
 	void (*vco_calibration) (struct rt2x00_dev *rt2x00dev);
+	void (*freq_calibration) (struct rt2x00_dev *rt2x00dev);
 
 	/*
 	 * Data queue handlers.
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00lib.h b/drivers/net/wireless/ralink/rt2x00/rt2x00lib.h
index 2f1385baa7ae9..a6f516ebdb2db 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00lib.h
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00lib.h
@@ -22,6 +22,8 @@
 #define LINK_TUNE_INTERVAL	round_jiffies_relative(LINK_TUNE_SECONDS * HZ)
 #define AGC_SECONDS		4
 #define VCO_SECONDS		10
+#define FREQ_SECONDS		10
+#define FREQ_ADJUST_SECONDS	1
 
 /*
  * rt2x00_rate: Per rate device information
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00link.c b/drivers/net/wireless/ralink/rt2x00/rt2x00link.c
index fb23d409fba87..2dc4edc12834f 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00link.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00link.c
@@ -343,6 +343,7 @@ static void rt2x00link_tuner(struct work_struct *work)
 	struct rt2x00_dev *rt2x00dev =
 	    container_of(work, struct rt2x00_dev, link.work.work);
 	struct link *link = &rt2x00dev->link;
+	unsigned int freq_interval;
 
 	/*
 	 * When the radio is shutting down we should
@@ -362,6 +363,14 @@ static void rt2x00link_tuner(struct work_struct *work)
 	    (link->count % (AGC_SECONDS / LINK_TUNE_SECONDS)) == 0)
 		rt2x00dev->ops->lib->gain_calibration(rt2x00dev);
 
+	freq_interval = link->freq_cal_adjusting ?
+			FREQ_ADJUST_SECONDS : FREQ_SECONDS;
+
+	/* VCO calibration should be done after frequency calibration */
+	if (rt2x00dev->ops->lib->freq_calibration &&
+	    (link->count % (freq_interval / LINK_TUNE_SECONDS)) == 0)
+		rt2x00dev->ops->lib->freq_calibration(rt2x00dev);
+
 	if (rt2x00dev->ops->lib->vco_calibration &&
 	    rt2x00_has_cap_vco_recalibration(rt2x00dev) &&
 	    (link->count % (VCO_SECONDS / LINK_TUNE_SECONDS)) == 0)
-- 
2.53.0

[PATCH wireless-next 3/8] wifi: rt2800: use the 3SS fallback table of RT3883 on RT3593

[Ho Jie Feng]

MCS 15 has worse reliability than MCS 16, since 16 and above
are 3 stream rates, and 15 is the top of the 2 stream rates.
Falling back from MCS 16 to MCS 15 does not make sense, so
make the fallback to MCS 8 like in RT3883

Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
---
 drivers/net/wireless/ralink/rt2x00/rt2800lib.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
index e7b6edf485099..9d88ee698b854 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
@@ -6303,7 +6303,7 @@ static int rt2800_init_registers(struct rt2x00_dev *rt2x00dev)
 	reg = rt2x00_rt(rt2x00dev, RT5592) ? 0x00000082 : 0x00000002;
 	rt2800_register_write(rt2x00dev, TXOP_HLDR_ET, reg);
 
-	if (rt2x00_rt(rt2x00dev, RT3883)) {
+	if (rt2x00_rt(rt2x00dev, RT3593) || rt2x00_rt(rt2x00dev, RT3883)) {
 		rt2800_register_write(rt2x00dev, TX_FBK_CFG_3S_0, 0x12111008);
 		rt2800_register_write(rt2x00dev, TX_FBK_CFG_3S_1, 0x16151413);
 	}
-- 
2.53.0

[PATCH wireless-next 4/8] wifi: rt2800: report the MCS16 to MCS8 fallback of RT3593 in TX status

[Ho Jie Feng]

The RT3593 retry chain is now MCS 16 to MCS 8, skipping 7 MCS
rates in between. Reflect that in the retry calculation code.

Test results:
   before after
TX 27.0   29.4
RX 70.0   70.0

Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
---
 drivers/net/wireless/ralink/rt2x00/rt2800lib.c |  8 ++++++++
 drivers/net/wireless/ralink/rt2x00/rt2x00dev.c | 13 +++++++++++--
 2 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
index 9d88ee698b854..de234bcc7a065 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
@@ -1116,6 +1116,14 @@ void rt2800_txdone_entry(struct queue_entry *entry, u32 status, __le32 *txwi,
 		 */
 		__set_bit(TXDONE_SUCCESS, &txdesc.flags);
 		txdesc.retry = ((mcs > real_mcs) ? mcs - real_mcs : 0);
+
+		/*
+		 * On RT3593 MCS16 falls back to MCS8, so a frame that was
+		 * sent below MCS16 after a three stream rate skipped MCS9
+		 * to MCS15.
+		 */
+		if (rt2x00_rt(rt2x00dev, RT3593) && mcs >= 16 && real_mcs < 16)
+			txdesc.retry -= 7;
 	} else {
 		/*
 		 * Transmission failed. The number of retries is
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
index 4d94b7062f440..823785088a308 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
@@ -329,10 +329,10 @@ static void rt2x00lib_fill_tx_status(struct rt2x00_dev *rt2x00dev,
 	 * lowest rate was used.
 	 */
 	for (i = 0; i < retry_rates && i < IEEE80211_TX_MAX_RATES; i++) {
-		tx_info->status.rates[i].idx = rate_idx - i;
+		tx_info->status.rates[i].idx = rate_idx;
 		tx_info->status.rates[i].flags = rate_flags;
 
