[PATCH net-next 00/18] net: group together hot data

[PATCH net-next 00/18] net: group together hot data

[Eric Dumazet]

While our recent structure reorganizations were focused
on increasing max throughput, there is still an
area where improvements are much needed.

In many cases, a cpu handles one packet at a time,
instead of a nice batch.

Hardware interrupt.
 -> Software interrupt.
   -> Network/Protocol stacks.

If the cpu was idle or busy in other layers,
it has to pull many cache lines.

This series adds a new net_hotdata structure, where
some critical (and read-mostly) data used in
rx and tx path is packed in a small number of cache lines.

Synthetic benchmarks will not see much difference,
but latency of single packet should improve.

net_hodata current size on 64bit is 416 bytes,
but might grow in the future.

Also move RPS definitions to a new include file.

Eric Dumazet (18):
  net: introduce struct net_hotdata
  net: move netdev_budget and netdev_budget to net_hotdata
  net: move netdev_tstamp_prequeue into net_hotdata
  net: move ptype_all into net_hotdata
  net: move netdev_max_backlog to net_hotdata
  net: move ip_packet_offload and ipv6_packet_offload to net_hotdata
  net: move tcpv4_offload and tcpv6_offload to net_hotdata
  net: move dev_tx_weight to net_hotdata
  net: move dev_rx_weight to net_hotdata
  net: move skbuff_cache(s) to net_hotdata
  udp: move udpv4_offload and udpv6_offload to net_hotdata
  ipv6: move tcpv6_protocol and udpv6_protocol to net_hotdata
  inet: move tcp_protocol and udp_protocol to net_hotdata
  inet: move inet_ehash_secret and udp_ehash_secret into net_hotdata
  ipv6: move inet6_ehash_secret and udp6_ehash_secret into net_hotdata
  ipv6: move tcp_ipv6_hash_secret and udp_ipv6_hash_secret to
    net_hotdata
  net: introduce include/net/rps.h
  net: move rps_sock_flow_table to net_hotdata

 drivers/net/ethernet/intel/ice/ice_arfs.c     |   1 +
 .../net/ethernet/mellanox/mlx4/en_netdev.c    |   1 +
 .../net/ethernet/mellanox/mlx5/core/en_arfs.c |   1 +
 drivers/net/ethernet/sfc/rx_common.c          |   1 +
 drivers/net/ethernet/sfc/siena/rx_common.c    |   1 +
 drivers/net/tun.c                             |   1 +
 include/linux/netdevice.h                     |  88 ------------
 include/linux/skbuff.h                        |   1 -
 include/net/gro.h                             |   5 +-
 include/net/hotdata.h                         |  52 ++++++++
 include/net/protocol.h                        |   3 +
 include/net/rps.h                             | 125 ++++++++++++++++++
 include/net/sock.h                            |  35 -----
 kernel/bpf/cpumap.c                           |   4 +-
 net/bpf/test_run.c                            |   4 +-
 net/core/Makefile                             |   1 +
 net/core/dev.c                                |  58 +++-----
 net/core/dev.h                                |   3 -
 net/core/gro.c                                |  15 +--
 net/core/gro_cells.c                          |   3 +-
 net/core/gso.c                                |   4 +-
 net/core/hotdata.c                            |  22 +++
 net/core/net-procfs.c                         |   7 +-
 net/core/net-sysfs.c                          |   1 +
 net/core/skbuff.c                             |  44 +++---
 net/core/sysctl_net_core.c                    |  25 ++--
 net/core/xdp.c                                |   5 +-
 net/ipv4/af_inet.c                            |  49 +++----
 net/ipv4/inet_hashtables.c                    |   3 +-
 net/ipv4/tcp.c                                |   1 +
 net/ipv4/tcp_offload.c                        |  17 ++-
 net/ipv4/udp.c                                |   2 -
 net/ipv4/udp_offload.c                        |  17 ++-
 net/ipv6/af_inet6.c                           |   1 +
 net/ipv6/inet6_hashtables.c                   |   8 +-
 net/ipv6/ip6_offload.c                        |  18 +--
 net/ipv6/tcp_ipv6.c                           |  17 +--
 net/ipv6/tcpv6_offload.c                      |  16 +--
 net/ipv6/udp.c                                |  19 ++-
 net/ipv6/udp_offload.c                        |  21 ++-
 net/sched/sch_generic.c                       |   3 +-
 net/sctp/socket.c                             |   1 +
 net/xfrm/espintcp.c                           |   4 +-
 net/xfrm/xfrm_input.c                         |   3 +-
 44 files changed, 391 insertions(+), 320 deletions(-)
 create mode 100644 include/net/hotdata.h
 create mode 100644 include/net/rps.h
 create mode 100644 net/core/hotdata.c

-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 01/18] net: introduce struct net_hotdata

[Eric Dumazet]

Instead of spreading networking critical fields
all over the places, add a custom net_hotdata
structure so that we can precisely control its layout.

In this first patch, move :

- gro_normal_batch used in rx (GRO stack)
- offload_base used in rx and tx (GRO and TSO stacks)

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/netdevice.h  |  1 -
 include/net/gro.h          |  5 ++---
 include/net/hotdata.h      | 15 +++++++++++++++
 net/core/Makefile          |  1 +
 net/core/gro.c             | 15 ++++++---------
 net/core/gso.c             |  4 ++--
 net/core/hotdata.c         |  9 +++++++++
 net/core/sysctl_net_core.c |  3 ++-
 8 files changed, 37 insertions(+), 16 deletions(-)
 create mode 100644 include/net/hotdata.h
 create mode 100644 net/core/hotdata.c

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index c41019f3417948d09ae9a50b57b856be1dc8ae42..15ce809e0541078bff7a48b8d7cb2cf2c1ac8a93 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -4790,7 +4790,6 @@ void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s);
 extern int		netdev_max_backlog;
 extern int		dev_rx_weight;
 extern int		dev_tx_weight;
-extern int		gro_normal_batch;
 
 enum {
 	NESTED_SYNC_IMM_BIT,
diff --git a/include/net/gro.h b/include/net/gro.h
index 2b58671a65492bf3f9dabf1e7a2d985cee007e11..d6fc8fbd37302338fc09ab01fead899002c5833f 100644
--- a/include/net/gro.h
+++ b/include/net/gro.h
@@ -9,6 +9,7 @@
 #include <net/ip6_checksum.h>
 #include <linux/skbuff.h>
 #include <net/udp.h>
+#include <net/hotdata.h>
 
 struct napi_gro_cb {
 	union {
@@ -446,7 +447,7 @@ static inline void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb,
 {
 	list_add_tail(&skb->list, &napi->rx_list);
 	napi->rx_count += segs;
-	if (napi->rx_count >= READ_ONCE(gro_normal_batch))
+	if (napi->rx_count >= READ_ONCE(net_hotdata.gro_normal_batch))
 		gro_normal_list(napi);
 }
 
@@ -493,6 +494,4 @@ static inline void inet6_get_iif_sdif(const struct sk_buff *skb, int *iif, int *
 #endif
 }
 
-extern struct list_head offload_base;
-
 #endif /* _NET_IPV6_GRO_H */
diff --git a/include/net/hotdata.h b/include/net/hotdata.h
new file mode 100644
index 0000000000000000000000000000000000000000..6ed32e4e34aa3bdc6e860f5a8a6cab69c36c7fad
--- /dev/null
+++ b/include/net/hotdata.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef _NET_HOTDATA_H
+#define _NET_HOTDATA_H
+
+#include <linux/types.h>
+
+/* Read mostly data used in network fast paths. */
+struct net_hotdata {
+	struct list_head	offload_base;
+	int			gro_normal_batch;
+};
+
+extern struct net_hotdata net_hotdata;
+
+#endif /* _NET_HOTDATA_H */
diff --git a/net/core/Makefile b/net/core/Makefile
index 821aec06abf1460d3504de4b6b66a328bba748d8..6e6548011fae570e345717e43eb3c1a6133571c7 100644
--- a/net/core/Makefile
+++ b/net/core/Makefile
@@ -18,6 +18,7 @@ obj-y		     += dev.o dev_addr_lists.o dst.o netevent.o \
 obj-$(CONFIG_NETDEV_ADDR_LIST_TEST) += dev_addr_lists_test.o
 
 obj-y += net-sysfs.o
+obj-y += hotdata.o
 obj-$(CONFIG_PAGE_POOL) += page_pool.o page_pool_user.o
 obj-$(CONFIG_PROC_FS) += net-procfs.o
 obj-$(CONFIG_NET_PKTGEN) += pktgen.o
diff --git a/net/core/gro.c b/net/core/gro.c
index 6a0edbd826a17573b51c5f71e20ff0c09364fc21..ee30d4f0c03876e78795397d1c495881a2c9e80f 100644
--- a/net/core/gro.c
+++ b/net/core/gro.c
@@ -10,9 +10,6 @@
 #define GRO_MAX_HEAD (MAX_HEADER + 128)
 
 static DEFINE_SPINLOCK(offload_lock);
-struct list_head offload_base __read_mostly = LIST_HEAD_INIT(offload_base);
-/* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
-int gro_normal_batch __read_mostly = 8;
 
 /**
  *	dev_add_offload - register offload handlers
@@ -31,7 +28,7 @@ void dev_add_offload(struct packet_offload *po)
 	struct packet_offload *elem;
 
 	spin_lock(&offload_lock);
-	list_for_each_entry(elem, &offload_base, list) {
+	list_for_each_entry(elem, &net_hotdata.offload_base, list) {
 		if (po->priority < elem->priority)
 			break;
 	}
@@ -55,7 +52,7 @@ EXPORT_SYMBOL(dev_add_offload);
  */
 static void __dev_remove_offload(struct packet_offload *po)
 {
-	struct list_head *head = &offload_base;
+	struct list_head *head = &net_hotdata.offload_base;
 	struct packet_offload *po1;
 
 	spin_lock(&offload_lock);
@@ -235,9 +232,9 @@ int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
 
 static void napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
 {
+	struct list_head *head = &net_hotdata.offload_base;
 	struct packet_offload *ptype;
 	__be16 type = skb->protocol;
-	struct list_head *head = &offload_base;
 	int err = -ENOENT;
 
 	BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
@@ -444,7 +441,7 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
 {
 	u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
 	struct gro_list *gro_list = &napi->gro_hash[bucket];
-	struct list_head *head = &offload_base;
+	struct list_head *head = &net_hotdata.offload_base;
 	struct packet_offload *ptype;
 	__be16 type = skb->protocol;
 	struct sk_buff *pp = NULL;
@@ -550,7 +547,7 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
 
 struct packet_offload *gro_find_receive_by_type(__be16 type)
 {
-	struct list_head *offload_head = &offload_base;
+	struct list_head *offload_head = &net_hotdata.offload_base;
 	struct packet_offload *ptype;
 
 	list_for_each_entry_rcu(ptype, offload_head, list) {
@@ -564,7 +561,7 @@ EXPORT_SYMBOL(gro_find_receive_by_type);
 
 struct packet_offload *gro_find_complete_by_type(__be16 type)
 {
-	struct list_head *offload_head = &offload_base;
+	struct list_head *offload_head = &net_hotdata.offload_base;
 	struct packet_offload *ptype;
 
 	list_for_each_entry_rcu(ptype, offload_head, list) {
diff --git a/net/core/gso.c b/net/core/gso.c
index 9e1803bfc9c6cac2fe7054661f8995909a6c28d9..bcd156372f4df080f83cc45fc96df1789125a8ae 100644
--- a/net/core/gso.c
+++ b/net/core/gso.c
@@ -17,7 +17,7 @@ struct sk_buff *skb_eth_gso_segment(struct sk_buff *skb,
 	struct packet_offload *ptype;
 
 	rcu_read_lock();
-	list_for_each_entry_rcu(ptype, &offload_base, list) {
+	list_for_each_entry_rcu(ptype, &net_hotdata.offload_base, list) {
 		if (ptype->type == type && ptype->callbacks.gso_segment) {
 			segs = ptype->callbacks.gso_segment(skb, features);
 			break;
@@ -48,7 +48,7 @@ struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
 	__skb_pull(skb, vlan_depth);
 
 	rcu_read_lock();
-	list_for_each_entry_rcu(ptype, &offload_base, list) {
+	list_for_each_entry_rcu(ptype, &net_hotdata.offload_base, list) {
 		if (ptype->type == type && ptype->callbacks.gso_segment) {
 			segs = ptype->callbacks.gso_segment(skb, features);
 			break;
diff --git a/net/core/hotdata.c b/net/core/hotdata.c
new file mode 100644
index 0000000000000000000000000000000000000000..abb8ad19d59acc0d7d6e1b06f4506afa42bde44b
--- /dev/null
+++ b/net/core/hotdata.c
@@ -0,0 +1,9 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+#include <net/hotdata.h>
+#include <linux/cache.h>
+#include <linux/list.h>
+
+struct net_hotdata net_hotdata __cacheline_aligned = {
+	.offload_base = LIST_HEAD_INIT(net_hotdata.offload_base),
+	.gro_normal_batch = 8,
+};
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index 986f15e5d6c41250c8b9099fc1d2883112e77ffb..0eb1242eabbe0d3ea58886b1db409c9d991ac672 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -23,6 +23,7 @@
 #include <net/net_ratelimit.h>
 #include <net/busy_poll.h>
 #include <net/pkt_sched.h>
+#include <net/hotdata.h>
 
 #include "dev.h"
 
@@ -632,7 +633,7 @@ static struct ctl_table net_core_table[] = {
 	},
 	{
 		.procname	= "gro_normal_batch",
-		.data		= &gro_normal_batch,
+		.data		= &net_hotdata.gro_normal_batch,
 		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec_minmax,
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 02/18] net: move netdev_budget and netdev_budget to net_hotdata

[Eric Dumazet]

netdev_budget and netdev_budget are used in rx path (net_rx_action())

Move them into net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h      | 2 ++
 net/core/dev.c             | 7 ++-----
 net/core/dev.h             | 2 --
 net/core/hotdata.c         | 6 ++++++
 net/core/sysctl_net_core.c | 4 ++--
 5 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 6ed32e4e34aa3bdc6e860f5a8a6cab69c36c7fad..72170223385ebe65cce42f762b3686c072291d36 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -8,6 +8,8 @@
 struct net_hotdata {
 	struct list_head	offload_base;
 	int			gro_normal_batch;
+	int			netdev_budget;
+	int			netdev_budget_usecs;
 };
 
 extern struct net_hotdata net_hotdata;
diff --git a/net/core/dev.c b/net/core/dev.c
index fe054cbd41e92cbca87f1c0640c6ebe4fb6b2d86..0102a1810e7b148f465b87886b743e3d12c0e578 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4410,9 +4410,6 @@ EXPORT_SYMBOL(netdev_max_backlog);
 
 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 */
@@ -6790,8 +6787,8 @@ 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);
+		usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs));
+	int budget = READ_ONCE(net_hotdata.netdev_budget);
 	LIST_HEAD(list);
 	LIST_HEAD(repoll);
 
diff --git a/net/core/dev.h b/net/core/dev.h
index 45892267848d7a35a09aea95f04cfd9b72204d3c..9a6170530850c78508f9234ec82b174f4bf5a4a3 100644
--- a/net/core/dev.h
+++ b/net/core/dev.h
@@ -38,8 +38,6 @@ int dev_addr_init(struct net_device *dev);
 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/hotdata.c b/net/core/hotdata.c
index abb8ad19d59acc0d7d6e1b06f4506afa42bde44b..907d69120397dfb8d5a901912b72580fe256c762 100644
--- a/net/core/hotdata.c
+++ b/net/core/hotdata.c
@@ -1,9 +1,15 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 #include <net/hotdata.h>
 #include <linux/cache.h>
+#include <linux/jiffies.h>
 #include <linux/list.h>
 
+
 struct net_hotdata net_hotdata __cacheline_aligned = {
 	.offload_base = LIST_HEAD_INIT(net_hotdata.offload_base),
 	.gro_normal_batch = 8,
+
+	.netdev_budget = 300,
+	/* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
+	.netdev_budget_usecs = 2 * USEC_PER_SEC / HZ,
 };
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index 0eb1242eabbe0d3ea58886b1db409c9d991ac672..a9c2d798b219506da75a5d0a30d490ff4011d668 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -577,7 +577,7 @@ static struct ctl_table net_core_table[] = {
 #endif
 	{
 		.procname	= "netdev_budget",
-		.data		= &netdev_budget,
+		.data		= &net_hotdata.netdev_budget,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec
@@ -600,7 +600,7 @@ static struct ctl_table net_core_table[] = {
 	},
 	{
 		.procname	= "netdev_budget_usecs",
-		.data		= &netdev_budget_usecs,
+		.data		= &net_hotdata.netdev_budget_usecs,
 		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec_minmax,
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 03/18] net: move netdev_tstamp_prequeue into net_hotdata

[Eric Dumazet]

netdev_tstamp_prequeue is used in rx path.

Move it to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h      |  1 +
 net/core/dev.c             | 10 +++++-----
 net/core/dev.h             |  1 -
 net/core/hotdata.c         |  2 ++
 net/core/sysctl_net_core.c |  2 +-
 5 files changed, 9 insertions(+), 7 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 72170223385ebe65cce42f762b3686c072291d36..149e56528537d8ed3365e46d6dc96e39c73a733a 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -10,6 +10,7 @@ struct net_hotdata {
 	int			gro_normal_batch;
 	int			netdev_budget;
 	int			netdev_budget_usecs;
+	int			tstamp_prequeue;
 };
 
 extern struct net_hotdata net_hotdata;
diff --git a/net/core/dev.c b/net/core/dev.c
index 0102a1810e7b148f465b87886b743e3d12c0e578..53ebdb55e8b7c0a6522eb3fdbb7bdd00948eb9a5 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4408,7 +4408,6 @@ EXPORT_SYMBOL(__dev_direct_xmit);
 int netdev_max_backlog __read_mostly = 1000;
 EXPORT_SYMBOL(netdev_max_backlog);
 
-int netdev_tstamp_prequeue __read_mostly = 1;
 unsigned int sysctl_skb_defer_max __read_mostly = 64;
 int weight_p __read_mostly = 64;           /* old backlog weight */
 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
@@ -5052,7 +5051,7 @@ static int netif_rx_internal(struct sk_buff *skb)
 {
 	int ret;
 
-	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
+	net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
 
 	trace_netif_rx(skb);
 
@@ -5344,7 +5343,7 @@ static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
 	int ret = NET_RX_DROP;
 	__be16 type;
 
-	net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
+	net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb);
 
 	trace_netif_receive_skb(skb);
 
@@ -5728,7 +5727,7 @@ static int netif_receive_skb_internal(struct sk_buff *skb)
 {
 	int ret;
 
-	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
+	net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
 
 	if (skb_defer_rx_timestamp(skb))
 		return NET_RX_SUCCESS;
@@ -5758,7 +5757,8 @@ void netif_receive_skb_list_internal(struct list_head *head)
 
 	INIT_LIST_HEAD(&sublist);
 	list_for_each_entry_safe(skb, next, head, list) {
-		net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
+		net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue),
+				    skb);
 		skb_list_del_init(skb);
 		if (!skb_defer_rx_timestamp(skb))
 			list_add_tail(&skb->list, &sublist);
diff --git a/net/core/dev.h b/net/core/dev.h
index 9a6170530850c78508f9234ec82b174f4bf5a4a3..2bcaf8eee50c179db2ca59880521b9be6ecd45c8 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 unsigned int	sysctl_skb_defer_max;
-extern int		netdev_tstamp_prequeue;
 extern int		netdev_unregister_timeout_secs;
 extern int		weight_p;
 extern int		dev_weight_rx_bias;
diff --git a/net/core/hotdata.c b/net/core/hotdata.c
index 907d69120397dfb8d5a901912b72580fe256c762..087c4c84987df07f11a87112a778a5cac608a654 100644
--- a/net/core/hotdata.c
+++ b/net/core/hotdata.c
@@ -12,4 +12,6 @@ struct net_hotdata net_hotdata __cacheline_aligned = {
 	.netdev_budget = 300,
 	/* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
 	.netdev_budget_usecs = 2 * USEC_PER_SEC / HZ,
+
+	.tstamp_prequeue = 1,
 };
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index a9c2d798b219506da75a5d0a30d490ff4011d668..bddd07da099886f2747f2ac4ba39a482b0f4231d 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -499,7 +499,7 @@ static struct ctl_table net_core_table[] = {
 #endif
 	{
 		.procname	= "netdev_tstamp_prequeue",
-		.data		= &netdev_tstamp_prequeue,
+		.data		= &net_hotdata.tstamp_prequeue,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 04/18] net: move ptype_all into net_hotdata

[Eric Dumazet]

ptype_all is used in rx/tx fast paths.

Move it to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/netdevice.h |  1 -
 include/net/hotdata.h     |  1 +
 net/core/dev.c            | 16 +++++++---------
 net/core/hotdata.c        |  1 +
 net/core/net-procfs.c     |  7 ++++---
 5 files changed, 13 insertions(+), 13 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 15ce809e0541078bff7a48b8d7cb2cf2c1ac8a93..e2e7d5a7ef34de165cd293eb71800e1e6b450432 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -5300,7 +5300,6 @@ static inline const char *netdev_reg_state(const struct net_device *dev)
 #define PTYPE_HASH_SIZE	(16)
 #define PTYPE_HASH_MASK	(PTYPE_HASH_SIZE - 1)
 
-extern struct list_head ptype_all __read_mostly;
 extern struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
 
 extern struct net_device *blackhole_netdev;
diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 149e56528537d8ed3365e46d6dc96e39c73a733a..d462cb8f16bad459b439c566274c01a0f18a95d0 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -7,6 +7,7 @@
 /* Read mostly data used in network fast paths. */
 struct net_hotdata {
 	struct list_head	offload_base;
+	struct list_head	ptype_all;
 	int			gro_normal_batch;
 	int			netdev_budget;
 	int			netdev_budget_usecs;
diff --git a/net/core/dev.c b/net/core/dev.c
index 53ebdb55e8b7c0a6522eb3fdbb7bdd00948eb9a5..119b7004a8e51ea6785c60e558988d369eec8935 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -161,7 +161,6 @@
 
 static DEFINE_SPINLOCK(ptype_lock);
 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
-struct list_head ptype_all __read_mostly;	/* Taps */
 
 static int netif_rx_internal(struct sk_buff *skb);
 static int call_netdevice_notifiers_extack(unsigned long val,
@@ -540,7 +539,7 @@ static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 static inline struct list_head *ptype_head(const struct packet_type *pt)
 {
 	if (pt->type == htons(ETH_P_ALL))
-		return pt->dev ? &pt->dev->ptype_all : &ptype_all;
+		return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all;
 	else
 		return pt->dev ? &pt->dev->ptype_specific :
 				 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
@@ -2226,7 +2225,8 @@ static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
  */
 bool dev_nit_active(struct net_device *dev)
 {
-	return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
+	return !list_empty(&net_hotdata.ptype_all) ||
+	       !list_empty(&dev->ptype_all);
 }
 EXPORT_SYMBOL_GPL(dev_nit_active);
 
@@ -2237,10 +2237,9 @@ EXPORT_SYMBOL_GPL(dev_nit_active);
 
 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
 {
-	struct packet_type *ptype;
+	struct list_head *ptype_list = &net_hotdata.ptype_all;
+	struct packet_type *ptype, *pt_prev = NULL;
 	struct sk_buff *skb2 = NULL;
-	struct packet_type *pt_prev = NULL;
-	struct list_head *ptype_list = &ptype_all;
 
 	rcu_read_lock();
 again:
@@ -2286,7 +2285,7 @@ void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
 		pt_prev = ptype;
 	}
 
-	if (ptype_list == &ptype_all) {
+	if (ptype_list == &net_hotdata.ptype_all) {
 		ptype_list = &dev->ptype_all;
 		goto again;
 	}
@@ -5387,7 +5386,7 @@ static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
 	if (pfmemalloc)
 		goto skip_taps;
 
-	list_for_each_entry_rcu(ptype, &ptype_all, list) {
+	list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) {
 		if (pt_prev)
 			ret = deliver_skb(skb, pt_prev, orig_dev);
 		pt_prev = ptype;
@@ -11771,7 +11770,6 @@ static int __init net_dev_init(void)
 	if (netdev_kobject_init())
 		goto out;
 
-	INIT_LIST_HEAD(&ptype_all);
 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
 		INIT_LIST_HEAD(&ptype_base[i]);
 
diff --git a/net/core/hotdata.c b/net/core/hotdata.c
index 087c4c84987df07f11a87112a778a5cac608a654..29fcfe89fd9a697120f826dbe2fd36a1617581a1 100644
--- a/net/core/hotdata.c
+++ b/net/core/hotdata.c
@@ -7,6 +7,7 @@
 
 struct net_hotdata net_hotdata __cacheline_aligned = {
 	.offload_base = LIST_HEAD_INIT(net_hotdata.offload_base),
+	.ptype_all = LIST_HEAD_INIT(net_hotdata.ptype_all),
 	.gro_normal_batch = 8,
 
 	.netdev_budget = 300,
diff --git a/net/core/net-procfs.c b/net/core/net-procfs.c
index 2e4e96d30ee1a7a51e49587378aab47aed1290da..a97eceb84e61ec347ad132ff0f22c8ce82f12e90 100644
--- a/net/core/net-procfs.c
+++ b/net/core/net-procfs.c
@@ -3,6 +3,7 @@
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
 #include <net/wext.h>
+#include <net/hotdata.h>
 
 #include "dev.h"
 
@@ -183,7 +184,7 @@ static void *ptype_get_idx(struct seq_file *seq, loff_t pos)
 		}
 	}
 
-	list_for_each_entry_rcu(pt, &ptype_all, list) {
+	list_for_each_entry_rcu(pt, &net_hotdata.ptype_all, list) {
 		if (i == pos)
 			return pt;
 		++i;
@@ -231,13 +232,13 @@ static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 			}
 		}
 
-		nxt = ptype_all.next;
+		nxt = net_hotdata.ptype_all.next;
 		goto ptype_all;
 	}
 
 	if (pt->type == htons(ETH_P_ALL)) {
 ptype_all:
-		if (nxt != &ptype_all)
+		if (nxt != &net_hotdata.ptype_all)
 			goto found;
 		hash = 0;
 		nxt = ptype_base[0].next;
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 05/18] net: move netdev_max_backlog to net_hotdata

[Eric Dumazet]

netdev_max_backlog is used in rx fat path.

Move it to net_hodata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/netdevice.h  | 1 -
 include/net/hotdata.h      | 1 +
 net/core/dev.c             | 8 +++-----
 net/core/gro_cells.c       | 3 ++-
 net/core/hotdata.c         | 2 ++
 net/core/sysctl_net_core.c | 2 +-
 net/xfrm/espintcp.c        | 4 +++-
 net/xfrm/xfrm_input.c      | 3 ++-
 8 files changed, 14 insertions(+), 10 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index e2e7d5a7ef34de165cd293eb71800e1e6b450432..044d6f5b2ace3e2decd4296e01c8d3e200c6c7dc 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -4787,7 +4787,6 @@ void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
 			   const struct pcpu_sw_netstats __percpu *netstats);
 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s);
 
-extern int		netdev_max_backlog;
 extern int		dev_rx_weight;
 extern int		dev_tx_weight;
 
diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index d462cb8f16bad459b439c566274c01a0f18a95d0..dc50b200a94b6b935cd79d8e0406a61209fdc68e 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -12,6 +12,7 @@ struct net_hotdata {
 	int			netdev_budget;
 	int			netdev_budget_usecs;
 	int			tstamp_prequeue;
+	int			max_backlog;
 };
 
 extern struct net_hotdata net_hotdata;
diff --git a/net/core/dev.c b/net/core/dev.c
index 119b7004a8e51ea6785c60e558988d369eec8935..1b112c4db983c2d7cd280bc8c2ebc621ea3c6145 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4404,9 +4404,6 @@ EXPORT_SYMBOL(__dev_direct_xmit);
  *			Receiver routines
  *************************************************************************/
 
-int netdev_max_backlog __read_mostly = 1000;
-EXPORT_SYMBOL(netdev_max_backlog);
-
 unsigned int sysctl_skb_defer_max __read_mostly = 64;
 int weight_p __read_mostly = 64;           /* old backlog weight */
 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
@@ -4713,7 +4710,7 @@ static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
 	struct softnet_data *sd;
 	unsigned int old_flow, new_flow;
 
-	if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
+	if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1))
 		return false;
 
 	sd = this_cpu_ptr(&softnet_data);
@@ -4761,7 +4758,8 @@ static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
 	if (!netif_running(skb->dev))
 		goto drop;
 	qlen = skb_queue_len(&sd->input_pkt_queue);
-	if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
+	if (qlen <= READ_ONCE(net_hotdata.max_backlog) &&
+	    !skb_flow_limit(skb, qlen)) {
 		if (qlen) {
 enqueue:
 			__skb_queue_tail(&sd->input_pkt_queue, skb);
diff --git a/net/core/gro_cells.c b/net/core/gro_cells.c
index ed5ec5de47f670753924bd0c72db1e3ceb9b9e7a..ff8e5b64bf6b76451a69e3eae132b593c60ee204 100644
--- a/net/core/gro_cells.c
+++ b/net/core/gro_cells.c
@@ -3,6 +3,7 @@
 #include <linux/slab.h>
 #include <linux/netdevice.h>
 #include <net/gro_cells.h>
+#include <net/hotdata.h>
 
 struct gro_cell {
 	struct sk_buff_head	napi_skbs;
@@ -26,7 +27,7 @@ int gro_cells_receive(struct gro_cells *gcells, struct sk_buff *skb)
 
 	cell = this_cpu_ptr(gcells->cells);
 
-	if (skb_queue_len(&cell->napi_skbs) > READ_ONCE(netdev_max_backlog)) {
+	if (skb_queue_len(&cell->napi_skbs) > READ_ONCE(net_hotdata.max_backlog)) {
 drop:
 		dev_core_stats_rx_dropped_inc(dev);
 		kfree_skb(skb);
diff --git a/net/core/hotdata.c b/net/core/hotdata.c
index 29fcfe89fd9a697120f826dbe2fd36a1617581a1..35ed5a83ecc7ebda513fe4fafc596e053f0252c5 100644
--- a/net/core/hotdata.c
+++ b/net/core/hotdata.c
@@ -15,4 +15,6 @@ struct net_hotdata net_hotdata __cacheline_aligned = {
 	.netdev_budget_usecs = 2 * USEC_PER_SEC / HZ,
 
 	.tstamp_prequeue = 1,
+	.max_backlog = 1000,
 };
+EXPORT_SYMBOL(net_hotdata);
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index bddd07da099886f2747f2ac4ba39a482b0f4231d..8eaeeb289914258f90cf940e906d5c6be0cc0cd6 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -440,7 +440,7 @@ static struct ctl_table net_core_table[] = {
 	},
 	{
 		.procname	= "netdev_max_backlog",
-		.data		= &netdev_max_backlog,
+		.data		= &net_hotdata.max_backlog,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec
diff --git a/net/xfrm/espintcp.c b/net/xfrm/espintcp.c
index d3b3f9e720b3b6c2a4ea89df4257a564100b2c4b..fe82e2d073006e5ab1b03868c851147c0422d26d 100644
--- a/net/xfrm/espintcp.c
+++ b/net/xfrm/espintcp.c
@@ -10,6 +10,7 @@
 #if IS_ENABLED(CONFIG_IPV6)
 #include <net/ipv6_stubs.h>
 #endif
+#include <net/hotdata.h>
 
 static void handle_nonesp(struct espintcp_ctx *ctx, struct sk_buff *skb,
 			  struct sock *sk)
@@ -169,7 +170,8 @@ int espintcp_queue_out(struct sock *sk, struct sk_buff *skb)
 {
 	struct espintcp_ctx *ctx = espintcp_getctx(sk);
 
-	if (skb_queue_len(&ctx->out_queue) >= READ_ONCE(netdev_max_backlog))
+	if (skb_queue_len(&ctx->out_queue) >=
+	    READ_ONCE(net_hotdata.max_backlog))
 		return -ENOBUFS;
 
 	__skb_queue_tail(&ctx->out_queue, skb);
diff --git a/net/xfrm/xfrm_input.c b/net/xfrm/xfrm_input.c
index bd4ce21d76d7551d8f0a4a11f4b75705a7f634c5..161f535c8b9495b01f6d9689e14c40e5c0885968 100644
--- a/net/xfrm/xfrm_input.c
+++ b/net/xfrm/xfrm_input.c
@@ -21,6 +21,7 @@
 #include <net/ip_tunnels.h>
 #include <net/ip6_tunnel.h>
 #include <net/dst_metadata.h>
+#include <net/hotdata.h>
 
 #include "xfrm_inout.h"
 
@@ -764,7 +765,7 @@ int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
 
 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
 
-	if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
+	if (skb_queue_len(&trans->queue) >= READ_ONCE(net_hotdata.max_backlog))
 		return -ENOBUFS;
 
 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 06/18] net: move ip_packet_offload and ipv6_packet_offload to net_hotdata

[Eric Dumazet]

These structures are used in GRO and GSO paths.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h  |  7 +++++++
 net/ipv4/af_inet.c     | 18 +++++++++---------
 net/ipv6/ip6_offload.c | 18 +++++++++---------
 3 files changed, 25 insertions(+), 18 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index dc50b200a94b6b935cd79d8e0406a61209fdc68e..ec752d234c5ed4b9c110c9e61c143fe8fc27089e 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -3,9 +3,16 @@
 #define _NET_HOTDATA_H
 
 #include <linux/types.h>
+#include <linux/netdevice.h>
 
 /* Read mostly data used in network fast paths. */
 struct net_hotdata {
+#if IS_ENABLED(CONFIG_INET)
+	struct packet_offload	ip_packet_offload;
+#endif
+#if IS_ENABLED(CONFIG_IPV6)
+	struct packet_offload	ipv6_packet_offload;
+#endif
 	struct list_head	offload_base;
 	struct list_head	ptype_all;
 	int			gro_normal_batch;
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index 5daebdcbca326aa1fc042e1e1ff1e82a18bd283d..08dda6955562ea6b89e02b8299b03ab52b342f27 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -1904,14 +1904,6 @@ static int ipv4_proc_init(void);
  *	IP protocol layer initialiser
  */
 
-static struct packet_offload ip_packet_offload __read_mostly = {
-	.type = cpu_to_be16(ETH_P_IP),
-	.callbacks = {
-		.gso_segment = inet_gso_segment,
-		.gro_receive = inet_gro_receive,
-		.gro_complete = inet_gro_complete,
-	},
-};
 
 static const struct net_offload ipip_offload = {
 	.callbacks = {
@@ -1938,7 +1930,15 @@ static int __init ipv4_offload_init(void)
 	if (ipip_offload_init() < 0)
 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
 
-	dev_add_offload(&ip_packet_offload);
+	net_hotdata.ip_packet_offload = (struct packet_offload) {
+		.type = cpu_to_be16(ETH_P_IP),
+		.callbacks = {
+			.gso_segment = inet_gso_segment,
+			.gro_receive = inet_gro_receive,
+			.gro_complete = inet_gro_complete,
+		},
+	};
+	dev_add_offload(&net_hotdata.ip_packet_offload);
 	return 0;
 }
 
diff --git a/net/ipv6/ip6_offload.c b/net/ipv6/ip6_offload.c
index cca64c7809bee9a0360cbfab6a645d3f8d2ffea3..b41e35af69ea2835aa47d6ca01d9b109d4092462 100644
--- a/net/ipv6/ip6_offload.c
+++ b/net/ipv6/ip6_offload.c
@@ -419,14 +419,6 @@ static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 	return inet_gro_complete(skb, nhoff);
 }
 
-static struct packet_offload ipv6_packet_offload __read_mostly = {
-	.type = cpu_to_be16(ETH_P_IPV6),
-	.callbacks = {
-		.gso_segment = ipv6_gso_segment,
-		.gro_receive = ipv6_gro_receive,
-		.gro_complete = ipv6_gro_complete,
-	},
-};
 
 static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
 				       netdev_features_t features)
@@ -486,7 +478,15 @@ static int __init ipv6_offload_init(void)
 	if (ipv6_exthdrs_offload_init() < 0)
 		pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 
-	dev_add_offload(&ipv6_packet_offload);
+	net_hotdata.ipv6_packet_offload = (struct packet_offload) {
+		.type = cpu_to_be16(ETH_P_IPV6),
+		.callbacks = {
+			.gso_segment = ipv6_gso_segment,
+			.gro_receive = ipv6_gro_receive,
+			.gro_complete = ipv6_gro_complete,
+		},
+	};
+	dev_add_offload(&net_hotdata.ipv6_packet_offload);
 
 	inet_add_offload(&sit_offload, IPPROTO_IPV6);
 	inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 07/18] net: move tcpv4_offload and tcpv6_offload to net_hotdata

[Eric Dumazet]

These are used in TCP fast paths.

Move them into net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h    |  3 +++
 net/ipv4/tcp_offload.c   | 17 ++++++++---------
 net/ipv6/tcpv6_offload.c | 16 ++++++++--------
 3 files changed, 19 insertions(+), 17 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index ec752d234c5ed4b9c110c9e61c143fe8fc27089e..a4a8df3bc0dea1b4c9589bd70f7ac457ebc5b634 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -4,14 +4,17 @@
 
 #include <linux/types.h>
 #include <linux/netdevice.h>
+#include <net/protocol.h>
 
 /* Read mostly data used in network fast paths. */
 struct net_hotdata {
 #if IS_ENABLED(CONFIG_INET)
 	struct packet_offload	ip_packet_offload;
+	struct net_offload	tcpv4_offload;
 #endif
 #if IS_ENABLED(CONFIG_IPV6)
 	struct packet_offload	ipv6_packet_offload;
+	struct net_offload	tcpv6_offload;
 #endif
 	struct list_head	offload_base;
 	struct list_head	ptype_all;
diff --git a/net/ipv4/tcp_offload.c b/net/ipv4/tcp_offload.c
index b955ab3b236d965a38054efa004fe12f03074c70..ebe4722bb0204433936e69724879779141288789 100644
--- a/net/ipv4/tcp_offload.c
+++ b/net/ipv4/tcp_offload.c
@@ -345,15 +345,14 @@ INDIRECT_CALLABLE_SCOPE int tcp4_gro_complete(struct sk_buff *skb, int thoff)
 	return 0;
 }
 
-static const struct net_offload tcpv4_offload = {
-	.callbacks = {
-		.gso_segment	=	tcp4_gso_segment,
-		.gro_receive	=	tcp4_gro_receive,
-		.gro_complete	=	tcp4_gro_complete,
-	},
-};
-
 int __init tcpv4_offload_init(void)
 {
-	return inet_add_offload(&tcpv4_offload, IPPROTO_TCP);
+	net_hotdata.tcpv4_offload = (struct net_offload) {
+		.callbacks = {
+			.gso_segment	=	tcp4_gso_segment,
+			.gro_receive	=	tcp4_gro_receive,
+			.gro_complete	=	tcp4_gro_complete,
+		},
+	};
+	return inet_add_offload(&net_hotdata.tcpv4_offload, IPPROTO_TCP);
 }
diff --git a/net/ipv6/tcpv6_offload.c b/net/ipv6/tcpv6_offload.c
index bf0c957e4b5eaaabc0ac3a7e55c7de6608cec156..4b07d1e6c952957419924c83c5d3ac0e9d0b2565 100644
--- a/net/ipv6/tcpv6_offload.c
+++ b/net/ipv6/tcpv6_offload.c
@@ -66,15 +66,15 @@ static struct sk_buff *tcp6_gso_segment(struct sk_buff *skb,
 
 	return tcp_gso_segment(skb, features);
 }
-static const struct net_offload tcpv6_offload = {
-	.callbacks = {
-		.gso_segment	=	tcp6_gso_segment,
-		.gro_receive	=	tcp6_gro_receive,
-		.gro_complete	=	tcp6_gro_complete,
-	},
-};
 
 int __init tcpv6_offload_init(void)
 {
-	return inet6_add_offload(&tcpv6_offload, IPPROTO_TCP);
+	net_hotdata.tcpv6_offload = (struct net_offload) {
+		.callbacks = {
+			.gso_segment	=	tcp6_gso_segment,
+			.gro_receive	=	tcp6_gro_receive,
+			.gro_complete	=	tcp6_gro_complete,
+		},
+	};
+	return inet6_add_offload(&net_hotdata.tcpv6_offload, IPPROTO_TCP);
 }
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 08/18] net: move dev_tx_weight to net_hotdata

[Eric Dumazet]

dev_tx_weight is used in tx fast path.

Move it to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/netdevice.h  | 1 -
 include/net/hotdata.h      | 1 +
 net/core/dev.c             | 1 -
 net/core/hotdata.c         | 1 +
 net/core/sysctl_net_core.c | 2 +-
 net/sched/sch_generic.c    | 3 ++-
 6 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 044d6f5b2ace3e2decd4296e01c8d3e200c6c7dc..c2a735edc44be95fbd4bbd1e234d883582bfde10 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -4788,7 +4788,6 @@ void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s);
 
 extern int		dev_rx_weight;
-extern int		dev_tx_weight;
 
 enum {
 	NESTED_SYNC_IMM_BIT,
diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index a4a8df3bc0dea1b4c9589bd70f7ac457ebc5b634..2b0eb6b7f1f2c9b1273b07e06ba0b5c12a2934bf 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -23,6 +23,7 @@ struct net_hotdata {
 	int			netdev_budget_usecs;
 	int			tstamp_prequeue;
 	int			max_backlog;
+	int			dev_tx_weight;
 };
 
 extern struct net_hotdata net_hotdata;
diff --git a/net/core/dev.c b/net/core/dev.c
index 1b112c4db983c2d7cd280bc8c2ebc621ea3c6145..3f8d451f42fb9ba621f1ea19e784c7f0be0bf2d8 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4409,7 +4409,6 @@ 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 */
 int dev_rx_weight __read_mostly = 64;
-int dev_tx_weight __read_mostly = 64;
 
 /* Called with irq disabled */
 static inline void ____napi_schedule(struct softnet_data *sd,
diff --git a/net/core/hotdata.c b/net/core/hotdata.c
index 35ed5a83ecc7ebda513fe4fafc596e053f0252c5..ec8c3b48e8fea57491c5870055cffb44c779db44 100644
--- a/net/core/hotdata.c
+++ b/net/core/hotdata.c
@@ -16,5 +16,6 @@ struct net_hotdata net_hotdata __cacheline_aligned = {
 
 	.tstamp_prequeue = 1,
 	.max_backlog = 1000,
+	.dev_tx_weight = 64,
 };
 EXPORT_SYMBOL(net_hotdata);
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index 8eaeeb289914258f90cf940e906d5c6be0cc0cd6..a30016a8660e09db89b3153e4103c185a800a2ef 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -302,7 +302,7 @@ static int proc_do_dev_weight(struct ctl_table *table, int write,
 	if (!ret && write) {
 		weight = READ_ONCE(weight_p);
 		WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
-		WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
+		WRITE_ONCE(net_hotdata.dev_tx_weight, weight * dev_weight_tx_bias);
 	}
 	mutex_unlock(&dev_weight_mutex);
 
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 9b3e9262040b6ef6516752c558c8997bf4054123..ff5336493777507242320d7e9214c637663f0734 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -27,6 +27,7 @@
 #include <net/sch_generic.h>
 #include <net/pkt_sched.h>
 #include <net/dst.h>
+#include <net/hotdata.h>
 #include <trace/events/qdisc.h>
 #include <trace/events/net.h>
 #include <net/xfrm.h>
@@ -409,7 +410,7 @@ static inline bool qdisc_restart(struct Qdisc *q, int *packets)
 
 void __qdisc_run(struct Qdisc *q)
 {
-	int quota = READ_ONCE(dev_tx_weight);
+	int quota = READ_ONCE(net_hotdata.dev_tx_weight);
 	int packets;
 
 	while (qdisc_restart(q, &packets)) {
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 09/18] net: move dev_rx_weight to net_hotdata

[Eric Dumazet]

dev_rx_weight is read from process_backlog().

Move it to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/netdevice.h  | 2 --
 include/net/hotdata.h      | 1 +
 net/core/dev.c             | 3 +--
 net/core/hotdata.c         | 1 +
 net/core/sysctl_net_core.c | 2 +-
 5 files changed, 4 insertions(+), 5 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index c2a735edc44be95fbd4bbd1e234d883582bfde10..d3d4d1052ecb0bc704e9bb40fffbea85cfb6ab03 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -4787,8 +4787,6 @@ void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
 			   const struct pcpu_sw_netstats __percpu *netstats);
 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s);
 
-extern int		dev_rx_weight;
-
 enum {
 	NESTED_SYNC_IMM_BIT,
 	NESTED_SYNC_TODO_BIT,
diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 2b0eb6b7f1f2c9b1273b07e06ba0b5c12a2934bf..f45085eddbc997c2a9686b5b4de9da4cc8c4ede8 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -24,6 +24,7 @@ struct net_hotdata {
 	int			tstamp_prequeue;
 	int			max_backlog;
 	int			dev_tx_weight;
+	int			dev_rx_weight;
 };
 
 extern struct net_hotdata net_hotdata;
diff --git a/net/core/dev.c b/net/core/dev.c
index 3f8d451f42fb9ba621f1ea19e784c7f0be0bf2d8..26676dbd6c8594371a4d30e0dd95d72342e226ca 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4408,7 +4408,6 @@ unsigned int sysctl_skb_defer_max __read_mostly = 64;
 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 */
-int dev_rx_weight __read_mostly = 64;
 
 /* Called with irq disabled */
 static inline void ____napi_schedule(struct softnet_data *sd,
@@ -5978,7 +5977,7 @@ static int process_backlog(struct napi_struct *napi, int quota)
 		net_rps_action_and_irq_enable(sd);
 	}
 
-	napi->weight = READ_ONCE(dev_rx_weight);
+	napi->weight = READ_ONCE(net_hotdata.dev_rx_weight);
 	while (again) {
 		struct sk_buff *skb;
 
diff --git a/net/core/hotdata.c b/net/core/hotdata.c
index ec8c3b48e8fea57491c5870055cffb44c779db44..c8a7a451c18a383d091e413a510d84d163473f2f 100644
--- a/net/core/hotdata.c
+++ b/net/core/hotdata.c
@@ -17,5 +17,6 @@ struct net_hotdata net_hotdata __cacheline_aligned = {
 	.tstamp_prequeue = 1,
 	.max_backlog = 1000,
 	.dev_tx_weight = 64,
+	.dev_rx_weight = 64,
 };
 EXPORT_SYMBOL(net_hotdata);
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index a30016a8660e09db89b3153e4103c185a800a2ef..8a4c698dad9c97636ca9cebfad925d4220e98f2a 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -301,7 +301,7 @@ static int proc_do_dev_weight(struct ctl_table *table, int write,
 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
 	if (!ret && write) {
 		weight = READ_ONCE(weight_p);
-		WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
+		WRITE_ONCE(net_hotdata.dev_rx_weight, weight * dev_weight_rx_bias);
 		WRITE_ONCE(net_hotdata.dev_tx_weight, weight * dev_weight_tx_bias);
 	}
 	mutex_unlock(&dev_weight_mutex);
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 10/18] net: move skbuff_cache(s) to net_hotdata

[Eric Dumazet]

skbuff_cache, skbuff_fclone_cache and skb_small_head_cache
are used in rx/tx fast paths.

Move them to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/skbuff.h |  1 -
 include/net/hotdata.h  |  3 +++
 kernel/bpf/cpumap.c    |  4 +++-
 net/bpf/test_run.c     |  4 +++-
 net/core/skbuff.c      | 44 +++++++++++++++++++-----------------------
 net/core/xdp.c         |  5 +++--
 6 files changed, 32 insertions(+), 29 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 3013355b63f5f29acfcb0331bfd1c3308aba034d..d0508f90bed50ca66850b29383033e11985dad34 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -1271,7 +1271,6 @@ static inline void consume_skb(struct sk_buff *skb)
 
 void __consume_stateless_skb(struct sk_buff *skb);
 void  __kfree_skb(struct sk_buff *skb);
-extern struct kmem_cache *skbuff_cache;
 
 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index f45085eddbc997c2a9686b5b4de9da4cc8c4ede8..728aee1cf07c8f0d85873d912248a99e148f84b1 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -18,6 +18,9 @@ struct net_hotdata {
 #endif
 	struct list_head	offload_base;
 	struct list_head	ptype_all;
+	struct kmem_cache	*skbuff_cache;
+	struct kmem_cache	*skbuff_fclone_cache;
+	struct kmem_cache	*skb_small_head_cache;
 	int			gro_normal_batch;
 	int			netdev_budget;
 	int			netdev_budget_usecs;
diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
index 8a0bb80fe48a344964e4029fec5e895ee512babf..33f4246cdf0de8232bfbba6a4300f158db3ca9d5 100644
--- a/kernel/bpf/cpumap.c
+++ b/kernel/bpf/cpumap.c
@@ -24,6 +24,7 @@
 #include <linux/filter.h>
 #include <linux/ptr_ring.h>
 #include <net/xdp.h>
+#include <net/hotdata.h>
 
 #include <linux/sched.h>
 #include <linux/workqueue.h>
@@ -326,7 +327,8 @@ static int cpu_map_kthread_run(void *data)
 		/* Support running another XDP prog on this CPU */
 		nframes = cpu_map_bpf_prog_run(rcpu, frames, xdp_n, &stats, &list);
 		if (nframes) {
-			m = kmem_cache_alloc_bulk(skbuff_cache, gfp, nframes, skbs);
+			m = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache,
+						  gfp, nframes, skbs);
 			if (unlikely(m == 0)) {
 				for (i = 0; i < nframes; i++)
 					skbs[i] = NULL; /* effect: xdp_return_frame */
diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c
index 5535f9adc6589d028138026ebff24286098dc46d..61efeadaff8db0529a62d074f441e2a7c35eaa9e 100644
--- a/net/bpf/test_run.c
+++ b/net/bpf/test_run.c
@@ -12,6 +12,7 @@
 #include <linux/rcupdate_trace.h>
 #include <linux/sched/signal.h>
 #include <net/bpf_sk_storage.h>
+#include <net/hotdata.h>
 #include <net/sock.h>
 #include <net/tcp.h>
 #include <net/net_namespace.h>
@@ -254,7 +255,8 @@ static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
 	int i, n;
 	LIST_HEAD(list);
 
-	n = kmem_cache_alloc_bulk(skbuff_cache, gfp, nframes, (void **)skbs);
+	n = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache, gfp, nframes,
+				  (void **)skbs);
 	if (unlikely(n == 0)) {
 		for (i = 0; i < nframes; i++)
 			xdp_return_frame(frames[i]);
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 43d7fc150acc9263760162c3f5778fa0a646bcc4..766219011aeaf5782df9d624696d273ef6c1577c 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -69,6 +69,7 @@
 #include <net/sock.h>
 #include <net/checksum.h>
 #include <net/gso.h>
+#include <net/hotdata.h>
 #include <net/ip6_checksum.h>
 #include <net/xfrm.h>
 #include <net/mpls.h>
@@ -88,15 +89,10 @@
 #include "dev.h"
 #include "sock_destructor.h"
 
-struct kmem_cache *skbuff_cache __ro_after_init;
-static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
 #ifdef CONFIG_SKB_EXTENSIONS
 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
 #endif
 
-
-static struct kmem_cache *skb_small_head_cache __ro_after_init;
-
 #define SKB_SMALL_HEAD_SIZE SKB_HEAD_ALIGN(MAX_TCP_HEADER)
 
 /* We want SKB_SMALL_HEAD_CACHE_SIZE to not be a power of two.
@@ -349,7 +345,7 @@ static struct sk_buff *napi_skb_cache_get(void)
 	struct sk_buff *skb;
 
 	if (unlikely(!nc->skb_count)) {
-		nc->skb_count = kmem_cache_alloc_bulk(skbuff_cache,
+		nc->skb_count = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache,
 						      GFP_ATOMIC,
 						      NAPI_SKB_CACHE_BULK,
 						      nc->skb_cache);
@@ -358,7 +354,7 @@ static struct sk_buff *napi_skb_cache_get(void)
 	}
 
 	skb = nc->skb_cache[--nc->skb_count];
-	kasan_mempool_unpoison_object(skb, kmem_cache_size(skbuff_cache));
+	kasan_mempool_unpoison_object(skb, kmem_cache_size(net_hotdata.skbuff_cache));
 
 	return skb;
 }
@@ -416,7 +412,7 @@ struct sk_buff *slab_build_skb(void *data)
 	struct sk_buff *skb;
 	unsigned int size;
 
-	skb = kmem_cache_alloc(skbuff_cache, GFP_ATOMIC);
+	skb = kmem_cache_alloc(net_hotdata.skbuff_cache, GFP_ATOMIC);
 	if (unlikely(!skb))
 		return NULL;
 
@@ -467,7 +463,7 @@ struct sk_buff *__build_skb(void *data, unsigned int frag_size)
 {
 	struct sk_buff *skb;
 
-	skb = kmem_cache_alloc(skbuff_cache, GFP_ATOMIC);
+	skb = kmem_cache_alloc(net_hotdata.skbuff_cache, GFP_ATOMIC);
 	if (unlikely(!skb))
 		return NULL;
 
@@ -578,7 +574,7 @@ static void *kmalloc_reserve(unsigned int *size, gfp_t flags, int node,
 	obj_size = SKB_HEAD_ALIGN(*size);
 	if (obj_size <= SKB_SMALL_HEAD_CACHE_SIZE &&
 	    !(flags & KMALLOC_NOT_NORMAL_BITS)) {
-		obj = kmem_cache_alloc_node(skb_small_head_cache,
+		obj = kmem_cache_alloc_node(net_hotdata.skb_small_head_cache,
 				flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
 				node);
 		*size = SKB_SMALL_HEAD_CACHE_SIZE;
@@ -586,7 +582,7 @@ static void *kmalloc_reserve(unsigned int *size, gfp_t flags, int node,
 			goto out;
 		/* Try again but now we are using pfmemalloc reserves */
 		ret_pfmemalloc = true;
-		obj = kmem_cache_alloc_node(skb_small_head_cache, flags, node);
+		obj = kmem_cache_alloc_node(net_hotdata.skb_small_head_cache, flags, node);
 		goto out;
 	}
 
@@ -649,7 +645,7 @@ struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 	u8 *data;
 
 	cache = (flags & SKB_ALLOC_FCLONE)
-		? skbuff_fclone_cache : skbuff_cache;
+		? net_hotdata.skbuff_fclone_cache : net_hotdata.skbuff_cache;
 
 	if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
 		gfp_mask |= __GFP_MEMALLOC;
@@ -1095,7 +1091,7 @@ static int skb_pp_frag_ref(struct sk_buff *skb)
 static void skb_kfree_head(void *head, unsigned int end_offset)
 {
 	if (end_offset == SKB_SMALL_HEAD_HEADROOM)
-		kmem_cache_free(skb_small_head_cache, head);
+		kmem_cache_free(net_hotdata.skb_small_head_cache, head);
 	else
 		kfree(head);
 }
@@ -1162,7 +1158,7 @@ static void kfree_skbmem(struct sk_buff *skb)
 
 	switch (skb->fclone) {
 	case SKB_FCLONE_UNAVAILABLE:
-		kmem_cache_free(skbuff_cache, skb);
+		kmem_cache_free(net_hotdata.skbuff_cache, skb);
 		return;
 
 	case SKB_FCLONE_ORIG:
@@ -1183,7 +1179,7 @@ static void kfree_skbmem(struct sk_buff *skb)
 	if (!refcount_dec_and_test(&fclones->fclone_ref))
 		return;
 fastpath:
-	kmem_cache_free(skbuff_fclone_cache, fclones);
+	kmem_cache_free(net_hotdata.skbuff_fclone_cache, fclones);
 }
 
 void skb_release_head_state(struct sk_buff *skb)
@@ -1280,7 +1276,7 @@ static void kfree_skb_add_bulk(struct sk_buff *skb,
 	sa->skb_array[sa->skb_count++] = skb;
 
 	if (unlikely(sa->skb_count == KFREE_SKB_BULK_SIZE)) {
-		kmem_cache_free_bulk(skbuff_cache, KFREE_SKB_BULK_SIZE,
+		kmem_cache_free_bulk(net_hotdata.skbuff_cache, KFREE_SKB_BULK_SIZE,
 				     sa->skb_array);
 		sa->skb_count = 0;
 	}
@@ -1305,7 +1301,7 @@ kfree_skb_list_reason(struct sk_buff *segs, enum skb_drop_reason reason)
 	}
 
 	if (sa.skb_count)
-		kmem_cache_free_bulk(skbuff_cache, sa.skb_count, sa.skb_array);
+		kmem_cache_free_bulk(net_hotdata.skbuff_cache, sa.skb_count, sa.skb_array);
 }
 EXPORT_SYMBOL(kfree_skb_list_reason);
 
@@ -1467,9 +1463,9 @@ static void napi_skb_cache_put(struct sk_buff *skb)
 	if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
 		for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
 			kasan_mempool_unpoison_object(nc->skb_cache[i],
-						kmem_cache_size(skbuff_cache));
+						kmem_cache_size(net_hotdata.skbuff_cache));
 
-		kmem_cache_free_bulk(skbuff_cache, NAPI_SKB_CACHE_HALF,
+		kmem_cache_free_bulk(net_hotdata.skbuff_cache, NAPI_SKB_CACHE_HALF,
 				     nc->skb_cache + NAPI_SKB_CACHE_HALF);
 		nc->skb_count = NAPI_SKB_CACHE_HALF;
 	}
@@ -2066,7 +2062,7 @@ struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
 		if (skb_pfmemalloc(skb))
 			gfp_mask |= __GFP_MEMALLOC;
 
-		n = kmem_cache_alloc(skbuff_cache, gfp_mask);
+		n = kmem_cache_alloc(net_hotdata.skbuff_cache, gfp_mask);
 		if (!n)
 			return NULL;
 
@@ -5005,7 +5001,7 @@ static void skb_extensions_init(void) {}
 
 void __init skb_init(void)
 {
-	skbuff_cache = kmem_cache_create_usercopy("skbuff_head_cache",
+	net_hotdata.skbuff_cache = kmem_cache_create_usercopy("skbuff_head_cache",
 					      sizeof(struct sk_buff),
 					      0,
 					      SLAB_HWCACHE_ALIGN|SLAB_PANIC|
@@ -5013,7 +5009,7 @@ void __init skb_init(void)
 					      offsetof(struct sk_buff, cb),
 					      sizeof_field(struct sk_buff, cb),
 					      NULL);
-	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
+	net_hotdata.skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
 						sizeof(struct sk_buff_fclones),
 						0,
 						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
@@ -5022,7 +5018,7 @@ void __init skb_init(void)
 	 * struct skb_shared_info is located at the end of skb->head,
 	 * and should not be copied to/from user.
 	 */
-	skb_small_head_cache = kmem_cache_create_usercopy("skbuff_small_head",
+	net_hotdata.skb_small_head_cache = kmem_cache_create_usercopy("skbuff_small_head",
 						SKB_SMALL_HEAD_CACHE_SIZE,
 						0,
 						SLAB_HWCACHE_ALIGN | SLAB_PANIC,
@@ -5895,7 +5891,7 @@ void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
 {
 	if (head_stolen) {
 		skb_release_head_state(skb);
-		kmem_cache_free(skbuff_cache, skb);
+		kmem_cache_free(net_hotdata.skbuff_cache, skb);
 	} else {
 		__kfree_skb(skb);
 	}
diff --git a/net/core/xdp.c b/net/core/xdp.c
index 0e3709a29175baf3ee90fdc9cc2d97d861da11a3..41693154e426f78656cc18fc97ca7e82e648270b 100644
--- a/net/core/xdp.c
+++ b/net/core/xdp.c
@@ -16,6 +16,7 @@
 #include <linux/bug.h>
 #include <net/page_pool/helpers.h>
 
+#include <net/hotdata.h>
 #include <net/xdp.h>
 #include <net/xdp_priv.h> /* struct xdp_mem_allocator */
 #include <trace/events/xdp.h>
@@ -589,7 +590,7 @@ EXPORT_SYMBOL_GPL(xdp_warn);
 
 int xdp_alloc_skb_bulk(void **skbs, int n_skb, gfp_t gfp)
 {
-	n_skb = kmem_cache_alloc_bulk(skbuff_cache, gfp, n_skb, skbs);
+	n_skb = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache, gfp, n_skb, skbs);
 	if (unlikely(!n_skb))
 		return -ENOMEM;
 
@@ -658,7 +659,7 @@ struct sk_buff *xdp_build_skb_from_frame(struct xdp_frame *xdpf,
 {
 	struct sk_buff *skb;
 
-	skb = kmem_cache_alloc(skbuff_cache, GFP_ATOMIC);
+	skb = kmem_cache_alloc(net_hotdata.skbuff_cache, GFP_ATOMIC);
 	if (unlikely(!skb))
 		return NULL;
 
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 11/18] udp: move udpv4_offload and udpv6_offload to net_hotdata

[Eric Dumazet]

These structures are used in GRO and GSO paths.
Move them to net_hodata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h  |  2 ++
 net/ipv4/udp_offload.c | 17 ++++++++---------
 net/ipv6/udp_offload.c | 21 ++++++++++-----------
 3 files changed, 20 insertions(+), 20 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 728aee1cf07c8f0d85873d912248a99e148f84b1..086114e4d3bcc9184304b024ed93c9024888fcf1 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -11,10 +11,12 @@ struct net_hotdata {
 #if IS_ENABLED(CONFIG_INET)
 	struct packet_offload	ip_packet_offload;
 	struct net_offload	tcpv4_offload;
+	struct net_offload 	udpv4_offload;
 #endif
 #if IS_ENABLED(CONFIG_IPV6)
 	struct packet_offload	ipv6_packet_offload;
 	struct net_offload	tcpv6_offload;
+	struct net_offload	udpv6_offload;
 #endif
 	struct list_head	offload_base;
 	struct list_head	ptype_all;
diff --git a/net/ipv4/udp_offload.c b/net/ipv4/udp_offload.c
index 6c95d28d0c4a7e56d587a986113b3711f8de964c..b9880743765c6c24c28bea095f16f0cf091664ce 100644
--- a/net/ipv4/udp_offload.c
+++ b/net/ipv4/udp_offload.c
@@ -737,15 +737,14 @@ INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
 	return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
 }
 
-static const struct net_offload udpv4_offload = {
-	.callbacks = {
-		.gso_segment = udp4_ufo_fragment,
-		.gro_receive  =	udp4_gro_receive,
-		.gro_complete =	udp4_gro_complete,
-	},
-};
-
 int __init udpv4_offload_init(void)
 {
-	return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
+	net_hotdata.udpv4_offload = (struct net_offload) {
+		.callbacks = {
+			.gso_segment = udp4_ufo_fragment,
+			.gro_receive  =	udp4_gro_receive,
+			.gro_complete =	udp4_gro_complete,
+		},
+	};
+	return inet_add_offload(&net_hotdata.udpv4_offload, IPPROTO_UDP);
 }
diff --git a/net/ipv6/udp_offload.c b/net/ipv6/udp_offload.c
index 6b95ba241ebe2af7e5f2760d8a9c1d78f08579c5..312bcaeea96fb78ac488124cf7795aa834392c64 100644
--- a/net/ipv6/udp_offload.c
+++ b/net/ipv6/udp_offload.c
@@ -192,20 +192,19 @@ INDIRECT_CALLABLE_SCOPE int udp6_gro_complete(struct sk_buff *skb, int nhoff)
 	return udp_gro_complete(skb, nhoff, udp6_lib_lookup_skb);
 }
 
-static const struct net_offload udpv6_offload = {
-	.callbacks = {
-		.gso_segment	=	udp6_ufo_fragment,
-		.gro_receive	=	udp6_gro_receive,
-		.gro_complete	=	udp6_gro_complete,
-	},
-};
-
-int udpv6_offload_init(void)
+int __init udpv6_offload_init(void)
 {
-	return inet6_add_offload(&udpv6_offload, IPPROTO_UDP);
+	net_hotdata.udpv6_offload = (struct net_offload) {
+		.callbacks = {
+			.gso_segment	=	udp6_ufo_fragment,
+			.gro_receive	=	udp6_gro_receive,
+			.gro_complete	=	udp6_gro_complete,
+		},
+	};
+	return inet6_add_offload(&net_hotdata.udpv6_offload, IPPROTO_UDP);
 }
 
 int udpv6_offload_exit(void)
 {
-	return inet6_del_offload(&udpv6_offload, IPPROTO_UDP);
+	return inet6_del_offload(&net_hotdata.udpv6_offload, IPPROTO_UDP);
 }
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 12/18] ipv6: move tcpv6_protocol and udpv6_protocol to net_hotdata

[Eric Dumazet]

These structures are read in rx path, move them to net_hotdata
for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h |  2 ++
 net/ipv6/tcp_ipv6.c   | 17 +++++++++--------
 net/ipv6/udp.c        | 16 ++++++++--------
 3 files changed, 19 insertions(+), 16 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 086114e4d3bcc9184304b024ed93c9024888fcf1..97617acb75e1f2141fe7170d93c06f9813c725a3 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -16,7 +16,9 @@ struct net_hotdata {
 #if IS_ENABLED(CONFIG_IPV6)
 	struct packet_offload	ipv6_packet_offload;
 	struct net_offload	tcpv6_offload;
+	struct inet6_protocol	tcpv6_protocol;
 	struct net_offload	udpv6_offload;
+	struct inet6_protocol	udpv6_protocol;
 #endif
 	struct list_head	offload_base;
 	struct list_head	ptype_all;
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index f677f0fa51968d00c3571d55ae7850742387f2d1..3f4cba49e9ee6520987993dcea082e6065b4688b 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -58,6 +58,7 @@
 #include <net/timewait_sock.h>
 #include <net/inet_common.h>
 #include <net/secure_seq.h>
+#include <net/hotdata.h>
 #include <net/busy_poll.h>
 
 #include <linux/proc_fs.h>
@@ -2367,11 +2368,6 @@ struct proto tcpv6_prot = {
 };
 EXPORT_SYMBOL_GPL(tcpv6_prot);
 
-static const struct inet6_protocol tcpv6_protocol = {
-	.handler	=	tcp_v6_rcv,
-	.err_handler	=	tcp_v6_err,
-	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
-};
 
 static struct inet_protosw tcpv6_protosw = {
 	.type		=	SOCK_STREAM,
@@ -2408,7 +2404,12 @@ int __init tcpv6_init(void)
 {
 	int ret;
 
-	ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
+	net_hotdata.tcpv6_protocol = (struct inet6_protocol) {
+		.handler     = tcp_v6_rcv,
+		.err_handler = tcp_v6_err,
+		.flags	     = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
+	};
+	ret = inet6_add_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
 	if (ret)
 		goto out;
 
@@ -2433,7 +2434,7 @@ int __init tcpv6_init(void)
 out_tcpv6_protosw:
 	inet6_unregister_protosw(&tcpv6_protosw);
 out_tcpv6_protocol:
-	inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
+	inet6_del_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
 	goto out;
 }
 
@@ -2441,5 +2442,5 @@ void tcpv6_exit(void)
 {
 	unregister_pernet_subsys(&tcpv6_net_ops);
 	inet6_unregister_protosw(&tcpv6_protosw);
-	inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
+	inet6_del_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
 }
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 3f2249b4cd5f6a594dd9768e29f20f0d9a57faed..97d86909aabb6588d0bba901f6df1f23a4f2e561 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -1702,11 +1702,6 @@ int udpv6_getsockopt(struct sock *sk, int level, int optname,
 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
 }
 
-static const struct inet6_protocol udpv6_protocol = {
-	.handler	=	udpv6_rcv,
-	.err_handler	=	udpv6_err,
-	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
-};
 
 /* ------------------------------------------------------------------------ */
 #ifdef CONFIG_PROC_FS
@@ -1803,7 +1798,12 @@ int __init udpv6_init(void)
 {
 	int ret;
 
-	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
+	net_hotdata.udpv6_protocol = (struct inet6_protocol) {
+		.handler     = udpv6_rcv,
+		.err_handler = udpv6_err,
+		.flags	     = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
+	};
+	ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
 	if (ret)
 		goto out;
 
@@ -1814,12 +1814,12 @@ int __init udpv6_init(void)
 	return ret;
 
 out_udpv6_protocol:
-	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
+	inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
 	goto out;
 }
 
 void udpv6_exit(void)
 {
 	inet6_unregister_protosw(&udpv6_protosw);
-	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
+	inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
 }
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 13/18] inet: move tcp_protocol and udp_protocol to net_hotdata

[Eric Dumazet]

These structures are read in rx path, move them to net_hotdata
for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h |  2 ++
 net/ipv4/af_inet.c    | 30 +++++++++++++++---------------
 2 files changed, 17 insertions(+), 15 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 97617acb75e1f2141fe7170d93c06f9813c725a3..4d1cb3c29d4edfbc18cf56c370a1e04e5fcb1cbd 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -11,7 +11,9 @@ struct net_hotdata {
 #if IS_ENABLED(CONFIG_INET)
 	struct packet_offload	ip_packet_offload;
 	struct net_offload	tcpv4_offload;
+	struct net_protocol	tcp_protocol;
 	struct net_offload 	udpv4_offload;
+	struct net_protocol	udp_protocol;
 #endif
 #if IS_ENABLED(CONFIG_IPV6)
 	struct packet_offload	ipv6_packet_offload;
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index 08dda6955562ea6b89e02b8299b03ab52b342f27..6f1cfd176e7b84f23d8a5e505bf8e13b2b755f06 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -1751,19 +1751,6 @@ static const struct net_protocol igmp_protocol = {
 };
 #endif
 
-static const struct net_protocol tcp_protocol = {
-	.handler	=	tcp_v4_rcv,
-	.err_handler	=	tcp_v4_err,
-	.no_policy	=	1,
-	.icmp_strict_tag_validation = 1,
-};
-
-static const struct net_protocol udp_protocol = {
-	.handler =	udp_rcv,
-	.err_handler =	udp_err,
-	.no_policy =	1,
-};
-
 static const struct net_protocol icmp_protocol = {
 	.handler =	icmp_rcv,
 	.err_handler =	icmp_err,
@@ -1992,9 +1979,22 @@ static int __init inet_init(void)
 
 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
-	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
+
+	net_hotdata.udp_protocol = (struct net_protocol) {
+		.handler =	udp_rcv,
+		.err_handler =	udp_err,
+		.no_policy =	1,
+	};
+	if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
-	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
+
+	net_hotdata.tcp_protocol = (struct net_protocol) {
+		.handler	=	tcp_v4_rcv,
+		.err_handler	=	tcp_v4_err,
+		.no_policy	=	1,
+		.icmp_strict_tag_validation = 1,
+	};
+	if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
 #ifdef CONFIG_IP_MULTICAST
 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 14/18] inet: move inet_ehash_secret and udp_ehash_secret into net_hotdata

[Eric Dumazet]

"struct net_protocol" has a 32bit hole in 32bit arches.

Use it to store the 32bit secret used by UDP and TCP,
to increase cache locality in rx path.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h      | 3 +++
 include/net/protocol.h     | 1 +
 net/ipv4/inet_hashtables.c | 3 +--
 net/ipv4/udp.c             | 2 --
 4 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 4d1cb3c29d4edfbc18cf56c370a1e04e5fcb1cbd..4dd86be99116ff83f2524461a006565b2ade2241 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -36,6 +36,9 @@ struct net_hotdata {
 	int			dev_rx_weight;
 };
 
+#define inet_ehash_secret	net_hotdata.tcp_protocol.secret
+#define udp_ehash_secret	net_hotdata.udp_protocol.secret
+
 extern struct net_hotdata net_hotdata;
 
 #endif /* _NET_HOTDATA_H */
diff --git a/include/net/protocol.h b/include/net/protocol.h
index 6aef8cb11cc8c409e5f7a2519f5e747be584c8d5..3ff26e66735cec98b08eadb1c3f129e011923cb0 100644
--- a/include/net/protocol.h
+++ b/include/net/protocol.h
@@ -46,6 +46,7 @@ struct net_protocol {
 				 * socket lookup?
 				 */
 				icmp_strict_tag_validation:1;
+	u32			secret;
 };
 
 #if IS_ENABLED(CONFIG_IPV6)
diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
index 308ff34002ea6b5e0620004f65ffd833087afbc1..7498af3201647fd937bf8177f04c200bea178a79 100644
--- a/net/ipv4/inet_hashtables.c
+++ b/net/ipv4/inet_hashtables.c
@@ -24,6 +24,7 @@
 #include <net/inet6_hashtables.h>
 #endif
 #include <net/secure_seq.h>
+#include <net/hotdata.h>
 #include <net/ip.h>
 #include <net/tcp.h>
 #include <net/sock_reuseport.h>
@@ -32,8 +33,6 @@ u32 inet_ehashfn(const struct net *net, const __be32 laddr,
 		 const __u16 lport, const __be32 faddr,
 		 const __be16 fport)
 {
-	static u32 inet_ehash_secret __read_mostly;
-
 	net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
 
 	return __inet_ehashfn(laddr, lport, faddr, fport,
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index a8acea17b4e5344d022ae8f8eb674d1a36f8035a..2beabf5b2d8628f1fed69a0212c57bd3cd638483 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -411,8 +411,6 @@ INDIRECT_CALLABLE_SCOPE
 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
 		const __be32 faddr, const __be16 fport)
 {
-	static u32 udp_ehash_secret __read_mostly;
-
 	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 
 	return __inet_ehashfn(laddr, lport, faddr, fport,
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 15/18] ipv6: move inet6_ehash_secret and udp6_ehash_secret into net_hotdata

[Eric Dumazet]

"struct inet6_protocol" has a 32bit hole in 32bit arches.

Use it to store the 32bit secret used by UDP and TCP,
to increase cache locality in rx path.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h       | 2 ++
 include/net/protocol.h      | 1 +
 net/ipv6/inet6_hashtables.c | 2 +-
 net/ipv6/udp.c              | 1 -
 4 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 4dd86be99116ff83f2524461a006565b2ade2241..e4dac2f859efd421b975a61360536af949046d0e 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -38,6 +38,8 @@ struct net_hotdata {
 
 #define inet_ehash_secret	net_hotdata.tcp_protocol.secret
 #define udp_ehash_secret	net_hotdata.udp_protocol.secret
+#define inet6_ehash_secret	net_hotdata.tcpv6_protocol.secret
+#define udp6_ehash_secret	net_hotdata.udpv6_protocol.secret
 
 extern struct net_hotdata net_hotdata;
 
diff --git a/include/net/protocol.h b/include/net/protocol.h
index 3ff26e66735cec98b08eadb1c3f129e011923cb0..213649d2ab098edc4eb9f44a5403540887c2b8ef 100644
--- a/include/net/protocol.h
+++ b/include/net/protocol.h
@@ -60,6 +60,7 @@ struct inet6_protocol {
 			       __be32 info);
 
 	unsigned int	flags;	/* INET6_PROTO_xxx */
+	u32		secret;
 };
 
 #define INET6_PROTO_NOPOLICY	0x1
diff --git a/net/ipv6/inet6_hashtables.c b/net/ipv6/inet6_hashtables.c
index b0e8d278e8a9b794d0001efdc0f43716f9a34f8f..0fee97f3166cf8326b3b714ac6bde48ca5188cec 100644
--- a/net/ipv6/inet6_hashtables.c
+++ b/net/ipv6/inet6_hashtables.c
@@ -14,6 +14,7 @@
 #include <linux/random.h>
 
 #include <net/addrconf.h>
+#include <net/hotdata.h>
 #include <net/inet_connection_sock.h>
 #include <net/inet_hashtables.h>
 #include <net/inet6_hashtables.h>
@@ -25,7 +26,6 @@ u32 inet6_ehashfn(const struct net *net,
 		  const struct in6_addr *laddr, const u16 lport,
 		  const struct in6_addr *faddr, const __be16 fport)
 {
-	static u32 inet6_ehash_secret __read_mostly;
 	static u32 ipv6_hash_secret __read_mostly;
 
 	u32 lhash, fhash;
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 97d86909aabb6588d0bba901f6df1f23a4f2e561..1e1c67a51675e8534a953a6d4d63211388d95ca9 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -79,7 +79,6 @@ u32 udp6_ehashfn(const struct net *net,
 		 const struct in6_addr *faddr,
 		 const __be16 fport)
 {
-	static u32 udp6_ehash_secret __read_mostly;
 	static u32 udp_ipv6_hash_secret __read_mostly;
 
 	u32 lhash, fhash;
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 16/18] ipv6: move tcp_ipv6_hash_secret and udp_ipv6_hash_secret to net_hotdata

[Eric Dumazet]

Use a 32bit hole in "struct net_offload" to store
the remaining 32bit secrets used by TCPv6 and UDPv6.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h       | 2 ++
 include/net/protocol.h      | 1 +
 net/ipv6/inet6_hashtables.c | 6 ++----
 net/ipv6/udp.c              | 2 --
 4 files changed, 5 insertions(+), 6 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index e4dac2f859efd421b975a61360536af949046d0e..0a0a9106b40030f56b04c1e48083c13498ce0939 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -39,7 +39,9 @@ struct net_hotdata {
 #define inet_ehash_secret	net_hotdata.tcp_protocol.secret
 #define udp_ehash_secret	net_hotdata.udp_protocol.secret
 #define inet6_ehash_secret	net_hotdata.tcpv6_protocol.secret
+#define tcp_ipv6_hash_secret	net_hotdata.tcpv6_offload.secret
 #define udp6_ehash_secret	net_hotdata.udpv6_protocol.secret
+#define udp_ipv6_hash_secret	net_hotdata.udpv6_offload.secret
 
 extern struct net_hotdata net_hotdata;
 
diff --git a/include/net/protocol.h b/include/net/protocol.h
index 213649d2ab098edc4eb9f44a5403540887c2b8ef..b2499f88f8f8199de28555d0cbd6b4523325205b 100644
--- a/include/net/protocol.h
+++ b/include/net/protocol.h
@@ -70,6 +70,7 @@ struct inet6_protocol {
 struct net_offload {
 	struct offload_callbacks callbacks;
 	unsigned int		 flags;	/* Flags used by IPv6 for now */
+	u32			 secret;
 };
 /* This should be set for any extension header which is compatible with GSO. */
 #define INET6_PROTO_GSO_EXTHDR	0x1
diff --git a/net/ipv6/inet6_hashtables.c b/net/ipv6/inet6_hashtables.c
index 0fee97f3166cf8326b3b714ac6bde48ca5188cec..2e81383b663b71b95719a295fd9629f1193e4225 100644
--- a/net/ipv6/inet6_hashtables.c
+++ b/net/ipv6/inet6_hashtables.c
@@ -26,15 +26,13 @@ u32 inet6_ehashfn(const struct net *net,
 		  const struct in6_addr *laddr, const u16 lport,
 		  const struct in6_addr *faddr, const __be16 fport)
 {
-	static u32 ipv6_hash_secret __read_mostly;
-
 	u32 lhash, fhash;
 
 	net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
-	net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
+	net_get_random_once(&tcp_ipv6_hash_secret, sizeof(tcp_ipv6_hash_secret));
 
 	lhash = (__force u32)laddr->s6_addr32[3];
-	fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
+	fhash = __ipv6_addr_jhash(faddr, tcp_ipv6_hash_secret);
 
 	return __inet6_ehashfn(lhash, lport, fhash, fport,
 			       inet6_ehash_secret + net_hash_mix(net));
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 1e1c67a51675e8534a953a6d4d63211388d95ca9..80ad8f436b179d7279cbbd5108c3494a9773c0d8 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -79,8 +79,6 @@ u32 udp6_ehashfn(const struct net *net,
 		 const struct in6_addr *faddr,
 		 const __be16 fport)
 {
-	static u32 udp_ipv6_hash_secret __read_mostly;
-
 	u32 lhash, fhash;
 
 	net_get_random_once(&udp6_ehash_secret,
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 17/18] net: introduce include/net/rps.h

[Eric Dumazet]

Move RPS related structures and helpers from include/linux/netdevice.h
and include/net/sock.h to a new include file.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 drivers/net/ethernet/intel/ice/ice_arfs.c     |   1 +
 .../net/ethernet/mellanox/mlx4/en_netdev.c    |   1 +
 .../net/ethernet/mellanox/mlx5/core/en_arfs.c |   1 +
 drivers/net/ethernet/sfc/rx_common.c          |   1 +
 drivers/net/ethernet/sfc/siena/rx_common.c    |   1 +
 drivers/net/tun.c                             |   1 +
 include/linux/netdevice.h                     |  82 -----------
 include/net/rps.h                             | 127 ++++++++++++++++++
 include/net/sock.h                            |  35 -----
 net/core/dev.c                                |   1 +
 net/core/net-sysfs.c                          |   1 +
 net/core/sysctl_net_core.c                    |   1 +
 net/ipv4/af_inet.c                            |   1 +
 net/ipv4/tcp.c                                |   1 +
 net/ipv6/af_inet6.c                           |   1 +
 net/sctp/socket.c                             |   1 +
 16 files changed, 140 insertions(+), 117 deletions(-)
 create mode 100644 include/net/rps.h

diff --git a/drivers/net/ethernet/intel/ice/ice_arfs.c b/drivers/net/ethernet/intel/ice/ice_arfs.c
index cca0e753f38ff92be24f8e8fc8963e2bf2416cfa..7cee365cc7d167865511c8b0dab08520814d4ec1 100644
--- a/drivers/net/ethernet/intel/ice/ice_arfs.c
+++ b/drivers/net/ethernet/intel/ice/ice_arfs.c
@@ -2,6 +2,7 @@
 /* Copyright (C) 2018-2020, Intel Corporation. */
 
 #include "ice.h"
+#include <net/rps.h>
 
 /**
  * ice_is_arfs_active - helper to check is aRFS is active
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
index d7da62cda821f70af530f95d23b63afd64423219..5d3fde63b273922f52f2466c7b339f3c9908e705 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
@@ -42,6 +42,7 @@
 #include <net/ip.h>
 #include <net/vxlan.h>
 #include <net/devlink.h>
+#include <net/rps.h>
 
 #include <linux/mlx4/driver.h>
 #include <linux/mlx4/device.h>
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c b/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
index e66f486faafe1a6b0cfc75f0f11b2e957b040842..c7f542d0b8f08c635a6fad868a364a8f5f91ba8c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
@@ -34,6 +34,7 @@
 #include <linux/mlx5/fs.h>
 #include <linux/ip.h>
 #include <linux/ipv6.h>
+#include <net/rps.h>
 #include "en.h"
 
 #define ARFS_HASH_SHIFT BITS_PER_BYTE
diff --git a/drivers/net/ethernet/sfc/rx_common.c b/drivers/net/ethernet/sfc/rx_common.c
index fac227d372db4c832a52a62144ff9c6c89995335..dcd901eccfc8f1f9c68ee391de548a2c1602ab94 100644
--- a/drivers/net/ethernet/sfc/rx_common.c
+++ b/drivers/net/ethernet/sfc/rx_common.c
@@ -11,6 +11,7 @@
 #include "net_driver.h"
 #include <linux/module.h>
 #include <linux/iommu.h>
+#include <net/rps.h>
 #include "efx.h"
 #include "nic.h"
 #include "rx_common.h"
diff --git a/drivers/net/ethernet/sfc/siena/rx_common.c b/drivers/net/ethernet/sfc/siena/rx_common.c
index 4579f43484c3675963de1ad8c7753a200c680fe1..219fb358a646d399dbd3c66cc34039f70bbd7341 100644
--- a/drivers/net/ethernet/sfc/siena/rx_common.c
+++ b/drivers/net/ethernet/sfc/siena/rx_common.c
@@ -11,6 +11,7 @@
 #include "net_driver.h"
 #include <linux/module.h>
 #include <linux/iommu.h>
+#include <net/rps.h>
 #include "efx.h"
 #include "nic.h"
 #include "rx_common.h"
diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index be37235af55d2d282b162adcd9d0e599cd5e6b3e..c1be5cec0f350d61c604d5d2776e047aa48c90b8 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -78,6 +78,7 @@
 #include <net/ax25.h>
 #include <net/rose.h>
 #include <net/6lowpan.h>
+#include <net/rps.h>
 
 #include <linux/uaccess.h>
 #include <linux/proc_fs.h>
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index d3d4d1052ecb0bc704e9bb40fffbea85cfb6ab03..aaacf86df56ab480485b34b19fecebfd00c34c59 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -227,12 +227,6 @@ struct net_device_core_stats {
 #include <linux/cache.h>
 #include <linux/skbuff.h>
 
-#ifdef CONFIG_RPS
-#include <linux/static_key.h>
-extern struct static_key_false rps_needed;
-extern struct static_key_false rfs_needed;
-#endif
-
 struct neighbour;
 struct neigh_parms;
 struct sk_buff;
@@ -732,86 +726,10 @@ static inline void netdev_queue_numa_node_write(struct netdev_queue *q, int node
 #endif
 }
 
-#ifdef CONFIG_RPS
-/*
- * This structure holds an RPS map which can be of variable length.  The
- * map is an array of CPUs.
- */
-struct rps_map {
-	unsigned int len;
-	struct rcu_head rcu;
-	u16 cpus[];
-};
-#define RPS_MAP_SIZE(_num) (sizeof(struct rps_map) + ((_num) * sizeof(u16)))
-
-/*
- * The rps_dev_flow structure contains the mapping of a flow to a CPU, the
- * tail pointer for that CPU's input queue at the time of last enqueue, and
- * a hardware filter index.
- */
-struct rps_dev_flow {
-	u16 cpu;
-	u16 filter;
-	unsigned int last_qtail;
-};
-#define RPS_NO_FILTER 0xffff
-
-/*
- * The rps_dev_flow_table structure contains a table of flow mappings.
- */
-struct rps_dev_flow_table {
-	unsigned int mask;
-	struct rcu_head rcu;
-	struct rps_dev_flow flows[];
-};
-#define RPS_DEV_FLOW_TABLE_SIZE(_num) (sizeof(struct rps_dev_flow_table) + \
-    ((_num) * sizeof(struct rps_dev_flow)))
-
-/*
- * The rps_sock_flow_table contains mappings of flows to the last CPU
- * on which they were processed by the application (set in recvmsg).
- * Each entry is a 32bit value. Upper part is the high-order bits
- * of flow hash, lower part is CPU number.
- * rps_cpu_mask is used to partition the space, depending on number of
- * possible CPUs : rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1
- * For example, if 64 CPUs are possible, rps_cpu_mask = 0x3f,
- * meaning we use 32-6=26 bits for the hash.
- */
-struct rps_sock_flow_table {
-	u32	mask;
-
-	u32	ents[] ____cacheline_aligned_in_smp;
-};
-#define	RPS_SOCK_FLOW_TABLE_SIZE(_num) (offsetof(struct rps_sock_flow_table, ents[_num]))
-
-#define RPS_NO_CPU 0xffff
-
-extern u32 rps_cpu_mask;
-extern struct rps_sock_flow_table __rcu *rps_sock_flow_table;
-
-static inline void rps_record_sock_flow(struct rps_sock_flow_table *table,
-					u32 hash)
-{
-	if (table && hash) {
-		unsigned int index = hash & table->mask;
-		u32 val = hash & ~rps_cpu_mask;
-
-		/* We only give a hint, preemption can change CPU under us */
-		val |= raw_smp_processor_id();
-
-		/* The following WRITE_ONCE() is paired with the READ_ONCE()
-		 * here, and another one in get_rps_cpu().
-		 */
-		if (READ_ONCE(table->ents[index]) != val)
-			WRITE_ONCE(table->ents[index], val);
-	}
-}
-
 #ifdef CONFIG_RFS_ACCEL
 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, u32 flow_id,
 			 u16 filter_id);
 #endif
-#endif /* CONFIG_RPS */
 
 /* XPS map type and offset of the xps map within net_device->xps_maps[]. */
 enum xps_map_type {
diff --git a/include/net/rps.h b/include/net/rps.h
new file mode 100644
index 0000000000000000000000000000000000000000..6081d817d2458b7b34036d87fbdef3fa6dc914ea
--- /dev/null
+++ b/include/net/rps.h
@@ -0,0 +1,127 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef _NET_RPS_H
+#define _NET_RPS_H
+
+#include <linux/types.h>
+#include <linux/static_key.h>
+#include <net/sock.h>
+
+#ifdef CONFIG_RPS
+
+extern struct static_key_false rps_needed;
+extern struct static_key_false rfs_needed;
+
+/*
+ * This structure holds an RPS map which can be of variable length.  The
+ * map is an array of CPUs.
+ */
+struct rps_map {
+	unsigned int	len;
+	struct rcu_head	rcu;
+	u16		cpus[];
+};
+#define RPS_MAP_SIZE(_num) (sizeof(struct rps_map) + ((_num) * sizeof(u16)))
+
+/*
+ * The rps_dev_flow structure contains the mapping of a flow to a CPU, the
+ * tail pointer for that CPU's input queue at the time of last enqueue, and
+ * a hardware filter index.
+ */
+struct rps_dev_flow {
+	u16		cpu;
+	u16		filter;
+	unsigned int	last_qtail;
+};
+#define RPS_NO_FILTER 0xffff
+
+/*
+ * The rps_dev_flow_table structure contains a table of flow mappings.
+ */
+struct rps_dev_flow_table {
+	unsigned int		mask;
+	struct rcu_head		rcu;
+	struct rps_dev_flow	flows[];
+};
+#define RPS_DEV_FLOW_TABLE_SIZE(_num) (sizeof(struct rps_dev_flow_table) + \
+    ((_num) * sizeof(struct rps_dev_flow)))
+
+/*
+ * The rps_sock_flow_table contains mappings of flows to the last CPU
+ * on which they were processed by the application (set in recvmsg).
+ * Each entry is a 32bit value. Upper part is the high-order bits
+ * of flow hash, lower part is CPU number.
+ * rps_cpu_mask is used to partition the space, depending on number of
+ * possible CPUs : rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1
+ * For example, if 64 CPUs are possible, rps_cpu_mask = 0x3f,
+ * meaning we use 32-6=26 bits for the hash.
+ */
+struct rps_sock_flow_table {
+	u32	mask;
+
+	u32	ents[] ____cacheline_aligned_in_smp;
+};
+#define	RPS_SOCK_FLOW_TABLE_SIZE(_num) (offsetof(struct rps_sock_flow_table, ents[_num]))
+
+#define RPS_NO_CPU 0xffff
+
+extern u32 rps_cpu_mask;
+extern struct rps_sock_flow_table __rcu *rps_sock_flow_table;
+
+static inline void rps_record_sock_flow(struct rps_sock_flow_table *table,
+					u32 hash)
+{
+	unsigned int index = hash & table->mask;
+	u32 val = hash & ~rps_cpu_mask;
+
+	/* We only give a hint, preemption can change CPU under us */
+	val |= raw_smp_processor_id();
+
+	/* The following WRITE_ONCE() is paired with the READ_ONCE()
+	 * here, and another one in get_rps_cpu().
+	 */
+	if (READ_ONCE(table->ents[index]) != val)
+		WRITE_ONCE(table->ents[index], val);
+}
+
+#endif /* CONFIG_RPS */
+
+static inline void sock_rps_record_flow_hash(__u32 hash)
+{
+#ifdef CONFIG_RPS
+	struct rps_sock_flow_table *sock_flow_table;
+
+	if (!hash)
+		return;
+	rcu_read_lock();
+	sock_flow_table = rcu_dereference(rps_sock_flow_table);
+	if (sock_flow_table)
+		rps_record_sock_flow(sock_flow_table, hash);
+	rcu_read_unlock();
+#endif
+}
+
+static inline void sock_rps_record_flow(const struct sock *sk)
+{
+#ifdef CONFIG_RPS
+	if (static_branch_unlikely(&rfs_needed)) {
+		/* Reading sk->sk_rxhash might incur an expensive cache line
+		 * miss.
+		 *
+		 * TCP_ESTABLISHED does cover almost all states where RFS
+		 * might be useful, and is cheaper [1] than testing :
+		 *	IPv4: inet_sk(sk)->inet_daddr
+		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
+		 * OR	an additional socket flag
+		 * [1] : sk_state and sk_prot are in the same cache line.
+		 */
+		if (sk->sk_state == TCP_ESTABLISHED) {
+			/* This READ_ONCE() is paired with the WRITE_ONCE()
+			 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
+			 */
+			sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
+		}
+	}
+#endif
+}
+
+#endif /* _NET_RPS_H */
diff --git a/include/net/sock.h b/include/net/sock.h
index 09a0cde8bf52286167f3b9e9b32f632f3d1b5487..b5e00702acc1f037df7eb8ad085d00e0b18079a8 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -1117,41 +1117,6 @@ static inline void sk_incoming_cpu_update(struct sock *sk)
 		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
 }
 
-static inline void sock_rps_record_flow_hash(__u32 hash)
-{
-#ifdef CONFIG_RPS
-	struct rps_sock_flow_table *sock_flow_table;
-
-	rcu_read_lock();
-	sock_flow_table = rcu_dereference(rps_sock_flow_table);
-	rps_record_sock_flow(sock_flow_table, hash);
-	rcu_read_unlock();
-#endif
-}
-
-static inline void sock_rps_record_flow(const struct sock *sk)
-{
-#ifdef CONFIG_RPS
-	if (static_branch_unlikely(&rfs_needed)) {
-		/* Reading sk->sk_rxhash might incur an expensive cache line
-		 * miss.
-		 *
-		 * TCP_ESTABLISHED does cover almost all states where RFS
-		 * might be useful, and is cheaper [1] than testing :
-		 *	IPv4: inet_sk(sk)->inet_daddr
-		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
-		 * OR	an additional socket flag
-		 * [1] : sk_state and sk_prot are in the same cache line.
-		 */
-		if (sk->sk_state == TCP_ESTABLISHED) {
-			/* This READ_ONCE() is paired with the WRITE_ONCE()
-			 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
-			 */
-			sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
-		}
-	}
-#endif
-}
 
 static inline void sock_rps_save_rxhash(struct sock *sk,
 					const struct sk_buff *skb)
diff --git a/net/core/dev.c b/net/core/dev.c
index 26676dbd6c8594371a4d30e0dd95d72342e226ca..e9f24a31ae121f713e6ef5a530a65218bbb457e8 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -155,6 +155,7 @@
 #include <net/netdev_rx_queue.h>
 #include <net/page_pool/types.h>
 #include <net/page_pool/helpers.h>
+#include <net/rps.h>
 
 #include "dev.h"
 #include "net-sysfs.h"
diff --git a/net/core/net-sysfs.c b/net/core/net-sysfs.c
index af238026ac3c6d81267bf0b53a7b240ee9ba32b1..5560083774b1ad4e2612637990dfa3f80f750c83 100644
--- a/net/core/net-sysfs.c
+++ b/net/core/net-sysfs.c
@@ -24,6 +24,7 @@
 #include <linux/of_net.h>
 #include <linux/cpu.h>
 #include <net/netdev_rx_queue.h>
+#include <net/rps.h>
 
 #include "dev.h"
 #include "net-sysfs.h"
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index 8a4c698dad9c97636ca9cebfad925d4220e98f2a..4b93e27404e83a5b3afaa23ebd18cf55b1fdc6e8 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -24,6 +24,7 @@
 #include <net/busy_poll.h>
 #include <net/pkt_sched.h>
 #include <net/hotdata.h>
+#include <net/rps.h>
 
 #include "dev.h"
 
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index 6f1cfd176e7b84f23d8a5e505bf8e13b2b755f06..55bd72997b31063b7baad350fdcff40e938aecb8 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -119,6 +119,7 @@
 #endif
 #include <net/l3mdev.h>
 #include <net/compat.h>
+#include <net/rps.h>
 
 #include <trace/events/sock.h>
 
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 7e1b848398d04f2da2a91c3af97b1e2e3895b8ee..c5b83875411aee6037d0afd060eb9d3e16f8b6b0 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -279,6 +279,7 @@
 #include <linux/uaccess.h>
 #include <asm/ioctls.h>
 #include <net/busy_poll.h>
+#include <net/rps.h>
 
 /* Track pending CMSGs. */
 enum {
diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
index b90d46533cdcc1ffb61ca483e6f67ab358ede55c..8041dc181bd42e5e1af2d9e9fe6af50057e5b58f 100644
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -64,6 +64,7 @@
 #include <net/xfrm.h>
 #include <net/ioam6.h>
 #include <net/rawv6.h>
+#include <net/rps.h>
 
 #include <linux/uaccess.h>
 #include <linux/mroute6.h>
diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index 6b9fcdb0952a0fe599ae5d1d1cc6fa9557a3a3bc..c67679a41044fc8e801d175b235249f2c8b99dc0 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c
@@ -67,6 +67,7 @@
 #include <net/sctp/sctp.h>
 #include <net/sctp/sm.h>
 #include <net/sctp/stream_sched.h>
+#include <net/rps.h>
 
 /* Forward declarations for internal helper functions. */
 static bool sctp_writeable(const struct sock *sk);
-- 
2.44.0.278.ge034bb2e1d-goog

[PATCH net-next 18/18] net: move rps_sock_flow_table to net_hotdata

[Eric Dumazet]

rps_sock_flow_table and rps_cpu_mask are used in fast path.

Move them to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/hotdata.h      |  4 ++++
 include/net/rps.h          |  8 +++-----
 net/core/dev.c             | 12 +++---------
 net/core/sysctl_net_core.c |  9 ++++++---
 4 files changed, 16 insertions(+), 17 deletions(-)

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 0a0a9106b40030f56b04c1e48083c13498ce0939..7bb6e46aec8f19deff42112041feb47724cdd538 100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -27,6 +27,10 @@ struct net_hotdata {
 	struct kmem_cache	*skbuff_cache;
 	struct kmem_cache	*skbuff_fclone_cache;
 	struct kmem_cache	*skb_small_head_cache;
+#ifdef CONFIG_RPS
+	struct rps_sock_flow_table __rcu *rps_sock_flow_table;
+	u32			rps_cpu_mask;
+#endif
 	int			gro_normal_batch;
 	int			netdev_budget;
 	int			netdev_budget_usecs;
diff --git a/include/net/rps.h b/include/net/rps.h
index 6081d817d2458b7b34036d87fbdef3fa6dc914ea..7660243e905b92651a41292e04caf72c5f12f26e 100644
--- a/include/net/rps.h
+++ b/include/net/rps.h
@@ -5,6 +5,7 @@
 #include <linux/types.h>
 #include <linux/static_key.h>
 #include <net/sock.h>
+#include <net/hotdata.h>
 
 #ifdef CONFIG_RPS
 
@@ -64,14 +65,11 @@ struct rps_sock_flow_table {
 
 #define RPS_NO_CPU 0xffff
 
-extern u32 rps_cpu_mask;
-extern struct rps_sock_flow_table __rcu *rps_sock_flow_table;
-
 static inline void rps_record_sock_flow(struct rps_sock_flow_table *table,
 					u32 hash)
 {
 	unsigned int index = hash & table->mask;
-	u32 val = hash & ~rps_cpu_mask;
+	u32 val = hash & ~net_hotdata.rps_cpu_mask;
 
 	/* We only give a hint, preemption can change CPU under us */
 	val |= raw_smp_processor_id();
@@ -93,7 +91,7 @@ static inline void sock_rps_record_flow_hash(__u32 hash)
 	if (!hash)
 		return;
 	rcu_read_lock();
-	sock_flow_table = rcu_dereference(rps_sock_flow_table);
+	sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table);
 	if (sock_flow_table)
 		rps_record_sock_flow(sock_flow_table, hash);
 	rcu_read_unlock();
diff --git a/net/core/dev.c b/net/core/dev.c
index e9f24a31ae121f713e6ef5a530a65218bbb457e8..9f54d8e61ec0f4a4ad1824e333f89ad08c5ec431 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4450,12 +4450,6 @@ static inline void ____napi_schedule(struct softnet_data *sd,
 
 #ifdef CONFIG_RPS
 
-/* One global table that all flow-based protocols share. */
-struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
-EXPORT_SYMBOL(rps_sock_flow_table);
-u32 rps_cpu_mask __read_mostly;
-EXPORT_SYMBOL(rps_cpu_mask);
-
 struct static_key_false rps_needed __read_mostly;
 EXPORT_SYMBOL(rps_needed);
 struct static_key_false rfs_needed __read_mostly;
@@ -4547,7 +4541,7 @@ static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 	if (!hash)
 		goto done;
 
-	sock_flow_table = rcu_dereference(rps_sock_flow_table);
+	sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table);
 	if (flow_table && sock_flow_table) {
 		struct rps_dev_flow *rflow;
 		u32 next_cpu;
@@ -4557,10 +4551,10 @@ static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 		 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
 		 */
 		ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
-		if ((ident ^ hash) & ~rps_cpu_mask)
+		if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask)
 			goto try_rps;
 
-		next_cpu = ident & rps_cpu_mask;
+		next_cpu = ident & net_hotdata.rps_cpu_mask;
 
 		/* OK, now we know there is a match,
 		 * we can look at the local (per receive queue) flow table
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index 4b93e27404e83a5b3afaa23ebd18cf55b1fdc6e8..6973dda3abda63e0924efa4b6b7026786e8bfb4f 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -140,7 +140,8 @@ static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
 
 	mutex_lock(&sock_flow_mutex);
 
-	orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
+	orig_sock_table = rcu_dereference_protected(
+					net_hotdata.rps_sock_flow_table,
 					lockdep_is_held(&sock_flow_mutex));
 	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
 
@@ -161,7 +162,8 @@ static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
 					mutex_unlock(&sock_flow_mutex);
 					return -ENOMEM;
 				}
-				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
+				net_hotdata.rps_cpu_mask =
+					roundup_pow_of_two(nr_cpu_ids) - 1;
 				sock_table->mask = size - 1;
 			} else
 				sock_table = orig_sock_table;
@@ -172,7 +174,8 @@ static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
 			sock_table = NULL;
 
 		if (sock_table != orig_sock_table) {
-			rcu_assign_pointer(rps_sock_flow_table, sock_table);
+			rcu_assign_pointer(net_hotdata.rps_sock_flow_table,
+					   sock_table);
 			if (sock_table) {
 				static_branch_inc(&rps_needed);
 				static_branch_inc(&rfs_needed);
-- 
2.44.0.278.ge034bb2e1d-goog

Re: [PATCH net-next 00/18] net: group together hot data

[Soheil Hassas Yeganeh]

On Tue, Mar 5, 2024 at 11:04 AM Eric Dumazet <edumazet@google.com> wrote:
>
> While our recent structure reorganizations were focused
> on increasing max throughput, there is still an
> area where improvements are much needed.
>
> In many cases, a cpu handles one packet at a time,
> instead of a nice batch.
>
> Hardware interrupt.
>  -> Software interrupt.
>    -> Network/Protocol stacks.
>
> If the cpu was idle or busy in other layers,
> it has to pull many cache lines.
>
> This series adds a new net_hotdata structure, where
> some critical (and read-mostly) data used in
> rx and tx path is packed in a small number of cache lines.
>
> Synthetic benchmarks will not see much difference,
> but latency of single packet should improve.
>
> net_hodata current size on 64bit is 416 bytes,
> but might grow in the future.
>
> Also move RPS definitions to a new include file.

Acked-by: Soheil Hassas Yeganeh <soheil@google.com>

Nice series!

> Eric Dumazet (18):
>   net: introduce struct net_hotdata
>   net: move netdev_budget and netdev_budget to net_hotdata
>   net: move netdev_tstamp_prequeue into net_hotdata
>   net: move ptype_all into net_hotdata
>   net: move netdev_max_backlog to net_hotdata
>   net: move ip_packet_offload and ipv6_packet_offload to net_hotdata
>   net: move tcpv4_offload and tcpv6_offload to net_hotdata
>   net: move dev_tx_weight to net_hotdata
>   net: move dev_rx_weight to net_hotdata
>   net: move skbuff_cache(s) to net_hotdata
>   udp: move udpv4_offload and udpv6_offload to net_hotdata
>   ipv6: move tcpv6_protocol and udpv6_protocol to net_hotdata
>   inet: move tcp_protocol and udp_protocol to net_hotdata
>   inet: move inet_ehash_secret and udp_ehash_secret into net_hotdata
>   ipv6: move inet6_ehash_secret and udp6_ehash_secret into net_hotdata
>   ipv6: move tcp_ipv6_hash_secret and udp_ipv6_hash_secret to
>     net_hotdata
>   net: introduce include/net/rps.h
>   net: move rps_sock_flow_table to net_hotdata
>
>  drivers/net/ethernet/intel/ice/ice_arfs.c     |   1 +
>  .../net/ethernet/mellanox/mlx4/en_netdev.c    |   1 +
>  .../net/ethernet/mellanox/mlx5/core/en_arfs.c |   1 +
>  drivers/net/ethernet/sfc/rx_common.c          |   1 +
>  drivers/net/ethernet/sfc/siena/rx_common.c    |   1 +
>  drivers/net/tun.c                             |   1 +
>  include/linux/netdevice.h                     |  88 ------------
>  include/linux/skbuff.h                        |   1 -
>  include/net/gro.h                             |   5 +-
>  include/net/hotdata.h                         |  52 ++++++++
>  include/net/protocol.h                        |   3 +
>  include/net/rps.h                             | 125 ++++++++++++++++++
>  include/net/sock.h                            |  35 -----
>  kernel/bpf/cpumap.c                           |   4 +-
>  net/bpf/test_run.c                            |   4 +-
>  net/core/Makefile                             |   1 +
>  net/core/dev.c                                |  58 +++-----
>  net/core/dev.h                                |   3 -
>  net/core/gro.c                                |  15 +--
>  net/core/gro_cells.c                          |   3 +-
>  net/core/gso.c                                |   4 +-
>  net/core/hotdata.c                            |  22 +++
>  net/core/net-procfs.c                         |   7 +-
>  net/core/net-sysfs.c                          |   1 +
>  net/core/skbuff.c                             |  44 +++---
>  net/core/sysctl_net_core.c                    |  25 ++--
>  net/core/xdp.c                                |   5 +-
>  net/ipv4/af_inet.c                            |  49 +++----
>  net/ipv4/inet_hashtables.c                    |   3 +-
>  net/ipv4/tcp.c                                |   1 +
>  net/ipv4/tcp_offload.c                        |  17 ++-
>  net/ipv4/udp.c                                |   2 -
>  net/ipv4/udp_offload.c                        |  17 ++-
>  net/ipv6/af_inet6.c                           |   1 +
>  net/ipv6/inet6_hashtables.c                   |   8 +-
>  net/ipv6/ip6_offload.c                        |  18 +--
>  net/ipv6/tcp_ipv6.c                           |  17 +--
>  net/ipv6/tcpv6_offload.c                      |  16 +--
>  net/ipv6/udp.c                                |  19 ++-
>  net/ipv6/udp_offload.c                        |  21 ++-
>  net/sched/sch_generic.c                       |   3 +-
>  net/sctp/socket.c                             |   1 +
>  net/xfrm/espintcp.c                           |   4 +-
>  net/xfrm/xfrm_input.c                         |   3 +-
>  44 files changed, 391 insertions(+), 320 deletions(-)
>  create mode 100644 include/net/hotdata.h
>  create mode 100644 include/net/rps.h
>  create mode 100644 net/core/hotdata.c
>
> --
> 2.44.0.278.ge034bb2e1d-goog
>

Re: [PATCH net-next 00/18] net: group together hot data

[David Ahern]

On 3/5/24 9:03 AM, Eric Dumazet wrote:
> While our recent structure reorganizations were focused
> on increasing max throughput, there is still an
> area where improvements are much needed.
> 
> In many cases, a cpu handles one packet at a time,
> instead of a nice batch.
> 
> Hardware interrupt.
>  -> Software interrupt.
>    -> Network/Protocol stacks.
> 
> If the cpu was idle or busy in other layers,
> it has to pull many cache lines.
> 
> This series adds a new net_hotdata structure, where
> some critical (and read-mostly) data used in
> rx and tx path is packed in a small number of cache lines.
> 
> Synthetic benchmarks will not see much difference,
> but latency of single packet should improve.
> 
> net_hodata current size on 64bit is 416 bytes,
> but might grow in the future.
> 
> Also move RPS definitions to a new include file.
> 

Interesting patch set. For the set:

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 06/18] net: move ip_packet_offload and ipv6_packet_offload to net_hotdata

[kernel test robot]

Hi Eric,

kernel test robot noticed the following build errors:

[auto build test ERROR on net-next/main]

url:    https://github.com/intel-lab-lkp/linux/commits/Eric-Dumazet/net-introduce-struct-net_hotdata/20240306-001407
base:   net-next/main
patch link:    https://lore.kernel.org/r/20240305160413.2231423-7-edumazet%40google.com
patch subject: [PATCH net-next 06/18] net: move ip_packet_offload and ipv6_packet_offload to net_hotdata
config: hexagon-defconfig (https://download.01.org/0day-ci/archive/20240306/202403061318.QMW92UEi-lkp@intel.com/config)
compiler: clang version 19.0.0git (https://github.com/llvm/llvm-project 325f51237252e6dab8e4e1ea1fa7acbb4faee1cd)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240306/202403061318.QMW92UEi-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202403061318.QMW92UEi-lkp@intel.com/

All errors (new ones prefixed by >>):

   In file included from include/linux/highmem.h:12:
   In file included from include/linux/hardirq.h:11:
   In file included from ./arch/hexagon/include/generated/asm/hardirq.h:1:
   In file included from include/asm-generic/hardirq.h:17:
   In file included from include/linux/irq.h:20:
   In file included from include/linux/io.h:13:
   In file included from arch/hexagon/include/asm/io.h:328:
   include/asm-generic/io.h:573:61: warning: performing pointer arithmetic on a null pointer has undefined behavior [-Wnull-pointer-arithmetic]
     573 |         val = __le32_to_cpu((__le32 __force)__raw_readl(PCI_IOBASE + addr));
         |                                                         ~~~~~~~~~~ ^
   include/uapi/linux/byteorder/little_endian.h:35:51: note: expanded from macro '__le32_to_cpu'
      35 | #define __le32_to_cpu(x) ((__force __u32)(__le32)(x))
         |                                                   ^
   In file included from net/ipv6/ip6_offload.c:9:
   In file included from include/linux/netdevice.h:38:
   In file included from include/net/net_namespace.h:43:
   In file included from include/linux/skbuff.h:17:
   In file included from include/linux/bvec.h:10:
   In file included from include/linux/highmem.h:12:
   In file included from include/linux/hardirq.h:11:
   In file included from ./arch/hexagon/include/generated/asm/hardirq.h:1:
   In file included from include/asm-generic/hardirq.h:17:
   In file included from include/linux/irq.h:20:
   In file included from include/linux/io.h:13:
   In file included from arch/hexagon/include/asm/io.h:328:
   include/asm-generic/io.h:584:33: warning: performing pointer arithmetic on a null pointer has undefined behavior [-Wnull-pointer-arithmetic]
     584 |         __raw_writeb(value, PCI_IOBASE + addr);
         |                             ~~~~~~~~~~ ^
   include/asm-generic/io.h:594:59: warning: performing pointer arithmetic on a null pointer has undefined behavior [-Wnull-pointer-arithmetic]
     594 |         __raw_writew((u16 __force)cpu_to_le16(value), PCI_IOBASE + addr);
         |                                                       ~~~~~~~~~~ ^
   include/asm-generic/io.h:604:59: warning: performing pointer arithmetic on a null pointer has undefined behavior [-Wnull-pointer-arithmetic]
     604 |         __raw_writel((u32 __force)cpu_to_le32(value), PCI_IOBASE + addr);
         |                                                       ~~~~~~~~~~ ^
   In file included from net/ipv6/ip6_offload.c:9:
   In file included from include/linux/netdevice.h:45:
   In file included from include/uapi/linux/neighbour.h:6:
   In file included from include/linux/netlink.h:9:
   In file included from include/net/scm.h:9:
   In file included from include/linux/security.h:35:
   include/linux/bpf.h:736:48: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_arg_type') [-Wenum-enum-conversion]
     736 |         ARG_PTR_TO_MAP_VALUE_OR_NULL    = PTR_MAYBE_NULL | ARG_PTR_TO_MAP_VALUE,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:737:43: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_arg_type') [-Wenum-enum-conversion]
     737 |         ARG_PTR_TO_MEM_OR_NULL          = PTR_MAYBE_NULL | ARG_PTR_TO_MEM,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~
   include/linux/bpf.h:738:43: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_arg_type') [-Wenum-enum-conversion]
     738 |         ARG_PTR_TO_CTX_OR_NULL          = PTR_MAYBE_NULL | ARG_PTR_TO_CTX,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~
   include/linux/bpf.h:739:45: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_arg_type') [-Wenum-enum-conversion]
     739 |         ARG_PTR_TO_SOCKET_OR_NULL       = PTR_MAYBE_NULL | ARG_PTR_TO_SOCKET,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:740:44: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_arg_type') [-Wenum-enum-conversion]
     740 |         ARG_PTR_TO_STACK_OR_NULL        = PTR_MAYBE_NULL | ARG_PTR_TO_STACK,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~
   include/linux/bpf.h:741:45: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_arg_type') [-Wenum-enum-conversion]
     741 |         ARG_PTR_TO_BTF_ID_OR_NULL       = PTR_MAYBE_NULL | ARG_PTR_TO_BTF_ID,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:745:38: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_arg_type') [-Wenum-enum-conversion]
     745 |         ARG_PTR_TO_UNINIT_MEM           = MEM_UNINIT | ARG_PTR_TO_MEM,
         |                                           ~~~~~~~~~~ ^ ~~~~~~~~~~~~~~
   include/linux/bpf.h:747:45: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_arg_type') [-Wenum-enum-conversion]
     747 |         ARG_PTR_TO_FIXED_SIZE_MEM       = MEM_FIXED_SIZE | ARG_PTR_TO_MEM,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~
   include/linux/bpf.h:770:48: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_return_type') [-Wenum-enum-conversion]
     770 |         RET_PTR_TO_MAP_VALUE_OR_NULL    = PTR_MAYBE_NULL | RET_PTR_TO_MAP_VALUE,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:771:45: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_return_type') [-Wenum-enum-conversion]
     771 |         RET_PTR_TO_SOCKET_OR_NULL       = PTR_MAYBE_NULL | RET_PTR_TO_SOCKET,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:772:47: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_return_type') [-Wenum-enum-conversion]
     772 |         RET_PTR_TO_TCP_SOCK_OR_NULL     = PTR_MAYBE_NULL | RET_PTR_TO_TCP_SOCK,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:773:50: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_return_type') [-Wenum-enum-conversion]
     773 |         RET_PTR_TO_SOCK_COMMON_OR_NULL  = PTR_MAYBE_NULL | RET_PTR_TO_SOCK_COMMON,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:775:49: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_return_type') [-Wenum-enum-conversion]
     775 |         RET_PTR_TO_DYNPTR_MEM_OR_NULL   = PTR_MAYBE_NULL | RET_PTR_TO_MEM,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~
   include/linux/bpf.h:776:45: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_return_type') [-Wenum-enum-conversion]
     776 |         RET_PTR_TO_BTF_ID_OR_NULL       = PTR_MAYBE_NULL | RET_PTR_TO_BTF_ID,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:777:43: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_return_type') [-Wenum-enum-conversion]
     777 |         RET_PTR_TO_BTF_ID_TRUSTED       = PTR_TRUSTED    | RET_PTR_TO_BTF_ID,
         |                                           ~~~~~~~~~~~    ^ ~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:888:44: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_reg_type') [-Wenum-enum-conversion]
     888 |         PTR_TO_MAP_VALUE_OR_NULL        = PTR_MAYBE_NULL | PTR_TO_MAP_VALUE,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~
   include/linux/bpf.h:889:42: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_reg_type') [-Wenum-enum-conversion]
     889 |         PTR_TO_SOCKET_OR_NULL           = PTR_MAYBE_NULL | PTR_TO_SOCKET,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~
   include/linux/bpf.h:890:46: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_reg_type') [-Wenum-enum-conversion]
     890 |         PTR_TO_SOCK_COMMON_OR_NULL      = PTR_MAYBE_NULL | PTR_TO_SOCK_COMMON,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~
   include/linux/bpf.h:891:44: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_reg_type') [-Wenum-enum-conversion]
     891 |         PTR_TO_TCP_SOCK_OR_NULL         = PTR_MAYBE_NULL | PTR_TO_TCP_SOCK,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~
   include/linux/bpf.h:892:42: warning: bitwise operation between different enumeration types ('enum bpf_type_flag' and 'enum bpf_reg_type') [-Wenum-enum-conversion]
     892 |         PTR_TO_BTF_ID_OR_NULL           = PTR_MAYBE_NULL | PTR_TO_BTF_ID,
         |                                           ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~
>> net/ipv6/ip6_offload.c:481:14: error: no member named 'ipv6_packet_offload' in 'struct net_hotdata'; did you mean 'ip_packet_offload'?
     481 |         net_hotdata.ipv6_packet_offload = (struct packet_offload) {
         |                     ^~~~~~~~~~~~~~~~~~~
         |                     ip_packet_offload
   include/net/hotdata.h:11:24: note: 'ip_packet_offload' declared here
      11 |         struct packet_offload   ip_packet_offload;
         |                                 ^
   net/ipv6/ip6_offload.c:489:31: error: no member named 'ipv6_packet_offload' in 'struct net_hotdata'; did you mean 'ip_packet_offload'?
     489 |         dev_add_offload(&net_hotdata.ipv6_packet_offload);
         |                                      ^~~~~~~~~~~~~~~~~~~
         |                                      ip_packet_offload
   include/net/hotdata.h:11:24: note: 'ip_packet_offload' declared here
      11 |         struct packet_offload   ip_packet_offload;
         |                                 ^
   27 warnings and 2 errors generated.


vim +481 net/ipv6/ip6_offload.c

   475	
   476		if (tcpv6_offload_init() < 0)
   477			pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
   478		if (ipv6_exthdrs_offload_init() < 0)
   479			pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
   480	
 > 481		net_hotdata.ipv6_packet_offload = (struct packet_offload) {
   482			.type = cpu_to_be16(ETH_P_IPV6),
   483			.callbacks = {
   484				.gso_segment = ipv6_gso_segment,
   485				.gro_receive = ipv6_gro_receive,
   486				.gro_complete = ipv6_gro_complete,
   487			},
   488		};
   489		dev_add_offload(&net_hotdata.ipv6_packet_offload);
   490	
   491		inet_add_offload(&sit_offload, IPPROTO_IPV6);
   492		inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
   493		inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
   494	
   495		return 0;
   496	}
   497	

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH net-next 06/18] net: move ip_packet_offload and ipv6_packet_offload to net_hotdata

[Eric Dumazet]

On Wed, Mar 6, 2024 at 7:09 AM kernel test robot <lkp@intel.com> wrote:
>
> Hi Eric,
>
> kernel test robot noticed the following build errors:
>
> [auto build test ERROR on net-next/main]
>
> url:    https://github.com/intel-lab-lkp/linux/commits/Eric-Dumazet/net-introduce-struct-net_hotdata/20240306-001407
> base:   net-next/main
> patch link:    https://lore.kernel.org/r/20240305160413.2231423-7-edumazet%40google.com
> patch subject: [PATCH net-next 06/18] net: move ip_packet_offload and ipv6_packet_offload to net_hotdata
> config: hexagon-defconfig (https://download.01.org/0day-ci/archive/20240306/202403061318.QMW92UEi-lkp@intel.com/config)
> compiler: clang version 19.0.0git (https://github.com/llvm/llvm-project 325f51237252e6dab8e4e1ea1fa7acbb4faee1cd)
> reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240306/202403061318.QMW92UEi-lkp@intel.com/reproduce)
>
> If you fix the issue in a separate patch/commit (i.e. not just a new version of
> the same patch/commit), kindly add following tags
> | Reported-by: kernel test robot <lkp@intel.com>
> | Closes: https://lore.kernel.org/oe-kbuild-all/202403061318.QMW92UEi-lkp@intel.com/

Right, all the offloads need to be there if CONFIG_INET=y, regardless
of CONFIG_IPV6

I will in v2 have changes so that the cumulative diff is :

diff --git a/include/net/hotdata.h b/include/net/hotdata.h
index 7bb6e46aec8f19deff42112041feb47724cdd538..9230052a453c1b959b9a72f1d1a4878d76f243b8
100644
--- a/include/net/hotdata.h
+++ b/include/net/hotdata.h
@@ -14,13 +14,13 @@ struct net_hotdata {
        struct net_protocol     tcp_protocol;
        struct net_offload      udpv4_offload;
        struct net_protocol     udp_protocol;
-#endif
-#if IS_ENABLED(CONFIG_IPV6)
        struct packet_offload   ipv6_packet_offload;
        struct net_offload      tcpv6_offload;
+#if defined (CONFIG_IPV6)
        struct inet6_protocol   tcpv6_protocol;
-       struct net_offload      udpv6_offload;
        struct inet6_protocol   udpv6_protocol;
+#endif
+       struct net_offload      udpv6_offload;
 #endif
        struct list_head        offload_base;
        struct list_head        ptype_all;

Re: [PATCH net-next 02/18] net: move netdev_budget and netdev_budget to net_hotdata

[Andrew Lunn]

On Tue, Mar 05, 2024 at 04:03:57PM +0000, Eric Dumazet wrote:
> netdev_budget and netdev_budget are used in rx path (net_rx_action())

I think one of these is missing _usecs ?

  Andrew