Netdev List
 help / color / mirror / Atom feed
From: Jiebin Sun <jiebin.sun@intel.com>
To: netdev@vger.kernel.org
Cc: edumazet@kernel.org, davem@davemloft.net, kuba@kernel.org,
	pabeni@redhat.com, horms@kernel.org,
	linux-kernel@vger.kernel.org, tim.c.chen@linux.intel.com,
	tianyou.li@intel.com, wangyang.guo@intel.com
Subject: [PATCH net-next v2] net: dst: avoid false sharing between tclassid and __rcuref
Date: Fri, 20 Nov 2026 14:16:04 +0800	[thread overview]
Message-ID: <20261120061604.3676605-1-jiebin.sun@intel.com> (raw)

On x86-64 with CONFIG_IP_ROUTE_CLASSID=y, ip_rcv_finish_core() reads
dst_entry::tclassid from the cache line containing __rcuref. Concurrent
reference updates cause false sharing with this load.

Swap expires with {error, __pad, tclassid} to move tclassid and error
into the first cache line without growing dst_entry. The structures
that embed it, including rtable, rt6_info, and metadata_dst, keep
their sizes.

pahole on x86-64 (sizes and offsets in bytes)

  +----------------------+--------+-------+
  | Struct or field      | Before | After |
  +----------------------+--------+-------+
  | sizeof(dst_entry)    |    136 |   136 |
  | dst_entry.expires    |     24 |    96 |
  | dst_entry.__rcuref   |     64 |    64 |
  | dst_entry.tclassid   |    100 |    28 |
  +----------------------+--------+-------+

The IPv4 loopback paths are shown below.

  TX: __dev_queue_xmit() -> skb_dst_force() -> dst_hold_safe()
      -> rcuref_get()
  RX: ip_rcv_finish_core() -> skb_dst(skb)->tclassid

At the tclassid load, separate 30-second perf c2c runs recorded:

  +----------+------------+-------------+
  | Kernel   | Local HITM | Remote HITM |
  +----------+------------+-------------+
  | Baseline |       2624 |        2611 |
  | Swap     |          1 |           0 |
  +----------+------------+-------------+

Memcached over IPv4 loopback (64 threads each, 5 x 60s,
20s warmup, mean ops/s)
Baseline: commit 6628e7d4c6b2 ("Linux 7.4-rc1")

  +-----------+-----------------+
  | Memcached | Improvement (%) |
  +-----------+-----------------+
  | default   |          +32.55 |
  | hit       |          +45.79 |
  | balanced  |           -0.12 |
  | large     |          +45.62 |
  +-----------+-----------------+

Default client

  memtier_benchmark -s 127.0.0.1 -p 11221 -P memcache_text
    -t 64 -c 1 --pipeline=16 --test-time=60 --ratio=1:100
    --data-size=32 --key-maximum=10000000

Server

  memcached -c 4096 -t 64 -u pnp -p 11221 -l 127.0.0.1

Client variants

  hit:      --key-maximum=1000
  balanced: --key-maximum=1000 --ratio=1:1
  large:    --key-maximum=1000 --data-size=1024

Suggested-by: Eric Dumazet <edumazet@kernel.org>
Suggested-by: Tim Chen <tim.c.chen@linux.intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
---
Changes in v2:
- Swap expires with error/__pad/tclassid and show the pahole results.
- Clarify the loopback path and keep the memcached and c2c tables.

 include/net/dst.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/include/net/dst.h b/include/net/dst.h
index dbedfe72e1fd..a692fe85f4bc 100644
--- a/include/net/dst.h
+++ b/include/net/dst.h
@@ -30,7 +30,9 @@ struct dst_entry {
 	};
 	struct  dst_ops	        *ops;
 	unsigned long		_metrics;
-	unsigned long           expires;
+	short			error;
+	short			__pad;
+	__u32			tclassid;
 #ifdef CONFIG_XFRM
 	struct xfrm_state	*xfrm;
 #else
@@ -74,9 +76,7 @@ struct dst_entry {
 	int			__use;
 	unsigned long		lastuse;
 	struct rcu_head		rcu_head;
-	short			error;
-	short			__pad;
-	__u32			tclassid;
+	unsigned long		expires;
 #ifndef CONFIG_64BIT
 	struct lwtunnel_state   *lwtstate;
 	rcuref_t		__rcuref;	/* 32-bit offset 64 */

base-commit: 66726ba54a01a2951688a56f5eb4d13d063acbea
-- 
2.52.0


                 reply	other threads:[~2026-11-20  6:11 UTC|newest]

Thread overview: [no followups] expand[flat|nested]  mbox.gz  Atom feed

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20261120061604.3676605-1-jiebin.sun@intel.com \
    --to=jiebin.sun@intel.com \
    --cc=davem@davemloft.net \
    --cc=edumazet@kernel.org \
    --cc=horms@kernel.org \
    --cc=kuba@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    --cc=tianyou.li@intel.com \
    --cc=tim.c.chen@linux.intel.com \
    --cc=wangyang.guo@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox