Re: [PATCH 7/9] softirq,net: Use softirq_needs_break()

[Jason Xing]

On Fri, May 5, 2023 at 7:27 PM Liu Jian <liujian56@xxxxxxxxxx> wrote:
>
> From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
>
> SoftIRQs provide their own timeout/break code now, use that.
>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> Signed-off-by: Liu Jian <liujian56@xxxxxxxxxx>
> ---
>  Documentation/admin-guide/sysctl/net.rst | 11 +----------
>  net/core/dev.c                           |  6 +-----
>  net/core/dev.h                           |  1 -
>  net/core/sysctl_net_core.c               |  8 --------
>  4 files changed, 2 insertions(+), 24 deletions(-)
>
> diff --git a/Documentation/admin-guide/sysctl/net.rst b/Documentation/admin-guide/sysctl/net.rst
> index 466c560b0c30..095c60788c61 100644
> --- a/Documentation/admin-guide/sysctl/net.rst
> +++ b/Documentation/admin-guide/sysctl/net.rst
> @@ -267,16 +267,7 @@ netdev_budget
>
>  Maximum number of packets taken from all interfaces in one polling cycle (NAPI
>  poll). In one polling cycle interfaces which are registered to polling are
> -probed in a round-robin manner. Also, a polling cycle may not exceed
> -netdev_budget_usecs microseconds, even if netdev_budget has not been
> -exhausted.
> -
> -netdev_budget_usecs
> ----------------------
> -
> -Maximum number of microseconds in one NAPI polling cycle. Polling
> -will exit when either netdev_budget_usecs have elapsed during the
> -poll cycle or the number of packets processed reaches netdev_budget.
> +probed in a round-robin manner.
>
>  netdev_max_backlog
>  ------------------
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 735096d42c1d..70b6726beee6 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -4321,7 +4321,6 @@ int netdev_tstamp_prequeue __read_mostly = 1;
>  unsigned int sysctl_skb_defer_max __read_mostly = 64;
>  int netdev_budget __read_mostly = 300;
>  /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
> -unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
>  int weight_p __read_mostly = 64;           /* old backlog weight */
>  int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
>  int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
> @@ -6659,8 +6658,6 @@ static int napi_threaded_poll(void *data)
>  static __latent_entropy void net_rx_action(struct softirq_action *h)
>  {
>         struct softnet_data *sd = this_cpu_ptr(&softnet_data);
> -       unsigned long time_limit = jiffies +
> -               usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
>         int budget = READ_ONCE(netdev_budget);
>         LIST_HEAD(list);
>         LIST_HEAD(repoll);
> @@ -6699,8 +6696,7 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
>                  * Allow this to run for 2 jiffies since which will allow
>                  * an average latency of 1.5/HZ.
>                  */
> -               if (unlikely(budget <= 0 ||
> -                            time_after_eq(jiffies, time_limit))) {
> +               if (unlikely(budget <= 0 || softirq_needs_break(h))) {
>                         sd->time_squeeze++;
>                         break;
>                 }
> diff --git a/net/core/dev.h b/net/core/dev.h
> index e075e198092c..e64a60c767ab 100644
> --- a/net/core/dev.h
> +++ b/net/core/dev.h
> @@ -39,7 +39,6 @@ void dev_addr_check(struct net_device *dev);
>
>  /* sysctls not referred to from outside net/core/ */
>  extern int             netdev_budget;
> -extern unsigned int    netdev_budget_usecs;
>  extern unsigned int    sysctl_skb_defer_max;
>  extern int             netdev_tstamp_prequeue;
>  extern int             netdev_unregister_timeout_secs;
> diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
> index 782273bb93c2..59765c805f5b 100644
> --- a/net/core/sysctl_net_core.c
> +++ b/net/core/sysctl_net_core.c
> @@ -595,14 +595,6 @@ static struct ctl_table net_core_table[] = {
>                 .extra1         = SYSCTL_ONE,
>                 .extra2         = &max_skb_frags,
>         },
> -       {
> -               .procname       = "netdev_budget_usecs",
> -               .data           = &netdev_budget_usecs,
> -               .maxlen         = sizeof(unsigned int),
> -               .mode           = 0644,
> -               .proc_handler   = proc_dointvec_minmax,
> -               .extra1         = SYSCTL_ZERO,
> -       },

I cannot help asking whether we really need to remove the sysctl knob
because it can let some users tune this. It's useful for some cases, I
believe. Do you have any evidence/number to prove we can get rid of
this?

Thanks,
Jason

>         {
>                 .procname       = "fb_tunnels_only_for_init_net",
>                 .data           = &sysctl_fb_tunnels_only_for_init_net,
> --
> 2.34.1
>
>