[PATCH net] net: busy-poll: use ktime_get_ns() instead of local_clock()

[PATCH net] net: busy-poll: use ktime_get_ns() instead of local_clock()

[Eric Dumazet]

Typically, busy-polling durations are below 100 usec.

When/if the busy-poller thread migrates to another cpu,
local_clock() can be off by +/-2msec or more for small
values of HZ, depending on the platform.

Use ktimer_get_ns() to ensure deterministic behavior,
which is the whole point of busy-polling.

Fixes: 060212928670 ("net: add low latency socket poll")
Fixes: 9a3c71aa8024 ("net: convert low latency sockets to sched_clock()")
Fixes: 37089834528b ("sched, net: Fixup busy_loop_us_clock()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Joe Damato <jdamato@fastly.com>
---
 include/net/busy_poll.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/net/busy_poll.h b/include/net/busy_poll.h
index 9b09acac538eed8dbaa2576bf2af926ecd98eb44..522f1da8b747ac73578d8fd93301d31835a6dae0 100644
--- a/include/net/busy_poll.h
+++ b/include/net/busy_poll.h
@@ -68,7 +68,7 @@ static inline bool sk_can_busy_loop(struct sock *sk)
 static inline unsigned long busy_loop_current_time(void)
 {
 #ifdef CONFIG_NET_RX_BUSY_POLL
-	return (unsigned long)(local_clock() >> 10);
+	return (unsigned long)(ktime_get_ns() >> 10);
 #else
 	return 0;
 #endif
-- 
2.46.0.295.g3b9ea8a38a-goog

Re: [PATCH net] net: busy-poll: use ktime_get_ns() instead of local_clock()

[Joe Damato]

On Tue, Aug 27, 2024 at 11:49:16AM +0000, Eric Dumazet wrote:
> Typically, busy-polling durations are below 100 usec.
> 
> When/if the busy-poller thread migrates to another cpu,
> local_clock() can be off by +/-2msec or more for small
> values of HZ, depending on the platform.
> 
> Use ktimer_get_ns() to ensure deterministic behavior,
> which is the whole point of busy-polling.
> 
> Fixes: 060212928670 ("net: add low latency socket poll")
> Fixes: 9a3c71aa8024 ("net: convert low latency sockets to sched_clock()")
> Fixes: 37089834528b ("sched, net: Fixup busy_loop_us_clock()")
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Cc: Mina Almasry <almasrymina@google.com>
> Cc: Willem de Bruijn <willemb@google.com>
> Cc: Joe Damato <jdamato@fastly.com>
> ---
>  include/net/busy_poll.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/include/net/busy_poll.h b/include/net/busy_poll.h
> index 9b09acac538eed8dbaa2576bf2af926ecd98eb44..522f1da8b747ac73578d8fd93301d31835a6dae0 100644
> --- a/include/net/busy_poll.h
> +++ b/include/net/busy_poll.h
> @@ -68,7 +68,7 @@ static inline bool sk_can_busy_loop(struct sock *sk)
>  static inline unsigned long busy_loop_current_time(void)
>  {
>  #ifdef CONFIG_NET_RX_BUSY_POLL
> -	return (unsigned long)(local_clock() >> 10);
> +	return (unsigned long)(ktime_get_ns() >> 10);
>  #else
>  	return 0;
>  #endif
> -- 
> 2.46.0.295.g3b9ea8a38a-goog

Makes sense to me, thanks.

Reviewed-by: Joe Damato <jdamato@fastly.com>

Re: [PATCH net] net: busy-poll: use ktime_get_ns() instead of local_clock()

[patchwork-bot+netdevbpf]

Hello:

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

On Tue, 27 Aug 2024 11:49:16 +0000 you wrote:
> Typically, busy-polling durations are below 100 usec.
> 
> When/if the busy-poller thread migrates to another cpu,
> local_clock() can be off by +/-2msec or more for small
> values of HZ, depending on the platform.
> 
> Use ktimer_get_ns() to ensure deterministic behavior,
> which is the whole point of busy-polling.
> 
> [...]

Here is the summary with links:
  - [net] net: busy-poll: use ktime_get_ns() instead of local_clock()
    https://git.kernel.org/netdev/net/c/0870b0d8b393

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