ath9k_htc: add new WMI_REG_RMW_CMDID command

Since usb bus add extra delay on each request, a command
with read + write requests is too expensive. We can dramtically
reduce usb load by moving this command to firmware.

In my tests, this patch will reduce channel scan time
for about 5-10 seconds.

Signed-off-by: Oleksij Rempel <linux@rempel-privat.de>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

drivers/net/wireless/ath/ath.h

@@ -131,6 +131,9 @@ struct ath_ops {
 	void (*enable_write_buffer)(void *);
 	void (*write_flush) (void *);
 	u32 (*rmw)(void *, u32 reg_offset, u32 set, u32 clr);
+	void (*enable_rmw_buffer)(void *);
+	void (*rmw_flush) (void *);
+
 };
 
 struct ath_common;

drivers/net/wireless/ath/ath9k/htc.h

@@ -444,6 +444,10 @@ static inline void ath9k_htc_stop_btcoex(struct ath9k_htc_priv *priv)
 #define OP_BT_SCAN                 BIT(4)
 #define OP_TSF_RESET               BIT(6)
 
+enum htc_op_flags {
+	HTC_FWFLAG_NO_RMW,
+};
+
 struct ath9k_htc_priv {
 	struct device *dev;
 	struct ieee80211_hw *hw;
@@ -482,6 +486,7 @@ struct ath9k_htc_priv {
 	bool reconfig_beacon;
 	unsigned int rxfilter;
 	unsigned long op_flags;
+	unsigned long fw_flags;
 
 	struct ath9k_hw_cal_data caldata;
 	struct ath_spec_scan_priv spec_priv;

drivers/net/wireless/ath/ath9k/htc_drv_init.c

@@ -376,17 +376,139 @@ static void ath9k_regwrite_flush(void *hw_priv)
 	mutex_unlock(&priv->wmi->multi_write_mutex);
 }
 
-static u32 ath9k_reg_rmw(void *hw_priv, u32 reg_offset, u32 set, u32 clr)
+static void ath9k_reg_rmw_buffer(void *hw_priv,
+				 u32 reg_offset, u32 set, u32 clr)
+{
+	struct ath_hw *ah = (struct ath_hw *) hw_priv;
+	struct ath_common *common = ath9k_hw_common(ah);
+	struct ath9k_htc_priv *priv = (struct ath9k_htc_priv *) common->priv;
+	u32 rsp_status;
+	int r;
+
+	mutex_lock(&priv->wmi->multi_rmw_mutex);
+
+	/* Store the register/value */
+	priv->wmi->multi_rmw[priv->wmi->multi_rmw_idx].reg =
+		cpu_to_be32(reg_offset);
+	priv->wmi->multi_rmw[priv->wmi->multi_rmw_idx].set =
+		cpu_to_be32(set);
+	priv->wmi->multi_rmw[priv->wmi->multi_rmw_idx].clr =
+		cpu_to_be32(clr);
+
+	priv->wmi->multi_rmw_idx++;
+
+	/* If the buffer is full, send it out. */
+	if (priv->wmi->multi_rmw_idx == MAX_RMW_CMD_NUMBER) {
+		r = ath9k_wmi_cmd(priv->wmi, WMI_REG_RMW_CMDID,
+			  (u8 *) &priv->wmi->multi_rmw,
+			  sizeof(struct register_write) * priv->wmi->multi_rmw_idx,
+			  (u8 *) &rsp_status, sizeof(rsp_status),
+			  100);
+		if (unlikely(r)) {
+			ath_dbg(common, WMI,
+				"REGISTER RMW FAILED, multi len: %d\n",
+				priv->wmi->multi_rmw_idx);
+		}
+		priv->wmi->multi_rmw_idx = 0;
+	}
+
+	mutex_unlock(&priv->wmi->multi_rmw_mutex);
+}
+
+static void ath9k_reg_rmw_flush(void *hw_priv)
 {
-	u32 val;
+	struct ath_hw *ah = (struct ath_hw *) hw_priv;
+	struct ath_common *common = ath9k_hw_common(ah);
+	struct ath9k_htc_priv *priv = (struct ath9k_htc_priv *) common->priv;
+	u32 rsp_status;
+	int r;
+
+	if (test_bit(HTC_FWFLAG_NO_RMW, &priv->fw_flags))
+		return;
+
+	atomic_dec(&priv->wmi->m_rmw_cnt);
 
-	val = ath9k_regread(hw_priv, reg_offset);
-	val &= ~clr;
-	val |= set;
-	ath9k_regwrite(hw_priv, val, reg_offset);
+	mutex_lock(&priv->wmi->multi_rmw_mutex);
+
+	if (priv->wmi->multi_rmw_idx) {
+		r = ath9k_wmi_cmd(priv->wmi, WMI_REG_RMW_CMDID,
+			  (u8 *) &priv->wmi->multi_rmw,
+			  sizeof(struct register_rmw) * priv->wmi->multi_rmw_idx,
+			  (u8 *) &rsp_status, sizeof(rsp_status),
+			  100);
+		if (unlikely(r)) {
+			ath_dbg(common, WMI,
+				"REGISTER RMW FAILED, multi len: %d\n",
+				priv->wmi->multi_rmw_idx);
+		}
+		priv->wmi->multi_rmw_idx = 0;
+	}
+
+	mutex_unlock(&priv->wmi->multi_rmw_mutex);
+}
+
+static void ath9k_enable_rmw_buffer(void *hw_priv)
+{
+	struct ath_hw *ah = (struct ath_hw *) hw_priv;
+	struct ath_common *common = ath9k_hw_common(ah);
+	struct ath9k_htc_priv *priv = (struct ath9k_htc_priv *) common->priv;
+
+	if (test_bit(HTC_FWFLAG_NO_RMW, &priv->fw_flags))
+		return;
+
+	atomic_inc(&priv->wmi->m_rmw_cnt);
+}
+
+static u32 ath9k_reg_rmw_single(void *hw_priv,
+				 u32 reg_offset, u32 set, u32 clr)
+{
+	struct ath_hw *ah = (struct ath_hw *) hw_priv;
+	struct ath_common *common = ath9k_hw_common(ah);
+	struct ath9k_htc_priv *priv = (struct ath9k_htc_priv *) common->priv;
+	struct register_rmw buf, buf_ret;
+	int ret;
+	u32 val = 0;
+
+	buf.reg = cpu_to_be32(reg_offset);
+	buf.set = cpu_to_be32(set);
+	buf.clr = cpu_to_be32(clr);
+
+	ret = ath9k_wmi_cmd(priv->wmi, WMI_REG_RMW_CMDID,
+			  (u8 *) &buf, sizeof(buf),
+			  (u8 *) &buf_ret, sizeof(buf_ret),
+			  100);
+	if (unlikely(ret)) {
+		ath_dbg(common, WMI, "REGISTER RMW FAILED:(0x%04x, %d)\n",
+			reg_offset, ret);
+	}
 	return val;
 }
 
+static u32 ath9k_reg_rmw(void *hw_priv, u32 reg_offset, u32 set, u32 clr)
+{
+	struct ath_hw *ah = (struct ath_hw *) hw_priv;
+	struct ath_common *common = ath9k_hw_common(ah);
+	struct ath9k_htc_priv *priv = (struct ath9k_htc_priv *) common->priv;
+
+	if (test_bit(HTC_FWFLAG_NO_RMW, &priv->fw_flags)) {
+		u32 val;
+
+		val = REG_READ(ah, reg_offset);
+		val &= ~clr;
+		val |= set;
+		REG_WRITE(ah, reg_offset, val);
+
+		return 0;
+	}
+
+	if (atomic_read(&priv->wmi->m_rmw_cnt))
+		ath9k_reg_rmw_buffer(hw_priv, reg_offset, set, clr);
+	else
+		ath9k_reg_rmw_single(hw_priv, reg_offset, set, clr);
+
+	return 0;
+}
+
 static void ath_usb_read_cachesize(struct ath_common *common, int *csz)
 {
 	*csz = L1_CACHE_BYTES >> 2;
@@ -501,6 +623,8 @@ static int ath9k_init_priv(struct ath9k_htc_priv *priv,
 	ah->reg_ops.write = ath9k_regwrite;
 	ah->reg_ops.enable_write_buffer = ath9k_enable_regwrite_buffer;
 	ah->reg_ops.write_flush = ath9k_regwrite_flush;
+	ah->reg_ops.enable_rmw_buffer = ath9k_enable_rmw_buffer;
+	ah->reg_ops.rmw_flush = ath9k_reg_rmw_flush;
 	ah->reg_ops.rmw = ath9k_reg_rmw;
 	priv->ah = ah;
 
