WSL2-Linux-Kernel/include/linux/netdevice.h

2291 строка
69 KiB
C
Исходник Обычный вид История

/*
* INET An implementation of the TCP/IP protocol suite for the LINUX
* operating system. INET is implemented using the BSD Socket
* interface as the means of communication with the user level.
*
* Definitions for the Interfaces handler.
*
* Version: @(#)dev.h 1.0.10 08/12/93
*
* Authors: Ross Biro
* Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
* Corey Minyard <wf-rch!minyard@relay.EU.net>
* Donald J. Becker, <becker@cesdis.gsfc.nasa.gov>
* Alan Cox, <alan@lxorguk.ukuu.org.uk>
* Bjorn Ekwall. <bj0rn@blox.se>
* Pekka Riikonen <priikone@poseidon.pspt.fi>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*
* Moved to /usr/include/linux for NET3
*/
#ifndef _LINUX_NETDEVICE_H
#define _LINUX_NETDEVICE_H
#include <linux/if.h>
#include <linux/if_ether.h>
#include <linux/if_packet.h>
#include <linux/if_link.h>
#ifdef __KERNEL__
#include <linux/timer.h>
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
#include <linux/delay.h>
#include <linux/mm.h>
#include <asm/atomic.h>
#include <asm/cache.h>
#include <asm/byteorder.h>
#include <linux/device.h>
#include <linux/percpu.h>
#include <linux/rculist.h>
#include <linux/dmaengine.h>
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
#include <linux/workqueue.h>
#include <linux/ethtool.h>
#include <net/net_namespace.h>
#include <net/dsa.h>
#ifdef CONFIG_DCB
#include <net/dcbnl.h>
#endif
struct vlan_group;
struct netpoll_info;
/* 802.11 specific */
struct wireless_dev;
/* source back-compat hooks */
#define SET_ETHTOOL_OPS(netdev,ops) \
( (netdev)->ethtool_ops = (ops) )
#define HAVE_ALLOC_NETDEV /* feature macro: alloc_xxxdev
functions are available. */
#define HAVE_FREE_NETDEV /* free_netdev() */
#define HAVE_NETDEV_PRIV /* netdev_priv() */
/* Backlog congestion levels */
#define NET_RX_SUCCESS 0 /* keep 'em coming, baby */
#define NET_RX_DROP 1 /* packet dropped */
/*
* Transmit return codes: transmit return codes originate from three different
* namespaces:
*
* - qdisc return codes
* - driver transmit return codes
* - errno values
*
* Drivers are allowed to return any one of those in their hard_start_xmit()
* function. Real network devices commonly used with qdiscs should only return
* the driver transmit return codes though - when qdiscs are used, the actual
* transmission happens asynchronously, so the value is not propagated to
* higher layers. Virtual network devices transmit synchronously, in this case
* the driver transmit return codes are consumed by dev_queue_xmit(), all
* others are propagated to higher layers.
*/
/* qdisc ->enqueue() return codes. */
#define NET_XMIT_SUCCESS 0x00
#define NET_XMIT_DROP 0x01 /* skb dropped */
#define NET_XMIT_CN 0x02 /* congestion notification */
#define NET_XMIT_POLICED 0x03 /* skb is shot by police */
#define NET_XMIT_MASK 0x0f /* qdisc flags in net/sch_generic.h */
/* NET_XMIT_CN is special. It does not guarantee that this packet is lost. It
* indicates that the device will soon be dropping packets, or already drops
* some packets of the same priority; prompting us to send less aggressively. */
#define net_xmit_eval(e) ((e) == NET_XMIT_CN ? 0 : (e))
#define net_xmit_errno(e) ((e) != NET_XMIT_CN ? -ENOBUFS : 0)
/* Driver transmit return codes */
#define NETDEV_TX_MASK 0xf0
enum netdev_tx {
__NETDEV_TX_MIN = INT_MIN, /* make sure enum is signed */
NETDEV_TX_OK = 0x00, /* driver took care of packet */
NETDEV_TX_BUSY = 0x10, /* driver tx path was busy*/
NETDEV_TX_LOCKED = 0x20, /* driver tx lock was already taken */
};
typedef enum netdev_tx netdev_tx_t;
/*
* Current order: NETDEV_TX_MASK > NET_XMIT_MASK >= 0 is significant;
* hard_start_xmit() return < NET_XMIT_MASK means skb was consumed.
*/
static inline bool dev_xmit_complete(int rc)
{
/*
* Positive cases with an skb consumed by a driver:
* - successful transmission (rc == NETDEV_TX_OK)
* - error while transmitting (rc < 0)
* - error while queueing to a different device (rc & NET_XMIT_MASK)
*/
if (likely(rc < NET_XMIT_MASK))
return true;
return false;
}
#endif
#define MAX_ADDR_LEN 32 /* Largest hardware address length */
#ifdef __KERNEL__
/*
* Compute the worst case header length according to the protocols
* used.
*/
#if defined(CONFIG_WLAN) || defined(CONFIG_AX25) || defined(CONFIG_AX25_MODULE)
# if defined(CONFIG_MAC80211_MESH)
# define LL_MAX_HEADER 128
# else
# define LL_MAX_HEADER 96
# endif
#elif defined(CONFIG_TR) || defined(CONFIG_TR_MODULE)
# define LL_MAX_HEADER 48
#else
# define LL_MAX_HEADER 32
#endif
#if !defined(CONFIG_NET_IPIP) && !defined(CONFIG_NET_IPIP_MODULE) && \
!defined(CONFIG_NET_IPGRE) && !defined(CONFIG_NET_IPGRE_MODULE) && \
!defined(CONFIG_IPV6_SIT) && !defined(CONFIG_IPV6_SIT_MODULE) && \
!defined(CONFIG_IPV6_TUNNEL) && !defined(CONFIG_IPV6_TUNNEL_MODULE)
#define MAX_HEADER LL_MAX_HEADER
#else
#define MAX_HEADER (LL_MAX_HEADER + 48)
#endif
#endif /* __KERNEL__ */
/*
* Network device statistics. Akin to the 2.0 ether stats but
* with byte counters.
*/
struct net_device_stats {
unsigned long rx_packets; /* total packets received */
unsigned long tx_packets; /* total packets transmitted */
unsigned long rx_bytes; /* total bytes received */
unsigned long tx_bytes; /* total bytes transmitted */
unsigned long rx_errors; /* bad packets received */
unsigned long tx_errors; /* packet transmit problems */
unsigned long rx_dropped; /* no space in linux buffers */
unsigned long tx_dropped; /* no space available in linux */
unsigned long multicast; /* multicast packets received */
unsigned long collisions;
/* detailed rx_errors: */
unsigned long rx_length_errors;
unsigned long rx_over_errors; /* receiver ring buff overflow */
unsigned long rx_crc_errors; /* recved pkt with crc error */
unsigned long rx_frame_errors; /* recv'd frame alignment error */
unsigned long rx_fifo_errors; /* recv'r fifo overrun */
unsigned long rx_missed_errors; /* receiver missed packet */
/* detailed tx_errors */
unsigned long tx_aborted_errors;
unsigned long tx_carrier_errors;
unsigned long tx_fifo_errors;
unsigned long tx_heartbeat_errors;
unsigned long tx_window_errors;
/* for cslip etc */
unsigned long rx_compressed;
unsigned long tx_compressed;
};
/* Media selection options. */
enum {
IF_PORT_UNKNOWN = 0,
IF_PORT_10BASE2,
IF_PORT_10BASET,
IF_PORT_AUI,
IF_PORT_100BASET,
IF_PORT_100BASETX,
IF_PORT_100BASEFX
};
#ifdef __KERNEL__
#include <linux/cache.h>
#include <linux/skbuff.h>
struct neighbour;
struct neigh_parms;
struct sk_buff;
struct netif_rx_stats {
unsigned total;
unsigned dropped;
unsigned time_squeeze;
unsigned cpu_collision;
rps: Receive Packet Steering This patch implements software receive side packet steering (RPS). RPS distributes the load of received packet processing across multiple CPUs. Problem statement: Protocol processing done in the NAPI context for received packets is serialized per device queue and becomes a bottleneck under high packet load. This substantially limits pps that can be achieved on a single queue NIC and provides no scaling with multiple cores. This solution queues packets early on in the receive path on the backlog queues of other CPUs. This allows protocol processing (e.g. IP and TCP) to be performed on packets in parallel. For each device (or each receive queue in a multi-queue device) a mask of CPUs is set to indicate the CPUs that can process packets. A CPU is selected on a per packet basis by hashing contents of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index into the CPU mask. The IPI mechanism is used to raise networking receive softirqs between CPUs. This effectively emulates in software what a multi-queue NIC can provide, but is generic requiring no device support. Many devices now provide a hash over the 4-tuple on a per packet basis (e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash in an skb field, and that value in turn is used to index into the RPS maps. Using the HW generated hash can avoid cache misses on the packet when steering it to a remote CPU. The CPU mask is set on a per device and per queue basis in the sysfs variable /sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical bit maps for receive queues in the device (numbered by <n>). If a device does not support multi-queue, a single variable is used for the device (rx-0). Generally, we have found this technique increases pps capabilities of a single queue device with good CPU utilization. Optimal settings for the CPU mask seem to depend on architectures and cache hierarcy. Below are some results running 500 instances of netperf TCP_RR test with 1 byte req. and resp. Results show cumulative transaction rate and system CPU utilization. e1000e on 8 core Intel Without RPS: 108K tps at 33% CPU With RPS: 311K tps at 64% CPU forcedeth on 16 core AMD Without RPS: 156K tps at 15% CPU With RPS: 404K tps at 49% CPU bnx2x on 16 core AMD Without RPS 567K tps at 61% CPU (4 HW RX queues) Without RPS 738K tps at 96% CPU (8 HW RX queues) With RPS: 854K tps at 76% CPU (4 HW RX queues) Caveats: - The benefits of this patch are dependent on architecture and cache hierarchy. Tuning the masks to get best performance is probably necessary. - This patch adds overhead in the path for processing a single packet. In a lightly loaded server this overhead may eliminate the advantages of increased parallelism, and possibly cause some relative performance degradation. We have found that masks that are cache aware (share same caches with the interrupting CPU) mitigate much of this. - The RPS masks can be changed dynamically, however whenever the mask is changed this introduces the possibility of generating out of order packets. It's probably best not change the masks too frequently. Signed-off-by: Tom Herbert <therbert@google.com> include/linux/netdevice.h | 32 ++++- include/linux/skbuff.h | 3 + net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++-------- net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++- net/core/skbuff.c | 2 + 5 files changed, 538 insertions(+), 59 deletions(-) Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-03-16 11:03:29 +03:00
unsigned received_rps;
};
DECLARE_PER_CPU(struct netif_rx_stats, netdev_rx_stat);
struct dev_addr_list {
struct dev_addr_list *next;
u8 da_addr[MAX_ADDR_LEN];
u8 da_addrlen;
u8 da_synced;
int da_users;
int da_gusers;
};
/*
* We tag multicasts with these structures.
*/
#define dev_mc_list dev_addr_list
#define dmi_addr da_addr
#define dmi_addrlen da_addrlen
#define dmi_users da_users
#define dmi_gusers da_gusers
struct netdev_hw_addr {
struct list_head list;
unsigned char addr[MAX_ADDR_LEN];
unsigned char type;
#define NETDEV_HW_ADDR_T_LAN 1
#define NETDEV_HW_ADDR_T_SAN 2
#define NETDEV_HW_ADDR_T_SLAVE 3
#define NETDEV_HW_ADDR_T_UNICAST 4
int refcount;
bool synced;
struct rcu_head rcu_head;
};
struct netdev_hw_addr_list {
struct list_head list;
int count;
};
#define netdev_uc_count(dev) ((dev)->uc.count)
#define netdev_uc_empty(dev) ((dev)->uc.count == 0)
#define netdev_for_each_uc_addr(ha, dev) \
list_for_each_entry(ha, &dev->uc.list, list)
#define netdev_mc_count(dev) ((dev)->mc_count)
#define netdev_mc_empty(dev) (netdev_mc_count(dev) == 0)
#define netdev_for_each_mc_addr(mclist, dev) \
for (mclist = dev->mc_list; mclist; mclist = mclist->next)
struct hh_cache {
struct hh_cache *hh_next; /* Next entry */
atomic_t hh_refcnt; /* number of users */
/*
* We want hh_output, hh_len, hh_lock and hh_data be a in a separate
* cache line on SMP.
* They are mostly read, but hh_refcnt may be changed quite frequently,
* incurring cache line ping pongs.
*/
__be16 hh_type ____cacheline_aligned_in_smp;
/* protocol identifier, f.e ETH_P_IP
* NOTE: For VLANs, this will be the
* encapuslated type. --BLG
*/
[NET]: Pack struct hh_cache [acme@newtoy net-2.6.20]$ pahole net/ipv4/tcp.o hh_cache /* /pub/scm/linux/kernel/git/acme/net-2.6.20/include/linux/netdevice.h:190 */ struct hh_cache { struct hh_cache * hh_next; /* 0 4 */ atomic_t hh_refcnt; /* 4 4 */ __be16 hh_type; /* 8 2 */ /* XXX 2 bytes hole, try to pack */ int hh_len; /* 12 4 */ int (*hh_output)(); /* 16 4 */ rwlock_t hh_lock; /* 20 36 */ long unsigned int hh_data[24]; /* 56 96 */ }; /* size: 152, sum members: 150, holes: 1, sum holes: 2 */ [acme@newtoy net-2.6.20]$ find net -name "*.[ch]" | xargs grep 'hh_len.\+=' | sort -u net/atm/br2684.c: hh->hh_len = PADLEN + ETH_HLEN; net/ethernet/eth.c: hh->hh_len = ETH_HLEN; net/ipv4/ipconfig.c: int hh_len = LL_RESERVED_SPACE(dev); net/ipv4/ip_output.c: hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); net/ipv4/ip_output.c: int hh_len = LL_RESERVED_SPACE(dev); net/ipv4/netfilter.c: hh_len = (*pskb)->dst->dev->hard_header_len; net/ipv4/raw.c: hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); net/ipv6/ip6_output.c: hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); net/ipv6/netfilter/ip6t_REJECT.c: hh_len = (dst->dev->hard_header_len + 15)&~15; net/ipv6/raw.c: hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); [acme@newtoy net-2.6.20]$ [acme@newtoy net-2.6.20]$ find include -name "*.h" | xargs grep 'define ETH_HLEN' include/linux/if_ether.h:#define ETH_HLEN 14 /* Total octets in header. */ (((dev)->hard_header_len&~(HH_DATA_MOD - 1)) + HH_DATA_MOD) [acme@newtoy net-2.6.20]$ pahole net/ipv4/tcp.o net_device | grep hard_header_len short unsigned int hard_header_len; /* 106 2 */ [acme@newtoy net-2.6.20]$ So I think we're safe in turning hh_len an u16, end result: [acme@newtoy net-2.6.20]$ codiff -sV /tmp/tcp.o.before net/ipv4/tcp.o /pub/scm/linux/kernel/git/acme/net-2.6.20/net/ipv4/tcp.c: struct hh_cache | -4 hh_len; from: int /* 12(0) 4(0) */ to: u16 /* 10(0) 2(0) */ 1 struct changed [acme@newtoy net-2.6.20]$ Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
2006-11-27 22:58:02 +03:00
u16 hh_len; /* length of header */
int (*hh_output)(struct sk_buff *skb);
seqlock_t hh_lock;
/* cached hardware header; allow for machine alignment needs. */
#define HH_DATA_MOD 16
#define HH_DATA_OFF(__len) \
(HH_DATA_MOD - (((__len - 1) & (HH_DATA_MOD - 1)) + 1))
#define HH_DATA_ALIGN(__len) \
(((__len)+(HH_DATA_MOD-1))&~(HH_DATA_MOD - 1))
unsigned long hh_data[HH_DATA_ALIGN(LL_MAX_HEADER) / sizeof(long)];
};
/* Reserve HH_DATA_MOD byte aligned hard_header_len, but at least that much.
* Alternative is:
* dev->hard_header_len ? (dev->hard_header_len +
* (HH_DATA_MOD - 1)) & ~(HH_DATA_MOD - 1) : 0
*
* We could use other alignment values, but we must maintain the
* relationship HH alignment <= LL alignment.
*
* LL_ALLOCATED_SPACE also takes into account the tailroom the device
* may need.
*/
#define LL_RESERVED_SPACE(dev) \
((((dev)->hard_header_len+(dev)->needed_headroom)&~(HH_DATA_MOD - 1)) + HH_DATA_MOD)
#define LL_RESERVED_SPACE_EXTRA(dev,extra) \
((((dev)->hard_header_len+(dev)->needed_headroom+(extra))&~(HH_DATA_MOD - 1)) + HH_DATA_MOD)
#define LL_ALLOCATED_SPACE(dev) \
((((dev)->hard_header_len+(dev)->needed_headroom+(dev)->needed_tailroom)&~(HH_DATA_MOD - 1)) + HH_DATA_MOD)
struct header_ops {
int (*create) (struct sk_buff *skb, struct net_device *dev,
unsigned short type, const void *daddr,
const void *saddr, unsigned len);
int (*parse)(const struct sk_buff *skb, unsigned char *haddr);
int (*rebuild)(struct sk_buff *skb);
#define HAVE_HEADER_CACHE
int (*cache)(const struct neighbour *neigh, struct hh_cache *hh);
void (*cache_update)(struct hh_cache *hh,
const struct net_device *dev,
const unsigned char *haddr);
};
/* These flag bits are private to the generic network queueing
* layer, they may not be explicitly referenced by any other
* code.
*/
enum netdev_state_t {
__LINK_STATE_START,
__LINK_STATE_PRESENT,
__LINK_STATE_NOCARRIER,
__LINK_STATE_LINKWATCH_PENDING,
__LINK_STATE_DORMANT,
};
/*
* This structure holds at boot time configured netdevice settings. They
* are then used in the device probing.
*/
struct netdev_boot_setup {
char name[IFNAMSIZ];
struct ifmap map;
};
#define NETDEV_BOOT_SETUP_MAX 8
extern int __init netdev_boot_setup(char *str);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/*
* Structure for NAPI scheduling similar to tasklet but with weighting
*/
struct napi_struct {
/* The poll_list must only be managed by the entity which
* changes the state of the NAPI_STATE_SCHED bit. This means
* whoever atomically sets that bit can add this napi_struct
* to the per-cpu poll_list, and whoever clears that bit
* can remove from the list right before clearing the bit.
*/
struct list_head poll_list;
unsigned long state;
int weight;
int (*poll)(struct napi_struct *, int);
#ifdef CONFIG_NETPOLL
spinlock_t poll_lock;
int poll_owner;
#endif
unsigned int gro_count;
struct net_device *dev;
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
struct list_head dev_list;
struct sk_buff *gro_list;
struct sk_buff *skb;
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
};
enum {
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
NAPI_STATE_SCHED, /* Poll is scheduled */
NAPI_STATE_DISABLE, /* Disable pending */
netpoll: fix race on poll_list resulting in garbage entry A few months back a race was discused between the netpoll napi service path, and the fast path through net_rx_action: http://kerneltrap.org/mailarchive/linux-netdev/2007/10/16/345470 A patch was submitted for that bug, but I think we missed a case. Consider the following scenario: INITIAL STATE CPU0 has one napi_struct A on its poll_list CPU1 is calling netpoll_send_skb and needs to call poll_napi on the same napi_struct A that CPU0 has on its list CPU0 CPU1 net_rx_action poll_napi !list_empty (returns true) locks poll_lock for A poll_one_napi napi->poll netif_rx_complete __napi_complete (removes A from poll_list) list_entry(list->next) In the above scenario, net_rx_action assumes that the per-cpu poll_list is exclusive to that cpu. netpoll of course violates that, and because the netpoll path can dequeue from the poll list, its possible for CPU0 to detect a non-empty list at the top of the while loop in net_rx_action, but have it become empty by the time it calls list_entry. Since the poll_list isn't surrounded by any other structure, the returned data from that list_entry call in this situation is garbage, and any number of crashes can result based on what exactly that garbage is. Given that its not fasible for performance reasons to place exclusive locks arround each cpus poll list to provide that mutal exclusion, I think the best solution is modify the netpoll path in such a way that we continue to guarantee that the poll_list for a cpu is in fact exclusive to that cpu. To do this I've implemented the patch below. It adds an additional bit to the state field in the napi_struct. When executing napi->poll from the netpoll_path, this bit will be set. When a driver calls netif_rx_complete, if that bit is set, it will not remove the napi_struct from the poll_list. That work will be saved for the next iteration of net_rx_action. I've tested this and it seems to work well. About the biggest drawback I can see to it is the fact that it might result in an extra loop through net_rx_action in the event that the device is actually contended for (i.e. the netpoll path actually preforms all the needed work no the device, and the call to net_rx_action winds up doing nothing, except removing the napi_struct from the poll_list. However I think this is probably a small price to pay, given that the alternative is a crash. Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-10 10:22:26 +03:00
NAPI_STATE_NPSVC, /* Netpoll - don't dequeue from poll_list */
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
};
enum gro_result {
GRO_MERGED,
GRO_MERGED_FREE,
GRO_HELD,
GRO_NORMAL,
GRO_DROP,
};
typedef enum gro_result gro_result_t;
extern void __napi_schedule(struct napi_struct *n);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
static inline int napi_disable_pending(struct napi_struct *n)
{
return test_bit(NAPI_STATE_DISABLE, &n->state);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* napi_schedule_prep - check if napi can be scheduled
* @n: napi context
*
* Test if NAPI routine is already running, and if not mark
* it as running. This is used as a condition variable
* insure only one NAPI poll instance runs. We also make
* sure there is no pending NAPI disable.
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
*/
static inline int napi_schedule_prep(struct napi_struct *n)
{
return !napi_disable_pending(n) &&
!test_and_set_bit(NAPI_STATE_SCHED, &n->state);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
}
/**
* napi_schedule - schedule NAPI poll
* @n: napi context
*
* Schedule NAPI poll routine to be called if it is not already
* running.
*/
static inline void napi_schedule(struct napi_struct *n)
{
if (napi_schedule_prep(n))
__napi_schedule(n);
}
/* Try to reschedule poll. Called by dev->poll() after napi_complete(). */
static inline int napi_reschedule(struct napi_struct *napi)
{
if (napi_schedule_prep(napi)) {
__napi_schedule(napi);
return 1;
}
return 0;
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* napi_complete - NAPI processing complete
* @n: napi context
*
* Mark NAPI processing as complete.
