From 52bd2d62ce6758d811edcbd2256eb9ea7f6a56cb Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:50 -0800
Subject: net: better skb->sender_cpu and skb->napi_id cohabitation

skb->sender_cpu and skb->napi_id share a common storage,
and we had various bugs about this.

We had to call skb_sender_cpu_clear() in some places to
not leave a prior skb->napi_id and fool netdev_pick_tx()

As suggested by Alexei, we could split the space so that
these errors can not happen.

0 value being reserved as the common (not initialized) value,
let's reserve [1 .. NR_CPUS] range for valid sender_cpu,
and [NR_CPUS+1 .. ~0U] for valid napi_id.

This will allow proper busy polling support over tunnels.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Suggested-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 include/linux/skbuff.h |  3 ---
 net/core/dev.c         | 33 ++++++++++++++++-----------------
 2 files changed, 16 insertions(+), 20 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 4355129fff91..c9c394bf0771 100644
--- a/include/linux/skbuff.h
+++ b/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
diff --git a/net/core/dev.c b/net/core/dev.c
index ae00b894e675..2582c24a75c6 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -182,7 +182,7 @@ 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 unsigned int napi_gen_id = NR_CPUS;
 static DEFINE_HASHTABLE(napi_hash, 8);
 
 static seqcount_t devnet_rename_seq;
@@ -3021,7 +3021,9 @@ 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
 
@@ -4676,25 +4678,22 @@ EXPORT_SYMBOL_GPL(napi_by_id);
 
 void napi_hash_add(struct napi_struct *napi)
 {
-	if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
+	if (test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
+		return;
 
-		spin_lock(&napi_hash_lock);
+	spin_lock(&napi_hash_lock);
 
-		/* 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;
-		}
+	/* 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)]);
+	hlist_add_head_rcu(&napi->napi_hash_node,
+			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
 
-		spin_unlock(&napi_hash_lock);
-	}
+	spin_unlock(&napi_hash_lock);
 }
 EXPORT_SYMBOL_GPL(napi_hash_add);
 
-- 
cgit 


From 5865316c9db31e863d4baef04dc86957d7fdae7f Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:51 -0800
Subject: mlx4: mlx4_en_low_latency_recv() called with BH disabled

mlx4_en_low_latency_recv() is called with BH disabled,
as other ndo_busy_poll() methods.

No need for spin_lock_bh()/spin_unlock_bh()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/mellanox/mlx4/mlx4_en.h | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h b/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
index c41f15102ae0..965c8f016ac4 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
@@ -660,11 +660,12 @@ static inline bool mlx4_en_cq_unlock_napi(struct mlx4_en_cq *cq)
 	return rc;
 }
 
-/* called from mlx4_en_low_latency_poll() */
+/* called from mlx4_en_low_latency_recv(), BH are disabled */
 static inline bool mlx4_en_cq_lock_poll(struct mlx4_en_cq *cq)
 {
 	int rc = true;
-	spin_lock_bh(&cq->poll_lock);
+
+	spin_lock(&cq->poll_lock);
 	if ((cq->state & MLX4_CQ_LOCKED)) {
 		struct net_device *dev = cq->dev;
 		struct mlx4_en_priv *priv = netdev_priv(dev);
@@ -676,7 +677,7 @@ static inline bool mlx4_en_cq_lock_poll(struct mlx4_en_cq *cq)
 	} else
 		/* preserve yield marks */
 		cq->state |= MLX4_EN_CQ_STATE_POLL;
-	spin_unlock_bh(&cq->poll_lock);
+	spin_unlock(&cq->poll_lock);
 	return rc;
 }
 
@@ -684,13 +685,14 @@ static inline bool mlx4_en_cq_lock_poll(struct mlx4_en_cq *cq)
 static inline bool mlx4_en_cq_unlock_poll(struct mlx4_en_cq *cq)
 {
 	int rc = false;
-	spin_lock_bh(&cq->poll_lock);
+
+	spin_lock(&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);
+	spin_unlock(&cq->poll_lock);
 	return rc;
 }
 
-- 
cgit 


From 02d62e86fe892c59a1259d089d4d16ac76977a37 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:52 -0800
Subject: net: un-inline sk_busy_loop()

There is really little gain from inlining this big function.
We'll soon make it even bigger in following patches.

This means we no longer need to export napi_by_id()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 include/linux/netdevice.h |  9 ---------
 include/net/busy_poll.h   | 45 +-----------------------------------------
 net/core/dev.c            | 50 +++++++++++++++++++++++++++++++++++++++++++++--
 3 files changed, 49 insertions(+), 55 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 67bfac1abfc1..2020a89df12b 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -460,15 +460,6 @@ 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
diff --git a/include/net/busy_poll.h b/include/net/busy_poll.h
index 1d67fb6b23a0..2fbeb1313c0f 100644
--- a/include/net/busy_poll.h
+++ b/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,
diff --git a/net/core/dev.c b/net/core/dev.c
index 2582c24a75c6..74a816b299df 100644
--- a/net/core/dev.c
+++ b/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>
@@ -4663,7 +4664,7 @@ 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;
@@ -4674,7 +4675,52 @@ struct napi_struct *napi_by_id(unsigned int napi_id)
 
 	return NULL;
 }
-EXPORT_SYMBOL_GPL(napi_by_id);
+
+#if defined(CONFIG_NET_RX_BUSY_POLL)
+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;
+}
+EXPORT_SYMBOL(sk_busy_loop);
+
+#endif /* CONFIG_NET_RX_BUSY_POLL */
 
 void napi_hash_add(struct napi_struct *napi)
 {
-- 
cgit 


From 2a028ecb76497d05e5cd4e3e8b09d965cac2e3f1 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:53 -0800
Subject: net: allow BH servicing in sk_busy_loop()

Instead of blocking BH in whole sk_busy_loop(), block them
only around ->ndo_busy_poll() calls.

This has many benefits.

1) allow tunneled traffic to use busy poll as well as native traffic.
   Tunnels handlers usually call netif_rx() and depend on net_rx_action()
   being run (from sofirq handler)

2) allow RFS/RPS being used (sending IPI to other cpus if needed)

3) use the 'lets burn cpu cycles' budget to do useful work
   (like TX completions, timers, RCU callbacks...)

4) reduce BH latencies, making busy poll a better citizen.

Tested:

Tested with SIT tunnel

lpaa5:~# echo 0 >/proc/sys/net/core/busy_read
lpaa5:~# ./netperf -H 2002:af6:786::1 -t TCP_RR
MIGRATED TCP REQUEST/RESPONSE TEST from ::0 (::) port 0 AF_INET6 to 2002:af6:786::1 () port 0 AF_INET6 : first burst 0
Local /Remote
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       10.00    37373.93
16384  87380

Now enable busy poll on both hosts

lpaa5:~# echo 70 >/proc/sys/net/core/busy_read
lpaa6:~# echo 70 >/proc/sys/net/core/busy_read

lpaa5:~# ./netperf -H 2002:af6:786::1 -t TCP_RR
MIGRATED TCP REQUEST/RESPONSE TEST from ::0 (::) port 0 AF_INET6 to 2002:af6:786::1 () port 0 AF_INET6 : first burst 0
Local /Remote
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       10.00    58314.77
16384  87380

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 net/core/dev.c | 18 +++++++-----------
 1 file changed, 7 insertions(+), 11 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 74a816b299df..2002eec2617d 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4684,11 +4684,7 @@ bool sk_busy_loop(struct sock *sk, int nonblock)
 	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();
+	rcu_read_lock();
 
 	napi = napi_by_id(sk->sk_napi_id);
 	if (!napi)
@@ -4699,23 +4695,23 @@ bool sk_busy_loop(struct sock *sk, int nonblock)
 		goto out;
 
 	do {
+		local_bh_disable();
 		rc = ops->ndo_busy_poll(napi);
+		if (rc > 0)
+			NET_ADD_STATS_BH(sock_net(sk),
+					 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
+		local_bh_enable();
 
 		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();
+	rcu_read_unlock();
 	return rc;
 }
 EXPORT_SYMBOL(sk_busy_loop);
-- 
cgit 


From ce6aea93f7510437dde625b77a7a2f4d20b72660 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:54 -0800
Subject: net: network drivers no longer need to implement ndo_busy_poll()

Instead of having to implement complex ndo_busy_poll() method,
drivers can simply rely on NAPI poll logic.

Busy polling gains are mainly coming from polling itself,
not on exact details on how we poll the device.

ndo_busy_poll() if implemented can avoid touching
napi state, but it adds extra synchronization between
normal napi->poll() and busy poll handler, slowing down
the common path (non busy polling) with extra atomic operations.
In practice few drivers ever got busy poll because of the complexity.

We could go one step further, and make busy polling
available for all NAPI drivers, but this would require
that all netif_napi_del() calls are done in process context
so that we can call synchronize_rcu().
Full audit would be required.

Before this is done, a driver still needs to call :

- skb_mark_napi_id() for each skb provided to the stack.
- napi_hash_add() and napi_hash_del() to allocate a napi_id per napi struct.
- Make sure RCU grace period is respected after napi_hash_del() before
  memory containing napi structure is freed.

Followup patch implements busy poll for mlx5 driver as an example.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 net/core/dev.c | 25 ++++++++++++++++++++-----
 1 file changed, 20 insertions(+), 5 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 2002eec2617d..93009610aee8 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4677,10 +4677,11 @@ static struct napi_struct *napi_by_id(unsigned int napi_id)
 }
 
 #if defined(CONFIG_NET_RX_BUSY_POLL)
+#define BUSY_POLL_BUDGET 8
 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;
+	int (*busy_poll)(struct napi_struct *dev);
 	struct napi_struct *napi;
 	int rc = false;
 
@@ -4690,13 +4691,27 @@ bool sk_busy_loop(struct sock *sk, int nonblock)
 	if (!napi)
 		goto out;
 
-	ops = napi->dev->netdev_ops;
-	if (!ops->ndo_busy_poll)
-		goto out;
+	/* 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();
-		rc = ops->ndo_busy_poll(napi);
+		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);
-- 
cgit 


From 7ae92ae588c9f78006c106bb3398d50274c5d7de Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:55 -0800
Subject: mlx5: add busy polling support

It is now easy to add busy polling support to a NAPI driver,
with very little impact on normal input path.

This patch serves as a reference implementation.

Note:

A followup patch will add proper napi_complete_done() in mlx5,
so that LINUX_MIB_BUSYPOLLRXPACKETS snmp counter is properly handled.

Tested:

Normal TCP_RR results without busy polling :

lpk51:~# echo 0 >/proc/sys/net/core/busy_read
lpk52:~# echo 0 >/proc/sys/net/core/busy_read

lpk51:~# ./netperf -H 192.168.4.52 -t TCP_RR -l 10
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.4.52 () port 0 AF_INET : first burst 0
Local /Remote
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       10.00    53509.49
16384  87380

Now enable busy polling :

lpk51:~# echo 70 >/proc/sys/net/core/busy_read
lpk52:~# echo 70 >/proc/sys/net/core/busy_read

lpk51:~# ./netperf -H 192.168.4.52 -t TCP_RR -l 10
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.4.52 () port 0 AF_INET : first burst 0
Local /Remote
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       10.00    97530.92
16384  87380

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 6 ++++++
 drivers/net/ethernet/mellanox/mlx5/core/en_rx.c   | 2 ++
 2 files changed, 8 insertions(+)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 1e52db32c73d..ffb1f9c1b973 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -982,6 +982,7 @@ static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
 	mlx5e_build_channeltc_to_txq_map(priv, ix);
 
 	netif_napi_add(netdev, &c->napi, mlx5e_napi_poll, 64);
+	napi_hash_add(&c->napi);
 
 	err = mlx5e_open_tx_cqs(c, cparam);
 	if (err)
@@ -1020,6 +1021,7 @@ err_close_tx_cqs:
 
 err_napi_del:
 	netif_napi_del(&c->napi);
+	napi_hash_del(&c->napi);
 	kfree(c);
 
 	return err;
@@ -1033,6 +1035,10 @@ 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);
 }
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index cf0098596e85..54800c61a563 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/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,
@@ -242,6 +243,7 @@ bool mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)
 		wqe            = mlx5_wq_ll_get_wqe(&rq->wq, wqe_counter);
 		skb            = rq->skb[wqe_counter];
 		prefetch(skb->data);
+		skb_mark_napi_id(skb, cq->napi);
 		rq->skb[wqe_counter] = NULL;
 
 		dma_unmap_single(rq->pdev,
-- 
cgit 


From 44fb6fbbac23ee304d113ceb5eed45666d8a32f9 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:56 -0800
Subject: mlx5: support napi_complete_done()

A NAPI poll handler should return number of RX packets processed,
instead of 0 / budget.

This allows proper busy poll accounting through LINUX_MIB_BUSYPOLLRXPACKETS
SNMP counter.

napi_complete_done() allows /sys/class/net/ethX/gro_flush_timeout
to be used for finer GRO aggregation control.

Tested:

Enabled busy polling, and checked TcpExtBusyPollRxPackets counter is increasing.

echo 70 >/proc/sys/net/core/busy_read
nstat >/dev/null
netperf -H target -t TCP_RR >/dev/null
nstat | grep TcpExtBusyPollRxPackets
TcpExtBusyPollRxPackets         490958             0.0

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Eli Cohen <eli@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h      |  2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_rx.c   | 14 ++++++--------
 drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 11 ++++++-----
 3 files changed, 13 insertions(+), 14 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 22e72bf1ae48..69f1c1a412b4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/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);
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 54800c61a563..fe752f8e24b9 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -216,16 +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;
@@ -271,10 +271,8 @@ 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;
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
index 2c7cb6755d1d..4ac8d716dbdd 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
+++ b/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)
-- 
cgit 


From b59768c6b4fc1c95b7679f2cfa9c13d8cc70ab46 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:57 -0800
Subject: bnx2x: remove bnx2x_low_latency_recv() support

Switch to native NAPI polling, as this reduces overhead and complexity.

Normal path is faster, since one cmpxchg() is not anymore requested,
and busy polling with the NAPI polling has same performance.

Tested:
lpk50:~# cat /proc/sys/net/core/busy_read
70
lpk50:~# nstat >/dev/null;./netperf -H lpk55 -t TCP_RR;nstat
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to lpk55.prod.google.com () port 0 AF_INET : first burst 0
Local /Remote
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       10.00    40095.07
16384  87380
IpInReceives                    401062             0.0
IpInDelivers                    401062             0.0
IpOutRequests                   401079             0.0
TcpActiveOpens                  7                  0.0
TcpPassiveOpens                 3                  0.0
TcpAttemptFails                 3                  0.0
TcpEstabResets                  5                  0.0
TcpInSegs                       401036             0.0
TcpOutSegs                      401052             0.0
TcpOutRsts                      38                 0.0
UdpInDatagrams                  26                 0.0
UdpOutDatagrams                 27                 0.0
Ip6OutNoRoutes                  1                  0.0
TcpExtDelayedACKs               1                  0.0
TcpExtTCPPrequeued              98                 0.0
TcpExtTCPDirectCopyFromPrequeue 98                 0.0
TcpExtTCPHPHits                 4                  0.0
TcpExtTCPHPHitsToUser           98                 0.0
TcpExtTCPPureAcks               5                  0.0
TcpExtTCPHPAcks                 101                0.0
TcpExtTCPAbortOnData            6                  0.0
TcpExtBusyPollRxPackets         400832             0.0
TcpExtTCPOrigDataSent           400983             0.0
IpExtInOctets                   21273867           0.0
IpExtOutOctets                  21261254           0.0
IpExtInNoECTPkts                401064             0.0

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bnx2x/bnx2x.h      | 113 -----------------------
 drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c  |  46 +--------
 drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h  |   7 --
 drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c |   3 -
 4 files changed, 2 insertions(+), 167 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x.h b/drivers/net/ethernet/broadcom/bnx2x/bnx2x.h
index b5e64b02200c..0b214b5d944a 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x.h
+++ b/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
 
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
index f8d7a2f06950..ca208a7eecd5 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
@@ -1096,10 +1096,7 @@ reuse_rx:
 
 		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 +1866,6 @@ 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 +1875,6 @@ 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 +1885,6 @@ 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 +1894,6 @@ 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 +3223,6 @@ 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 +3231,10 @@ 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 +3278,6 @@ 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
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h
index b7d32e8412f1..4cbb03f87b5a 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h
+++ b/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
  *
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
index c9b036789184..2273576404b4 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
+++ b/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,
-- 
cgit 


From 868fdb06063d92f4373be4974d1b651a32bb9442 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:58 -0800
Subject: mlx4: remove mlx4_en_low_latency_recv()

Busy polling can now be handled in generic NAPI poll infrastructure.
This removes complexity and fast path overhead :

mlx4 used two spin_lock()/spin_unlock() pair per napi->poll() call
in mlx4_en_cq_lock_napi()/mlx4_en_cq_unlock_napi()

Tested:

Without busy polling :

lpaa23:~# echo 0 >/proc/sys/net/core/busy_read
lpaa24:~# echo 0 >/proc/sys/net/core/busy_read
lpaa23:~# ./netperf -H lpaa24 -t TCP_RR
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to lpaa24.prod.google.com () port 0 AF_INET : first burst 0
Local /Remote
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       10.00    47330.78

With busy polling :

lpaa23:~# echo 70 >/proc/sys/net/core/busy_read
lpaa24:~# echo 70 >/proc/sys/net/core/busy_read
lpaa23:~# ./netperf -H lpaa24 -t TCP_RR
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to lpaa24.prod.google.com () port 0 AF_INET : first burst 0
Local /Remote
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       10.00    97643.55

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/mellanox/mlx4/en_ethtool.c |  17 ----
 drivers/net/ethernet/mellanox/mlx4/en_netdev.c  |  40 --------
 drivers/net/ethernet/mellanox/mlx4/en_rx.c      |  15 +--
 drivers/net/ethernet/mellanox/mlx4/mlx4_en.h    | 128 ------------------------
 4 files changed, 2 insertions(+), 198 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
index ddb5541882f5..dd84cabb2a51 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
+++ b/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:
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
index 886e1bc86374..659209ff7af6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
+++ b/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
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
index e7a5000aa12c..1feead34093b 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
+++ b/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;
@@ -992,11 +990,7 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 
 		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 +1032,8 @@ 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;
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h b/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
index 965c8f016ac4..35de7d2e6b34 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
+++ b/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;
 };
 
@@ -622,117 +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_recv(), BH are disabled */
-static inline bool mlx4_en_cq_lock_poll(struct mlx4_en_cq *cq)
-{
-	int rc = true;
-
-	spin_lock(&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(&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(&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(&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,
-- 
cgit 


From 93f93a4404159ecf7e9148f5ad0718ec702ac4cb Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:30:59 -0800
Subject: net: move skb_mark_napi_id() into core networking stack

We would like to automatically provide busy polling support
to all NAPI drivers, without them having to implement anything.

skb_mark_napi_id() can be called from napi_gro_receive() and
napi_get_frags().

Few drivers are still calling skb_mark_napi_id() because
they use netif_receive_skb(). They should eventually call
napi_gro_receive() instead. I will leave this to drivers
maintainers.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/amd/xgbe/xgbe-drv.c         | 1 -
 drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c  | 2 --
 drivers/net/ethernet/chelsio/cxgb4/sge.c         | 1 -
 drivers/net/ethernet/emulex/benet/be_main.c      | 1 -
 drivers/net/ethernet/intel/i40e/i40e_txrx.c      | 1 -
 drivers/net/ethernet/intel/i40evf/i40e_txrx.c    | 1 -
 drivers/net/ethernet/intel/ixgbe/ixgbe_main.c    | 2 +-
 drivers/net/ethernet/mellanox/mlx4/en_rx.c       | 3 ---
 drivers/net/ethernet/mellanox/mlx5/core/en_rx.c  | 1 -
 drivers/net/ethernet/myricom/myri10ge/myri10ge.c | 2 +-
 drivers/net/ethernet/sfc/rx.c                    | 1 -
 drivers/net/virtio_net.c                         | 2 --
 net/core/dev.c                                   | 2 ++
 13 files changed, 4 insertions(+), 16 deletions(-)

diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-drv.c b/drivers/net/ethernet/amd/xgbe/xgbe-drv.c
index 53ce1222b11d..8a9b493566c9 100644
--- a/drivers/net/ethernet/amd/xgbe/xgbe-drv.c
+++ b/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);
 
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
index ca208a7eecd5..ab9222924bd9 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
@@ -1094,8 +1094,6 @@ 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);
-
 		napi_gro_receive(&fp->napi, skb);
 next_rx:
 		rx_buf->data = NULL;
diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c
index b7b93e7a643d..f650f295f264 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/sge.c
+++ b/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),
diff --git a/drivers/net/ethernet/emulex/benet/be_main.c b/drivers/net/ethernet/emulex/benet/be_main.c
index b6ad02909d6b..c29d62496ad9 100644
--- a/drivers/net/ethernet/emulex/benet/be_main.c
+++ b/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);
diff --git a/drivers/net/ethernet/intel/i40e/i40e_txrx.c b/drivers/net/ethernet/intel/i40e/i40e_txrx.c
index 635b3ac17877..6649ce4ba2de 100644
--- a/drivers/net/ethernet/intel/i40e/i40e_txrx.c
+++ b/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;
diff --git a/drivers/net/ethernet/intel/i40evf/i40e_txrx.c b/drivers/net/ethernet/intel/i40evf/i40e_txrx.c
index 47e9a90d6b10..77968b184b1f 100644
--- a/drivers/net/ethernet/intel/i40evf/i40e_txrx.c
+++ b/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;
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
index 47395ff5d908..4089d776d01a 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
+++ b/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 */
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
index 1feead34093b..41440b2b20a3 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
@@ -925,7 +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);
@@ -988,8 +987,6 @@ 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);
-
 		napi_gro_receive(&cq->napi, skb);
 next:
 		for (nr = 0; nr < priv->num_frags; nr++)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index fe752f8e24b9..7c8c4088d1be 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -243,7 +243,6 @@ int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)
 		wqe            = mlx5_wq_ll_get_wqe(&rq->wq, wqe_counter);
 		skb            = rq->skb[wqe_counter];
 		prefetch(skb->data);
-		skb_mark_napi_id(skb, cq->napi);
 		rq->skb[wqe_counter] = NULL;
 
 		dma_unmap_single(rq->pdev,
diff --git a/drivers/net/ethernet/myricom/myri10ge/myri10ge.c b/drivers/net/ethernet/myricom/myri10ge/myri10ge.c
index 83651ac8ddb9..acf866147d65 100644
--- a/drivers/net/ethernet/myricom/myri10ge/myri10ge.c
+++ b/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
diff --git a/drivers/net/ethernet/sfc/rx.c b/drivers/net/ethernet/sfc/rx.c
index 809ea4610a77..8956995b2fe7 100644
--- a/drivers/net/ethernet/sfc/rx.c
+++ b/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;
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index d8838dedb7a4..d1d14cecf450 100644
--- a/drivers/net/virtio_net.c
+++ b/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;
 
diff --git a/net/core/dev.c b/net/core/dev.c
index 93009610aee8..83b48747928c 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4356,6 +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);
@@ -4390,6 +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;
 }
-- 
cgit 


From d64b5e85bfe2fe4c790abcbd16d9ae32391ddd7e Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:31:00 -0800
Subject: net: add netif_tx_napi_add()

netif_tx_napi_add() is a variant of netif_napi_add()

It should be used by drivers that use a napi structure
to exclusively poll TX.

We do not want to add this kind of napi in napi_hash[] in following
patches, adding generic busy polling to all NAPI drivers.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bcmsysport.c         |  2 +-
 drivers/net/ethernet/broadcom/genet/bcmgenet.c     |  4 ++--
 .../net/ethernet/freescale/fs_enet/fs_enet-main.c  |  2 +-
 drivers/net/ethernet/freescale/gianfar.c           |  4 ++--
 drivers/net/ethernet/mellanox/mlx4/en_cq.c         |  4 ++--
 drivers/net/ethernet/qlogic/qlcnic/qlcnic_io.c     |  4 ++--
 drivers/net/ethernet/rocker/rocker.c               |  2 +-
 drivers/net/ethernet/ti/cpsw.c                     |  2 +-
 drivers/net/ethernet/ti/netcp_core.c               |  2 +-
 drivers/net/wireless/ath/wil6210/netdev.c          |  2 +-
 include/linux/netdevice.h                          | 23 +++++++++++++++++++++-
 net/core/dev.c                                     |  3 ++-
 12 files changed, 38 insertions(+), 16 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
index 858106352ce9..993c780bdfab 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.c
+++ b/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;
diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.c b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
index 17f017ab4dac..b15a60d787c7 100644
--- a/drivers/net/ethernet/broadcom/genet/bcmgenet.c
+++ b/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)
diff --git a/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c b/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c
index cf8e54652df9..48a9c176e0d1 100644
--- a/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c
+++ b/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;
 