@@ -686,6 +810,12 @@ static int ath9k_init_firmware_version(struct ath9k_htc_priv *priv)
 		return -EINVAL;
 	}
 
+	if (priv->fw_version_major == 1 && priv->fw_version_minor < 4)
+		set_bit(HTC_FWFLAG_NO_RMW, &priv->fw_flags);
+
+	dev_info(priv->dev, "FW RMW support: %s\n",
+		test_bit(HTC_FWFLAG_NO_RMW, &priv->fw_flags) ? "Off" : "On");
+
 	return 0;
 }
 

drivers/net/wireless/ath/ath9k/hw.h

@@ -100,6 +100,18 @@
 			(_ah)->reg_ops.write_flush((_ah));	\
 	} while (0)
 
+#define ENABLE_REG_RMW_BUFFER(_ah)					\
+	do {								\
+		if ((_ah)->reg_ops.enable_rmw_buffer)	\
+			(_ah)->reg_ops.enable_rmw_buffer((_ah)); \
+	} while (0)
+
+#define REG_RMW_BUFFER_FLUSH(_ah)					\
+	do {								\
+		if ((_ah)->reg_ops.rmw_flush)		\
+			(_ah)->reg_ops.rmw_flush((_ah));	\
+	} while (0)
+
 #define PR_EEP(_s, _val)						\
 	do {								\
 		len += scnprintf(buf + len, size - len, "%20s : %10d\n",\

drivers/net/wireless/ath/ath9k/wmi.c

@@ -61,6 +61,8 @@ static const char *wmi_cmd_to_name(enum wmi_cmd_id wmi_cmd)
 		return "WMI_REG_READ_CMDID";
 	case WMI_REG_WRITE_CMDID:
 		return "WMI_REG_WRITE_CMDID";
+	case WMI_REG_RMW_CMDID:
+		return "WMI_REG_RMW_CMDID";
 	case WMI_RC_STATE_CHANGE_CMDID:
 		return "WMI_RC_STATE_CHANGE_CMDID";
 	case WMI_RC_RATE_UPDATE_CMDID:
@@ -101,6 +103,7 @@ struct wmi *ath9k_init_wmi(struct ath9k_htc_priv *priv)
 	spin_lock_init(&wmi->event_lock);
 	mutex_init(&wmi->op_mutex);
 	mutex_init(&wmi->multi_write_mutex);
+	mutex_init(&wmi->multi_rmw_mutex);
 	init_completion(&wmi->cmd_wait);
 	INIT_LIST_HEAD(&wmi->pending_tx_events);
 	tasklet_init(&wmi->wmi_event_tasklet, ath9k_wmi_event_tasklet,

drivers/net/wireless/ath/ath9k/wmi.h

@@ -112,6 +112,7 @@ enum wmi_cmd_id {
 	WMI_TX_STATS_CMDID,
 	WMI_RX_STATS_CMDID,
 	WMI_BITRATE_MASK_CMDID,
+	WMI_REG_RMW_CMDID,
 };
 
 enum wmi_event_id {
@@ -125,12 +126,19 @@ enum wmi_event_id {
 };
 
 #define MAX_CMD_NUMBER 62
+#define MAX_RMW_CMD_NUMBER 15
 
 struct register_write {
 	__be32 reg;
 	__be32 val;
 };
 
+struct register_rmw {
+	__be32 reg;
+	__be32 set;
+	__be32 clr;
+} __packed;
+
 struct ath9k_htc_tx_event {
 	int count;
 	struct __wmi_event_txstatus txs;
@@ -156,10 +164,18 @@ struct wmi {
 
 	spinlock_t wmi_lock;
 
+	/* multi write section */
 	atomic_t mwrite_cnt;
 	struct register_write multi_write[MAX_CMD_NUMBER];
 	u32 multi_write_idx;
 	struct mutex multi_write_mutex;
+
+	/* multi rmw section */
+	atomic_t m_rmw_cnt;
+	struct register_rmw multi_rmw[MAX_RMW_CMD_NUMBER];
+	u32 multi_rmw_idx;
+	struct mutex multi_rmw_mutex;
+
 };
 
 struct wmi *ath9k_init_wmi(struct ath9k_htc_priv *priv);