*/
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
extern void __napi_complete(struct napi_struct *n);
extern void napi_complete(struct napi_struct *n);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* napi_disable - prevent NAPI from scheduling
* @n: napi context
*
* Stop NAPI from being scheduled on this context.
* Waits till any outstanding processing completes.
*/
static inline void napi_disable(struct napi_struct *n)
{
set_bit(NAPI_STATE_DISABLE, &n->state);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
msleep(1);
clear_bit(NAPI_STATE_DISABLE, &n->state);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
}
/**
* napi_enable - enable NAPI scheduling
* @n: napi context
*
* Resume NAPI from being scheduled on this context.
* Must be paired with napi_disable.
*/
static inline void napi_enable(struct napi_struct *n)
{
BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
smp_mb__before_clear_bit();
clear_bit(NAPI_STATE_SCHED, &n->state);
}
#ifdef CONFIG_SMP
/**
* napi_synchronize - wait until NAPI is not running
* @n: napi context
*
* Wait until NAPI is done being scheduled on this context.
* Waits till any outstanding processing completes but
* does not disable future activations.
*/
static inline void napi_synchronize(const struct napi_struct *n)
{
while (test_bit(NAPI_STATE_SCHED, &n->state))
msleep(1);
}
#else
# define napi_synchronize(n) barrier()
#endif
enum netdev_queue_state_t {
__QUEUE_STATE_XOFF,
__QUEUE_STATE_FROZEN,
};
struct netdev_queue {
/*
* read mostly part
*/
struct net_device *dev;
struct Qdisc *qdisc;
unsigned long state;
struct Qdisc *qdisc_sleeping;
/*
* write mostly part
*/
spinlock_t _xmit_lock ____cacheline_aligned_in_smp;
int xmit_lock_owner;
/*
* please use this field instead of dev->trans_start
*/
unsigned long trans_start;
unsigned long tx_bytes;
unsigned long tx_packets;
unsigned long tx_dropped;
} ____cacheline_aligned_in_smp;
#ifdef CONFIG_RPS
rps: Receive Packet Steering This patch implements software receive side packet steering (RPS). RPS distributes the load of received packet processing across multiple CPUs. Problem statement: Protocol processing done in the NAPI context for received packets is serialized per device queue and becomes a bottleneck under high packet load. This substantially limits pps that can be achieved on a single queue NIC and provides no scaling with multiple cores. This solution queues packets early on in the receive path on the backlog queues of other CPUs. This allows protocol processing (e.g. IP and TCP) to be performed on packets in parallel. For each device (or each receive queue in a multi-queue device) a mask of CPUs is set to indicate the CPUs that can process packets. A CPU is selected on a per packet basis by hashing contents of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index into the CPU mask. The IPI mechanism is used to raise networking receive softirqs between CPUs. This effectively emulates in software what a multi-queue NIC can provide, but is generic requiring no device support. Many devices now provide a hash over the 4-tuple on a per packet basis (e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash in an skb field, and that value in turn is used to index into the RPS maps. Using the HW generated hash can avoid cache misses on the packet when steering it to a remote CPU. The CPU mask is set on a per device and per queue basis in the sysfs variable /sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical bit maps for receive queues in the device (numbered by <n>). If a device does not support multi-queue, a single variable is used for the device (rx-0). Generally, we have found this technique increases pps capabilities of a single queue device with good CPU utilization. Optimal settings for the CPU mask seem to depend on architectures and cache hierarcy. Below are some results running 500 instances of netperf TCP_RR test with 1 byte req. and resp. Results show cumulative transaction rate and system CPU utilization. e1000e on 8 core Intel Without RPS: 108K tps at 33% CPU With RPS: 311K tps at 64% CPU forcedeth on 16 core AMD Without RPS: 156K tps at 15% CPU With RPS: 404K tps at 49% CPU bnx2x on 16 core AMD Without RPS 567K tps at 61% CPU (4 HW RX queues) Without RPS 738K tps at 96% CPU (8 HW RX queues) With RPS: 854K tps at 76% CPU (4 HW RX queues) Caveats: - The benefits of this patch are dependent on architecture and cache hierarchy. Tuning the masks to get best performance is probably necessary. - This patch adds overhead in the path for processing a single packet. In a lightly loaded server this overhead may eliminate the advantages of increased parallelism, and possibly cause some relative performance degradation. We have found that masks that are cache aware (share same caches with the interrupting CPU) mitigate much of this. - The RPS masks can be changed dynamically, however whenever the mask is changed this introduces the possibility of generating out of order packets. It's probably best not change the masks too frequently. Signed-off-by: Tom Herbert <therbert@google.com> include/linux/netdevice.h | 32 ++++- include/linux/skbuff.h | 3 + net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++-------- net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++- net/core/skbuff.c | 2 + 5 files changed, 538 insertions(+), 59 deletions(-) Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-03-16 11:03:29 +03:00
/*
* 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[0];
};
#define RPS_MAP_SIZE(_num) (sizeof(struct rps_map) + (_num * sizeof(u16)))
/* This structure contains an instance of an RX queue. */
struct netdev_rx_queue {
struct rps_map *rps_map;
struct kobject kobj;
struct netdev_rx_queue *first;
atomic_t count;
} ____cacheline_aligned_in_smp;
#endif
/*
* This structure defines the management hooks for network devices.
* The following hooks can be defined; unless noted otherwise, they are
* optional and can be filled with a null pointer.
*
* int (*ndo_init)(struct net_device *dev);
* This function is called once when network device is registered.
* The network device can use this to any late stage initializaton
* or semantic validattion. It can fail with an error code which will
* be propogated back to register_netdev
*
* void (*ndo_uninit)(struct net_device *dev);
* This function is called when device is unregistered or when registration
* fails. It is not called if init fails.
*
* int (*ndo_open)(struct net_device *dev);
* This function is called when network device transistions to the up
* state.
*
* int (*ndo_stop)(struct net_device *dev);
* This function is called when network device transistions to the down
* state.
*
* netdev_tx_t (*ndo_start_xmit)(struct sk_buff *skb,
* struct net_device *dev);
* Called when a packet needs to be transmitted.
* Must return NETDEV_TX_OK , NETDEV_TX_BUSY.
* (can also return NETDEV_TX_LOCKED iff NETIF_F_LLTX)
* Required can not be NULL.
*
* u16 (*ndo_select_queue)(struct net_device *dev, struct sk_buff *skb);
* Called to decide which queue to when device supports multiple
* transmit queues.
*
* void (*ndo_change_rx_flags)(struct net_device *dev, int flags);
* This function is called to allow device receiver to make
* changes to configuration when multicast or promiscious is enabled.
*
* void (*ndo_set_rx_mode)(struct net_device *dev);
* This function is called device changes address list filtering.
*
* void (*ndo_set_multicast_list)(struct net_device *dev);
* This function is called when the multicast address list changes.
*
* int (*ndo_set_mac_address)(struct net_device *dev, void *addr);
* This function is called when the Media Access Control address
* needs to be changed. If this interface is not defined, the
* mac address can not be changed.
*
* int (*ndo_validate_addr)(struct net_device *dev);
* Test if Media Access Control address is valid for the device.
*
* int (*ndo_do_ioctl)(struct net_device *dev, struct ifreq *ifr, int cmd);
* Called when a user request an ioctl which can't be handled by
* the generic interface code. If not defined ioctl's return
* not supported error code.
*
* int (*ndo_set_config)(struct net_device *dev, struct ifmap *map);
* Used to set network devices bus interface parameters. This interface
* is retained for legacy reason, new devices should use the bus
* interface (PCI) for low level management.
*
* int (*ndo_change_mtu)(struct net_device *dev, int new_mtu);
* Called when a user wants to change the Maximum Transfer Unit
* of a device. If not defined, any request to change MTU will
* will return an error.
*
* void (*ndo_tx_timeout)(struct net_device *dev);
* Callback uses when the transmitter has not made any progress
* for dev->watchdog ticks.
*
* struct net_device_stats* (*ndo_get_stats)(struct net_device *dev);
* Called when a user wants to get the network device usage
* statistics. If not defined, the counters in dev->stats will
* be used.
*
* void (*ndo_vlan_rx_register)(struct net_device *dev, struct vlan_group *grp);
* If device support VLAN receive accleration
* (ie. dev->features & NETIF_F_HW_VLAN_RX), then this function is called
* when vlan groups for the device changes. Note: grp is NULL
* if no vlan's groups are being used.
*
* void (*ndo_vlan_rx_add_vid)(struct net_device *dev, unsigned short vid);
* If device support VLAN filtering (dev->features & NETIF_F_HW_VLAN_FILTER)
* this function is called when a VLAN id is registered.
*
* void (*ndo_vlan_rx_kill_vid)(struct net_device *dev, unsigned short vid);
* If device support VLAN filtering (dev->features & NETIF_F_HW_VLAN_FILTER)
* this function is called when a VLAN id is unregistered.
*
* void (*ndo_poll_controller)(struct net_device *dev);
*
* SR-IOV management functions.
* int (*ndo_set_vf_mac)(struct net_device *dev, int vf, u8* mac);
* int (*ndo_set_vf_vlan)(struct net_device *dev, int vf, u16 vlan, u8 qos);
* int (*ndo_set_vf_tx_rate)(struct net_device *dev, int vf, int rate);
* int (*ndo_get_vf_config)(struct net_device *dev,
* int vf, struct ifla_vf_info *ivf);
*/
#define HAVE_NET_DEVICE_OPS
struct net_device_ops {
int (*ndo_init)(struct net_device *dev);
void (*ndo_uninit)(struct net_device *dev);
int (*ndo_open)(struct net_device *dev);
int (*ndo_stop)(struct net_device *dev);
netdev_tx_t (*ndo_start_xmit) (struct sk_buff *skb,
struct net_device *dev);
u16 (*ndo_select_queue)(struct net_device *dev,
struct sk_buff *skb);
void (*ndo_change_rx_flags)(struct net_device *dev,
int flags);
void (*ndo_set_rx_mode)(struct net_device *dev);
void (*ndo_set_multicast_list)(struct net_device *dev);
int (*ndo_set_mac_address)(struct net_device *dev,
void *addr);
int (*ndo_validate_addr)(struct net_device *dev);
int (*ndo_do_ioctl)(struct net_device *dev,
struct ifreq *ifr, int cmd);
int (*ndo_set_config)(struct net_device *dev,
struct ifmap *map);
int (*ndo_change_mtu)(struct net_device *dev,
int new_mtu);
int (*ndo_neigh_setup)(struct net_device *dev,
struct neigh_parms *);
void (*ndo_tx_timeout) (struct net_device *dev);
struct net_device_stats* (*ndo_get_stats)(struct net_device *dev);
void (*ndo_vlan_rx_register)(struct net_device *dev,
struct vlan_group *grp);
void (*ndo_vlan_rx_add_vid)(struct net_device *dev,
unsigned short vid);
void (*ndo_vlan_rx_kill_vid)(struct net_device *dev,
unsigned short vid);
#ifdef CONFIG_NET_POLL_CONTROLLER
void (*ndo_poll_controller)(struct net_device *dev);
#endif
int (*ndo_set_vf_mac)(struct net_device *dev,
int queue, u8 *mac);
int (*ndo_set_vf_vlan)(struct net_device *dev,
int queue, u16 vlan, u8 qos);
int (*ndo_set_vf_tx_rate)(struct net_device *dev,
int vf, int rate);
int (*ndo_get_vf_config)(struct net_device *dev,
int vf,
struct ifla_vf_info *ivf);
#if defined(CONFIG_FCOE) || defined(CONFIG_FCOE_MODULE)
int (*ndo_fcoe_enable)(struct net_device *dev);
int (*ndo_fcoe_disable)(struct net_device *dev);
int (*ndo_fcoe_ddp_setup)(struct net_device *dev,
u16 xid,
struct scatterlist *sgl,
unsigned int sgc);
int (*ndo_fcoe_ddp_done)(struct net_device *dev,
u16 xid);
#define NETDEV_FCOE_WWNN 0
#define NETDEV_FCOE_WWPN 1
int (*ndo_fcoe_get_wwn)(struct net_device *dev,
u64 *wwn, int type);
#endif
};
/*
* The DEVICE structure.
