[net-next,v5] atl1c: move tx cleanup processing out of interrupt

Tx queue cleanup happens in interrupt handler on same core as rx queue
processing. Both can take considerable amount of processing in high
packet-per-second scenarios.

Sending big amounts of packets can stall the rx processing which is 
unfair
and also can lead to out-of-memory condition since __dev_kfree_skb_irq
queues the skbs for later kfree in softirq which is not allowed to 
happen
with heavy load in interrupt handler.

This puts tx cleanup in its own napi and enables threaded napi to allow
the rx/tx queue processing to happen on different cores.

The ability to sustain equal amounts of tx/rx traffic increased:
from 280Kpps to 1130Kpps on Threadripper 3960X with upcoming
Mikrotik 10/25G NIC,
from 520Kpps to 850Kpps on Intel i3-3320 with Mikrotik RB44Ge adapter.

Signed-off-by: Gatis Peisenieks <gatis@mikrotik.com>
---
Reposting for net-next as requested by David Miller,

v5:
	- EXPORT_SYMBOL(dev_set_threaded) not needed, already there
v4:
	- made scripts/checkpatch.pl happy
	- moved the new intr_mask_lock to be besides the intr_mask it
	  protects so they are more likely to be on same cacheline
v3:
	- addressed comments from Eric Dumazet
	- added EXPORT_SYMBOL for dev_set_threaded
---
  drivers/net/ethernet/atheros/atl1c/atl1c.h    |  2 +
  .../net/ethernet/atheros/atl1c/atl1c_main.c   | 44 ++++++++++++++-----
  2 files changed, 36 insertions(+), 10 deletions(-)

  	return 0;
@@ -1530,20 +1531,19 @@ static inline void atl1c_clear_phy_int(struct 
atl1c_adapter *adapter)
  	spin_unlock(&adapter->mdio_lock);
  }

-static bool atl1c_clean_tx_irq(struct atl1c_adapter *adapter,
-				enum atl1c_trans_queue type)
+static int atl1c_clean_tx(struct napi_struct *napi, int budget)
  {
-	struct atl1c_tpd_ring *tpd_ring = &adapter->tpd_ring[type];
+	struct atl1c_adapter *adapter =
+		container_of(napi, struct atl1c_adapter, tx_napi);
+	struct atl1c_tpd_ring *tpd_ring = 
&adapter->tpd_ring[atl1c_trans_normal];
  	struct atl1c_buffer *buffer_info;
  	struct pci_dev *pdev = adapter->pdev;
  	u16 next_to_clean = atomic_read(&tpd_ring->next_to_clean);
  	u16 hw_next_to_clean;
-	u16 reg;
  	unsigned int total_bytes = 0, total_packets = 0;
+	unsigned long flags;

-	reg = type == atl1c_trans_high ? REG_TPD_PRI1_CIDX : 
REG_TPD_PRI0_CIDX;
-
-	AT_READ_REGW(&adapter->hw, reg, &hw_next_to_clean);
+	AT_READ_REGW(&adapter->hw, REG_TPD_PRI0_CIDX, &hw_next_to_clean);

  	while (next_to_clean != hw_next_to_clean) {
  		buffer_info = &tpd_ring->buffer_info[next_to_clean];
@@ -1564,7 +1564,15 @@ static bool atl1c_clean_tx_irq(struct 
atl1c_adapter *adapter,
  		netif_wake_queue(adapter->netdev);
  	}

-	return true;
+	if (total_packets < budget) {
+		napi_complete_done(napi, total_packets);
+		spin_lock_irqsave(&adapter->hw.intr_mask_lock, flags);
+		adapter->hw.intr_mask |= ISR_TX_PKT;
+		AT_WRITE_REG(&adapter->hw, REG_IMR, adapter->hw.intr_mask);
+		spin_unlock_irqrestore(&adapter->hw.intr_mask_lock, flags);
+		return total_packets;
+	}
+	return budget;
  }

  /**
@@ -1599,13 +1607,22 @@ static irqreturn_t atl1c_intr(int irq, void 
*data)
  		AT_WRITE_REG(hw, REG_ISR, status | ISR_DIS_INT);
  		if (status & ISR_RX_PKT) {
  			if (likely(napi_schedule_prep(&adapter->napi))) {
+				spin_lock(&hw->intr_mask_lock);
  				hw->intr_mask &= ~ISR_RX_PKT;
  				AT_WRITE_REG(hw, REG_IMR, hw->intr_mask);
+				spin_unlock(&hw->intr_mask_lock);
  				__napi_schedule(&adapter->napi);
  			}
  		}
-		if (status & ISR_TX_PKT)
-			atl1c_clean_tx_irq(adapter, atl1c_trans_normal);
+		if (status & ISR_TX_PKT) {
+			if (napi_schedule_prep(&adapter->tx_napi)) {
+				spin_lock(&hw->intr_mask_lock);
+				hw->intr_mask &= ~ISR_TX_PKT;
+				AT_WRITE_REG(hw, REG_IMR, hw->intr_mask);
+				spin_unlock(&hw->intr_mask_lock);
+				__napi_schedule(&adapter->tx_napi);
+			}
+		}

  		handled = IRQ_HANDLED;
  		/* check if PCIE PHY Link down */
@@ -1876,6 +1893,7 @@ static int atl1c_clean(struct napi_struct *napi, 
int budget)
  	struct atl1c_adapter *adapter =
  			container_of(napi, struct atl1c_adapter, napi);
  	int work_done = 0;
+	unsigned long flags;

  	/* Keep link state information with original netdev */
  	if (!netif_carrier_ok(adapter->netdev))
@@ -1886,8 +1904,10 @@ static int atl1c_clean(struct napi_struct *napi, 
int budget)
  	if (work_done < budget) {
  quit_polling:
  		napi_complete_done(napi, work_done);
+		spin_lock_irqsave(&adapter->hw.intr_mask_lock, flags);
  		adapter->hw.intr_mask |= ISR_RX_PKT;
  		AT_WRITE_REG(&adapter->hw, REG_IMR, adapter->hw.intr_mask);
+		spin_unlock_irqrestore(&adapter->hw.intr_mask_lock, flags);
  	}
  	return work_done;
  }
@@ -2325,6 +2345,7 @@ static int atl1c_up(struct atl1c_adapter *adapter)
  	atl1c_check_link_status(adapter);
  	clear_bit(__AT_DOWN, &adapter->flags);
  	napi_enable(&adapter->napi);
+	napi_enable(&adapter->tx_napi);
  	atl1c_irq_enable(adapter);
  	netif_start_queue(netdev);
  	return err;
@@ -2345,6 +2366,7 @@ static void atl1c_down(struct atl1c_adapter 
*adapter)
  	set_bit(__AT_DOWN, &adapter->flags);
  	netif_carrier_off(netdev);
  	napi_disable(&adapter->napi);
+	napi_disable(&adapter->tx_napi);
  	atl1c_irq_disable(adapter);
  	atl1c_free_irq(adapter);
  	/* disable ASPM if device inactive */
@@ -2593,7 +2615,9 @@ static int atl1c_probe(struct pci_dev *pdev, const 
struct pci_device_id *ent)
  	adapter->mii.mdio_write = atl1c_mdio_write;
  	adapter->mii.phy_id_mask = 0x1f;
  	adapter->mii.reg_num_mask = MDIO_CTRL_REG_MASK;
+	dev_set_threaded(netdev, true);
  	netif_napi_add(netdev, &adapter->napi, atl1c_clean, 64);
+	netif_napi_add(netdev, &adapter->tx_napi, atl1c_clean_tx, 64);
  	timer_setup(&adapter->phy_config_timer, atl1c_phy_config, 0);
  	/* setup the private structure */
  	err = atl1c_sw_init(adapter);