Merge branch 'net-generic-busy-polling'

Eric Dumazet says:

====================
net: extend busy polling support

This patch series extends busy polling range to tunnels devices,
and adds busy polling generic support to all NAPI drivers.

No need to provide ndo_busy_poll() method and extra synchronization
between ndo_busy_poll() and normal napi->poll() method.
This was proven very difficult and bug prone.

mlx5 driver is changed to support busy polling using this new method,
and a second mlx5 patch adds napi_complete_done() support and proper
SNMP accounting.

bnx2x and mlx4 drivers are converted to new infrastructure,
reducing kernel bloat and improving performance.

Latest patch, adding generic support, adds a new requirement :

 -free_netdev() and netif_napi_del() must be called from process context.

Since this might not be the case in some drivers, we might have to
either : fix the non conformant drivers (by disabling busy polling on them)
or revert this last patch.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/amd/xgbe/xgbe-drv.c

@@ -2024,7 +2024,6 @@ read_again:
 		skb->dev = netdev;
 		skb->protocol = eth_type_trans(skb, netdev);
 		skb_record_rx_queue(skb, channel->queue_index);
-		skb_mark_napi_id(skb, napi);
 
 		napi_gro_receive(napi, skb);
 

drivers/net/ethernet/broadcom/bcmsysport.c

@@ -1216,7 +1216,7 @@ static int bcm_sysport_init_tx_ring(struct bcm_sysport_priv *priv,
 	/* Initialize SW view of the ring */
 	spin_lock_init(&ring->lock);
 	ring->priv = priv;
-	netif_napi_add(priv->netdev, &ring->napi, bcm_sysport_tx_poll, 64);
+	netif_tx_napi_add(priv->netdev, &ring->napi, bcm_sysport_tx_poll, 64);
 	ring->index = index;
 	ring->size = size;
 	ring->alloc_size = ring->size;

drivers/net/ethernet/broadcom/bnx2x/bnx2x.h

@@ -540,10 +540,6 @@ struct bnx2x_fastpath {
 
 	struct napi_struct	napi;
 
-#ifdef CONFIG_NET_RX_BUSY_POLL
-	unsigned long		busy_poll_state;
-#endif
-
 	union host_hc_status_block	status_blk;
 	/* chip independent shortcuts into sb structure */
 	__le16			*sb_index_values;
@@ -617,115 +613,6 @@ struct bnx2x_fastpath {
 #define bnx2x_fp_stats(bp, fp)	(&((bp)->fp_stats[(fp)->index]))
 #define bnx2x_fp_qstats(bp, fp)	(&((bp)->fp_stats[(fp)->index].eth_q_stats))
 
-#ifdef CONFIG_NET_RX_BUSY_POLL
-
-enum bnx2x_fp_state {
-	BNX2X_STATE_FP_NAPI	= BIT(0), /* NAPI handler owns the queue */
-
-	BNX2X_STATE_FP_NAPI_REQ_BIT = 1, /* NAPI would like to own the queue */
-	BNX2X_STATE_FP_NAPI_REQ = BIT(1),
-
-	BNX2X_STATE_FP_POLL_BIT = 2,
-	BNX2X_STATE_FP_POLL     = BIT(2), /* busy_poll owns the queue */
-
-	BNX2X_STATE_FP_DISABLE_BIT = 3, /* queue is dismantled */
-};
-
-static inline void bnx2x_fp_busy_poll_init(struct bnx2x_fastpath *fp)
-{
-	WRITE_ONCE(fp->busy_poll_state, 0);
-}
-
-/* called from the device poll routine to get ownership of a FP */
-static inline bool bnx2x_fp_lock_napi(struct bnx2x_fastpath *fp)
-{
-	unsigned long prev, old = READ_ONCE(fp->busy_poll_state);
-
-	while (1) {
-		switch (old) {
-		case BNX2X_STATE_FP_POLL:
-			/* make sure bnx2x_fp_lock_poll() wont starve us */
-			set_bit(BNX2X_STATE_FP_NAPI_REQ_BIT,
-				&fp->busy_poll_state);
-			/* fallthrough */
-		case BNX2X_STATE_FP_POLL | BNX2X_STATE_FP_NAPI_REQ:
-			return false;
-		default:
-			break;
-		}
-		prev = cmpxchg(&fp->busy_poll_state, old, BNX2X_STATE_FP_NAPI);
-		if (unlikely(prev != old)) {
-			old = prev;
-			continue;
-		}
-		return true;
-	}
-}
-
-static inline void bnx2x_fp_unlock_napi(struct bnx2x_fastpath *fp)
-{
-	smp_wmb();
-	fp->busy_poll_state = 0;
-}
-
-/* called from bnx2x_low_latency_poll() */
-static inline bool bnx2x_fp_lock_poll(struct bnx2x_fastpath *fp)
-{
-	return cmpxchg(&fp->busy_poll_state, 0, BNX2X_STATE_FP_POLL) == 0;
-}
-
-static inline void bnx2x_fp_unlock_poll(struct bnx2x_fastpath *fp)
-{
-	smp_mb__before_atomic();
-	clear_bit(BNX2X_STATE_FP_POLL_BIT, &fp->busy_poll_state);
-}
-
-/* true if a socket is polling */
-static inline bool bnx2x_fp_ll_polling(struct bnx2x_fastpath *fp)
-{
-	return READ_ONCE(fp->busy_poll_state) & BNX2X_STATE_FP_POLL;
-}
-
-/* false if fp is currently owned */
-static inline bool bnx2x_fp_ll_disable(struct bnx2x_fastpath *fp)
-{
-	set_bit(BNX2X_STATE_FP_DISABLE_BIT, &fp->busy_poll_state);
-	return !bnx2x_fp_ll_polling(fp);
-
-}
-#else
-static inline void bnx2x_fp_busy_poll_init(struct bnx2x_fastpath *fp)
-{
-}
-
-static inline bool bnx2x_fp_lock_napi(struct bnx2x_fastpath *fp)
-{
-	return true;
-}
-
-static inline void bnx2x_fp_unlock_napi(struct bnx2x_fastpath *fp)
-{
-}
-
-static inline bool bnx2x_fp_lock_poll(struct bnx2x_fastpath *fp)
-{
-	return false;
-}
-
-static inline void bnx2x_fp_unlock_poll(struct bnx2x_fastpath *fp)
-{
-}
-
-static inline bool bnx2x_fp_ll_polling(struct bnx2x_fastpath *fp)
-{
-	return false;
-}
-static inline bool bnx2x_fp_ll_disable(struct bnx2x_fastpath *fp)
-{
-	return true;
-}
-#endif /* CONFIG_NET_RX_BUSY_POLL */
-
 /* Use 2500 as a mini-jumbo MTU for FCoE */
 #define BNX2X_FCOE_MINI_JUMBO_MTU	2500
 

drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c

@@ -46,7 +46,6 @@ static void bnx2x_add_all_napi_cnic(struct bnx2x *bp)
 	for_each_rx_queue_cnic(bp, i) {
 		netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
 			       bnx2x_poll, NAPI_POLL_WEIGHT);
-		napi_hash_add(&bnx2x_fp(bp, i, napi));
 	}
 }
 
@@ -58,7 +57,6 @@ static void bnx2x_add_all_napi(struct bnx2x *bp)
 	for_each_eth_queue(bp, i) {
 		netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
 			       bnx2x_poll, NAPI_POLL_WEIGHT);
-		napi_hash_add(&bnx2x_fp(bp, i, napi));
 	}
 }
 
@@ -1094,12 +1092,7 @@ reuse_rx:
 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 					       le16_to_cpu(cqe_fp->vlan_tag));
 
-		skb_mark_napi_id(skb, &fp->napi);
-
-		if (bnx2x_fp_ll_polling(fp))
-			netif_receive_skb(skb);
-		else
-			napi_gro_receive(&fp->napi, skb);
+		napi_gro_receive(&fp->napi, skb);
 next_rx:
 		rx_buf->data = NULL;
 
@@ -1869,7 +1862,6 @@ static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
 	int i;
 
 	for_each_rx_queue_cnic(bp, i) {
-		bnx2x_fp_busy_poll_init(&bp->fp[i]);
 		napi_enable(&bnx2x_fp(bp, i, napi));
 	}
 }
@@ -1879,7 +1871,6 @@ static void bnx2x_napi_enable(struct bnx2x *bp)
 	int i;
 
 	for_each_eth_queue(bp, i) {
-		bnx2x_fp_busy_poll_init(&bp->fp[i]);
 		napi_enable(&bnx2x_fp(bp, i, napi));
 	}
 }
@@ -1890,8 +1881,6 @@ static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
 
 	for_each_rx_queue_cnic(bp, i) {
 		napi_disable(&bnx2x_fp(bp, i, napi));
-		while (!bnx2x_fp_ll_disable(&bp->fp[i]))
-			usleep_range(1000, 2000);
 	}
 }
 
@@ -1901,8 +1890,6 @@ static void bnx2x_napi_disable(struct bnx2x *bp)
 
 	for_each_eth_queue(bp, i) {
 		napi_disable(&bnx2x_fp(bp, i, napi));
-		while (!bnx2x_fp_ll_disable(&bp->fp[i]))
-			usleep_range(1000, 2000);
 	}
 }
 
@@ -3232,9 +3219,6 @@ static int bnx2x_poll(struct napi_struct *napi, int budget)
 			return 0;
 		}
 #endif
-		if (!bnx2x_fp_lock_napi(fp))
-			return budget;
-
 		for_each_cos_in_tx_queue(fp, cos)
 			if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
 				bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
@@ -3243,14 +3227,10 @@ static int bnx2x_poll(struct napi_struct *napi, int budget)
 			work_done += bnx2x_rx_int(fp, budget - work_done);
 
 			/* must not complete if we consumed full budget */
-			if (work_done >= budget) {
-				bnx2x_fp_unlock_napi(fp);
+			if (work_done >= budget)
 				break;
-			}
 		}
 
-		bnx2x_fp_unlock_napi(fp);
-
 		/* Fall out from the NAPI loop if needed */
 		if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
 
@@ -3294,32 +3274,6 @@ static int bnx2x_poll(struct napi_struct *napi, int budget)
 	return work_done;
 }
 
-#ifdef CONFIG_NET_RX_BUSY_POLL
-/* must be called with local_bh_disable()d */
-int bnx2x_low_latency_recv(struct napi_struct *napi)
-{
-	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
-						 napi);
-	struct bnx2x *bp = fp->bp;
-	int found = 0;
-
-	if ((bp->state == BNX2X_STATE_CLOSED) ||
-	    (bp->state == BNX2X_STATE_ERROR) ||
-	    (bp->dev->features & (NETIF_F_LRO | NETIF_F_GRO)))
-		return LL_FLUSH_FAILED;
-
-	if (!bnx2x_fp_lock_poll(fp))
-		return LL_FLUSH_BUSY;
-
-	if (bnx2x_has_rx_work(fp))
-		found = bnx2x_rx_int(fp, 4);
-
-	bnx2x_fp_unlock_poll(fp);
-
-	return found;
-}
-#endif
-
 /* we split the first BD into headers and data BDs
  * to ease the pain of our fellow microcode engineers
  * we use one mapping for both BDs

drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h

@@ -569,13 +569,6 @@ int bnx2x_enable_msix(struct bnx2x *bp);
  */
 int bnx2x_enable_msi(struct bnx2x *bp);
 
-/**
- * bnx2x_low_latency_recv - LL callback
- *
- * @napi:	napi structure
- */
-int bnx2x_low_latency_recv(struct napi_struct *napi);
-
 /**
  * bnx2x_alloc_mem_bp - allocate memories outsize main driver structure
  *

drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c

@@ -13004,9 +13004,6 @@ static const struct net_device_ops bnx2x_netdev_ops = {
 	.ndo_fcoe_get_wwn	= bnx2x_fcoe_get_wwn,
 #endif
 
-#ifdef CONFIG_NET_RX_BUSY_POLL
-	.ndo_busy_poll		= bnx2x_low_latency_recv,
-#endif
 	.ndo_get_phys_port_id	= bnx2x_get_phys_port_id,
 	.ndo_set_vf_link_state	= bnx2x_set_vf_link_state,
 	.ndo_features_check	= bnx2x_features_check,

drivers/net/ethernet/broadcom/bnxt/bnxt.c

@@ -4227,12 +4227,10 @@ static void bnxt_init_napi(struct bnxt *bp)
 			bnapi = bp->bnapi[i];
 			netif_napi_add(bp->dev, &bnapi->napi,
 				       bnxt_poll, 64);
-			napi_hash_add(&bnapi->napi);
 		}
 	} else {
 		bnapi = bp->bnapi[0];
 		netif_napi_add(bp->dev, &bnapi->napi, bnxt_poll, 64);
-		napi_hash_add(&bnapi->napi);
 	}
 }
 

drivers/net/ethernet/broadcom/genet/bcmgenet.c

@@ -2041,11 +2041,11 @@ static void bcmgenet_init_tx_napi(struct bcmgenet_priv *priv)
 
 	for (i = 0; i < priv->hw_params->tx_queues; ++i) {
 		ring = &priv->tx_rings[i];
-		netif_napi_add(priv->dev, &ring->napi, bcmgenet_tx_poll, 64);
+		netif_tx_napi_add(priv->dev, &ring->napi, bcmgenet_tx_poll, 64);
 	}
 
 	ring = &priv->tx_rings[DESC_INDEX];
-	netif_napi_add(priv->dev, &ring->napi, bcmgenet_tx_poll, 64);
+	netif_tx_napi_add(priv->dev, &ring->napi, bcmgenet_tx_poll, 64);
 }
 
 static void bcmgenet_enable_tx_napi(struct bcmgenet_priv *priv)

drivers/net/ethernet/chelsio/cxgb4/sge.c

@@ -1864,7 +1864,6 @@ static void do_gro(struct sge_eth_rxq *rxq, const struct pkt_gl *gl,
 	skb->truesize += skb->data_len;
 	skb->ip_summed = CHECKSUM_UNNECESSARY;
 	skb_record_rx_queue(skb, rxq->rspq.idx);
-	skb_mark_napi_id(skb, &rxq->rspq.napi);
 	pi = netdev_priv(skb->dev);
 	if (pi->rxtstamp)
 		cxgb4_sgetim_to_hwtstamp(adapter, skb_hwtstamps(skb),
@@ -2528,7 +2527,6 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
 		goto err;
 
