net: enetc: take the MDIO lock only once per NAPI poll cycle

stable inclusion
from stable-5.10.24
commit bf9c564716a13dde6a990d3b02c27cd6e39608bf
bugzilla: 51348

--------------------------------

commit 6d36ecdb upstream.

The workaround for the ENETC MDIO erratum caused a performance
degradation of 82 Kpps (seen with IP forwarding of two 1Gbps streams of
64B packets). This is due to excessive locking and unlocking in the fast
path, which can be avoided.

By taking the MDIO read-side lock only once per NAPI poll cycle, we are
able to regain 54 Kpps (65%) of the performance hit. The rest of the
performance degradation comes from the TX data path, but unfortunately
it doesn't look like we can optimize that away easily, even with
netdev_xmit_more(), there just isn't any skb batching done, to help with
taking the MDIO lock less often than once per packet.

We need to change the register accessor type for enetc_get_tx_tstamp,
because it now runs under the enetc_lock_mdio as per the new call path
detailed below:

enetc_msix
-> napi_schedule
   -> enetc_poll
      -> enetc_lock_mdio
      -> enetc_clean_tx_ring
         -> enetc_get_tx_tstamp
      -> enetc_clean_rx_ring
      -> enetc_unlock_mdio

Fixes: fd5736bf ("enetc: Workaround for MDIO register access issue")
Signed-off-by: NVladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: NChen Jun <chenjun102@huawei.com>
Acked-by: N  Weilong Chen <chenweilong@huawei.com>
Signed-off-by: NZheng Zengkai <zhengzengkai@huawei.com>

drivers/net/ethernet/freescale/enetc/enetc.c

@@ -321,6 +321,8 @@ static int enetc_poll(struct napi_struct *napi, int budget)
 	int work_done;
 	int i;
 
+	enetc_lock_mdio();
+
 	for (i = 0; i < v->count_tx_rings; i++)
 		if (!enetc_clean_tx_ring(&v->tx_ring[i], budget))
 			complete = false;
@@ -331,8 +333,10 @@ static int enetc_poll(struct napi_struct *napi, int budget)
 	if (work_done)
 		v->rx_napi_work = true;
 
-	if (!complete)
+	if (!complete) {
+		enetc_unlock_mdio();
 		return budget;
+	}
 
 	napi_complete_done(napi, work_done);
 
@@ -341,8 +345,6 @@ static int enetc_poll(struct napi_struct *napi, int budget)
 
 	v->rx_napi_work = false;
 
-	enetc_lock_mdio();
-
 	/* enable interrupts */
 	enetc_wr_reg_hot(v->rbier, ENETC_RBIER_RXTIE);
 
@@ -367,8 +369,8 @@ static void enetc_get_tx_tstamp(struct enetc_hw *hw, union enetc_tx_bd *txbd,
 {
 	u32 lo, hi, tstamp_lo;
 
-	lo = enetc_rd(hw, ENETC_SICTR0);
-	hi = enetc_rd(hw, ENETC_SICTR1);
+	lo = enetc_rd_hot(hw, ENETC_SICTR0);
+	hi = enetc_rd_hot(hw, ENETC_SICTR1);
 	tstamp_lo = le32_to_cpu(txbd->wb.tstamp);
 	if (lo <= tstamp_lo)
 		hi -= 1;
@@ -398,9 +400,7 @@ static bool enetc_clean_tx_ring(struct enetc_bdr *tx_ring, int napi_budget)
 	i = tx_ring->next_to_clean;
 	tx_swbd = &tx_ring->tx_swbd[i];
 
-	enetc_lock_mdio();
 	bds_to_clean = enetc_bd_ready_count(tx_ring, i);
-	enetc_unlock_mdio();
 
 	do_tstamp = false;
 
@@ -443,8 +443,6 @@ static bool enetc_clean_tx_ring(struct enetc_bdr *tx_ring, int napi_budget)
 			tx_swbd = tx_ring->tx_swbd;
 		}
 
-		enetc_lock_mdio();
-
 		/* BD iteration loop end */
 		if (is_eof) {
 			tx_frm_cnt++;
@@ -455,8 +453,6 @@ static bool enetc_clean_tx_ring(struct enetc_bdr *tx_ring, int napi_budget)
 
 		if (unlikely(!bds_to_clean))
 			bds_to_clean = enetc_bd_ready_count(tx_ring, i);
-
-		enetc_unlock_mdio();
 	}
 
 	tx_ring->next_to_clean = i;
@@ -700,8 +696,6 @@ static int enetc_clean_rx_ring(struct enetc_bdr *rx_ring,
 		u32 bd_status;
 		u16 size;
 
-		enetc_lock_mdio();
-
 		if (cleaned_cnt >= ENETC_RXBD_BUNDLE) {
 			int count = enetc_refill_rx_ring(rx_ring, cleaned_cnt);
 
@@ -712,19 +706,15 @@ static int enetc_clean_rx_ring(struct enetc_bdr *rx_ring,
 
 		rxbd = enetc_rxbd(rx_ring, i);
 		bd_status = le32_to_cpu(rxbd->r.lstatus);
-		if (!bd_status) {
-			enetc_unlock_mdio();
+		if (!bd_status)
 			break;
-		}
 
 		enetc_wr_reg_hot(rx_ring->idr, BIT(rx_ring->index));
 		dma_rmb(); /* for reading other rxbd fields */
 		size = le16_to_cpu(rxbd->r.buf_len);
 		skb = enetc_map_rx_buff_to_skb(rx_ring, i, size);
-		if (!skb) {
-			enetc_unlock_mdio();
+		if (!skb)
 			break;
-		}
 
 		enetc_get_offloads(rx_ring, rxbd, skb);
 
@@ -736,7 +726,6 @@ static int enetc_clean_rx_ring(struct enetc_bdr *rx_ring,
 
 		if (unlikely(bd_status &
 			     ENETC_RXBD_LSTATUS(ENETC_RXBD_ERR_MASK))) {
-			enetc_unlock_mdio();
 			dev_kfree_skb(skb);
 			while (!(bd_status & ENETC_RXBD_LSTATUS_F)) {
 				dma_rmb();
@@ -776,8 +765,6 @@ static int enetc_clean_rx_ring(struct enetc_bdr *rx_ring,
 
 		enetc_process_skb(rx_ring, skb);
 
-		enetc_unlock_mdio();
-
 		napi_gro_receive(napi, skb);
 
 		rx_frm_cnt++;

drivers/net/ethernet/freescale/enetc/enetc_hw.h

@@ -453,6 +453,8 @@ static inline u64 _enetc_rd_reg64_wa(void __iomem *reg)
 #define enetc_wr_reg(reg, val)		_enetc_wr_reg_wa((reg), (val))
 #define enetc_rd(hw, off)		enetc_rd_reg((hw)->reg + (off))
 #define enetc_wr(hw, off, val)		enetc_wr_reg((hw)->reg + (off), val)
+#define enetc_rd_hot(hw, off)		enetc_rd_reg_hot((hw)->reg + (off))
+#define enetc_wr_hot(hw, off, val)	enetc_wr_reg_hot((hw)->reg + (off), val)
 #define enetc_rd64(hw, off)		_enetc_rd_reg64_wa((hw)->reg + (off))
 /* port register accessors - PF only */
 #define enetc_port_rd(hw, off)		enetc_rd_reg((hw)->port + (off))