ath9k_hw: optimize tx status descriptor processing

Disassembly shows, that at least on MIPS, the compiler generates a lot of
memory accesses to the same location in the descriptor field parsing.
Since it is operating on uncached memory, this can be quite expensive in
this hot path.
Change the code a bit to help the compiler optimize it properly, and get
rid of some unused fields in the ath_tx_status struct.
Signed-off-by: NFelix Fietkau <nbd@openwrt.org>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

drivers/net/wireless/ath/ath9k/ar9002_mac.c

@@ -208,77 +208,68 @@ static int ar9002_hw_proc_txdesc(struct ath_hw *ah, void *ds,
 				 struct ath_tx_status *ts)
 {
 	struct ar5416_desc *ads = AR5416DESC(ds);
+	u32 status;
 
-	if ((ads->ds_txstatus9 & AR_TxDone) == 0)
+	status = ACCESS_ONCE(ads->ds_txstatus9);
+	if ((status & AR_TxDone) == 0)
 		return -EINPROGRESS;
 
-	ts->ts_seqnum = MS(ads->ds_txstatus9, AR_SeqNum);
 	ts->ts_tstamp = ads->AR_SendTimestamp;
 	ts->ts_status = 0;
 	ts->ts_flags = 0;
 
-	if (ads->ds_txstatus1 & AR_FrmXmitOK)
+	if (status & AR_TxOpExceeded)
+		ts->ts_status |= ATH9K_TXERR_XTXOP;
+	ts->tid = MS(status, AR_TxTid);
+	ts->ts_rateindex = MS(status, AR_FinalTxIdx);
+	ts->ts_seqnum = MS(status, AR_SeqNum);
+
+	status = ACCESS_ONCE(ads->ds_txstatus0);
+	ts->ts_rssi_ctl0 = MS(status, AR_TxRSSIAnt00);
+	ts->ts_rssi_ctl1 = MS(status, AR_TxRSSIAnt01);
+	ts->ts_rssi_ctl2 = MS(status, AR_TxRSSIAnt02);
+	if (status & AR_TxBaStatus) {
+		ts->ts_flags |= ATH9K_TX_BA;
+		ts->ba_low = ads->AR_BaBitmapLow;
+		ts->ba_high = ads->AR_BaBitmapHigh;
+	}
+
+	status = ACCESS_ONCE(ads->ds_txstatus1);
+	if (status & AR_FrmXmitOK)
 		ts->ts_status |= ATH9K_TX_ACKED;
-	if (ads->ds_txstatus1 & AR_ExcessiveRetries)
+	if (status & AR_ExcessiveRetries)
 		ts->ts_status |= ATH9K_TXERR_XRETRY;
-	if (ads->ds_txstatus1 & AR_Filtered)
+	if (status & AR_Filtered)
 		ts->ts_status |= ATH9K_TXERR_FILT;
-	if (ads->ds_txstatus1 & AR_FIFOUnderrun) {
+	if (status & AR_FIFOUnderrun) {
 		ts->ts_status |= ATH9K_TXERR_FIFO;
 		ath9k_hw_updatetxtriglevel(ah, true);
 	}
-	if (ads->ds_txstatus9 & AR_TxOpExceeded)
-		ts->ts_status |= ATH9K_TXERR_XTXOP;
-	if (ads->ds_txstatus1 & AR_TxTimerExpired)
+	if (status & AR_TxTimerExpired)
 		ts->ts_status |= ATH9K_TXERR_TIMER_EXPIRED;
-
-	if (ads->ds_txstatus1 & AR_DescCfgErr)
+	if (status & AR_DescCfgErr)
 		ts->ts_flags |= ATH9K_TX_DESC_CFG_ERR;
-	if (ads->ds_txstatus1 & AR_TxDataUnderrun) {
+	if (status & AR_TxDataUnderrun) {
 		ts->ts_flags |= ATH9K_TX_DATA_UNDERRUN;
 		ath9k_hw_updatetxtriglevel(ah, true);
 	}
-	if (ads->ds_txstatus1 & AR_TxDelimUnderrun) {
+	if (status & AR_TxDelimUnderrun) {
 		ts->ts_flags |= ATH9K_TX_DELIM_UNDERRUN;
 		ath9k_hw_updatetxtriglevel(ah, true);
 	}
-	if (ads->ds_txstatus0 & AR_TxBaStatus) {
-		ts->ts_flags |= ATH9K_TX_BA;
-		ts->ba_low = ads->AR_BaBitmapLow;
-		ts->ba_high = ads->AR_BaBitmapHigh;
-	}
+	ts->ts_shortretry = MS(status, AR_RTSFailCnt);
+	ts->ts_longretry = MS(status, AR_DataFailCnt);
+	ts->ts_virtcol = MS(status, AR_VirtRetryCnt);
 