 	netif_napi_add(dev, &iq->napi, napi_rx_handler, 64);
-	napi_hash_add(&iq->napi);
 	iq->cur_desc = iq->desc;
 	iq->cidx = 0;
 	iq->gen = 1;

drivers/net/ethernet/cisco/enic/enic_main.c

@@ -2458,13 +2458,11 @@ static int enic_dev_init(struct enic *enic)
 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
 	default:
 		netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
-		napi_hash_add(&enic->napi[0]);
 		break;
 	case VNIC_DEV_INTR_MODE_MSIX:
 		for (i = 0; i < enic->rq_count; i++) {
 			netif_napi_add(netdev, &enic->napi[i],
 				enic_poll_msix_rq, NAPI_POLL_WEIGHT);
-			napi_hash_add(&enic->napi[i]);
 		}
 		for (i = 0; i < enic->wq_count; i++)
 			netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],

drivers/net/ethernet/emulex/benet/be_main.c

@@ -2184,7 +2184,6 @@ static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
 		skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
 
 	skb->csum_level = rxcp->tunneled;
-	skb_mark_napi_id(skb, napi);
 
 	if (rxcp->vlanf)
 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
@@ -2631,7 +2630,6 @@ static int be_evt_queues_create(struct be_adapter *adapter)
 				eqo->affinity_mask);
 		netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
 			       BE_NAPI_WEIGHT);
-		napi_hash_add(&eqo->napi);
 	}
 	return 0;
 }

drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c

@@ -1050,7 +1050,7 @@ static int fs_enet_probe(struct platform_device *ofdev)
 	ndev->netdev_ops = &fs_enet_netdev_ops;
 	ndev->watchdog_timeo = 2 * HZ;
 	netif_napi_add(ndev, &fep->napi, fs_enet_rx_napi, fpi->napi_weight);
-	netif_napi_add(ndev, &fep->napi_tx, fs_enet_tx_napi, 2);
+	netif_tx_napi_add(ndev, &fep->napi_tx, fs_enet_tx_napi, 2);
 
 	ndev->ethtool_ops = &fs_ethtool_ops;
 

drivers/net/ethernet/freescale/gianfar.c

@@ -1347,12 +1347,12 @@ static int gfar_probe(struct platform_device *ofdev)
 		if (priv->poll_mode == GFAR_SQ_POLLING) {
 			netif_napi_add(dev, &priv->gfargrp[i].napi_rx,
 				       gfar_poll_rx_sq, GFAR_DEV_WEIGHT);
-			netif_napi_add(dev, &priv->gfargrp[i].napi_tx,
+			netif_tx_napi_add(dev, &priv->gfargrp[i].napi_tx,
 				       gfar_poll_tx_sq, 2);
 		} else {
 			netif_napi_add(dev, &priv->gfargrp[i].napi_rx,
 				       gfar_poll_rx, GFAR_DEV_WEIGHT);
-			netif_napi_add(dev, &priv->gfargrp[i].napi_tx,
+			netif_tx_napi_add(dev, &priv->gfargrp[i].napi_tx,
 				       gfar_poll_tx, 2);
 		}
 	}

drivers/net/ethernet/intel/i40e/i40e_txrx.c

@@ -1632,7 +1632,6 @@ static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget)
 			continue;
 		}
 #endif
-		skb_mark_napi_id(skb, &rx_ring->q_vector->napi);
 		i40e_receive_skb(rx_ring, skb, vlan_tag);
 
 		rx_desc->wb.qword1.status_error_len = 0;

drivers/net/ethernet/intel/i40evf/i40e_txrx.c

@@ -1090,7 +1090,6 @@ static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget)
 			continue;
 		}
 #endif
-		skb_mark_napi_id(skb, &rx_ring->q_vector->napi);
 		i40e_receive_skb(rx_ring, skb, vlan_tag);
 
 		rx_desc->wb.qword1.status_error_len = 0;

drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c

@@ -844,7 +844,6 @@ static int ixgbe_alloc_q_vector(struct ixgbe_adapter *adapter,
 	/* initialize NAPI */
 	netif_napi_add(adapter->netdev, &q_vector->napi,
 		       ixgbe_poll, 64);
-	napi_hash_add(&q_vector->napi);
 
 #ifdef CONFIG_NET_RX_BUSY_POLL
 	/* initialize busy poll */

drivers/net/ethernet/intel/ixgbe/ixgbe_main.c

@@ -1659,6 +1659,7 @@ static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
 			 struct sk_buff *skb)
 {
+	skb_mark_napi_id(skb, &q_vector->napi);
 	if (ixgbe_qv_busy_polling(q_vector))
 		netif_receive_skb(skb);
 	else
@@ -2123,7 +2124,6 @@ static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 		}
 
 #endif /* IXGBE_FCOE */
-		skb_mark_napi_id(skb, &q_vector->napi);
 		ixgbe_rx_skb(q_vector, skb);
 
 		/* update budget accounting */

drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c

@@ -2483,9 +2483,6 @@ static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
 		q_vector->v_idx = q_idx;
 		netif_napi_add(adapter->netdev, &q_vector->napi,
 			       ixgbevf_poll, 64);
-#ifdef CONFIG_NET_RX_BUSY_POLL
-		napi_hash_add(&q_vector->napi);
-#endif
 		adapter->q_vector[q_idx] = q_vector;
 	}
 

drivers/net/ethernet/mellanox/mlx4/en_cq.c

@@ -155,13 +155,11 @@ int mlx4_en_activate_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq,
 	cq->mcq.comp  = cq->is_tx ? mlx4_en_tx_irq : mlx4_en_rx_irq;
 	cq->mcq.event = mlx4_en_cq_event;
 
-	if (cq->is_tx) {
-		netif_napi_add(cq->dev, &cq->napi, mlx4_en_poll_tx_cq,
-			       NAPI_POLL_WEIGHT);
-	} else {
+	if (cq->is_tx)
+		netif_tx_napi_add(cq->dev, &cq->napi, mlx4_en_poll_tx_cq,
+				  NAPI_POLL_WEIGHT);
+	else
 		netif_napi_add(cq->dev, &cq->napi, mlx4_en_poll_rx_cq, 64);
-		napi_hash_add(&cq->napi);
-	}
 
 	napi_enable(&cq->napi);
 

drivers/net/ethernet/mellanox/mlx4/en_ethtool.c

@@ -337,11 +337,7 @@ static int mlx4_en_get_sset_count(struct net_device *dev, int sset)
 	case ETH_SS_STATS:
 		return bitmap_iterator_count(&it) +
 			(priv->tx_ring_num * 2) +
-#ifdef CONFIG_NET_RX_BUSY_POLL
-			(priv->rx_ring_num * 5);
-#else
 			(priv->rx_ring_num * 2);
-#endif
 	case ETH_SS_TEST:
 		return MLX4_EN_NUM_SELF_TEST - !(priv->mdev->dev->caps.flags
 					& MLX4_DEV_CAP_FLAG_UC_LOOPBACK) * 2;
@@ -408,11 +404,6 @@ static void mlx4_en_get_ethtool_stats(struct net_device *dev,
 	for (i = 0; i < priv->rx_ring_num; i++) {
 		data[index++] = priv->rx_ring[i]->packets;
 		data[index++] = priv->rx_ring[i]->bytes;
-#ifdef CONFIG_NET_RX_BUSY_POLL
-		data[index++] = priv->rx_ring[i]->yields;
-		data[index++] = priv->rx_ring[i]->misses;
-		data[index++] = priv->rx_ring[i]->cleaned;
-#endif
 	}
 	spin_unlock_bh(&priv->stats_lock);
 