diff --git a/drivers/net/ethernet/freescale/gianfar.c b/drivers/net/ethernet/freescale/gianfar.c
index 3e6b9b437497..c8bc43e99a35 100644
--- a/drivers/net/ethernet/freescale/gianfar.c
+++ b/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);
 		}
 	}
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_cq.c b/drivers/net/ethernet/mellanox/mlx4/en_cq.c
index eb8a4988de63..3a6176fea78d 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_cq.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_cq.c
@@ -156,8 +156,8 @@ int mlx4_en_activate_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq,
 	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);
+		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);
diff --git a/drivers/net/ethernet/qlogic/qlcnic/qlcnic_io.c b/drivers/net/ethernet/qlogic/qlcnic/qlcnic_io.c
index d4b5085a21fa..7bd6f25b4625 100644
--- a/drivers/net/ethernet/qlogic/qlcnic/qlcnic_io.c
+++ b/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);
 		}
diff --git a/drivers/net/ethernet/rocker/rocker.c b/drivers/net/ethernet/rocker/rocker.c
index e9f2349e98bc..a4ab71d43e4e 100644
--- a/drivers/net/ethernet/rocker/rocker.c
+++ b/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);
diff --git a/drivers/net/ethernet/ti/cpsw.c b/drivers/net/ethernet/ti/cpsw.c
index 48b92c9de12a..15322c08de80 100644
--- a/drivers/net/ethernet/ti/cpsw.c
+++ b/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);
diff --git a/drivers/net/ethernet/ti/netcp_core.c b/drivers/net/ethernet/ti/netcp_core.c
index 37b9b39192ec..e5e20e734f21 100644
--- a/drivers/net/ethernet/ti/netcp_core.c
+++ b/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;
diff --git a/drivers/net/wireless/ath/wil6210/netdev.c b/drivers/net/wireless/ath/wil6210/netdev.c
index e3b3c8fb4605..56aaa2d4fb0e 100644
--- a/drivers/net/wireless/ath/wil6210/netdev.c
+++ b/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);
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 2020a89df12b..838935d1cdbb 100644
--- a/include/linux/netdevice.h
+++ b/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 {
@@ -1938,6 +1939,26 @@ 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
diff --git a/net/core/dev.c b/net/core/dev.c
index 83b48747928c..ff58a8bc5e3c 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4737,7 +4737,8 @@ EXPORT_SYMBOL(sk_busy_loop);
 
 void napi_hash_add(struct napi_struct *napi)
 {
-	if (test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
+	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);
-- 
cgit 


From 6180d9de61a5c461f9e3efef5417a844701dbbb2 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:31:01 -0800
Subject: net: move napi_hash[] into read mostly section

We do not often add/delete a napi context.
Moving napi_hash[] into read mostly section avoids potential false sharing.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 include/linux/hashtable.h | 4 ++++
 net/core/dev.c            | 2 +-
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/include/linux/hashtable.h b/include/linux/hashtable.h
index 519b6e2d769e..661e5c2a8e2a 100644
--- a/include/linux/hashtable.h
+++ b/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)]
 
diff --git a/net/core/dev.c b/net/core/dev.c
index ff58a8bc5e3c..02dfbd91a8e4 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -184,7 +184,7 @@ EXPORT_SYMBOL(dev_base_lock);
 static DEFINE_SPINLOCK(napi_hash_lock);
 
 static unsigned int napi_gen_id = NR_CPUS;
-static DEFINE_HASHTABLE(napi_hash, 8);
+static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
 
 static seqcount_t devnet_rename_seq;
 
-- 
cgit 


From 34cbe27e811c591c854a39c0dee1b461bb796953 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:31:02 -0800
Subject: net: napi_hash_del() returns a boolean status

napi_hash_del() will soon be used from both drivers (if they want)
or core networking stack.

Callers are responsibles to ensure an RCU grace period is respected
before freeing napi structure : napi_hash_del() can signal if
this RCU grace period is needed or not.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 include/linux/netdevice.h |  5 +++--
 net/core/dev.c            | 10 +++++++---
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 838935d1cdbb..e5c33b29471b 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -474,9 +474,10 @@ 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.
  */
-void napi_hash_del(struct napi_struct *napi);
+bool napi_hash_del(struct napi_struct *napi);
 
 /**
  *	napi_disable - prevent NAPI from scheduling
diff --git a/net/core/dev.c b/net/core/dev.c
index 02dfbd91a8e4..59dddac1c2e7 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4760,14 +4760,18 @@ 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);
 
-- 
cgit 


From 93d05d4a320cb16712bb3d57a9658f395d8cecb9 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 18 Nov 2015 06:31:03 -0800
Subject: net: provide generic busy polling to all NAPI drivers

NAPI drivers no longer need to observe a particular protocol
to benefit from busy polling (CONFIG_NET_RX_BUSY_POLL=y)

napi_hash_add() and napi_hash_del() are automatically called
from core networking stack, respectively from
netif_napi_add() and netif_napi_del()

This patch depends on free_netdev() and netif_napi_del() being
called from process context, which seems to be the norm.

Drivers might still prefer to call napi_hash_del() on their
own, since they might combine all the rcu grace periods into
a single one, knowing their NAPI structures lifetime, while
core networking stack has no idea of a possible combining.

Once this patch proves to not bring serious regressions,
we will cleanup drivers to either remove napi_hash_del()
or provide appropriate rcu grace periods combining.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c   | 2 --
 drivers/net/ethernet/broadcom/bnxt/bnxt.c         | 2 --
 drivers/net/ethernet/chelsio/cxgb4/sge.c          | 1 -
 drivers/net/ethernet/cisco/enic/enic_main.c       | 2 --
 drivers/net/ethernet/emulex/benet/be_main.c       | 1 -
 drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c      | 1 -
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c | 3 ---
 drivers/net/ethernet/mellanox/mlx4/en_cq.c        | 6 ++----
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 1 -
 drivers/net/ethernet/myricom/myri10ge/myri10ge.c  | 1 -
 drivers/net/ethernet/sfc/efx.c                    | 1 -
 drivers/net/virtio_net.c                          | 1 -
 include/linux/netdevice.h                         | 7 +++++++
 net/core/dev.c                                    | 7 +++++++
 14 files changed, 16 insertions(+), 20 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
index ab9222924bd9..d9add7c02e42 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
+++ b/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));
 	}
 }
 
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index db15c5ee09c5..f2d0dc9b1c41 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/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);
 	}
 }
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c
index f650f295f264..48d8fbb1c220 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/sge.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/sge.c
@@ -2527,7 +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;
diff --git a/drivers/net/ethernet/cisco/enic/enic_main.c b/drivers/net/ethernet/cisco/enic/enic_main.c
index b36643ef0593..b2182d3ba3cc 100644
--- a/drivers/net/ethernet/cisco/enic/enic_main.c
+++ b/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)],
diff --git a/drivers/net/ethernet/emulex/benet/be_main.c b/drivers/net/ethernet/emulex/benet/be_main.c
index c29d62496ad9..4cab8879f5ae 100644
--- a/drivers/net/ethernet/emulex/benet/be_main.c
+++ b/drivers/net/ethernet/emulex/benet/be_main.c
@@ -2630,7 +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;
 }
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c
index f3168bcc7d87..e771e764daa3 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c
+++ b/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 */
diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index 592ff237d692..2955186cd4f6 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/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;
 	}
 
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_cq.c b/drivers/net/ethernet/mellanox/mlx4/en_cq.c
index 3a6176fea78d..af975a2b74c6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_cq.c
+++ b/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) {
+	if (cq->is_tx)
 		netif_tx_napi_add(cq->dev, &cq->napi, mlx4_en_poll_tx_cq,
 				  NAPI_POLL_WEIGHT);
-	} else {
+	else
 		netif_napi_add(cq->dev, &cq->napi, mlx4_en_poll_rx_cq, 64);
-		napi_hash_add(&cq->napi);
-	}
 
 	napi_enable(&cq->napi);
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index ffb1f9c1b973..f6a8cc787603 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -982,7 +982,6 @@ static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
 	mlx5e_build_channeltc_to_txq_map(priv, ix);
 
 	netif_napi_add(netdev, &c->napi, mlx5e_napi_poll, 64);
-	napi_hash_add(&c->napi);
 
 	err = mlx5e_open_tx_cqs(c, cparam);
 	if (err)
diff --git a/drivers/net/ethernet/myricom/myri10ge/myri10ge.c b/drivers/net/ethernet/myricom/myri10ge/myri10ge.c
index acf866147d65..270c9eeb7ab6 100644
--- a/drivers/net/ethernet/myricom/myri10ge/myri10ge.c
+++ b/drivers/net/ethernet/myricom/myri10ge/myri10ge.c
@@ -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:
diff --git a/drivers/net/ethernet/sfc/efx.c b/drivers/net/ethernet/sfc/efx.c
index a3c42a376741..4e82bcfbe3e0 100644
--- a/drivers/net/ethernet/sfc/efx.c
+++ b/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);
 }
 
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index d1d14cecf450..b1ae4cbf2453 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -1610,7 +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);
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index e5c33b29471b..7d2d1d7aaec7 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -466,6 +466,9 @@ static inline void napi_complete(struct napi_struct *n)
  *	@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);
 
@@ -476,6 +479,10 @@ void napi_hash_add(struct napi_struct *napi);
  * Warning: caller must observe rcu grace period
  * 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.
  */
 bool napi_hash_del(struct napi_struct *napi);
 
diff --git a/net/core/dev.c b/net/core/dev.c
index 59dddac1c2e7..41cef3e3f558 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4807,6 +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);
 
@@ -4826,8 +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);
 
@@ -7227,11 +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);
-- 
cgit