-		if (rate_idx - i == 0) {
+		if (rate_idx == 0) {
 			/*
 			 * The lowest rate (index 0) was used until the
 			 * number of max retries was reached.
@@ -340,8 +340,17 @@ static void rt2x00lib_fill_tx_status(struct rt2x00_dev *rt2x00dev,
 			tx_info->status.rates[i].count = retry_rates - i;
 			i++;
 			break;
+		} else if (rate_idx == 16) {
+			/*
+			 * TX_FBK_CFG_3S_0 makes MCS16 fall back to MCS8 on
+			 * RT3593, the decrement below shifts it back to 8.
+			 */
+			if (rt2x00_rt(rt2x00dev, RT3593) &&
+			    (rate_flags & IEEE80211_TX_RC_MCS))
+				rate_idx = 9;
 		}
 		tx_info->status.rates[i].count = 1;
+		rate_idx--;
 	}
 	if (i < (IEEE80211_TX_MAX_RATES - 1))
 		tx_info->status.rates[i].idx = -1; /* terminate */
-- 
2.53.0

[PATCH wireless-next 5/8] wifi: rt2800: don't report an A-MPDU length to minstrel_ht

[Ho Jie Feng]

Set IEEE80211_HW_TX_STATUS_NO_AMPDU_LEN, which was added for drivers
that can't report the proper AMPDU length and record this workaround
below the TODO comment.

Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
---
 drivers/net/wireless/ralink/rt2x00/rt2800lib.c | 1 +
 drivers/net/wireless/ralink/rt2x00/rt2x00dev.c | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
index de234bcc7a065..032cd6ac345f0 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
@@ -11860,6 +11860,7 @@ static int rt2800_probe_hw_mode(struct rt2x00_dev *rt2x00dev)
 	ieee80211_hw_set(rt2x00dev->hw, PS_NULLFUNC_STACK);
 	ieee80211_hw_set(rt2x00dev->hw, SIGNAL_DBM);
 	ieee80211_hw_set(rt2x00dev->hw, SUPPORTS_PS);
+	ieee80211_hw_set(rt2x00dev->hw, TX_STATUS_NO_AMPDU_LEN);
 
 	/*
 	 * Don't set IEEE80211_HW_HOST_BROADCAST_PS_BUFFERING for USB devices
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
index 823785088a308..66094da07325f 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
@@ -373,6 +373,8 @@ static void rt2x00lib_fill_tx_status(struct rt2x00_dev *rt2x00dev,
 	 * by the hw we could provide the real ampdu_len to mac80211
 	 * which would allow the rc algorithm to better decide on
 	 * which rates are suitable.
+	 *
+	 * Currently, we work around this by setting TX_STATUS_NO_AMPDU_LEN.
 	 */
 	if (test_bit(TXDONE_AMPDU, &txdesc->flags) ||
 	    tx_info->flags & IEEE80211_TX_CTL_AMPDU) {
-- 
2.53.0

[PATCH wireless-next 6/8] wifi: rt2800: set the mac80211 fast-xmit and aggregation flags

[Ho Jie Feng]

Enable AMSDU TX/RX support for this driver, which can increase
throughput.

Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
---
 drivers/net/wireless/ralink/rt2x00/rt2800lib.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
index 032cd6ac345f0..8b4458a74d33d 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
@@ -11861,6 +11861,9 @@ static int rt2800_probe_hw_mode(struct rt2x00_dev *rt2x00dev)
 	ieee80211_hw_set(rt2x00dev->hw, SIGNAL_DBM);
 	ieee80211_hw_set(rt2x00dev->hw, SUPPORTS_PS);
 	ieee80211_hw_set(rt2x00dev->hw, TX_STATUS_NO_AMPDU_LEN);
+	ieee80211_hw_set(rt2x00dev->hw, SUPPORT_FAST_XMIT);
+	ieee80211_hw_set(rt2x00dev->hw, TX_AMSDU);
+	ieee80211_hw_set(rt2x00dev->hw, SUPPORTS_AMSDU_IN_AMPDU);
 
 	/*
 	 * Don't set IEEE80211_HW_HOST_BROADCAST_PS_BUFFERING for USB devices
-- 
2.53.0

Re: [PATCH wireless-next 6/8] wifi: rt2800: set the mac80211 fast-xmit and aggregation flags

[Stanislaw Gruszka]

On Thu, Oct 29, 2026 at 09:47:47PM +0800, Ho Jie Feng wrote:
> Enable AMSDU TX/RX support for this driver, which can increase
> throughput.
> 
> Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
> ---
>  drivers/net/wireless/ralink/rt2x00/rt2800lib.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
> index 032cd6ac345f0..8b4458a74d33d 100644
> --- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
> +++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
> @@ -11861,6 +11861,9 @@ static int rt2800_probe_hw_mode(struct rt2x00_dev *rt2x00dev)
>  	ieee80211_hw_set(rt2x00dev->hw, SIGNAL_DBM);
>  	ieee80211_hw_set(rt2x00dev->hw, SUPPORTS_PS);
>  	ieee80211_hw_set(rt2x00dev->hw, TX_STATUS_NO_AMPDU_LEN);
> +	ieee80211_hw_set(rt2x00dev->hw, SUPPORT_FAST_XMIT);

IFICT this flag will not take the effect since we have SUPPORTS_PS.
Also for WAP3 (802.11w) we fallback to software encryption - 
that will prevent fast xmit as well. To truly enable
fast xmit, I think we would need add hardware encryption support for 
WPA3 and also compromise by remove Power Save support (which is disabled by
default, so that's possibly fine).

Regards
Stanislaw

> +	ieee80211_hw_set(rt2x00dev->hw, TX_AMSDU);
> +	ieee80211_hw_set(rt2x00dev->hw, SUPPORTS_AMSDU_IN_AMPDU);
>  
>  	/*
>  	 * Don't set IEEE80211_HW_HOST_BROADCAST_PS_BUFFERING for USB devices
> -- 
> 2.53.0
> 

Re: [PATCH wireless-next 6/8] wifi: rt2800: set the mac80211 fast-xmit and aggregation flags

[Jie Feng]

On Fri, 30 Oct 2026 at 17:17, Stanislaw Gruszka <stf_xl@wp.pl> wrote:
>
> On Thu, Oct 29, 2026 at 09:47:47PM +0800, Ho Jie Feng wrote:
> > Enable AMSDU TX/RX support for this driver, which can increase
> > throughput.
> >
> > Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
> > ---
> >  drivers/net/wireless/ralink/rt2x00/rt2800lib.c | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> > diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
> > index 032cd6ac345f0..8b4458a74d33d 100644
> > --- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
> > +++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
> > @@ -11861,6 +11861,9 @@ static int rt2800_probe_hw_mode(struct rt2x00_dev *rt2x00dev)
> >       ieee80211_hw_set(rt2x00dev->hw, SIGNAL_DBM);
> >       ieee80211_hw_set(rt2x00dev->hw, SUPPORTS_PS);
> >       ieee80211_hw_set(rt2x00dev->hw, TX_STATUS_NO_AMPDU_LEN);
> > +     ieee80211_hw_set(rt2x00dev->hw, SUPPORT_FAST_XMIT);
>
> IFICT this flag will not take the effect since we have SUPPORTS_PS.
> Also for WAP3 (802.11w) we fallback to software encryption -
> that will prevent fast xmit as well. To truly enable
Yeah, that is true. I also have a patch that enables hwcrypto in that case.
> fast xmit, I think we would need add hardware encryption support for
> WPA3 and also compromise by remove Power Save support (which is disabled by
> default, so that's possibly fine).
I think maybe a more general patch to mac80211 to enable fast-xmit when PS
is disabled and not only checking the feature flag will be better?

> Regards
> Stanislaw
>
> > +     ieee80211_hw_set(rt2x00dev->hw, TX_AMSDU);
> > +     ieee80211_hw_set(rt2x00dev->hw, SUPPORTS_AMSDU_IN_AMPDU);
Although I think this flag will still be useful for this patch.
> >
> >       /*
> >        * Don't set IEEE80211_HW_HOST_BROADCAST_PS_BUFFERING for USB devices
> > --
> > 2.53.0
> >

[PATCH wireless-next 7/8] wifi: rt2800usb: enable RX bulk aggregation

[Ho Jie Feng]

Enable RX bulk aggregation on high speed USB, to receive more
frames per URB.

RX buffers are now 24KB, so reduce it to 16 buffers instead of 128,
so that they take 512 KB in total instead of 4 MB.

Test results:
   before after
TX 30.5   30.5
RX 67.8   80.9

This code references the mt7601u driver.

Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
---
 .../net/wireless/ralink/rt2x00/rt2800usb.c    | 32 +++++--
 .../net/wireless/ralink/rt2x00/rt2800usb.h    |  6 ++
 drivers/net/wireless/ralink/rt2x00/rt2x00.h   |  3 +-
 .../net/wireless/ralink/rt2x00/rt2x00dev.c    | 84 +++++++++++++------
 .../net/wireless/ralink/rt2x00/rt2x00mmio.c   |  2 +-
 .../net/wireless/ralink/rt2x00/rt2x00usb.c    |  5 +-
 6 files changed, 100 insertions(+), 32 deletions(-)

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800usb.c b/drivers/net/wireless/ralink/rt2x00/rt2800usb.c
index 387c07e2251b7..2e4277b1e5d1b 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800usb.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800usb.c
@@ -301,15 +301,15 @@ static int rt2800usb_enable_radio(struct rt2x00_dev *rt2x00dev)
 		return -EIO;
 
 	rt2x00_set_field32(&reg, USB_DMA_CFG_PHY_CLEAR, 0);
-	rt2x00_set_field32(&reg, USB_DMA_CFG_RX_BULK_AGG_EN, 0);
+	rt2x00_set_field32(&reg, USB_DMA_CFG_RX_BULK_AGG_EN,
+			   rt2x00dev->rx->usb_maxpacket == 512);
 	rt2x00_set_field32(&reg, USB_DMA_CFG_RX_BULK_AGG_TIMEOUT, 128);
 	/*
 	 * Total room for RX frames in kilobytes, PBF might still exceed
 	 * this limit so reduce the number to prevent errors.
 	 */
 	rt2x00_set_field32(&reg, USB_DMA_CFG_RX_BULK_AGG_LIMIT,
-			   ((rt2x00dev->rx->limit * DATA_FRAME_SIZE)
-			    / 1024) - 3);
+			   (rt2x00dev->rx->data_size / 1024) - 4);
 	rt2x00_set_field32(&reg, USB_DMA_CFG_RX_BULK_EN, 1);
 	rt2x00_set_field32(&reg, USB_DMA_CFG_TX_BULK_EN, 1);
 	rt2x00usb_register_write(rt2x00dev, USB_DMA_CFG, reg);
@@ -478,6 +478,27 @@ static void rt2800usb_work_txdone(struct work_struct *work)
 /*
  * RX control handlers
  */
+static unsigned int rt2800usb_get_rx_seg_len(void *data, unsigned int len)
+{
+	unsigned int min_seg_len = RXINFO_DESC_SIZE + RXWI_DESC_SIZE_4WORDS +
+				   RXD_DESC_SIZE;
+	unsigned int rx_pkt_len;
+	u32 word;
+
+	if (len < min_seg_len)
+		return 0;
+
+	word = rt2x00_desc_read(data, 0);
+	rx_pkt_len = rt2x00_get_field32(word, RXINFO_W0_USB_DMA_RX_PKT_LEN);
+
+	if (!rx_pkt_len ||
+	    RXINFO_DESC_SIZE + rx_pkt_len + RXD_DESC_SIZE > len ||
+	    rx_pkt_len & 0x3)
+		return 0;
+
+	return RXINFO_DESC_SIZE + rx_pkt_len + RXD_DESC_SIZE;
+}
+
 static void rt2800usb_fill_rxdone(struct queue_entry *entry,
 				  struct rxdone_entry_desc *rxdesc)
 {
@@ -706,6 +727,7 @@ static const struct rt2x00lib_ops rt2800usb_rt2x00_ops = {
 	.clear_beacon		= rt2800_clear_beacon,
 	.get_tx_data_len	= rt2800usb_get_tx_data_len,
 	.fill_rxdone		= rt2800usb_fill_rxdone,
+	.get_rx_seg_len		= rt2800usb_get_rx_seg_len,
 	.config_shared_key	= rt2800_config_shared_key,
 	.config_pairwise_key	= rt2800_config_pairwise_key,
 	.config_filter		= rt2800_config_filter,
@@ -725,8 +747,8 @@ static void rt2800usb_queue_init(struct data_queue *queue)
 
 	switch (queue->qid) {
 	case QID_RX:
-		queue->limit = 128;
-		queue->data_size = AGGREGATION_SIZE;
+		queue->limit = 16;
+		queue->data_size = MAX_RXBULK_SIZE;
 		queue->desc_size = RXINFO_DESC_SIZE;
 		queue->winfo_size = rxwi_size;
 		queue->priv_size = sizeof(struct queue_entry_priv_usb);
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800usb.h b/drivers/net/wireless/ralink/rt2x00/rt2800usb.h
index 9e180e9e41d92..2fae79072fd58 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800usb.h
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800usb.h
@@ -29,6 +29,12 @@
  */
 #define TXINFO_DESC_SIZE		(1 * sizeof(__le32))
 #define RXINFO_DESC_SIZE		(1 * sizeof(__le32))
+#define RXD_DESC_SIZE			(1 * sizeof(__le32))
+
+/*
+ * RX buffer size
+ */
+#define MAX_RXBULK_SIZE			24576
 
 /*
  * TX Info structure
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00.h b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
index a80508db32d9c..5bc077fce9cbf 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00.h
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
@@ -593,6 +593,7 @@ struct rt2x00lib_ops {
 	 */
 	void (*fill_rxdone) (struct queue_entry *entry,
 			     struct rxdone_entry_desc *rxdesc);
+	unsigned int (*get_rx_seg_len) (void *data, unsigned int len);
 
 	/*
 	 * Configuration handlers.
@@ -1448,7 +1449,7 @@ void rt2x00lib_txdone(struct queue_entry *entry,
 void rt2x00lib_txdone_nomatch(struct queue_entry *entry,
 			      struct txdone_entry_desc *txdesc);
 void rt2x00lib_txdone_noinfo(struct queue_entry *entry, u32 status);
-void rt2x00lib_rxdone(struct queue_entry *entry, gfp_t gfp);
+void rt2x00lib_rxdone(struct queue_entry *entry, unsigned int len, gfp_t gfp);
 
 /*
  * mac80211 handlers.
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
index 66094da07325f..351e7da178279 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
@@ -745,35 +745,14 @@ static int rt2x00lib_rxdone_read_signal(struct rt2x00_dev *rt2x00dev,
 	return 0;
 }
 
-void rt2x00lib_rxdone(struct queue_entry *entry, gfp_t gfp)
+static void rt2x00lib_rxdone_skb(struct queue_entry *entry)
 {
 	struct rt2x00_dev *rt2x00dev = entry->queue->rt2x00dev;
 	struct rxdone_entry_desc rxdesc;
-	struct sk_buff *skb;
 	struct ieee80211_rx_status *rx_status;
 	unsigned int header_length;
 	int rate_idx;
 
-	if (!test_bit(DEVICE_STATE_PRESENT, &rt2x00dev->flags) ||
-	    !test_bit(DEVICE_STATE_ENABLED_RADIO, &rt2x00dev->flags))
-		goto submit_entry;
-
-	if (test_bit(ENTRY_DATA_IO_FAILED, &entry->flags))
-		goto submit_entry;
-
-	/*
-	 * Allocate a new sk_buffer. If no new buffer available, drop the
-	 * received frame and reuse the existing buffer.
-	 */
-	skb = rt2x00queue_alloc_rxskb(entry, gfp);
-	if (!skb)
-		goto submit_entry;
-
-	/*
-	 * Unmap the skb.
-	 */
-	rt2x00queue_unmap_skb(entry);
-
 	/*
 	 * Extract the RXD details.
 	 */
@@ -789,7 +768,7 @@ void rt2x00lib_rxdone(struct queue_entry *entry, gfp_t gfp)
 		rt2x00_err(rt2x00dev, "Wrong frame size %d max %d\n",
 			   rxdesc.size, entry->queue->data_size);
 		dev_kfree_skb(entry->skb);
-		goto renew_skb;
+		return;
 	}
 
 	/*
@@ -868,8 +847,65 @@ void rt2x00lib_rxdone(struct queue_entry *entry, gfp_t gfp)
 	rx_status->antenna = rt2x00dev->link.ant.active.rx;
 
 	ieee80211_rx_ni(rt2x00dev->hw, entry->skb);
+}
+
+void rt2x00lib_rxdone(struct queue_entry *entry, unsigned int len, gfp_t gfp)
+{
+	struct rt2x00_dev *rt2x00dev = entry->queue->rt2x00dev;
+	struct skb_frame_desc *skbdesc, *frame_skbdesc;
+	struct sk_buff *skb;
+	unsigned int seg_len;
+	u8 *data;
+
+	if (!test_bit(DEVICE_STATE_PRESENT, &rt2x00dev->flags) ||
+	    !test_bit(DEVICE_STATE_ENABLED_RADIO, &rt2x00dev->flags))
+		goto submit_entry;
+
+	if (test_bit(ENTRY_DATA_IO_FAILED, &entry->flags))
+		goto submit_entry;
+
+	if (rt2x00dev->ops->lib->get_rx_seg_len) {
+		skb = entry->skb;
+		skbdesc = get_skb_frame_desc(skb);
+		data = skb->data;
+
+		while ((seg_len = rt2x00dev->ops->lib->get_rx_seg_len(data, len))) {
+			entry->skb = __dev_alloc_skb(seg_len, gfp);
+			if (!entry->skb)
+				break;
+
+			skb_put_data(entry->skb, data, seg_len);
+
+			frame_skbdesc = get_skb_frame_desc(entry->skb);
+			memset(frame_skbdesc, 0, sizeof(*frame_skbdesc));
+			frame_skbdesc->desc = skbdesc->desc;
+			frame_skbdesc->desc_len = skbdesc->desc_len;
+
+			rt2x00lib_rxdone_skb(entry);
+
+			data += seg_len;
+			len -= seg_len;
+		}
+
+		entry->skb = skb;
+		goto submit_entry;
+	}
+
+	/*
+	 * Allocate a new sk_buffer. If no new buffer available, drop the
+	 * received frame and reuse the existing buffer.
+	 */
+	skb = rt2x00queue_alloc_rxskb(entry, gfp);
+	if (!skb)
+		goto submit_entry;
+
+	/*
+	 * Unmap the skb.
+	 */
+	rt2x00queue_unmap_skb(entry);
+
+	rt2x00lib_rxdone_skb(entry);
 
-renew_skb:
 	/*
 	 * Replace the skb with the freshly allocated one.
 	 */
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00mmio.c b/drivers/net/wireless/ralink/rt2x00/rt2x00mmio.c
index 93f76acf3dc7d..a1d8b8d13daca 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00mmio.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00mmio.c
@@ -78,7 +78,7 @@ bool rt2x00mmio_rxdone(struct rt2x00_dev *rt2x00dev)
 		/*
 		 * Send the frame to rt2x00lib for further processing.
 		 */
-		rt2x00lib_rxdone(entry, GFP_ATOMIC);
+		rt2x00lib_rxdone(entry, entry->skb->len, GFP_ATOMIC);
 	}
 
 	return !max_rx;
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00usb.c b/drivers/net/wireless/ralink/rt2x00/rt2x00usb.c
index c9fbc3710fd8d..4519a20a7f4bc 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00usb.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00usb.c
@@ -345,6 +345,7 @@ static void rt2x00usb_work_rxdone(struct work_struct *work)
 {
 	struct rt2x00_dev *rt2x00dev =
 	    container_of(work, struct rt2x00_dev, rxdone_work);
+	struct queue_entry_priv_usb *entry_priv;
 	struct queue_entry *entry;
 	struct skb_frame_desc *skbdesc;
 	u8 rxd[32];
@@ -365,7 +366,9 @@ static void rt2x00usb_work_rxdone(struct work_struct *work)
 		/*
 		 * Send the frame to rt2x00lib for further processing.
 		 */
-		rt2x00lib_rxdone(entry, GFP_KERNEL);
+		entry_priv = entry->priv_data;
+		rt2x00lib_rxdone(entry, entry_priv->urb->actual_length,
+				 GFP_KERNEL);
 	}
 }
 
-- 
2.53.0

[PATCH wireless-next 8/8] wifi: rt2x00usb: use a threaded NAPI with GRO for RX

[Ho Jie Feng]

GRO makes sense now that RX aggregation is enabled, since multiple
frames arrive at once.

This code is referenced from commit 5323d3e50c20
("wifi: mt76: mt76u: use a threaded NAPI for the RX path")

Test results:
   before after
TX 30.5   30.5
RX 80.9   161.7

Signed-off-by: Ho Jie Feng <hjf3108@gmail.com>
---
 drivers/net/wireless/ralink/rt2x00/rt2x00.h   |  7 ++-
 .../net/wireless/ralink/rt2x00/rt2x00dev.c    |  6 ++-
 .../net/wireless/ralink/rt2x00/rt2x00usb.c    | 49 ++++++++++++++++---
 3 files changed, 52 insertions(+), 10 deletions(-)

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00.h b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
index 5bc077fce9cbf..a120adb18f4b8 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00.h
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
@@ -955,9 +955,14 @@ struct rt2x00_dev {
 	/**
 	 * Scheduled work for TX/RX done handling (USB devices)
 	 */
-	struct work_struct rxdone_work;
 	struct work_struct txdone_work;
 
+	/*
+	 * NAPI context for RX done handling (USB devices).
+	 */
+	struct net_device *napi_dev;
+	struct napi_struct napi;
+
 	/*
 	 * Powersaving work
 	 */
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
index 351e7da178279..4527fdf905b15 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
@@ -846,7 +846,11 @@ static void rt2x00lib_rxdone_skb(struct queue_entry *entry)
 	rx_status->bw = rxdesc.bw;
 	rx_status->antenna = rt2x00dev->link.ant.active.rx;
 
-	ieee80211_rx_ni(rt2x00dev->hw, entry->skb);
+	if (rt2x00dev->napi_dev)
+		ieee80211_rx_napi(rt2x00dev->hw, NULL, entry->skb,
+				  &rt2x00dev->napi);
+	else
+		ieee80211_rx_ni(rt2x00dev->hw, entry->skb);
 }
 
 void rt2x00lib_rxdone(struct queue_entry *entry, unsigned int len, gfp_t gfp)
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00usb.c b/drivers/net/wireless/ralink/rt2x00/rt2x00usb.c
index 4519a20a7f4bc..d11244a2c4f2e 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00usb.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00usb.c
@@ -341,10 +341,10 @@ static bool rt2x00usb_kick_tx_entry(struct queue_entry *entry, void *data)
 /*
  * RX data handlers.
  */
-static void rt2x00usb_work_rxdone(struct work_struct *work)
+static int rt2x00usb_poll_rxdone(struct napi_struct *napi, int budget)
 {
 	struct rt2x00_dev *rt2x00dev =
-	    container_of(work, struct rt2x00_dev, rxdone_work);
+	    container_of(napi, struct rt2x00_dev, napi);
 	struct queue_entry_priv_usb *entry_priv;
 	struct queue_entry *entry;
 	struct skb_frame_desc *skbdesc;
@@ -368,8 +368,12 @@ static void rt2x00usb_work_rxdone(struct work_struct *work)
 		 */
 		entry_priv = entry->priv_data;
 		rt2x00lib_rxdone(entry, entry_priv->urb->actual_length,
-				 GFP_KERNEL);
+				 GFP_ATOMIC);
 	}
+
+	napi_complete(napi);
+
+	return 0;
 }
 
 static void rt2x00usb_interrupt_rxdone(struct urb *urb)
@@ -396,7 +400,7 @@ static void rt2x00usb_interrupt_rxdone(struct urb *urb)
 	/*
 	 * Schedule the delayed work for processing RX data
 	 */
-	queue_work(rt2x00dev->workqueue, &rt2x00dev->rxdone_work);
+	napi_schedule(&rt2x00dev->napi);
 }
 
 static bool rt2x00usb_kick_rx_entry(struct queue_entry *entry, void *data)
@@ -496,7 +500,7 @@ void rt2x00usb_flush_queue(struct data_queue *queue, bool drop)
 		completion = &queue->rt2x00dev->txdone_work;
 		break;
 	case QID_RX:
-		completion = &queue->rt2x00dev->rxdone_work;
+		completion = NULL;
 		break;
 	default:
 		return;
@@ -515,7 +519,13 @@ void rt2x00usb_flush_queue(struct data_queue *queue, bool drop)
 		 * Schedule the completion handler manually, when this
 		 * worker function runs, it should cleanup the queue.
 		 */
-		queue_work(queue->rt2x00dev->workqueue, completion);
+		if (completion) {
+			queue_work(queue->rt2x00dev->workqueue, completion);
+		} else {
+			local_bh_disable();
+			napi_schedule(&queue->rt2x00dev->napi);
+			local_bh_enable();
+		}
 