@@ -486,14 +477,6 @@ static void mlx4_en_get_strings(struct net_device *dev,
 				"rx%d_packets", i);
 			sprintf(data + (index++) * ETH_GSTRING_LEN,
 				"rx%d_bytes", i);
-#ifdef CONFIG_NET_RX_BUSY_POLL
-			sprintf(data + (index++) * ETH_GSTRING_LEN,
-				"rx%d_napi_yield", i);
-			sprintf(data + (index++) * ETH_GSTRING_LEN,
-				"rx%d_misses", i);
-			sprintf(data + (index++) * ETH_GSTRING_LEN,
-				"rx%d_cleaned", i);
-#endif
 		}
 		break;
 	case ETH_SS_PRIV_FLAGS:

drivers/net/ethernet/mellanox/mlx4/en_netdev.c

@@ -69,34 +69,6 @@ int mlx4_en_setup_tc(struct net_device *dev, u8 up)
 	return 0;
 }
 
-#ifdef CONFIG_NET_RX_BUSY_POLL
-/* must be called with local_bh_disable()d */
-static int mlx4_en_low_latency_recv(struct napi_struct *napi)
-{
-	struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
-	struct net_device *dev = cq->dev;
-	struct mlx4_en_priv *priv = netdev_priv(dev);
-	struct mlx4_en_rx_ring *rx_ring = priv->rx_ring[cq->ring];
-	int done;
-
-	if (!priv->port_up)
-		return LL_FLUSH_FAILED;
-
-	if (!mlx4_en_cq_lock_poll(cq))
-		return LL_FLUSH_BUSY;
-
-	done = mlx4_en_process_rx_cq(dev, cq, 4);
-	if (likely(done))
-		rx_ring->cleaned += done;
-	else
-		rx_ring->misses++;
-
-	mlx4_en_cq_unlock_poll(cq);
-
-	return done;
-}
-#endif	/* CONFIG_NET_RX_BUSY_POLL */
-
 #ifdef CONFIG_RFS_ACCEL
 
 struct mlx4_en_filter {
@@ -1561,8 +1533,6 @@ int mlx4_en_start_port(struct net_device *dev)
 	for (i = 0; i < priv->rx_ring_num; i++) {
 		cq = priv->rx_cq[i];
 
-		mlx4_en_cq_init_lock(cq);
-
 		err = mlx4_en_init_affinity_hint(priv, i);
 		if (err) {
 			en_err(priv, "Failed preparing IRQ affinity hint\n");
@@ -1859,13 +1829,6 @@ void mlx4_en_stop_port(struct net_device *dev, int detach)
 	for (i = 0; i < priv->rx_ring_num; i++) {
 		struct mlx4_en_cq *cq = priv->rx_cq[i];
 
-		local_bh_disable();
-		while (!mlx4_en_cq_lock_napi(cq)) {
-			pr_info("CQ %d locked\n", i);
-			mdelay(1);
-		}
-		local_bh_enable();
-
 		napi_synchronize(&cq->napi);
 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
 		mlx4_en_deactivate_cq(priv, cq);
@@ -2503,9 +2466,6 @@ static const struct net_device_ops mlx4_netdev_ops = {
 	.ndo_setup_tc		= mlx4_en_setup_tc,
 #ifdef CONFIG_RFS_ACCEL
 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
-#endif
-#ifdef CONFIG_NET_RX_BUSY_POLL
-	.ndo_busy_poll		= mlx4_en_low_latency_recv,
 #endif
 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
 #ifdef CONFIG_MLX4_EN_VXLAN

drivers/net/ethernet/mellanox/mlx4/en_rx.c

@@ -873,10 +873,8 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 		 * - TCP/IP (v4)
 		 * - without IP options
 		 * - not an IP fragment
-		 * - no LLS polling in progress
 		 */
-		if (!mlx4_en_cq_busy_polling(cq) &&
-		    (dev->features & NETIF_F_GRO)) {
+		if (dev->features & NETIF_F_GRO) {
 			struct sk_buff *gro_skb = napi_get_frags(&cq->napi);
 			if (!gro_skb)
 				goto next;
@@ -927,7 +925,6 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 						PKT_HASH_TYPE_L3);
 
 			skb_record_rx_queue(gro_skb, cq->ring);
-			skb_mark_napi_id(gro_skb, &cq->napi);
 
 			if (ring->hwtstamp_rx_filter == HWTSTAMP_FILTER_ALL) {
 				timestamp = mlx4_en_get_cqe_ts(cqe);
@@ -990,13 +987,7 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 					       timestamp);
 		}
 
-		skb_mark_napi_id(skb, &cq->napi);
-
-		if (!mlx4_en_cq_busy_polling(cq))
-			napi_gro_receive(&cq->napi, skb);
-		else
-			netif_receive_skb(skb);
-
+		napi_gro_receive(&cq->napi, skb);
 next:
 		for (nr = 0; nr < priv->num_frags; nr++)
 			mlx4_en_free_frag(priv, frags, nr);
@@ -1038,13 +1029,8 @@ int mlx4_en_poll_rx_cq(struct napi_struct *napi, int budget)
 	struct mlx4_en_priv *priv = netdev_priv(dev);
 	int done;
 
-	if (!mlx4_en_cq_lock_napi(cq))
-		return budget;
-
 	done = mlx4_en_process_rx_cq(dev, cq, budget);
 
-	mlx4_en_cq_unlock_napi(cq);
-
 	/* If we used up all the quota - we're probably not done yet... */
 	if (done == budget) {
 		const struct cpumask *aff;

drivers/net/ethernet/mellanox/mlx4/mlx4_en.h

@@ -320,11 +320,6 @@ struct mlx4_en_rx_ring {
 	void *rx_info;
 	unsigned long bytes;
 	unsigned long packets;
-#ifdef CONFIG_NET_RX_BUSY_POLL
-	unsigned long yields;
-	unsigned long misses;
-	unsigned long cleaned;
-#endif
 	unsigned long csum_ok;
 	unsigned long csum_none;
 	unsigned long csum_complete;
@@ -347,18 +342,6 @@ struct mlx4_en_cq {
 	struct mlx4_cqe *buf;
 #define MLX4_EN_OPCODE_ERROR	0x1e
 
-#ifdef CONFIG_NET_RX_BUSY_POLL
-	unsigned int state;
-#define MLX4_EN_CQ_STATE_IDLE        0
-#define MLX4_EN_CQ_STATE_NAPI     1    /* NAPI owns this CQ */
-#define MLX4_EN_CQ_STATE_POLL     2    /* poll owns this CQ */
-#define MLX4_CQ_LOCKED (MLX4_EN_CQ_STATE_NAPI | MLX4_EN_CQ_STATE_POLL)
-#define MLX4_EN_CQ_STATE_NAPI_YIELD  4    /* NAPI yielded this CQ */
-#define MLX4_EN_CQ_STATE_POLL_YIELD  8    /* poll yielded this CQ */
-#define CQ_YIELD (MLX4_EN_CQ_STATE_NAPI_YIELD | MLX4_EN_CQ_STATE_POLL_YIELD)
-#define CQ_USER_PEND (MLX4_EN_CQ_STATE_POLL | MLX4_EN_CQ_STATE_POLL_YIELD)
-	spinlock_t poll_lock; /* protects from LLS/napi conflicts */
-#endif  /* CONFIG_NET_RX_BUSY_POLL */
 	struct irq_desc *irq_desc;
 };
 
@@ -622,115 +605,6 @@ static inline struct mlx4_cqe *mlx4_en_get_cqe(void *buf, int idx, int cqe_sz)
 	return buf + idx * cqe_sz;
 }
 
-#ifdef CONFIG_NET_RX_BUSY_POLL
-static inline void mlx4_en_cq_init_lock(struct mlx4_en_cq *cq)
-{
-	spin_lock_init(&cq->poll_lock);
-	cq->state = MLX4_EN_CQ_STATE_IDLE;
-}
-
-/* called from the device poll rutine to get ownership of a cq */
-static inline bool mlx4_en_cq_lock_napi(struct mlx4_en_cq *cq)
-{
-	int rc = true;
-	spin_lock(&cq->poll_lock);
-	if (cq->state & MLX4_CQ_LOCKED) {
-		WARN_ON(cq->state & MLX4_EN_CQ_STATE_NAPI);
-		cq->state |= MLX4_EN_CQ_STATE_NAPI_YIELD;
-		rc = false;
-	} else
-		/* we don't care if someone yielded */
-		cq->state = MLX4_EN_CQ_STATE_NAPI;
-	spin_unlock(&cq->poll_lock);
-	return rc;
-}
-
-/* returns true is someone tried to get the cq while napi had it */
-static inline bool mlx4_en_cq_unlock_napi(struct mlx4_en_cq *cq)
-{
-	int rc = false;
-	spin_lock(&cq->poll_lock);
-	WARN_ON(cq->state & (MLX4_EN_CQ_STATE_POLL |
-			       MLX4_EN_CQ_STATE_NAPI_YIELD));
-
-	if (cq->state & MLX4_EN_CQ_STATE_POLL_YIELD)
-		rc = true;
-	cq->state = MLX4_EN_CQ_STATE_IDLE;
-	spin_unlock(&cq->poll_lock);
-	return rc;
-}
-
-/* called from mlx4_en_low_latency_poll() */
-static inline bool mlx4_en_cq_lock_poll(struct mlx4_en_cq *cq)
-{
-	int rc = true;
-	spin_lock_bh(&cq->poll_lock);
-	if ((cq->state & MLX4_CQ_LOCKED)) {
-		struct net_device *dev = cq->dev;
-		struct mlx4_en_priv *priv = netdev_priv(dev);
-		struct mlx4_en_rx_ring *rx_ring = priv->rx_ring[cq->ring];
-
-		cq->state |= MLX4_EN_CQ_STATE_POLL_YIELD;
-		rc = false;
-		rx_ring->yields++;
-	} else
-		/* preserve yield marks */
-		cq->state |= MLX4_EN_CQ_STATE_POLL;
-	spin_unlock_bh(&cq->poll_lock);
-	return rc;
-}
-
-/* returns true if someone tried to get the cq while it was locked */
-static inline bool mlx4_en_cq_unlock_poll(struct mlx4_en_cq *cq)
-{
-	int rc = false;
-	spin_lock_bh(&cq->poll_lock);
-	WARN_ON(cq->state & (MLX4_EN_CQ_STATE_NAPI));
-
-	if (cq->state & MLX4_EN_CQ_STATE_POLL_YIELD)
-		rc = true;
-	cq->state = MLX4_EN_CQ_STATE_IDLE;
-	spin_unlock_bh(&cq->poll_lock);
-	return rc;
-}
-
-/* true if a socket is polling, even if it did not get the lock */
-static inline bool mlx4_en_cq_busy_polling(struct mlx4_en_cq *cq)
-{
-	WARN_ON(!(cq->state & MLX4_CQ_LOCKED));
-	return cq->state & CQ_USER_PEND;
-}
-#else
-static inline void mlx4_en_cq_init_lock(struct mlx4_en_cq *cq)
-{
-}
-
-static inline bool mlx4_en_cq_lock_napi(struct mlx4_en_cq *cq)
-{
-	return true;
-}
-
-static inline bool mlx4_en_cq_unlock_napi(struct mlx4_en_cq *cq)
-{
-	return false;
-}
-
-static inline bool mlx4_en_cq_lock_poll(struct mlx4_en_cq *cq)
-{
-	return false;
-}
-
-static inline bool mlx4_en_cq_unlock_poll(struct mlx4_en_cq *cq)
-{
-	return false;
-}
-
-static inline bool mlx4_en_cq_busy_polling(struct mlx4_en_cq *cq)
-{
-	return false;
-}
-#endif /* CONFIG_NET_RX_BUSY_POLL */
-
 #define MLX4_EN_WOL_DO_MODIFY (1ULL << 63)
 
 void mlx4_en_update_loopback_state(struct net_device *dev,

drivers/net/ethernet/mellanox/mlx5/core/en.h

@@ -564,7 +564,7 @@ void mlx5e_completion_event(struct mlx5_core_cq *mcq);
 void mlx5e_cq_error_event(struct mlx5_core_cq *mcq, enum mlx5_event event);
 int mlx5e_napi_poll(struct napi_struct *napi, int budget);
 bool mlx5e_poll_tx_cq(struct mlx5e_cq *cq);
-bool mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget);
+int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget);
 bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq);
 struct mlx5_cqe64 *mlx5e_get_cqe(struct mlx5e_cq *cq);
 

drivers/net/ethernet/mellanox/mlx5/core/en_main.c

@@ -1020,6 +1020,7 @@ err_close_tx_cqs:
 
 err_napi_del:
 	netif_napi_del(&c->napi);
+	napi_hash_del(&c->napi);
 	kfree(c);
 
 	return err;
@@ -1033,6 +1034,10 @@ static void mlx5e_close_channel(struct mlx5e_channel *c)
 	mlx5e_close_cq(&c->rq.cq);
 	mlx5e_close_tx_cqs(c);
 	netif_napi_del(&c->napi);
+
+	napi_hash_del(&c->napi);
+	synchronize_rcu();
+
 	kfree(c);
 }
 

drivers/net/ethernet/mellanox/mlx5/core/en_rx.c

@@ -33,6 +33,7 @@
 #include <linux/ip.h>
 #include <linux/ipv6.h>
 #include <linux/tcp.h>
+#include <net/busy_poll.h>
 #include "en.h"
 
 static inline int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq,
@@ -215,16 +216,16 @@ static inline void mlx5e_build_rx_skb(struct mlx5_cqe64 *cqe,
 				       be16_to_cpu(cqe->vlan_info));
 }
 
-bool mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)
+int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)
 {
 	struct mlx5e_rq *rq = container_of(cq, struct mlx5e_rq, cq);
-	int i;
+	int work_done;
 
 	/* avoid accessing cq (dma coherent memory) if not needed */
 	if (!test_and_clear_bit(MLX5E_CQ_HAS_CQES, &cq->flags))
-		return false;
+		return 0;
 