-	ts->ts_rateindex = MS(ads->ds_txstatus9, AR_FinalTxIdx);
-	switch (ts->ts_rateindex) {
-	case 0:
-		ts->ts_ratecode = MS(ads->ds_ctl3, AR_XmitRate0);
-		break;
-	case 1:
-		ts->ts_ratecode = MS(ads->ds_ctl3, AR_XmitRate1);
-		break;
-	case 2:
-		ts->ts_ratecode = MS(ads->ds_ctl3, AR_XmitRate2);
-		break;
-	case 3:
-		ts->ts_ratecode = MS(ads->ds_ctl3, AR_XmitRate3);
-		break;
-	}
+	status = ACCESS_ONCE(ads->ds_txstatus5);
+	ts->ts_rssi = MS(status, AR_TxRSSICombined);
+	ts->ts_rssi_ext0 = MS(status, AR_TxRSSIAnt10);
+	ts->ts_rssi_ext1 = MS(status, AR_TxRSSIAnt11);
+	ts->ts_rssi_ext2 = MS(status, AR_TxRSSIAnt12);
 
-	ts->ts_rssi = MS(ads->ds_txstatus5, AR_TxRSSICombined);
-	ts->ts_rssi_ctl0 = MS(ads->ds_txstatus0, AR_TxRSSIAnt00);
-	ts->ts_rssi_ctl1 = MS(ads->ds_txstatus0, AR_TxRSSIAnt01);
-	ts->ts_rssi_ctl2 = MS(ads->ds_txstatus0, AR_TxRSSIAnt02);
-	ts->ts_rssi_ext0 = MS(ads->ds_txstatus5, AR_TxRSSIAnt10);
-	ts->ts_rssi_ext1 = MS(ads->ds_txstatus5, AR_TxRSSIAnt11);
-	ts->ts_rssi_ext2 = MS(ads->ds_txstatus5, AR_TxRSSIAnt12);
 	ts->evm0 = ads->AR_TxEVM0;
 	ts->evm1 = ads->AR_TxEVM1;
 	ts->evm2 = ads->AR_TxEVM2;
-	ts->ts_shortretry = MS(ads->ds_txstatus1, AR_RTSFailCnt);
-	ts->ts_longretry = MS(ads->ds_txstatus1, AR_DataFailCnt);
-	ts->ts_virtcol = MS(ads->ds_txstatus1, AR_VirtRetryCnt);
-	ts->tid = MS(ads->ds_txstatus9, AR_TxTid);
-	ts->ts_antenna = 0;
 
 	return 0;
 }

drivers/net/wireless/ath/ath9k/ar9003_mac.c

@@ -237,10 +237,12 @@ static int ar9003_hw_proc_txdesc(struct ath_hw *ah, void *ds,
 				 struct ath_tx_status *ts)
 {
 	struct ar9003_txs *ads;
+	u32 status;
 
 	ads = &ah->ts_ring[ah->ts_tail];
 
-	if ((ads->status8 & AR_TxDone) == 0)
+	status = ACCESS_ONCE(ads->status8);
+	if ((status & AR_TxDone) == 0)
 		return -EINPROGRESS;
 
 	ah->ts_tail = (ah->ts_tail + 1) % ah->ts_size;
@@ -253,57 +255,58 @@ static int ar9003_hw_proc_txdesc(struct ath_hw *ah, void *ds,
 		return -EIO;
 	}
 
+	if (status & AR_TxOpExceeded)
+		ts->ts_status |= ATH9K_TXERR_XTXOP;
+	ts->ts_rateindex = MS(status, AR_FinalTxIdx);
+	ts->ts_seqnum = MS(status, AR_SeqNum);
+	ts->tid = MS(status, AR_TxTid);
+
 	ts->qid = MS(ads->ds_info, AR_TxQcuNum);
 	ts->desc_id = MS(ads->status1, AR_TxDescId);
-	ts->ts_seqnum = MS(ads->status8, AR_SeqNum);
 	ts->ts_tstamp = ads->status4;
 	ts->ts_status = 0;
 	ts->ts_flags  = 0;
 
-	if (ads->status3 & AR_ExcessiveRetries)
+	status = ACCESS_ONCE(ads->status2);
+	ts->ts_rssi_ctl0 = MS(status, AR_TxRSSIAnt00);
+	ts->ts_rssi_ctl1 = MS(status, AR_TxRSSIAnt01);
+	ts->ts_rssi_ctl2 = MS(status, AR_TxRSSIAnt02);
+	if (status & AR_TxBaStatus) {
+		ts->ts_flags |= ATH9K_TX_BA;
+		ts->ba_low = ads->status5;
+		ts->ba_high = ads->status6;
+	}
+
+	status = ACCESS_ONCE(ads->status3);
+	if (status & AR_ExcessiveRetries)
 		ts->ts_status |= ATH9K_TXERR_XRETRY;
-	if (ads->status3 & AR_Filtered)
+	if (status & AR_Filtered)
 		ts->ts_status |= ATH9K_TXERR_FILT;
-	if (ads->status3 & AR_FIFOUnderrun) {
+	if (status & AR_FIFOUnderrun) {
 		ts->ts_status |= ATH9K_TXERR_FIFO;
 		ath9k_hw_updatetxtriglevel(ah, true);
 	}
-	if (ads->status8 & AR_TxOpExceeded)
-		ts->ts_status |= ATH9K_TXERR_XTXOP;
-	if (ads->status3 & AR_TxTimerExpired)
+	if (status & AR_TxTimerExpired)
 		ts->ts_status |= ATH9K_TXERR_TIMER_EXPIRED;
-
-	if (ads->status3 & AR_DescCfgErr)
+	if (status & AR_DescCfgErr)
 		ts->ts_flags |= ATH9K_TX_DESC_CFG_ERR;
-	if (ads->status3 & AR_TxDataUnderrun) {
+	if (status & AR_TxDataUnderrun) {
 		ts->ts_flags |= ATH9K_TX_DATA_UNDERRUN;
 		ath9k_hw_updatetxtriglevel(ah, true);
 	}
-	if (ads->status3 & AR_TxDelimUnderrun) {
+	if (status & AR_TxDelimUnderrun) {
 		ts->ts_flags |= ATH9K_TX_DELIM_UNDERRUN;
 		ath9k_hw_updatetxtriglevel(ah, true);
 	}
-	if (ads->status2 & AR_TxBaStatus) {
-		ts->ts_flags |= ATH9K_TX_BA;
-		ts->ba_low = ads->status5;
-		ts->ba_high = ads->status6;
-	}
-
-	ts->ts_rateindex = MS(ads->status8, AR_FinalTxIdx);
-
-	ts->ts_rssi = MS(ads->status7, AR_TxRSSICombined);
-	ts->ts_rssi_ctl0 = MS(ads->status2, AR_TxRSSIAnt00);
-	ts->ts_rssi_ctl1 = MS(ads->status2, AR_TxRSSIAnt01);
-	ts->ts_rssi_ctl2 = MS(ads->status2, AR_TxRSSIAnt02);
-	ts->ts_rssi_ext0 = MS(ads->status7, AR_TxRSSIAnt10);
-	ts->ts_rssi_ext1 = MS(ads->status7, AR_TxRSSIAnt11);
-	ts->ts_rssi_ext2 = MS(ads->status7, AR_TxRSSIAnt12);
-	ts->ts_shortretry = MS(ads->status3, AR_RTSFailCnt);
-	ts->ts_longretry = MS(ads->status3, AR_DataFailCnt);
-	ts->ts_virtcol = MS(ads->status3, AR_VirtRetryCnt);
-	ts->ts_antenna = 0;
-
-	ts->tid = MS(ads->status8, AR_TxTid);
+	ts->ts_shortretry = MS(status, AR_RTSFailCnt);
+	ts->ts_longretry = MS(status, AR_DataFailCnt);
+	ts->ts_virtcol = MS(status, AR_VirtRetryCnt);
+
+	status = ACCESS_ONCE(ads->status7);
+	ts->ts_rssi = MS(status, AR_TxRSSICombined);
+	ts->ts_rssi_ext0 = MS(status, AR_TxRSSIAnt10);
+	ts->ts_rssi_ext1 = MS(status, AR_TxRSSIAnt11);
+	ts->ts_rssi_ext2 = MS(status, AR_TxRSSIAnt12);
 
 	memset(ads, 0, sizeof(*ads));
 

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

@@ -104,13 +104,11 @@ struct ath_tx_status {
 	u32 ts_tstamp;
 	u16 ts_seqnum;
 	u8 ts_status;
-	u8 ts_ratecode;
 	u8 ts_rateindex;
 	int8_t ts_rssi;
 	u8 ts_shortretry;
 	u8 ts_longretry;
 	u8 ts_virtcol;
-	u8 ts_antenna;
 	u8 ts_flags;
 	int8_t ts_rssi_ctl0;
 	int8_t ts_rssi_ctl1;
@@ -121,7 +119,6 @@ struct ath_tx_status {
 	u8 qid;
 	u16 desc_id;
 	u8 tid;
-	u8 pad[2];
 	u32 ba_low;
 	u32 ba_high;
 	u32 evm0;