 		/*
 		 * Wait for a little while to give the driver
@@ -750,7 +760,7 @@ void rt2x00usb_uninitialize(struct rt2x00_dev *rt2x00dev)
 
 	usb_kill_anchored_urbs(rt2x00dev->anchor);
 	hrtimer_cancel(&rt2x00dev->txstatus_timer);
-	cancel_work_sync(&rt2x00dev->rxdone_work);
+	napi_synchronize(&rt2x00dev->napi);
 	cancel_work_sync(&rt2x00dev->txdone_work);
 
 	queue_for_each(rt2x00dev, queue)
@@ -758,6 +768,14 @@ void rt2x00usb_uninitialize(struct rt2x00_dev *rt2x00dev)
 }
 EXPORT_SYMBOL_GPL(rt2x00usb_uninitialize);
 
+static void rt2x00usb_free_napi(struct rt2x00_dev *rt2x00dev)
+{
+	napi_disable(&rt2x00dev->napi);
+	netif_napi_del(&rt2x00dev->napi);
+	free_netdev(rt2x00dev->napi_dev);
+	rt2x00dev->napi_dev = NULL;
+}
+
 int rt2x00usb_probe(struct usb_interface *usb_intf,
 		    const struct rt2x00_ops *ops)
 {
@@ -801,9 +819,21 @@ int rt2x00usb_probe(struct usb_interface *usb_intf,
 
 	rt2x00_set_chip_intf(rt2x00dev, RT2X00_CHIP_INTF_USB);
 
-	INIT_WORK(&rt2x00dev->rxdone_work, rt2x00usb_work_rxdone);
 	INIT_WORK(&rt2x00dev->txdone_work, rt2x00usb_work_txdone);
 
+	rt2x00dev->napi_dev = alloc_netdev_dummy(0);
+	if (!rt2x00dev->napi_dev) {
+		retval = -ENOMEM;
+		goto exit_free_hw;
+	}
+
+	strscpy(rt2x00dev->napi_dev->name, wiphy_name(hw->wiphy),
+		sizeof(rt2x00dev->napi_dev->name));
+	netif_napi_add(rt2x00dev->napi_dev, &rt2x00dev->napi,
+		       rt2x00usb_poll_rxdone);
+	netif_threaded_enable(rt2x00dev->napi_dev);
+	napi_enable(&rt2x00dev->napi);
+
 	init_usb_anchor(rt2x00dev->anchor);
 
 	retval = rt2x00lib_probe_dev(rt2x00dev);
@@ -814,6 +844,8 @@ int rt2x00usb_probe(struct usb_interface *usb_intf,
 
 exit_free_anchor:
 	usb_kill_anchored_urbs(rt2x00dev->anchor);
+	rt2x00usb_free_napi(rt2x00dev);
+exit_free_hw:
 	ieee80211_free_hw(hw);
 	usb_set_intfdata(usb_intf, NULL);
 
@@ -830,6 +862,7 @@ void rt2x00usb_disconnect(struct usb_interface *usb_intf)
 	 * Free all allocated data.
 	 */
 	rt2x00lib_remove_dev(rt2x00dev);
+	rt2x00usb_free_napi(rt2x00dev);
 	ieee80211_free_hw(hw);
 
 	usb_set_intfdata(usb_intf, NULL);
-- 
2.53.0

Re: [PATCH wireless-next 0/8] wifi: rt2x00: improve the throughput of rt2800usb

[Stanislaw Gruszka]

On Thu, Oct 29, 2026 at 09:47:41PM +0800, Ho Jie Feng wrote:
> This series improves the throughput of rt2800usb.
> 
> Implements
> - Frequency offset calibration
> - Better fallback rates
> - Aggregation
> - GRO
> 
> Tested on WNDA4100 RT3593, 5 GHz with perf3
> 
>   patch                          before        after
>                                  TX    RX      TX    RX
>   2 crystal calibration          27.7  31.6    28.5  69.7
>   4 MCS16 to MCS8 TX status      27.0  70.0    29.4  70.0
>   7 RX bulk aggregation          30.5  67.8    30.5  80.9
>   8 threaded NAPI with GRO       30.5  80.9    30.5  161.7

Nice work! On what host hardware did you take the measurements,
PC laptop/desktop or some embedded board ?

Regards
Stanislaw

> RT3070 did not have any visible change for both TX and RX.
> 
> Ho Jie Feng (8):
>   wifi: rt2x00: start the link tuner when an interface is added
>   wifi: rt2800: calibrate the crystal of RT3593 with the beacons
>   wifi: rt2800: use the 3SS fallback table of RT3883 on RT3593
>   wifi: rt2800: report the MCS16 to MCS8 fallback of RT3593 in TX status
>   wifi: rt2800: don't report an A-MPDU length to minstrel_ht
>   wifi: rt2800: set the mac80211 fast-xmit and aggregation flags
>   wifi: rt2800usb: enable RX bulk aggregation
>   wifi: rt2x00usb: use a threaded NAPI with GRO for RX
> 
>  drivers/net/wireless/ralink/rt2x00/rt2800.h   |   1 +
>  .../net/wireless/ralink/rt2x00/rt2800lib.c    | 115 +++++++++++++++++-
>  .../net/wireless/ralink/rt2x00/rt2800lib.h    |   4 +
>  .../net/wireless/ralink/rt2x00/rt2800usb.c    |  33 ++++-
>  .../net/wireless/ralink/rt2x00/rt2800usb.h    |   6 +
>  drivers/net/wireless/ralink/rt2x00/rt2x00.h   |  16 ++-
>  .../net/wireless/ralink/rt2x00/rt2x00dev.c    | 105 ++++++++++++----
>  .../net/wireless/ralink/rt2x00/rt2x00lib.h    |   2 +
>  .../net/wireless/ralink/rt2x00/rt2x00link.c   |   9 ++
>  .../net/wireless/ralink/rt2x00/rt2x00mac.c    |   6 +
>  .../net/wireless/ralink/rt2x00/rt2x00mmio.c   |   2 +-
>  .../net/wireless/ralink/rt2x00/rt2x00usb.c    |  52 ++++++--
>  12 files changed, 307 insertions(+), 44 deletions(-)
> 
> 
> base-commit: 734bef548a96e8774d6eef5dfa2026c6edc4e80b
> -- 
> 2.53.0
> 

Re: [PATCH wireless-next 0/8] wifi: rt2x00: improve the throughput of rt2800usb

[Jie Feng]

On Fri, 30 Oct 2026 at 17:10, Stanislaw Gruszka <stf_xl@wp.pl> wrote:
>
> On Thu, Oct 29, 2026 at 09:47:41PM +0800, Ho Jie Feng wrote:
> > This series improves the throughput of rt2800usb.
> >
> > Implements
> > - Frequency offset calibration
> > - Better fallback rates
> > - Aggregation
> > - GRO
> >
> > Tested on WNDA4100 RT3593, 5 GHz with perf3
> >
> >   patch                          before        after
> >                                  TX    RX      TX    RX
> >   2 crystal calibration          27.7  31.6    28.5  69.7
> >   4 MCS16 to MCS8 TX status      27.0  70.0    29.4  70.0
> >   7 RX bulk aggregation          30.5  67.8    30.5  80.9
> >   8 threaded NAPI with GRO       30.5  80.9    30.5  161.7
>
> Nice work! On what host hardware did you take the measurements,
> PC laptop/desktop or some embedded board ?

I used a VM on a desktop, with the usb xHCI controller passed
through into the VM, so that kernel panics don't bring down the
entire host. The TX speed is limited by the latency of
control transfers to read the TX status, and through the
passthrough and a hub it can only read ~2000-3000statuses/s.

Desktop is a mini pc with Ryzen 7 7735HS.

I think it's not that bad though, the slowness of the VM
allows me to see the improvement more clearly than
testing it on bare-metal.

> Regards
> Stanislaw
>
> > RT3070 did not have any visible change for both TX and RX.
> >
> > Ho Jie Feng (8):
> >   wifi: rt2x00: start the link tuner when an interface is added
> >   wifi: rt2800: calibrate the crystal of RT3593 with the beacons
> >   wifi: rt2800: use the 3SS fallback table of RT3883 on RT3593
> >   wifi: rt2800: report the MCS16 to MCS8 fallback of RT3593 in TX status
> >   wifi: rt2800: don't report an A-MPDU length to minstrel_ht
> >   wifi: rt2800: set the mac80211 fast-xmit and aggregation flags
> >   wifi: rt2800usb: enable RX bulk aggregation
> >   wifi: rt2x00usb: use a threaded NAPI with GRO for RX
> >
> >  drivers/net/wireless/ralink/rt2x00/rt2800.h   |   1 +
> >  .../net/wireless/ralink/rt2x00/rt2800lib.c    | 115 +++++++++++++++++-
> >  .../net/wireless/ralink/rt2x00/rt2800lib.h    |   4 +
> >  .../net/wireless/ralink/rt2x00/rt2800usb.c    |  33 ++++-
> >  .../net/wireless/ralink/rt2x00/rt2800usb.h    |   6 +
> >  drivers/net/wireless/ralink/rt2x00/rt2x00.h   |  16 ++-
> >  .../net/wireless/ralink/rt2x00/rt2x00dev.c    | 105 ++++++++++++----
> >  .../net/wireless/ralink/rt2x00/rt2x00lib.h    |   2 +
> >  .../net/wireless/ralink/rt2x00/rt2x00link.c   |   9 ++
> >  .../net/wireless/ralink/rt2x00/rt2x00mac.c    |   6 +
> >  .../net/wireless/ralink/rt2x00/rt2x00mmio.c   |   2 +-
> >  .../net/wireless/ralink/rt2x00/rt2x00usb.c    |  52 ++++++--
> >  12 files changed, 307 insertions(+), 44 deletions(-)
> >
> >
> > base-commit: 734bef548a96e8774d6eef5dfa2026c6edc4e80b
> > --
> > 2.53.0
> >