-	for (i = 0; i < budget; i++) {
+	for (work_done = 0; work_done < budget; work_done++) {
 		struct mlx5e_rx_wqe *wqe;
 		struct mlx5_cqe64 *cqe;
 		struct sk_buff *skb;
@@ -269,10 +270,8 @@ wq_ll_pop:
 	/* ensure cq space is freed before enabling more cqes */
 	wmb();
 
-	if (i == budget) {
+	if (work_done == budget)
 		set_bit(MLX5E_CQ_HAS_CQES, &cq->flags);
-		return true;
-	}
 
-	return false;
+	return work_done;
 }

drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c

@@ -54,6 +54,7 @@ int mlx5e_napi_poll(struct napi_struct *napi, int budget)
 	struct mlx5e_channel *c = container_of(napi, struct mlx5e_channel,
 					       napi);
 	bool busy = false;
+	int work_done;
 	int i;
 
 	clear_bit(MLX5E_CHANNEL_NAPI_SCHED, &c->flags);
@@ -61,26 +62,26 @@ int mlx5e_napi_poll(struct napi_struct *napi, int budget)
 	for (i = 0; i < c->num_tc; i++)
 		busy |= mlx5e_poll_tx_cq(&c->sq[i].cq);
 
-	busy |= mlx5e_poll_rx_cq(&c->rq.cq, budget);
-
+	work_done = mlx5e_poll_rx_cq(&c->rq.cq, budget);
+	busy |= work_done == budget;
 	busy |= mlx5e_post_rx_wqes(&c->rq);
 
 	if (busy)
 		return budget;
 
-	napi_complete(napi);
+	napi_complete_done(napi, work_done);
 
 	/* avoid losing completion event during/after polling cqs */
 	if (test_bit(MLX5E_CHANNEL_NAPI_SCHED, &c->flags)) {
 		napi_schedule(napi);
-		return 0;
+		return work_done;
 	}
 
 	for (i = 0; i < c->num_tc; i++)
 		mlx5e_cq_arm(&c->sq[i].cq);
 	mlx5e_cq_arm(&c->rq.cq);
 
-	return 0;
+	return work_done;
 }
 
 void mlx5e_completion_event(struct mlx5_core_cq *mcq)

drivers/net/ethernet/myricom/myri10ge/myri10ge.c

@@ -1488,7 +1488,6 @@ myri10ge_rx_done(struct myri10ge_slice_state *ss, int len, __wsum csum)
 	}
 	myri10ge_vlan_rx(mgp->dev, va, skb);
 	skb_record_rx_queue(skb, ss - &mgp->ss[0]);
-	skb_mark_napi_id(skb, &ss->napi);
 
 	if (polling) {
 		int hlen;
@@ -1506,6 +1505,7 @@ myri10ge_rx_done(struct myri10ge_slice_state *ss, int len, __wsum csum)
 		skb->data_len -= hlen;
 		skb->tail += hlen;
 		skb->protocol = eth_type_trans(skb, dev);
+		skb_mark_napi_id(skb, &ss->napi);
 		netif_receive_skb(skb);
 	}
 	else
@@ -3814,7 +3814,6 @@ static int myri10ge_alloc_slices(struct myri10ge_priv *mgp)
 		ss->dev = mgp->dev;
 		netif_napi_add(ss->dev, &ss->napi, myri10ge_poll,
 			       myri10ge_napi_weight);
-		napi_hash_add(&ss->napi);
 	}
 	return 0;
 abort:

drivers/net/ethernet/qlogic/qlcnic/qlcnic_io.c

@@ -1604,7 +1604,7 @@ int qlcnic_82xx_napi_add(struct qlcnic_adapter *adapter,
 	if (qlcnic_check_multi_tx(adapter) && !adapter->ahw->diag_test) {
 		for (ring = 0; ring < adapter->drv_tx_rings; ring++) {
 			tx_ring = &adapter->tx_ring[ring];
-			netif_napi_add(netdev, &tx_ring->napi, qlcnic_tx_poll,
+			netif_tx_napi_add(netdev, &tx_ring->napi, qlcnic_tx_poll,
 				       NAPI_POLL_WEIGHT);
 		}
 	}
@@ -2135,7 +2135,7 @@ int qlcnic_83xx_napi_add(struct qlcnic_adapter *adapter,
 	    !(adapter->flags & QLCNIC_TX_INTR_SHARED)) {
 		for (ring = 0; ring < adapter->drv_tx_rings; ring++) {
 			tx_ring = &adapter->tx_ring[ring];
-			netif_napi_add(netdev, &tx_ring->napi,
+			netif_tx_napi_add(netdev, &tx_ring->napi,
 				       qlcnic_83xx_msix_tx_poll,
 				       NAPI_POLL_WEIGHT);
 		}

drivers/net/ethernet/rocker/rocker.c

@@ -4998,7 +4998,7 @@ static int rocker_probe_port(struct rocker *rocker, unsigned int port_number)
 	dev->netdev_ops = &rocker_port_netdev_ops;
 	dev->ethtool_ops = &rocker_port_ethtool_ops;
 	dev->switchdev_ops = &rocker_port_switchdev_ops;
-	netif_napi_add(dev, &rocker_port->napi_tx, rocker_port_poll_tx,
+	netif_tx_napi_add(dev, &rocker_port->napi_tx, rocker_port_poll_tx,
 		       NAPI_POLL_WEIGHT);
 	netif_napi_add(dev, &rocker_port->napi_rx, rocker_port_poll_rx,
 		       NAPI_POLL_WEIGHT);

drivers/net/ethernet/sfc/efx.c

@@ -2059,7 +2059,6 @@ static void efx_init_napi_channel(struct efx_channel *channel)
 	channel->napi_dev = efx->net_dev;
 	netif_napi_add(channel->napi_dev, &channel->napi_str,
 		       efx_poll, napi_weight);
-	napi_hash_add(&channel->napi_str);
 	efx_channel_busy_poll_init(channel);
 }
 

drivers/net/ethernet/sfc/rx.c

@@ -463,7 +463,6 @@ efx_rx_packet_gro(struct efx_channel *channel, struct efx_rx_buffer *rx_buf,
 
 	skb_record_rx_queue(skb, channel->rx_queue.core_index);
 
-	skb_mark_napi_id(skb, &channel->napi_str);
 	gro_result = napi_gro_frags(napi);
 	if (gro_result != GRO_DROP)
 		channel->irq_mod_score += 2;

drivers/net/ethernet/ti/cpsw.c

@@ -2469,7 +2469,7 @@ static int cpsw_probe(struct platform_device *pdev)
 	ndev->netdev_ops = &cpsw_netdev_ops;
 	ndev->ethtool_ops = &cpsw_ethtool_ops;
 	netif_napi_add(ndev, &priv->napi_rx, cpsw_rx_poll, CPSW_POLL_WEIGHT);
-	netif_napi_add(ndev, &priv->napi_tx, cpsw_tx_poll, CPSW_POLL_WEIGHT);
+	netif_tx_napi_add(ndev, &priv->napi_tx, cpsw_tx_poll, CPSW_POLL_WEIGHT);
 
 	/* register the network device */
 	SET_NETDEV_DEV(ndev, &pdev->dev);

drivers/net/ethernet/ti/netcp_core.c

@@ -1990,7 +1990,7 @@ static int netcp_create_interface(struct netcp_device *netcp_device,
 
 	/* NAPI register */
 	netif_napi_add(ndev, &netcp->rx_napi, netcp_rx_poll, NETCP_NAPI_WEIGHT);
-	netif_napi_add(ndev, &netcp->tx_napi, netcp_tx_poll, NETCP_NAPI_WEIGHT);
+	netif_tx_napi_add(ndev, &netcp->tx_napi, netcp_tx_poll, NETCP_NAPI_WEIGHT);
 
 	/* Register the network device */
 	ndev->dev_id		= 0;

drivers/net/virtio_net.c

@@ -516,8 +516,6 @@ static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
 		skb_shinfo(skb)->gso_segs = 0;
 	}
 
-	skb_mark_napi_id(skb, &rq->napi);
-
 	napi_gro_receive(&rq->napi, skb);
 	return;
 
@@ -1612,7 +1610,6 @@ static int virtnet_alloc_queues(struct virtnet_info *vi)
 		vi->rq[i].pages = NULL;
 		netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
 			       napi_weight);
-		napi_hash_add(&vi->rq[i].napi);
 
 		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
 		ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);

drivers/net/wireless/ath/wil6210/netdev.c

@@ -183,7 +183,7 @@ void *wil_if_alloc(struct device *dev)
 
 	netif_napi_add(ndev, &wil->napi_rx, wil6210_netdev_poll_rx,
 		       WIL6210_NAPI_BUDGET);
-	netif_napi_add(ndev, &wil->napi_tx, wil6210_netdev_poll_tx,
+	netif_tx_napi_add(ndev, &wil->napi_tx, wil6210_netdev_poll_tx,
 		       WIL6210_NAPI_BUDGET);
 
 	netif_tx_stop_all_queues(ndev);

include/linux/hashtable.h

@@ -16,6 +16,10 @@
 	struct hlist_head name[1 << (bits)] =					\
 			{ [0 ... ((1 << (bits)) - 1)] = HLIST_HEAD_INIT }
 
+#define DEFINE_READ_MOSTLY_HASHTABLE(name, bits)				\
+	struct hlist_head name[1 << (bits)] __read_mostly =			\
+			{ [0 ... ((1 << (bits)) - 1)] = HLIST_HEAD_INIT }
+
 #define DECLARE_HASHTABLE(name, bits)                                   	\
 	struct hlist_head name[1 << (bits)]
 

include/linux/netdevice.h

@@ -326,7 +326,8 @@ enum {
 	NAPI_STATE_SCHED,	/* Poll is scheduled */
 	NAPI_STATE_DISABLE,	/* Disable pending */
 	NAPI_STATE_NPSVC,	/* Netpoll - don't dequeue from poll_list */
-	NAPI_STATE_HASHED,	/* In NAPI hash */
+	NAPI_STATE_HASHED,	/* In NAPI hash (busy polling possible) */
+	NAPI_STATE_NO_BUSY_POLL,/* Do not add in napi_hash, no busy polling */
 };
 
 enum gro_result {
@@ -460,20 +461,14 @@ static inline void napi_complete(struct napi_struct *n)
 	return napi_complete_done(n, 0);
 }
 
-/**
- *	napi_by_id - lookup a NAPI by napi_id
- *	@napi_id: hashed napi_id
- *
- * lookup @napi_id in napi_hash table
- * must be called under rcu_read_lock()
- */
-struct napi_struct *napi_by_id(unsigned int napi_id);
-
 /**
  *	napi_hash_add - add a NAPI to global hashtable
  *	@napi: napi context
  *
  * generate a new napi_id and store a @napi under it in napi_hash
+ * Used for busy polling (CONFIG_NET_RX_BUSY_POLL)
+ * Note: This is normally automatically done from netif_napi_add(),
+ * so might disappear in a future linux version.
  */
 void napi_hash_add(struct napi_struct *napi);
 
@@ -482,9 +477,14 @@ void napi_hash_add(struct napi_struct *napi);
  *	@napi: napi context
  *
  * Warning: caller must observe rcu grace period
- * before freeing memory containing @napi
+ * before freeing memory containing @napi, if
+ * this function returns true.
+ * Note: core networking stack automatically calls it
+ * from netif_napi_del()
+ * Drivers might want to call this helper to combine all
+ * the needed rcu grace periods into a single one.
  */
-void napi_hash_del(struct napi_struct *napi);
+bool napi_hash_del(struct napi_struct *napi);
 
 /**
  *	napi_disable - prevent NAPI from scheduling
@@ -1947,6 +1947,26 @@ static inline void *netdev_priv(const struct net_device *dev)
 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
 		    int (*poll)(struct napi_struct *, int), int weight);
 
+/**
+ *	netif_tx_napi_add - initialize a napi context
+ *	@dev:  network device
+ *	@napi: napi context
+ *	@poll: polling function
+ *	@weight: default weight
+ *
+ * This variant of netif_napi_add() should be used from drivers using NAPI
+ * to exclusively poll a TX queue.
+ * This will avoid we add it into napi_hash[], thus polluting this hash table.
+ */
+static inline void netif_tx_napi_add(struct net_device *dev,
+				     struct napi_struct *napi,
+				     int (*poll)(struct napi_struct *, int),
+				     int weight)
+{
+	set_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state);
+	netif_napi_add(dev, napi, poll, weight);
+}
+
 /**
  *  netif_napi_del - remove a napi context
  *  @napi: napi context

include/linux/skbuff.h

@@ -1082,9 +1082,6 @@ static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
 
 static inline void skb_sender_cpu_clear(struct sk_buff *skb)
 {
-#ifdef CONFIG_XPS
-	skb->sender_cpu = 0;
-#endif
 }
 
 #ifdef NET_SKBUFF_DATA_USES_OFFSET

include/net/busy_poll.h

@@ -72,50 +72,7 @@ static inline bool busy_loop_timeout(unsigned long end_time)
 	return time_after(now, end_time);
 }
 
-/* when used in sock_poll() nonblock is known at compile time to be true
- * so the loop and end_time will be optimized out
- */
-static inline bool sk_busy_loop(struct sock *sk, int nonblock)
-{
-	unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
-	const struct net_device_ops *ops;
-	struct napi_struct *napi;
-	int rc = false;
-
-	/*
-	 * rcu read lock for napi hash
-	 * bh so we don't race with net_rx_action
-	 */
-	rcu_read_lock_bh();
-
-	napi = napi_by_id(sk->sk_napi_id);
-	if (!napi)
-		goto out;
-
-	ops = napi->dev->netdev_ops;
-	if (!ops->ndo_busy_poll)
-		goto out;
-
-	do {
-		rc = ops->ndo_busy_poll(napi);
-
-		if (rc == LL_FLUSH_FAILED)
-			break; /* permanent failure */
-
-		if (rc > 0)
-			/* local bh are disabled so it is ok to use _BH */
-			NET_ADD_STATS_BH(sock_net(sk),
-					 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
-		cpu_relax();
-
-	} while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
-		 !need_resched() && !busy_loop_timeout(end_time));
-
-	rc = !skb_queue_empty(&sk->sk_receive_queue);
-out:
-	rcu_read_unlock_bh();
-	return rc;
-}
+bool sk_busy_loop(struct sock *sk, int nonblock);
 
 /* used in the NIC receive handler to mark the skb */
 static inline void skb_mark_napi_id(struct sk_buff *skb,

net/core/dev.c

@@ -96,6 +96,7 @@
 #include <linux/skbuff.h>
 #include <net/net_namespace.h>
 #include <net/sock.h>
+#include <net/busy_poll.h>
 #include <linux/rtnetlink.h>
 #include <linux/stat.h>
 #include <net/dst.h>
@@ -182,8 +183,8 @@ EXPORT_SYMBOL(dev_base_lock);
 /* protects napi_hash addition/deletion and napi_gen_id */
 static DEFINE_SPINLOCK(napi_hash_lock);
 
-static unsigned int napi_gen_id;
-static DEFINE_HASHTABLE(napi_hash, 8);
+static unsigned int napi_gen_id = NR_CPUS;
+static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
 
 static seqcount_t devnet_rename_seq;
 
@@ -3021,7 +3022,9 @@ struct netdev_queue *netdev_pick_tx(struct net_device *dev,
 	int queue_index = 0;
 
 #ifdef CONFIG_XPS
-	if (skb->sender_cpu == 0)
+	u32 sender_cpu = skb->sender_cpu - 1;
+
+	if (sender_cpu >= (u32)NR_CPUS)
 		skb->sender_cpu = raw_smp_processor_id() + 1;
 #endif
 
@@ -4353,6 +4356,7 @@ static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
 
 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
 {
+	skb_mark_napi_id(skb, napi);
 	trace_napi_gro_receive_entry(skb);
 
 	skb_gro_reset_offset(skb);
@@ -4387,6 +4391,7 @@ struct sk_buff *napi_get_frags(struct napi_struct *napi)
 	if (!skb) {
 		skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
 		napi->skb = skb;
+		skb_mark_napi_id(skb, napi);
 	}
 	return skb;
 }
@@ -4661,7 +4666,7 @@ void napi_complete_done(struct napi_struct *n, int work_done)
 EXPORT_SYMBOL(napi_complete_done);
 
 /* must be called under rcu_read_lock(), as we dont take a reference */
-struct napi_struct *napi_by_id(unsigned int napi_id)
+static struct napi_struct *napi_by_id(unsigned int napi_id)
 {
 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
 	struct napi_struct *napi;
@@ -4672,43 +4677,101 @@ struct napi_struct *napi_by_id(unsigned int napi_id)
 
 	return NULL;
 }
-EXPORT_SYMBOL_GPL(napi_by_id);
 
-void napi_hash_add(struct napi_struct *napi)
+#if defined(CONFIG_NET_RX_BUSY_POLL)
+#define BUSY_POLL_BUDGET 8
+bool sk_busy_loop(struct sock *sk, int nonblock)
 {
-	if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
+	unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
+	int (*busy_poll)(struct napi_struct *dev);
+	struct napi_struct *napi;
+	int rc = false;
+
+	rcu_read_lock();
 
-		spin_lock(&napi_hash_lock);
+	napi = napi_by_id(sk->sk_napi_id);
+	if (!napi)
+		goto out;
 
-		/* 0 is not a valid id, we also skip an id that is taken
-		 * we expect both events to be extremely rare
-		 */
-		napi->napi_id = 0;
-		while (!napi->napi_id) {
-			napi->napi_id = ++napi_gen_id;
-			if (napi_by_id(napi->napi_id))
-				napi->napi_id = 0;
+	/* Note: ndo_busy_poll method is optional in linux-4.5 */
+	busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
+
+	do {
+		rc = 0;
+		local_bh_disable();
+		if (busy_poll) {
+			rc = busy_poll(napi);
+		} else if (napi_schedule_prep(napi)) {
+			void *have = netpoll_poll_lock(napi);
+
+			if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
+				rc = napi->poll(napi, BUSY_POLL_BUDGET);
+				trace_napi_poll(napi);
+				if (rc == BUSY_POLL_BUDGET) {
+					napi_complete_done(napi, rc);
+					napi_schedule(napi);
+				}
+			}
+			netpoll_poll_unlock(have);
 		}
+		if (rc > 0)
+			NET_ADD_STATS_BH(sock_net(sk),
+					 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
+		local_bh_enable();
 
-		hlist_add_head_rcu(&napi->napi_hash_node,
-			&napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
+		if (rc == LL_FLUSH_FAILED)
+			break; /* permanent failure */
 
-		spin_unlock(&napi_hash_lock);
-	}
+		cpu_relax();
+	} while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
+		 !need_resched() && !busy_loop_timeout(end_time));
+
+	rc = !skb_queue_empty(&sk->sk_receive_queue);
+out:
+	rcu_read_unlock();
+	return rc;
+}
+EXPORT_SYMBOL(sk_busy_loop);
+
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
+void napi_hash_add(struct napi_struct *napi)
+{
+	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
+	    test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
+		return;
+
+	spin_lock(&napi_hash_lock);
+
+	/* 0..NR_CPUS+1 range is reserved for sender_cpu use */
+	do {
+		if (unlikely(++napi_gen_id < NR_CPUS + 1))
+			napi_gen_id = NR_CPUS + 1;
+	} while (napi_by_id(napi_gen_id));
+	napi->napi_id = napi_gen_id;
+
+	hlist_add_head_rcu(&napi->napi_hash_node,
+			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
+
+	spin_unlock(&napi_hash_lock);
 }
 EXPORT_SYMBOL_GPL(napi_hash_add);
 
 /* Warning : caller is responsible to make sure rcu grace period
  * is respected before freeing memory containing @napi
  */
-void napi_hash_del(struct napi_struct *napi)
+bool napi_hash_del(struct napi_struct *napi)
 {
+	bool rcu_sync_needed = false;
+
 	spin_lock(&napi_hash_lock);
 
-	if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
+	if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
+		rcu_sync_needed = true;
 		hlist_del_rcu(&napi->napi_hash_node);
-
+	}
 	spin_unlock(&napi_hash_lock);
+	return rcu_sync_needed;
 }
 EXPORT_SYMBOL_GPL(napi_hash_del);
 
@@ -4744,6 +4807,7 @@ void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
 	napi->poll_owner = -1;
 #endif
 	set_bit(NAPI_STATE_SCHED, &napi->state);
+	napi_hash_add(napi);
 }
 EXPORT_SYMBOL(netif_napi_add);
 
@@ -4763,8 +4827,12 @@ void napi_disable(struct napi_struct *n)
 }
 EXPORT_SYMBOL(napi_disable);
 
+/* Must be called in process context */
 void netif_napi_del(struct napi_struct *napi)
 {
+	might_sleep();
+	if (napi_hash_del(napi))
+		synchronize_net();
 	list_del_init(&napi->dev_list);
 	napi_free_frags(napi);
 
@@ -7164,11 +7232,13 @@ EXPORT_SYMBOL(alloc_netdev_mqs);
  *	This function does the last stage of destroying an allocated device
  * 	interface. The reference to the device object is released.
  *	If this is the last reference then it will be freed.
+ *	Must be called in process context.
  */
 void free_netdev(struct net_device *dev)
 {
 	struct napi_struct *p, *n;
 
+	might_sleep();
 	netif_free_tx_queues(dev);
 #ifdef CONFIG_SYSFS
 	kvfree(dev->_rx);