public inbox for netdev@vger.kernel.org
 help / color / mirror / Atom feed
* [PATCH v1 net-next] bonding: Optimise is_netpoll_tx_blocked().
@ 2026-02-23 23:07 Kuniyuki Iwashima
  2026-02-25  1:42 ` Eric Dumazet
  2026-02-25  4:09 ` patchwork-bot+netdevbpf
  0 siblings, 2 replies; 3+ messages in thread
From: Kuniyuki Iwashima @ 2026-02-23 23:07 UTC (permalink / raw)
  To: Jay Vosburgh, Andrew Lunn, David S. Miller, Eric Dumazet,
	Jakub Kicinski, Paolo Abeni
  Cc: Simon Horman, Kuniyuki Iwashima, Kuniyuki Iwashima, netdev

bond_start_xmit() spends some cycles in is_netpoll_tx_blocked():

  if (unlikely(is_netpoll_tx_blocked(dev)))
      return NETDEV_TX_BUSY;

because of the "pushf;pop reg" sequence (aka irqs_disabled()).

Let's swap the conditions in is_netpoll_tx_blocked() and
convert netpoll_block_tx to a static key.

Before:

   1.23 │       mov    %gs:0x28,%rax
   1.24 │       mov    %rax,0x18(%rsp)
  29.45 │       pushfq
   0.50 │       pop    %rax
   0.47 │       test   $0x200,%eax
        │     ↓ je     1b4
   0.49 │ 32:   lea    0x980(%rsi),%rbx

After:

   0.72 │       mov    %gs:0x28,%rax
   0.81 │       mov    %rax,0x18(%rsp)
   0.82 │       nop
   2.77 │ 2a:   lea    0x980(%rsi),%rbx

Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
---
 drivers/net/bonding/bond_main.c |  4 ++--
 include/net/bonding.h           | 10 +++++-----
 2 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index 55a960da42b5..2f2dddcf9d30 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -206,7 +206,7 @@ MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 /*----------------------------- Global variables ----------------------------*/
 
 #ifdef CONFIG_NET_POLL_CONTROLLER
-atomic_t netpoll_block_tx = ATOMIC_INIT(0);
+DEFINE_STATIC_KEY_FALSE(netpoll_block_tx);
 #endif
 
 unsigned int bond_net_id __read_mostly;
@@ -6589,7 +6589,7 @@ static void __exit bonding_exit(void)
 
 #ifdef CONFIG_NET_POLL_CONTROLLER
 	/* Make sure we don't have an imbalance on our netpoll blocking */
-	WARN_ON(atomic_read(&netpoll_block_tx));
+	WARN_ON(static_branch_unlikely(&netpoll_block_tx));
 #endif
 }
 
diff --git a/include/net/bonding.h b/include/net/bonding.h
index 4ad5521e7731..7f8fda2fdfcb 100644
--- a/include/net/bonding.h
+++ b/include/net/bonding.h
@@ -91,22 +91,22 @@
 			    NETIF_F_GSO_ESP)
 
 #ifdef CONFIG_NET_POLL_CONTROLLER
-extern atomic_t netpoll_block_tx;
+DECLARE_STATIC_KEY_FALSE(netpoll_block_tx);
 
 static inline void block_netpoll_tx(void)
 {
-	atomic_inc(&netpoll_block_tx);
+	static_branch_inc(&netpoll_block_tx);
 }
 
 static inline void unblock_netpoll_tx(void)
 {
-	atomic_dec(&netpoll_block_tx);
+	static_branch_dec(&netpoll_block_tx);
 }
 
 static inline int is_netpoll_tx_blocked(struct net_device *dev)
 {
-	if (unlikely(netpoll_tx_running(dev)))
-		return atomic_read(&netpoll_block_tx);
+	if (static_branch_unlikely(&netpoll_block_tx))
+		return netpoll_tx_running(dev);
 	return 0;
 }
 #else
-- 
2.53.0.414.gf7e9f6c205-goog


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH v1 net-next] bonding: Optimise is_netpoll_tx_blocked().
  2026-02-23 23:07 [PATCH v1 net-next] bonding: Optimise is_netpoll_tx_blocked() Kuniyuki Iwashima
@ 2026-02-25  1:42 ` Eric Dumazet
  2026-02-25  4:09 ` patchwork-bot+netdevbpf
  1 sibling, 0 replies; 3+ messages in thread
From: Eric Dumazet @ 2026-02-25  1:42 UTC (permalink / raw)
  To: Kuniyuki Iwashima
  Cc: Jay Vosburgh, Andrew Lunn, David S. Miller, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Kuniyuki Iwashima, netdev

On Tue, Feb 24, 2026 at 12:07 AM Kuniyuki Iwashima <kuniyu@google.com> wrote:
>
> bond_start_xmit() spends some cycles in is_netpoll_tx_blocked():
>
>   if (unlikely(is_netpoll_tx_blocked(dev)))
>       return NETDEV_TX_BUSY;
>
> because of the "pushf;pop reg" sequence (aka irqs_disabled()).
>
> Let's swap the conditions in is_netpoll_tx_blocked() and
> convert netpoll_block_tx to a static key.
>

> Suggested-by: Eric Dumazet <edumazet@google.com>
> Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>

Reviewed-by: Eric Dumazet <edumazet@google.com>

Thanks!

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH v1 net-next] bonding: Optimise is_netpoll_tx_blocked().
  2026-02-23 23:07 [PATCH v1 net-next] bonding: Optimise is_netpoll_tx_blocked() Kuniyuki Iwashima
  2026-02-25  1:42 ` Eric Dumazet
@ 2026-02-25  4:09 ` patchwork-bot+netdevbpf
  1 sibling, 0 replies; 3+ messages in thread
From: patchwork-bot+netdevbpf @ 2026-02-25  4:09 UTC (permalink / raw)
  To: Kuniyuki Iwashima
  Cc: jv, andrew+netdev, davem, edumazet, kuba, pabeni, horms, kuni1840,
	netdev

Hello:

This patch was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Mon, 23 Feb 2026 23:07:18 +0000 you wrote:
> bond_start_xmit() spends some cycles in is_netpoll_tx_blocked():
> 
>   if (unlikely(is_netpoll_tx_blocked(dev)))
>       return NETDEV_TX_BUSY;
> 
> because of the "pushf;pop reg" sequence (aka irqs_disabled()).
> 
> [...]

Here is the summary with links:
  - [v1,net-next] bonding: Optimise is_netpoll_tx_blocked().
    https://git.kernel.org/netdev/net-next/c/fc1f97929ada

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html



^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2026-02-25  4:09 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-02-23 23:07 [PATCH v1 net-next] bonding: Optimise is_netpoll_tx_blocked() Kuniyuki Iwashima
2026-02-25  1:42 ` Eric Dumazet
2026-02-25  4:09 ` patchwork-bot+netdevbpf

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox