提交 28af22c6 编写于 作者: J Jesper Dangaard Brouer 提交者: Jakub Kicinski

net: adjust net_device layout for cacheline usage

The current layout of net_device is not optimal for cacheline usage.

The member adj_list.lower linked list is split between cacheline 2 and 3.
The ifindex is placed together with stats (struct net_device_stats),
although most modern drivers don't update this stats member.

The members netdev_ops, mtu and hard_header_len are placed on three
different cachelines. These members are accessed for XDP redirect into
devmap, which were noticeably with perf tool. When not using the map
redirect variant (like TC-BPF does), then ifindex is also used, which is
placed on a separate fourth cacheline. These members are also accessed
during forwarding with regular network stack. The members priv_flags and
flags are on fast-path for network stack transmit path in __dev_queue_xmit
(currently located together with mtu cacheline).

This patch creates a read mostly cacheline, with the purpose of keeping the
above mentioned members on the same cacheline.

Some netdev_features_t members also becomes part of this cacheline, which is
on purpose, as function netif_skb_features() is on fast-path via
validate_xmit_skb().
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Link: https://lore.kernel.org/r/161168277983.410784.12401225493601624417.stgit@firesoulSigned-off-by: NJakub Kicinski <kuba@kernel.org>
上级 46251282
......@@ -1858,7 +1858,6 @@ struct net_device {
unsigned long mem_end;
unsigned long mem_start;
unsigned long base_addr;
int irq;
/*
* Some hardware also needs these fields (state,dev_list,
......@@ -1880,6 +1879,23 @@ struct net_device {
struct list_head lower;
} adj_list;
/* Read-mostly cache-line for fast-path access */
unsigned int flags;
unsigned int priv_flags;
const struct net_device_ops *netdev_ops;
int ifindex;
unsigned short gflags;
unsigned short hard_header_len;
/* Note : dev->mtu is often read without holding a lock.
* Writers usually hold RTNL.
* It is recommended to use READ_ONCE() to annotate the reads,
* and to use WRITE_ONCE() to annotate the writes.
*/
unsigned int mtu;
unsigned short needed_headroom;
unsigned short needed_tailroom;
netdev_features_t features;
netdev_features_t hw_features;
netdev_features_t wanted_features;
......@@ -1888,10 +1904,15 @@ struct net_device {
netdev_features_t mpls_features;
netdev_features_t gso_partial_features;
int ifindex;
unsigned int min_mtu;
unsigned int max_mtu;
unsigned short type;
unsigned char min_header_len;
unsigned char name_assign_type;
int group;
struct net_device_stats stats;
struct net_device_stats stats; /* not used by modern drivers */
atomic_long_t rx_dropped;
atomic_long_t tx_dropped;
......@@ -1905,7 +1926,6 @@ struct net_device {
const struct iw_handler_def *wireless_handlers;
struct iw_public_data *wireless_data;
#endif
const struct net_device_ops *netdev_ops;
const struct ethtool_ops *ethtool_ops;
#ifdef CONFIG_NET_L3_MASTER_DEV
const struct l3mdev_ops *l3mdev_ops;
......@@ -1924,34 +1944,12 @@ struct net_device {
const struct header_ops *header_ops;
unsigned int flags;
unsigned int priv_flags;
unsigned short gflags;
unsigned short padded;
unsigned char operstate;
unsigned char link_mode;
unsigned char if_port;
unsigned char dma;
/* Note : dev->mtu is often read without holding a lock.
* Writers usually hold RTNL.
* It is recommended to use READ_ONCE() to annotate the reads,
* and to use WRITE_ONCE() to annotate the writes.
*/
unsigned int mtu;
unsigned int min_mtu;
unsigned int max_mtu;
unsigned short type;
unsigned short hard_header_len;
unsigned char min_header_len;
unsigned char name_assign_type;
unsigned short needed_headroom;
unsigned short needed_tailroom;
/* Interface address info. */
unsigned char perm_addr[MAX_ADDR_LEN];
unsigned char addr_assign_type;
......@@ -1962,7 +1960,10 @@ struct net_device {
unsigned short neigh_priv_len;
unsigned short dev_id;
unsigned short dev_port;
unsigned short padded;
spinlock_t addr_list_lock;
int irq;
struct netdev_hw_addr_list uc;
struct netdev_hw_addr_list mc;
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册