netdev - Re: [PATCH net-next 3/3] net: add NUMA awareness to skb_attempt_defer

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <CAAVpQUDiGuCtUzXeJw0004VxDHFketF7J+Xu6tqEod_o=3r7Kg@mail.gmail.com>
Date: Sat, 27 Sep 2025 13:28:20 -0700
From: Kuniyuki Iwashima <kuniyu@...gle.com>
To: Eric Dumazet <edumazet@...gle.com>
Cc: "David S . Miller" <davem@...emloft.net>, Jakub Kicinski <kuba@...nel.org>, 
	Paolo Abeni <pabeni@...hat.com>, Simon Horman <horms@...nel.org>, 
	Willem de Bruijn <willemb@...gle.com>, netdev@...r.kernel.org, eric.dumazet@...il.com
Subject: Re: [PATCH net-next 3/3] net: add NUMA awareness to skb_attempt_defer_free()

On Fri, Sep 26, 2025 at 8:13 AM Eric Dumazet <edumazet@...gle.com> wrote:
>
> Instead of sharing sd->defer_list & sd->defer_count with
> many cpus, add one pair for each NUMA node.
>
> Signed-off-by: Eric Dumazet <edumazet@...gle.com>
> ---
>  include/linux/netdevice.h |  4 ----
>  include/net/hotdata.h     |  7 +++++++
>  net/core/dev.c            | 37 +++++++++++++++++++++++++------------
>  net/core/dev.h            |  2 +-
>  net/core/skbuff.c         | 11 ++++++-----
>  5 files changed, 39 insertions(+), 22 deletions(-)
>
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index 5c9aa16933d197f70746d64e5f44cae052d9971c..d1a687444b275d45d105e336d2ede264fd310f1b 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -3536,10 +3536,6 @@ struct softnet_data {
>
>         struct numa_drop_counters drop_counters;
>
> -       /* Another possibly contended cache line */
> -       struct llist_head       defer_list ____cacheline_aligned_in_smp;
> -       atomic_long_t           defer_count;
> -
>         int                     defer_ipi_scheduled ____cacheline_aligned_in_smp;
>         call_single_data_t      defer_csd;
>  };
> diff --git a/include/net/hotdata.h b/include/net/hotdata.h
> index fda94b2647ffa242c256c95ae929d9ef25e54f96..4acec191c54ab367ca12fff590d1f8c8aad64651 100644
> --- a/include/net/hotdata.h
> +++ b/include/net/hotdata.h
> @@ -2,10 +2,16 @@
>  #ifndef _NET_HOTDATA_H
>  #define _NET_HOTDATA_H
>
> +#include <linux/llist.h>
>  #include <linux/types.h>
>  #include <linux/netdevice.h>
>  #include <net/protocol.h>
>
> +struct skb_defer_node {
> +       struct llist_head       defer_list;
> +       atomic_long_t           defer_count;
> +} ____cacheline_aligned_in_smp;
> +
>  /* Read mostly data used in network fast paths. */
>  struct net_hotdata {
>  #if IS_ENABLED(CONFIG_INET)
> @@ -30,6 +36,7 @@ struct net_hotdata {
>         struct rps_sock_flow_table __rcu *rps_sock_flow_table;
>         u32                     rps_cpu_mask;
>  #endif
> +       struct skb_defer_node __percpu *skb_defer_nodes;
>         int                     gro_normal_batch;
>         int                     netdev_budget;
>         int                     netdev_budget_usecs;
> diff --git a/net/core/dev.c b/net/core/dev.c
> index fb67372774de10b0b112ca71c7c7a13819c2325b..afcf07352eaa3b9a563173106c84167ebe1ab387 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -5180,8 +5180,9 @@ static void napi_schedule_rps(struct softnet_data *sd)
>         __napi_schedule_irqoff(&mysd->backlog);
>  }
>
> -void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu)
> +void kick_defer_list_purge(unsigned int cpu)
>  {
> +       struct softnet_data *sd = &per_cpu(softnet_data, cpu);
>         unsigned long flags;
>
>         if (use_backlog_threads()) {
> @@ -6715,18 +6716,26 @@ bool napi_complete_done(struct napi_struct *n, int work_done)
>  }
>  EXPORT_SYMBOL(napi_complete_done);
>
> -static void skb_defer_free_flush(struct softnet_data *sd)
> +struct skb_defer_node __percpu *skb_defer_nodes;

seems this is unused, given it's close to the end of cycle
maybe this can be fixed up while applying ? :)

Reviewed-by: Kuniyuki Iwashima <kuniyu@...gle.com>

Thanks!