* Actually, this whole structure is a big mistake. It mixes I/O
* data with strictly "high-level" data, and it has to know about
* almost every data structure used in the INET module.
*
* FIXME: cleanup struct net_device such that network protocol info
* moves out.
*/
struct net_device {
/*
* This is the first field of the "visible" part of this structure
* (i.e. as seen by users in the "Space.c" file). It is the name
* the interface.
*/
char name[IFNAMSIZ];
/* device name hash chain */
struct hlist_node name_hlist;
/* snmp alias */
char *ifalias;
/*
* I/O specific fields
* FIXME: Merge these and struct ifmap into one
*/
unsigned long mem_end; /* shared mem end */
unsigned long mem_start; /* shared mem start */
unsigned long base_addr; /* device I/O address */
unsigned int irq; /* device IRQ number */
/*
* Some hardware also needs these fields, but they are not
* part of the usual set specified in Space.c.
*/
unsigned char if_port; /* Selectable AUI, TP,..*/
unsigned char dma; /* DMA channel */
unsigned long state;
struct list_head dev_list;
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
struct list_head napi_list;
struct list_head unreg_list;
/* Net device features */
unsigned long features;
#define NETIF_F_SG 1 /* Scatter/gather IO. */
#define NETIF_F_IP_CSUM 2 /* Can checksum TCP/UDP over IPv4. */
#define NETIF_F_NO_CSUM 4 /* Does not require checksum. F.e. loopack. */
#define NETIF_F_HW_CSUM 8 /* Can checksum all the packets. */
#define NETIF_F_IPV6_CSUM 16 /* Can checksum TCP/UDP over IPV6 */
#define NETIF_F_HIGHDMA 32 /* Can DMA to high memory. */
#define NETIF_F_FRAGLIST 64 /* Scatter/gather IO. */
#define NETIF_F_HW_VLAN_TX 128 /* Transmit VLAN hw acceleration */
#define NETIF_F_HW_VLAN_RX 256 /* Receive VLAN hw acceleration */
#define NETIF_F_HW_VLAN_FILTER 512 /* Receive filtering on VLAN */
#define NETIF_F_VLAN_CHALLENGED 1024 /* Device cannot handle VLAN packets */
#define NETIF_F_GSO 2048 /* Enable software GSO. */
#define NETIF_F_LLTX 4096 /* LockLess TX - deprecated. Please */
/* do not use LLTX in new drivers */
#define NETIF_F_NETNS_LOCAL 8192 /* Does not change network namespaces */
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
#define NETIF_F_GRO 16384 /* Generic receive offload */
#define NETIF_F_LRO 32768 /* large receive offload */
/* the GSO_MASK reserves bits 16 through 23 */
#define NETIF_F_FCOE_CRC (1 << 24) /* FCoE CRC32 */
#define NETIF_F_SCTP_CSUM (1 << 25) /* SCTP checksum offload */
#define NETIF_F_FCOE_MTU (1 << 26) /* Supports max FCoE MTU, 2158 bytes*/
#define NETIF_F_NTUPLE (1 << 27) /* N-tuple filters supported */
#define NETIF_F_RXHASH (1 << 28) /* Receive hashing offload */
/* Segmentation offload features */
#define NETIF_F_GSO_SHIFT 16
#define NETIF_F_GSO_MASK 0x00ff0000
#define NETIF_F_TSO (SKB_GSO_TCPV4 << NETIF_F_GSO_SHIFT)
[IPV6]: Added GSO support for TCPv6 This patch adds GSO support for IPv6 and TCPv6. This is based on a patch by Ananda Raju <Ananda.Raju@neterion.com>. His original description is: This patch enables TSO over IPv6. Currently Linux network stacks restricts TSO over IPv6 by clearing of the NETIF_F_TSO bit from "dev->features". This patch will remove this restriction. This patch will introduce a new flag NETIF_F_TSO6 which will be used to check whether device supports TSO over IPv6. If device support TSO over IPv6 then we don't clear of NETIF_F_TSO and which will make the TCP layer to create TSO packets. Any device supporting TSO over IPv6 will set NETIF_F_TSO6 flag in "dev->features" along with NETIF_F_TSO. In case when user disables TSO using ethtool, NETIF_F_TSO will get cleared from "dev->features". So even if we have NETIF_F_TSO6 we don't get TSO packets created by TCP layer. SKB_GSO_TCPV4 renamed to SKB_GSO_TCP to make it generic GSO packet. SKB_GSO_UDPV4 renamed to SKB_GSO_UDP as UFO is not a IPv4 feature. UFO is supported over IPv6 also The following table shows there is significant improvement in throughput with normal frames and CPU usage for both normal and jumbo. -------------------------------------------------- | | 1500 | 9600 | | ------------------|-------------------| | | thru CPU | thru CPU | -------------------------------------------------- | TSO OFF | 2.00 5.5% id | 5.66 20.0% id | -------------------------------------------------- | TSO ON | 2.63 78.0 id | 5.67 39.0% id | -------------------------------------------------- Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-07-01 00:37:03 +04:00
#define NETIF_F_UFO (SKB_GSO_UDP << NETIF_F_GSO_SHIFT)
#define NETIF_F_GSO_ROBUST (SKB_GSO_DODGY << NETIF_F_GSO_SHIFT)
[IPV6]: Added GSO support for TCPv6 This patch adds GSO support for IPv6 and TCPv6. This is based on a patch by Ananda Raju <Ananda.Raju@neterion.com>. His original description is: This patch enables TSO over IPv6. Currently Linux network stacks restricts TSO over IPv6 by clearing of the NETIF_F_TSO bit from "dev->features". This patch will remove this restriction. This patch will introduce a new flag NETIF_F_TSO6 which will be used to check whether device supports TSO over IPv6. If device support TSO over IPv6 then we don't clear of NETIF_F_TSO and which will make the TCP layer to create TSO packets. Any device supporting TSO over IPv6 will set NETIF_F_TSO6 flag in "dev->features" along with NETIF_F_TSO. In case when user disables TSO using ethtool, NETIF_F_TSO will get cleared from "dev->features". So even if we have NETIF_F_TSO6 we don't get TSO packets created by TCP layer. SKB_GSO_TCPV4 renamed to SKB_GSO_TCP to make it generic GSO packet. SKB_GSO_UDPV4 renamed to SKB_GSO_UDP as UFO is not a IPv4 feature. UFO is supported over IPv6 also The following table shows there is significant improvement in throughput with normal frames and CPU usage for both normal and jumbo. -------------------------------------------------- | | 1500 | 9600 | | ------------------|-------------------| | | thru CPU | thru CPU | -------------------------------------------------- | TSO OFF | 2.00 5.5% id | 5.66 20.0% id | -------------------------------------------------- | TSO ON | 2.63 78.0 id | 5.67 39.0% id | -------------------------------------------------- Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-07-01 00:37:03 +04:00
#define NETIF_F_TSO_ECN (SKB_GSO_TCP_ECN << NETIF_F_GSO_SHIFT)
#define NETIF_F_TSO6 (SKB_GSO_TCPV6 << NETIF_F_GSO_SHIFT)
#define NETIF_F_FSO (SKB_GSO_FCOE << NETIF_F_GSO_SHIFT)
/* List of features with software fallbacks. */
#define NETIF_F_GSO_SOFTWARE (NETIF_F_TSO | NETIF_F_TSO_ECN | NETIF_F_TSO6)
#define NETIF_F_GEN_CSUM (NETIF_F_NO_CSUM | NETIF_F_HW_CSUM)
#define NETIF_F_V4_CSUM (NETIF_F_GEN_CSUM | NETIF_F_IP_CSUM)
#define NETIF_F_V6_CSUM (NETIF_F_GEN_CSUM | NETIF_F_IPV6_CSUM)
#define NETIF_F_ALL_CSUM (NETIF_F_V4_CSUM | NETIF_F_V6_CSUM)
/*
* If one device supports one of these features, then enable them
* for all in netdev_increment_features.
*/
#define NETIF_F_ONE_FOR_ALL (NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ROBUST | \
NETIF_F_SG | NETIF_F_HIGHDMA | \
NETIF_F_FRAGLIST)
/* Interface index. Unique device identifier */
int ifindex;
int iflink;
struct net_device_stats stats;
#ifdef CONFIG_WIRELESS_EXT
/* List of functions to handle Wireless Extensions (instead of ioctl).
* See <net/iw_handler.h> for details. Jean II */
const struct iw_handler_def * wireless_handlers;
/* Instance data managed by the core of Wireless Extensions. */
struct iw_public_data * wireless_data;
#endif
/* Management operations */
const struct net_device_ops *netdev_ops;
const struct ethtool_ops *ethtool_ops;
/* Hardware header description */
const struct header_ops *header_ops;
unsigned int flags; /* interface flags (a la BSD) */
unsigned short gflags;
unsigned short priv_flags; /* Like 'flags' but invisible to userspace. */
unsigned short padded; /* How much padding added by alloc_netdev() */
unsigned char operstate; /* RFC2863 operstate */
unsigned char link_mode; /* mapping policy to operstate */
unsigned mtu; /* interface MTU value */
unsigned short type; /* interface hardware type */
unsigned short hard_header_len; /* hardware hdr length */
/* extra head- and tailroom the hardware may need, but not in all cases
* can this be guaranteed, especially tailroom. Some cases also use
* LL_MAX_HEADER instead to allocate the skb.
*/
unsigned short needed_headroom;
unsigned short needed_tailroom;
struct net_device *master; /* Pointer to master device of a group,
* which this device is member of.
*/
/* Interface address info. */
unsigned char perm_addr[MAX_ADDR_LEN]; /* permanent hw address */
unsigned char addr_len; /* hardware address length */
unsigned short dev_id; /* for shared network cards */
struct netdev_hw_addr_list uc; /* Secondary unicast
mac addresses */
int uc_promisc;
spinlock_t addr_list_lock;
struct dev_addr_list *mc_list; /* Multicast mac addresses */
int mc_count; /* Number of installed mcasts */
unsigned int promiscuity;
unsigned int allmulti;
/* Protocol specific pointers */
net: Distributed Switch Architecture protocol support Distributed Switch Architecture is a protocol for managing hardware switch chips. It consists of a set of MII management registers and commands to configure the switch, and an ethernet header format to signal which of the ports of the switch a packet was received from or is intended to be sent to. The switches that this driver supports are typically embedded in access points and routers, and a typical setup with a DSA switch looks something like this: +-----------+ +-----------+ | | RGMII | | | +-------+ +------ 1000baseT MDI ("WAN") | | | 6-port +------ 1000baseT MDI ("LAN1") | CPU | | ethernet +------ 1000baseT MDI ("LAN2") | |MIImgmt| switch +------ 1000baseT MDI ("LAN3") | +-------+ w/5 PHYs +------ 1000baseT MDI ("LAN4") | | | | +-----------+ +-----------+ The switch driver presents each port on the switch as a separate network interface to Linux, polls the switch to maintain software link state of those ports, forwards MII management interface accesses to those network interfaces (e.g. as done by ethtool) to the switch, and exposes the switch's hardware statistics counters via the appropriate Linux kernel interfaces. This initial patch supports the MII management interface register layout of the Marvell 88E6123, 88E6161 and 88E6165 switch chips, and supports the "Ethertype DSA" packet tagging format. (There is no officially registered ethertype for the Ethertype DSA packet format, so we just grab a random one. The ethertype to use is programmed into the switch, and the switch driver uses the value of ETH_P_EDSA for this, so this define can be changed at any time in the future if the one we chose is allocated to another protocol or if Ethertype DSA gets its own officially registered ethertype, and everything will continue to work.) Signed-off-by: Lennert Buytenhek <buytenh@marvell.com> Tested-by: Nicolas Pitre <nico@marvell.com> Tested-by: Byron Bradley <byron.bbradley@gmail.com> Tested-by: Tim Ellis <tim.ellis@mac.com> Tested-by: Peter van Valderen <linux@ddcrew.com> Tested-by: Dirk Teurlings <dirk@upexia.nl> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-10-07 17:44:02 +04:00
#ifdef CONFIG_NET_DSA
void *dsa_ptr; /* dsa specific data */
#endif
void *atalk_ptr; /* AppleTalk link */
void *ip_ptr; /* IPv4 specific data */
void *dn_ptr; /* DECnet specific data */
void *ip6_ptr; /* IPv6 specific data */
void *ec_ptr; /* Econet specific data */
void *ax25_ptr; /* AX.25 specific data */
struct wireless_dev *ieee80211_ptr; /* IEEE 802.11 specific data,
assign before registering */
/*
* Cache line mostly used on receive path (including eth_type_trans())
*/
unsigned long last_rx; /* Time of last Rx */
/* Interface address info used in eth_type_trans() */
unsigned char *dev_addr; /* hw address, (before bcast
because most packets are
unicast) */
struct netdev_hw_addr_list dev_addrs; /* list of device
hw addresses */
unsigned char broadcast[MAX_ADDR_LEN]; /* hw bcast add */
#ifdef CONFIG_RPS
rps: Receive Packet Steering This patch implements software receive side packet steering (RPS). RPS distributes the load of received packet processing across multiple CPUs. Problem statement: Protocol processing done in the NAPI context for received packets is serialized per device queue and becomes a bottleneck under high packet load. This substantially limits pps that can be achieved on a single queue NIC and provides no scaling with multiple cores. This solution queues packets early on in the receive path on the backlog queues of other CPUs. This allows protocol processing (e.g. IP and TCP) to be performed on packets in parallel. For each device (or each receive queue in a multi-queue device) a mask of CPUs is set to indicate the CPUs that can process packets. A CPU is selected on a per packet basis by hashing contents of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index into the CPU mask. The IPI mechanism is used to raise networking receive softirqs between CPUs. This effectively emulates in software what a multi-queue NIC can provide, but is generic requiring no device support. Many devices now provide a hash over the 4-tuple on a per packet basis (e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash in an skb field, and that value in turn is used to index into the RPS maps. Using the HW generated hash can avoid cache misses on the packet when steering it to a remote CPU. The CPU mask is set on a per device and per queue basis in the sysfs variable /sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical bit maps for receive queues in the device (numbered by <n>). If a device does not support multi-queue, a single variable is used for the device (rx-0). Generally, we have found this technique increases pps capabilities of a single queue device with good CPU utilization. Optimal settings for the CPU mask seem to depend on architectures and cache hierarcy. Below are some results running 500 instances of netperf TCP_RR test with 1 byte req. and resp. Results show cumulative transaction rate and system CPU utilization. e1000e on 8 core Intel Without RPS: 108K tps at 33% CPU With RPS: 311K tps at 64% CPU forcedeth on 16 core AMD Without RPS: 156K tps at 15% CPU With RPS: 404K tps at 49% CPU bnx2x on 16 core AMD Without RPS 567K tps at 61% CPU (4 HW RX queues) Without RPS 738K tps at 96% CPU (8 HW RX queues) With RPS: 854K tps at 76% CPU (4 HW RX queues) Caveats: - The benefits of this patch are dependent on architecture and cache hierarchy. Tuning the masks to get best performance is probably necessary. - This patch adds overhead in the path for processing a single packet. In a lightly loaded server this overhead may eliminate the advantages of increased parallelism, and possibly cause some relative performance degradation. We have found that masks that are cache aware (share same caches with the interrupting CPU) mitigate much of this. - The RPS masks can be changed dynamically, however whenever the mask is changed this introduces the possibility of generating out of order packets. It's probably best not change the masks too frequently. Signed-off-by: Tom Herbert <therbert@google.com> include/linux/netdevice.h | 32 ++++- include/linux/skbuff.h | 3 + net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++-------- net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++- net/core/skbuff.c | 2 + 5 files changed, 538 insertions(+), 59 deletions(-) Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-03-16 11:03:29 +03:00
struct kset *queues_kset;
struct netdev_rx_queue *_rx;
/* Number of RX queues allocated at alloc_netdev_mq() time */
unsigned int num_rx_queues;
#endif
rps: Receive Packet Steering This patch implements software receive side packet steering (RPS). RPS distributes the load of received packet processing across multiple CPUs. Problem statement: Protocol processing done in the NAPI context for received packets is serialized per device queue and becomes a bottleneck under high packet load. This substantially limits pps that can be achieved on a single queue NIC and provides no scaling with multiple cores. This solution queues packets early on in the receive path on the backlog queues of other CPUs. This allows protocol processing (e.g. IP and TCP) to be performed on packets in parallel. For each device (or each receive queue in a multi-queue device) a mask of CPUs is set to indicate the CPUs that can process packets. A CPU is selected on a per packet basis by hashing contents of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index into the CPU mask. The IPI mechanism is used to raise networking receive softirqs between CPUs. This effectively emulates in software what a multi-queue NIC can provide, but is generic requiring no device support. Many devices now provide a hash over the 4-tuple on a per packet basis (e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash in an skb field, and that value in turn is used to index into the RPS maps. Using the HW generated hash can avoid cache misses on the packet when steering it to a remote CPU. The CPU mask is set on a per device and per queue basis in the sysfs variable /sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical bit maps for receive queues in the device (numbered by <n>). If a device does not support multi-queue, a single variable is used for the device (rx-0). Generally, we have found this technique increases pps capabilities of a single queue device with good CPU utilization. Optimal settings for the CPU mask seem to depend on architectures and cache hierarcy. Below are some results running 500 instances of netperf TCP_RR test with 1 byte req. and resp. Results show cumulative transaction rate and system CPU utilization. e1000e on 8 core Intel Without RPS: 108K tps at 33% CPU With RPS: 311K tps at 64% CPU forcedeth on 16 core AMD Without RPS: 156K tps at 15% CPU With RPS: 404K tps at 49% CPU bnx2x on 16 core AMD Without RPS 567K tps at 61% CPU (4 HW RX queues) Without RPS 738K tps at 96% CPU (8 HW RX queues) With RPS: 854K tps at 76% CPU (4 HW RX queues) Caveats: - The benefits of this patch are dependent on architecture and cache hierarchy. Tuning the masks to get best performance is probably necessary. - This patch adds overhead in the path for processing a single packet. In a lightly loaded server this overhead may eliminate the advantages of increased parallelism, and possibly cause some relative performance degradation. We have found that masks that are cache aware (share same caches with the interrupting CPU) mitigate much of this. - The RPS masks can be changed dynamically, however whenever the mask is changed this introduces the possibility of generating out of order packets. It's probably best not change the masks too frequently. Signed-off-by: Tom Herbert <therbert@google.com> include/linux/netdevice.h | 32 ++++- include/linux/skbuff.h | 3 + net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++-------- net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++- net/core/skbuff.c | 2 + 5 files changed, 538 insertions(+), 59 deletions(-) Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-03-16 11:03:29 +03:00
struct netdev_queue rx_queue;
struct netdev_queue *_tx ____cacheline_aligned_in_smp;
/* Number of TX queues allocated at alloc_netdev_mq() time */
unsigned int num_tx_queues;
/* Number of TX queues currently active in device */
unsigned int real_num_tx_queues;
/* root qdisc from userspace point of view */
struct Qdisc *qdisc;
unsigned long tx_queue_len; /* Max frames per queue allowed */
spinlock_t tx_global_lock;
/*
* One part is mostly used on xmit path (device)
*/
/* These may be needed for future network-power-down code. */
/*
* trans_start here is expensive for high speed devices on SMP,
* please use netdev_queue->trans_start instead.
*/
unsigned long trans_start; /* Time (in jiffies) of last Tx */
int watchdog_timeo; /* used by dev_watchdog() */
struct timer_list watchdog_timer;
/* Number of references to this device */
atomic_t refcnt ____cacheline_aligned_in_smp;
/* delayed register/unregister */
struct list_head todo_list;
/* device index hash chain */
struct hlist_node index_hlist;
linkwatch: linkwatch_forget_dev() to speedup device dismantle Herbert Xu a écrit : > On Tue, Nov 17, 2009 at 04:26:04AM -0800, David Miller wrote: >> Really, the link watch stuff is just due for a redesign. I don't >> think a simple hack is going to cut it this time, sorry Eric :-) > > I have no objections against any redesigns, but since the only > caller of linkwatch_forget_dev runs in process context with the > RTNL, it could also legally emit those events. Thanks guys, here an updated version then, before linkwatch surgery ? In this version, I force the event to be sent synchronously. [PATCH net-next-2.6] linkwatch: linkwatch_forget_dev() to speedup device dismantle time ip link del eth3.103 ; time ip link del eth3.104 ; time ip link del eth3.105 real 0m0.266s user 0m0.000s sys 0m0.001s real 0m0.770s user 0m0.000s sys 0m0.000s real 0m1.022s user 0m0.000s sys 0m0.000s One problem of current schem in vlan dismantle phase is the holding of device done by following chain : vlan_dev_stop() -> netif_carrier_off(dev) -> linkwatch_fire_event(dev) -> dev_hold() ... And __linkwatch_run_queue() runs up to one second later... A generic fix to this problem is to add a linkwatch_forget_dev() method to unlink the device from the list of watched devices. dev->link_watch_next becomes dev->link_watch_list (and use a bit more memory), to be able to unlink device in O(1). After patch : time ip link del eth3.103 ; time ip link del eth3.104 ; time ip link del eth3.105 real 0m0.024s user 0m0.000s sys 0m0.000s real 0m0.032s user 0m0.000s sys 0m0.001s real 0m0.033s user 0m0.000s sys 0m0.000s Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-11-17 08:59:21 +03:00
struct list_head link_watch_list;
/* register/unregister state machine */
enum { NETREG_UNINITIALIZED=0,
NETREG_REGISTERED, /* completed register_netdevice */
NETREG_UNREGISTERING, /* called unregister_netdevice */
NETREG_UNREGISTERED, /* completed unregister todo */
NETREG_RELEASED, /* called free_netdev */
NETREG_DUMMY, /* dummy device for NAPI poll */
} reg_state:16;
enum {
RTNL_LINK_INITIALIZED,
RTNL_LINK_INITIALIZING,
} rtnl_link_state:16;
/* Called from unregister, can be used to call free_netdev */
void (*destructor)(struct net_device *dev);
#ifdef CONFIG_NETPOLL
struct netpoll_info *npinfo;
#endif
#ifdef CONFIG_NET_NS
/* Network namespace this network device is inside */
struct net *nd_net;
#endif
/* mid-layer private */
void *ml_priv;
/* bridge stuff */
struct net_bridge_port *br_port;
/* macvlan */
struct macvlan_port *macvlan_port;
/* GARP */
struct garp_port *garp_port;
/* class/net/name entry */
struct device dev;
/* space for optional device, statistics, and wireless sysfs groups */
const struct attribute_group *sysfs_groups[4];
/* rtnetlink link ops */
const struct rtnl_link_ops *rtnl_link_ops;
/* VLAN feature mask */
unsigned long vlan_features;
[NET]: Add per-connection option to set max TSO frame size Update: My mailer ate one of Jarek's feedback mails... Fixed the parameter in netif_set_gso_max_size() to be u32, not u16. Fixed the whitespace issue due to a patch import botch. Changed the types from u32 to unsigned int to be more consistent with other variables in the area. Also brought the patch up to the latest net-2.6.26 tree. Update: Made gso_max_size container 32 bits, not 16. Moved the location of gso_max_size within netdev to be less hotpath. Made more consistent names between the sock and netdev layers, and added a define for the max GSO size. Update: Respun for net-2.6.26 tree. Update: changed max_gso_frame_size and sk_gso_max_size from signed to unsigned - thanks Stephen! This patch adds the ability for device drivers to control the size of the TSO frames being sent to them, per TCP connection. By setting the netdevice's gso_max_size value, the socket layer will set the GSO frame size based on that value. This will propogate into the TCP layer, and send TSO's of that size to the hardware. This can be desirable to help tune the bursty nature of TSO on a per-adapter basis, where one may have 1 GbE and 10 GbE devices coexisting in a system, one running multiqueue and the other not, etc. This can also be desirable for devices that cannot support full 64 KB TSO's, but still want to benefit from some level of segmentation offloading. Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-03-21 13:43:19 +03:00
/* for setting kernel sock attribute on TCP connection setup */
#define GSO_MAX_SIZE 65536
unsigned int gso_max_size;
#ifdef CONFIG_DCB
/* Data Center Bridging netlink ops */
const struct dcbnl_rtnl_ops *dcbnl_ops;
#endif
#if defined(CONFIG_FCOE) || defined(CONFIG_FCOE_MODULE)
/* max exchange id for FCoE LRO by ddp */
unsigned int fcoe_ddp_xid;
#endif
/* n-tuple filter list attached to this device */
struct ethtool_rx_ntuple_list ethtool_ntuple_list;
};
#define to_net_dev(d) container_of(d, struct net_device, dev)
#define NETDEV_ALIGN 32
static inline
struct netdev_queue *netdev_get_tx_queue(const struct net_device *dev,
unsigned int index)
{
return &dev->_tx[index];
}
static inline void netdev_for_each_tx_queue(struct net_device *dev,
void (*f)(struct net_device *,
struct netdev_queue *,
void *),
void *arg)
{
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++)
f(dev, &dev->_tx[i], arg);
}
/*
* Net namespace inlines
*/
static inline
struct net *dev_net(const struct net_device *dev)
{
#ifdef CONFIG_NET_NS
return dev->nd_net;
#else
return &init_net;
#endif
}
static inline
void dev_net_set(struct net_device *dev, struct net *net)
{
#ifdef CONFIG_NET_NS
release_net(dev->nd_net);
dev->nd_net = hold_net(net);
#endif
}
static inline bool netdev_uses_dsa_tags(struct net_device *dev)
{
#ifdef CONFIG_NET_DSA_TAG_DSA
if (dev->dsa_ptr != NULL)
return dsa_uses_dsa_tags(dev->dsa_ptr);
#endif
return 0;
}
#ifndef CONFIG_NET_NS
static inline void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
{
skb->dev = dev;
}
#else /* CONFIG_NET_NS */
void skb_set_dev(struct sk_buff *skb, struct net_device *dev);
#endif
static inline bool netdev_uses_trailer_tags(struct net_device *dev)
{
#ifdef CONFIG_NET_DSA_TAG_TRAILER
if (dev->dsa_ptr != NULL)
return dsa_uses_trailer_tags(dev->dsa_ptr);
#endif
return 0;
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netdev_priv - access network device private data
* @dev: network device
*
* Get network device private data
*/
static inline void *netdev_priv(const struct net_device *dev)
{
return (char *)dev + ALIGN(sizeof(struct net_device), NETDEV_ALIGN);
}
/* Set the sysfs physical device reference for the network logical device
* if set prior to registration will cause a symlink during initialization.
*/
#define SET_NETDEV_DEV(net, pdev) ((net)->dev.parent = (pdev))
/* Set the sysfs device type for the network logical device to allow
* fin grained indentification of different network device types. For
* example Ethernet, Wirelss LAN, Bluetooth, WiMAX etc.
*/
#define SET_NETDEV_DEVTYPE(net, devtype) ((net)->dev.type = (devtype))
/**
* netif_napi_add - initialize a napi context
* @dev: network device
* @napi: napi context
* @poll: polling function
* @weight: default weight
*
* netif_napi_add() must be used to initialize a napi context prior to calling
* *any* of the other napi related functions.
*/
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
int (*poll)(struct napi_struct *, int), int weight);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_napi_del - remove a napi context
* @napi: napi context
*
* netif_napi_del() removes a napi context from the network device napi list
*/
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
void netif_napi_del(struct napi_struct *napi);
struct napi_gro_cb {
/* Virtual address of skb_shinfo(skb)->frags[0].page + offset. */
void *frag0;
/* Length of frag0. */
unsigned int frag0_len;
/* This indicates where we are processing relative to skb->data. */
int data_offset;
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
/* This is non-zero if the packet may be of the same flow. */
int same_flow;
/* This is non-zero if the packet cannot be merged with the new skb. */
int flush;
/* Number of segments aggregated. */
int count;
/* Free the skb? */
int free;
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
};
#define NAPI_GRO_CB(skb) ((struct napi_gro_cb *)(skb)->cb)
struct packet_type {
__be16 type; /* This is really htons(ether_type). */
struct net_device *dev; /* NULL is wildcarded here */
int (*func) (struct sk_buff *,
struct net_device *,
struct packet_type *,
struct net_device *);
struct sk_buff *(*gso_segment)(struct sk_buff *skb,
int features);
int (*gso_send_check)(struct sk_buff *skb);
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
struct sk_buff **(*gro_receive)(struct sk_buff **head,
struct sk_buff *skb);
int (*gro_complete)(struct sk_buff *skb);
void *af_packet_priv;
struct list_head list;
};
#include <linux/interrupt.h>
#include <linux/notifier.h>
extern rwlock_t dev_base_lock; /* Device list lock */
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
#define for_each_netdev(net, d) \
list_for_each_entry(d, &(net)->dev_base_head, dev_list)
#define for_each_netdev_reverse(net, d) \
list_for_each_entry_reverse(d, &(net)->dev_base_head, dev_list)
#define for_each_netdev_rcu(net, d) \
list_for_each_entry_rcu(d, &(net)->dev_base_head, dev_list)
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
#define for_each_netdev_safe(net, d, n) \
list_for_each_entry_safe(d, n, &(net)->dev_base_head, dev_list)
#define for_each_netdev_continue(net, d) \
list_for_each_entry_continue(d, &(net)->dev_base_head, dev_list)
#define for_each_netdev_continue_rcu(net, d) \
list_for_each_entry_continue_rcu(d, &(net)->dev_base_head, dev_list)
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
#define net_device_entry(lh) list_entry(lh, struct net_device, dev_list)
static inline struct net_device *next_net_device(struct net_device *dev)
{
struct list_head *lh;
struct net *net;
net = dev_net(dev);
lh = dev->dev_list.next;
return lh == &net->dev_base_head ? NULL : net_device_entry(lh);
}
static inline struct net_device *next_net_device_rcu(struct net_device *dev)
{
struct list_head *lh;
struct net *net;
net = dev_net(dev);
lh = rcu_dereference(dev->dev_list.next);
return lh == &net->dev_base_head ? NULL : net_device_entry(lh);
}
static inline struct net_device *first_net_device(struct net *net)
{
return list_empty(&net->dev_base_head) ? NULL :
net_device_entry(net->dev_base_head.next);
}
extern int netdev_boot_setup_check(struct net_device *dev);
extern unsigned long netdev_boot_base(const char *prefix, int unit);
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
extern struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *hwaddr);
extern struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type);
extern struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type);
extern void dev_add_pack(struct packet_type *pt);
extern void dev_remove_pack(struct packet_type *pt);
extern void __dev_remove_pack(struct packet_type *pt);
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
extern struct net_device *dev_get_by_flags(struct net *net, unsigned short flags,
unsigned short mask);
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
extern struct net_device *dev_get_by_name(struct net *net, const char *name);
extern struct net_device *dev_get_by_name_rcu(struct net *net, const char *name);
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
extern struct net_device *__dev_get_by_name(struct net *net, const char *name);
extern int dev_alloc_name(struct net_device *dev, const char *name);
extern int dev_open(struct net_device *dev);
extern int dev_close(struct net_device *dev);
extern void dev_disable_lro(struct net_device *dev);
extern int dev_queue_xmit(struct sk_buff *skb);
extern int register_netdevice(struct net_device *dev);
extern void unregister_netdevice_queue(struct net_device *dev,
struct list_head *head);
extern void unregister_netdevice_many(struct list_head *head);
static inline void unregister_netdevice(struct net_device *dev)
{
unregister_netdevice_queue(dev, NULL);
}
extern void free_netdev(struct net_device *dev);
extern void synchronize_net(void);
extern int register_netdevice_notifier(struct notifier_block *nb);
extern int unregister_netdevice_notifier(struct notifier_block *nb);
extern int init_dummy_netdev(struct net_device *dev);
extern void netdev_resync_ops(struct net_device *dev);
extern int call_netdevice_notifiers(unsigned long val, struct net_device *dev);
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
extern struct net_device *dev_get_by_index(struct net *net, int ifindex);
extern struct net_device *__dev_get_by_index(struct net *net, int ifindex);
extern struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex);
extern int dev_restart(struct net_device *dev);
#ifdef CONFIG_NETPOLL_TRAP
extern int netpoll_trap(void);
#endif
extern int skb_gro_receive(struct sk_buff **head,
struct sk_buff *skb);
extern void skb_gro_reset_offset(struct sk_buff *skb);
static inline unsigned int skb_gro_offset(const struct sk_buff *skb)
{
return NAPI_GRO_CB(skb)->data_offset;
}
static inline unsigned int skb_gro_len(const struct sk_buff *skb)
{
return skb->len - NAPI_GRO_CB(skb)->data_offset;
}
static inline void skb_gro_pull(struct sk_buff *skb, unsigned int len)
{
NAPI_GRO_CB(skb)->data_offset += len;
}
static inline void *skb_gro_header_fast(struct sk_buff *skb,
unsigned int offset)
{
return NAPI_GRO_CB(skb)->frag0 + offset;
}
static inline int skb_gro_header_hard(struct sk_buff *skb, unsigned int hlen)
{
return NAPI_GRO_CB(skb)->frag0_len < hlen;
}
static inline void *skb_gro_header_slow(struct sk_buff *skb, unsigned int hlen,
unsigned int offset)
{
NAPI_GRO_CB(skb)->frag0 = NULL;
NAPI_GRO_CB(skb)->frag0_len = 0;
return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL;
}
static inline void *skb_gro_mac_header(struct sk_buff *skb)
{
return NAPI_GRO_CB(skb)->frag0 ?: skb_mac_header(skb);
}
static inline void *skb_gro_network_header(struct sk_buff *skb)
{
return (NAPI_GRO_CB(skb)->frag0 ?: skb->data) +
skb_network_offset(skb);
}
static inline int dev_hard_header(struct sk_buff *skb, struct net_device *dev,
unsigned short type,
const void *daddr, const void *saddr,
unsigned len)
{
if (!dev->header_ops || !dev->header_ops->create)
return 0;
return dev->header_ops->create(skb, dev, type, daddr, saddr, len);
}
static inline int dev_parse_header(const struct sk_buff *skb,
unsigned char *haddr)
{
const struct net_device *dev = skb->dev;
if (!dev->header_ops || !dev->header_ops->parse)
return 0;
return dev->header_ops->parse(skb, haddr);
}
typedef int gifconf_func_t(struct net_device * dev, char __user * bufptr, int len);
extern int register_gifconf(unsigned int family, gifconf_func_t * gifconf);
static inline int unregister_gifconf(unsigned int family)
{
return register_gifconf(family, NULL);
}
/*
* Incoming packets are placed on per-cpu queues so that
* no locking is needed.
*/
struct softnet_data {
struct Qdisc *output_queue;
struct list_head poll_list;
struct sk_buff *completion_queue;
rps: Receive Packet Steering This patch implements software receive side packet steering (RPS). RPS distributes the load of received packet processing across multiple CPUs. Problem statement: Protocol processing done in the NAPI context for received packets is serialized per device queue and becomes a bottleneck under high packet load. This substantially limits pps that can be achieved on a single queue NIC and provides no scaling with multiple cores. This solution queues packets early on in the receive path on the backlog queues of other CPUs. This allows protocol processing (e.g. IP and TCP) to be performed on packets in parallel. For each device (or each receive queue in a multi-queue device) a mask of CPUs is set to indicate the CPUs that can process packets. A CPU is selected on a per packet basis by hashing contents of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index into the CPU mask. The IPI mechanism is used to raise networking receive softirqs between CPUs. This effectively emulates in software what a multi-queue NIC can provide, but is generic requiring no device support. Many devices now provide a hash over the 4-tuple on a per packet basis (e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash in an skb field, and that value in turn is used to index into the RPS maps. Using the HW generated hash can avoid cache misses on the packet when steering it to a remote CPU. The CPU mask is set on a per device and per queue basis in the sysfs variable /sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical bit maps for receive queues in the device (numbered by <n>). If a device does not support multi-queue, a single variable is used for the device (rx-0). Generally, we have found this technique increases pps capabilities of a single queue device with good CPU utilization. Optimal settings for the CPU mask seem to depend on architectures and cache hierarcy. Below are some results running 500 instances of netperf TCP_RR test with 1 byte req. and resp. Results show cumulative transaction rate and system CPU utilization. e1000e on 8 core Intel Without RPS: 108K tps at 33% CPU With RPS: 311K tps at 64% CPU forcedeth on 16 core AMD Without RPS: 156K tps at 15% CPU With RPS: 404K tps at 49% CPU bnx2x on 16 core AMD Without RPS 567K tps at 61% CPU (4 HW RX queues) Without RPS 738K tps at 96% CPU (8 HW RX queues) With RPS: 854K tps at 76% CPU (4 HW RX queues) Caveats: - The benefits of this patch are dependent on architecture and cache hierarchy. Tuning the masks to get best performance is probably necessary. - This patch adds overhead in the path for processing a single packet. In a lightly loaded server this overhead may eliminate the advantages of increased parallelism, and possibly cause some relative performance degradation. We have found that masks that are cache aware (share same caches with the interrupting CPU) mitigate much of this. - The RPS masks can be changed dynamically, however whenever the mask is changed this introduces the possibility of generating out of order packets. It's probably best not change the masks too frequently. Signed-off-by: Tom Herbert <therbert@google.com> include/linux/netdevice.h | 32 ++++- include/linux/skbuff.h | 3 + net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++-------- net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++- net/core/skbuff.c | 2 + 5 files changed, 538 insertions(+), 59 deletions(-) Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-03-16 11:03:29 +03:00
/* Elements below can be accessed between CPUs for RPS */
#ifdef CONFIG_SMP
rps: Receive Packet Steering This patch implements software receive side packet steering (RPS). RPS distributes the load of received packet processing across multiple CPUs. Problem statement: Protocol processing done in the NAPI context for received packets is serialized per device queue and becomes a bottleneck under high packet load. This substantially limits pps that can be achieved on a single queue NIC and provides no scaling with multiple cores. This solution queues packets early on in the receive path on the backlog queues of other CPUs. This allows protocol processing (e.g. IP and TCP) to be performed on packets in parallel. For each device (or each receive queue in a multi-queue device) a mask of CPUs is set to indicate the CPUs that can process packets. A CPU is selected on a per packet basis by hashing contents of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index into the CPU mask. The IPI mechanism is used to raise networking receive softirqs between CPUs. This effectively emulates in software what a multi-queue NIC can provide, but is generic requiring no device support. Many devices now provide a hash over the 4-tuple on a per packet basis (e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash in an skb field, and that value in turn is used to index into the RPS maps. Using the HW generated hash can avoid cache misses on the packet when steering it to a remote CPU. The CPU mask is set on a per device and per queue basis in the sysfs variable /sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical bit maps for receive queues in the device (numbered by <n>). If a device does not support multi-queue, a single variable is used for the device (rx-0). Generally, we have found this technique increases pps capabilities of a single queue device with good CPU utilization. Optimal settings for the CPU mask seem to depend on architectures and cache hierarcy. Below are some results running 500 instances of netperf TCP_RR test with 1 byte req. and resp. Results show cumulative transaction rate and system CPU utilization. e1000e on 8 core Intel Without RPS: 108K tps at 33% CPU With RPS: 311K tps at 64% CPU forcedeth on 16 core AMD Without RPS: 156K tps at 15% CPU With RPS: 404K tps at 49% CPU bnx2x on 16 core AMD Without RPS 567K tps at 61% CPU (4 HW RX queues) Without RPS 738K tps at 96% CPU (8 HW RX queues) With RPS: 854K tps at 76% CPU (4 HW RX queues) Caveats: - The benefits of this patch are dependent on architecture and cache hierarchy. Tuning the masks to get best performance is probably necessary. - This patch adds overhead in the path for processing a single packet. In a lightly loaded server this overhead may eliminate the advantages of increased parallelism, and possibly cause some relative performance degradation. We have found that masks that are cache aware (share same caches with the interrupting CPU) mitigate much of this. - The RPS masks can be changed dynamically, however whenever the mask is changed this introduces the possibility of generating out of order packets. It's probably best not change the masks too frequently. Signed-off-by: Tom Herbert <therbert@google.com> include/linux/netdevice.h | 32 ++++- include/linux/skbuff.h | 3 + net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++-------- net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++- net/core/skbuff.c | 2 + 5 files changed, 538 insertions(+), 59 deletions(-) Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-03-16 11:03:29 +03:00
struct call_single_data csd ____cacheline_aligned_in_smp;
#endif
rps: Receive Packet Steering This patch implements software receive side packet steering (RPS). RPS distributes the load of received packet processing across multiple CPUs. Problem statement: Protocol processing done in the NAPI context for received packets is serialized per device queue and becomes a bottleneck under high packet load. This substantially limits pps that can be achieved on a single queue NIC and provides no scaling with multiple cores. This solution queues packets early on in the receive path on the backlog queues of other CPUs. This allows protocol processing (e.g. IP and TCP) to be performed on packets in parallel. For each device (or each receive queue in a multi-queue device) a mask of CPUs is set to indicate the CPUs that can process packets. A CPU is selected on a per packet basis by hashing contents of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index into the CPU mask. The IPI mechanism is used to raise networking receive softirqs between CPUs. This effectively emulates in software what a multi-queue NIC can provide, but is generic requiring no device support. Many devices now provide a hash over the 4-tuple on a per packet basis (e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash in an skb field, and that value in turn is used to index into the RPS maps. Using the HW generated hash can avoid cache misses on the packet when steering it to a remote CPU. The CPU mask is set on a per device and per queue basis in the sysfs variable /sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical bit maps for receive queues in the device (numbered by <n>). If a device does not support multi-queue, a single variable is used for the device (rx-0). Generally, we have found this technique increases pps capabilities of a single queue device with good CPU utilization. Optimal settings for the CPU mask seem to depend on architectures and cache hierarcy. Below are some results running 500 instances of netperf TCP_RR test with 1 byte req. and resp. Results show cumulative transaction rate and system CPU utilization. e1000e on 8 core Intel Without RPS: 108K tps at 33% CPU With RPS: 311K tps at 64% CPU forcedeth on 16 core AMD Without RPS: 156K tps at 15% CPU With RPS: 404K tps at 49% CPU bnx2x on 16 core AMD Without RPS 567K tps at 61% CPU (4 HW RX queues) Without RPS 738K tps at 96% CPU (8 HW RX queues) With RPS: 854K tps at 76% CPU (4 HW RX queues) Caveats: - The benefits of this patch are dependent on architecture and cache hierarchy. Tuning the masks to get best performance is probably necessary. - This patch adds overhead in the path for processing a single packet. In a lightly loaded server this overhead may eliminate the advantages of increased parallelism, and possibly cause some relative performance degradation. We have found that masks that are cache aware (share same caches with the interrupting CPU) mitigate much of this. - The RPS masks can be changed dynamically, however whenever the mask is changed this introduces the possibility of generating out of order packets. It's probably best not change the masks too frequently. Signed-off-by: Tom Herbert <therbert@google.com> include/linux/netdevice.h | 32 ++++- include/linux/skbuff.h | 3 + net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++-------- net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++- net/core/skbuff.c | 2 + 5 files changed, 538 insertions(+), 59 deletions(-) Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-03-16 11:03:29 +03:00
struct sk_buff_head input_pkt_queue;
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
struct napi_struct backlog;
};
rps: Receive Packet Steering This patch implements software receive side packet steering (RPS). RPS distributes the load of received packet processing across multiple CPUs. Problem statement: Protocol processing done in the NAPI context for received packets is serialized per device queue and becomes a bottleneck under high packet load. This substantially limits pps that can be achieved on a single queue NIC and provides no scaling with multiple cores. This solution queues packets early on in the receive path on the backlog queues of other CPUs. This allows protocol processing (e.g. IP and TCP) to be performed on packets in parallel. For each device (or each receive queue in a multi-queue device) a mask of CPUs is set to indicate the CPUs that can process packets. A CPU is selected on a per packet basis by hashing contents of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index into the CPU mask. The IPI mechanism is used to raise networking receive softirqs between CPUs. This effectively emulates in software what a multi-queue NIC can provide, but is generic requiring no device support. Many devices now provide a hash over the 4-tuple on a per packet basis (e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash in an skb field, and that value in turn is used to index into the RPS maps. Using the HW generated hash can avoid cache misses on the packet when steering it to a remote CPU. The CPU mask is set on a per device and per queue basis in the sysfs variable /sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical bit maps for receive queues in the device (numbered by <n>). If a device does not support multi-queue, a single variable is used for the device (rx-0). Generally, we have found this technique increases pps capabilities of a single queue device with good CPU utilization. Optimal settings for the CPU mask seem to depend on architectures and cache hierarcy. Below are some results running 500 instances of netperf TCP_RR test with 1 byte req. and resp. Results show cumulative transaction rate and system CPU utilization. e1000e on 8 core Intel Without RPS: 108K tps at 33% CPU With RPS: 311K tps at 64% CPU forcedeth on 16 core AMD Without RPS: 156K tps at 15% CPU With RPS: 404K tps at 49% CPU bnx2x on 16 core AMD Without RPS 567K tps at 61% CPU (4 HW RX queues) Without RPS 738K tps at 96% CPU (8 HW RX queues) With RPS: 854K tps at 76% CPU (4 HW RX queues) Caveats: - The benefits of this patch are dependent on architecture and cache hierarchy. Tuning the masks to get best performance is probably necessary. - This patch adds overhead in the path for processing a single packet. In a lightly loaded server this overhead may eliminate the advantages of increased parallelism, and possibly cause some relative performance degradation. We have found that masks that are cache aware (share same caches with the interrupting CPU) mitigate much of this. - The RPS masks can be changed dynamically, however whenever the mask is changed this introduces the possibility of generating out of order packets. It's probably best not change the masks too frequently. Signed-off-by: Tom Herbert <therbert@google.com> include/linux/netdevice.h | 32 ++++- include/linux/skbuff.h | 3 + net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++-------- net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++- net/core/skbuff.c | 2 + 5 files changed, 538 insertions(+), 59 deletions(-) Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-03-16 11:03:29 +03:00
DECLARE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
#define HAVE_NETIF_QUEUE
extern void __netif_schedule(struct Qdisc *q);
static inline void netif_schedule_queue(struct netdev_queue *txq)
{
if (!test_bit(__QUEUE_STATE_XOFF, &txq->state))
__netif_schedule(txq->qdisc);
}
static inline void netif_tx_schedule_all(struct net_device *dev)
{
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++)
netif_schedule_queue(netdev_get_tx_queue(dev, i));
}
static inline void netif_tx_start_queue(struct netdev_queue *dev_queue)
{
clear_bit(__QUEUE_STATE_XOFF, &dev_queue->state);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_start_queue - allow transmit
* @dev: network device
*
* Allow upper layers to call the device hard_start_xmit routine.
*/
static inline void netif_start_queue(struct net_device *dev)
{
netif_tx_start_queue(netdev_get_tx_queue(dev, 0));
}
static inline void netif_tx_start_all_queues(struct net_device *dev)
{
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
netif_tx_start_queue(txq);
}
}
static inline void netif_tx_wake_queue(struct netdev_queue *dev_queue)
{
#ifdef CONFIG_NETPOLL_TRAP
if (netpoll_trap()) {
netif_tx_start_queue(dev_queue);
return;
}
#endif
if (test_and_clear_bit(__QUEUE_STATE_XOFF, &dev_queue->state))
__netif_schedule(dev_queue->qdisc);
}
/**
* netif_wake_queue - restart transmit
* @dev: network device
*
* Allow upper layers to call the device hard_start_xmit routine.
* Used for flow control when transmit resources are available.
*/
static inline void netif_wake_queue(struct net_device *dev)
{
netif_tx_wake_queue(netdev_get_tx_queue(dev, 0));
}
static inline void netif_tx_wake_all_queues(struct net_device *dev)
{
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
netif_tx_wake_queue(txq);
}
}
static inline void netif_tx_stop_queue(struct netdev_queue *dev_queue)
{
set_bit(__QUEUE_STATE_XOFF, &dev_queue->state);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_stop_queue - stop transmitted packets
* @dev: network device
*
* Stop upper layers calling the device hard_start_xmit routine.
* Used for flow control when transmit resources are unavailable.
*/
static inline void netif_stop_queue(struct net_device *dev)
{
netif_tx_stop_queue(netdev_get_tx_queue(dev, 0));
}
static inline void netif_tx_stop_all_queues(struct net_device *dev)
{
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
netif_tx_stop_queue(txq);
}
}
static inline int netif_tx_queue_stopped(const struct netdev_queue *dev_queue)
{
return test_bit(__QUEUE_STATE_XOFF, &dev_queue->state);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_queue_stopped - test if transmit queue is flowblocked
* @dev: network device
*
* Test if transmit queue on device is currently unable to send.
*/
static inline int netif_queue_stopped(const struct net_device *dev)
{
return netif_tx_queue_stopped(netdev_get_tx_queue(dev, 0));
}
static inline int netif_tx_queue_frozen(const struct netdev_queue *dev_queue)
{
return test_bit(__QUEUE_STATE_FROZEN, &dev_queue->state);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_running - test if up
* @dev: network device
*
* Test if the device has been brought up.
*/
static inline int netif_running(const struct net_device *dev)
{
return test_bit(__LINK_STATE_START, &dev->state);
}
/*
* Routines to manage the subqueues on a device. We only need start
* stop, and a check if it's stopped. All other device management is
* done at the overall netdevice level.
* Also test the device if we're multiqueue.
*/
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_start_subqueue - allow sending packets on subqueue
* @dev: network device
* @queue_index: sub queue index
*
* Start individual transmit queue of a device with multiple transmit queues.
*/
static inline void netif_start_subqueue(struct net_device *dev, u16 queue_index)
{
struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
netif_tx_start_queue(txq);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_stop_subqueue - stop sending packets on subqueue
* @dev: network device
* @queue_index: sub queue index
*
* Stop individual transmit queue of a device with multiple transmit queues.
*/
static inline void netif_stop_subqueue(struct net_device *dev, u16 queue_index)
{
struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
#ifdef CONFIG_NETPOLL_TRAP
if (netpoll_trap())
return;
#endif
netif_tx_stop_queue(txq);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_subqueue_stopped - test status of subqueue
* @dev: network device
* @queue_index: sub queue index
*
* Check individual transmit queue of a device with multiple transmit queues.
*/
static inline int __netif_subqueue_stopped(const struct net_device *dev,
u16 queue_index)
{
struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
return netif_tx_queue_stopped(txq);
}
static inline int netif_subqueue_stopped(const struct net_device *dev,
struct sk_buff *skb)
{
return __netif_subqueue_stopped(dev, skb_get_queue_mapping(skb));
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_wake_subqueue - allow sending packets on subqueue
* @dev: network device
* @queue_index: sub queue index
*
* Resume individual transmit queue of a device with multiple transmit queues.
*/
static inline void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
{
struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
#ifdef CONFIG_NETPOLL_TRAP
if (netpoll_trap())
return;
#endif
if (test_and_clear_bit(__QUEUE_STATE_XOFF, &txq->state))
__netif_schedule(txq->qdisc);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_is_multiqueue - test if device has multiple transmit queues
* @dev: network device
*
* Check if device has multiple transmit queues
*/
static inline int netif_is_multiqueue(const struct net_device *dev)
{
return (dev->num_tx_queues > 1);
}
/* Use this variant when it is known for sure that it
* is executing from hardware interrupt context or with hardware interrupts
* disabled.
*/
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
extern void dev_kfree_skb_irq(struct sk_buff *skb);
/* Use this variant in places where it could be invoked
* from either hardware interrupt or other context, with hardware interrupts
* either disabled or enabled.
*/
[NET]: Deinline some larger functions from netdevice.h On a allyesconfig'ured kernel: Size Uses Wasted Name and definition ===== ==== ====== ================================================ 95 162 12075 netif_wake_queue include/linux/netdevice.h 129 86 9265 dev_kfree_skb_any include/linux/netdevice.h 127 56 5885 netif_device_attach include/linux/netdevice.h 73 86 4505 dev_kfree_skb_irq include/linux/netdevice.h 46 60 1534 netif_device_detach include/linux/netdevice.h 119 16 1485 __netif_rx_schedule include/linux/netdevice.h 143 5 492 netif_rx_schedule include/linux/netdevice.h 81 7 366 netif_schedule include/linux/netdevice.h netif_wake_queue is big because __netif_schedule is a big inline: static inline void __netif_schedule(struct net_device *dev) { if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) { unsigned long flags; struct softnet_data *sd; local_irq_save(flags); sd = &__get_cpu_var(softnet_data); dev->next_sched = sd->output_queue; sd->output_queue = dev; raise_softirq_irqoff(NET_TX_SOFTIRQ); local_irq_restore(flags); } } static inline void netif_wake_queue(struct net_device *dev) { #ifdef CONFIG_NETPOLL_TRAP if (netpoll_trap()) return; #endif if (test_and_clear_bit(__LINK_STATE_XOFF, &dev->state)) __netif_schedule(dev); } By de-inlining __netif_schedule we are saving a lot of text at each callsite of netif_wake_queue and netif_schedule. __netif_rx_schedule is also big, and it makes more sense to keep both of them out of line. Patch also deinlines dev_kfree_skb_any. We can deinline dev_kfree_skb_irq instead... oh well. netif_device_attach/detach are not hot paths, we can deinline them too. Signed-off-by: Denis Vlasenko <vda@ilport.com.ua> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-03-30 03:57:29 +04:00
extern void dev_kfree_skb_any(struct sk_buff *skb);
#define HAVE_NETIF_RX 1
extern int netif_rx(struct sk_buff *skb);
extern int netif_rx_ni(struct sk_buff *skb);
#define HAVE_NETIF_RECEIVE_SKB 1
extern int netif_receive_skb(struct sk_buff *skb);
extern gro_result_t dev_gro_receive(struct napi_struct *napi,
struct sk_buff *skb);
extern gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb);
extern gro_result_t napi_gro_receive(struct napi_struct *napi,
net: Add Generic Receive Offload infrastructure This patch adds the top-level GRO (Generic Receive Offload) infrastructure. This is pretty similar to LRO except that this is protocol-independent. Instead of holding packets in an lro_mgr structure, they're now held in napi_struct. For drivers that intend to use this, they can set the NETIF_F_GRO bit and call napi_gro_receive instead of netif_receive_skb or just call netif_rx. The latter will call napi_receive_skb automatically. When napi_gro_receive is used, the driver must either call napi_complete/napi_rx_complete, or call napi_gro_flush in softirq context if the driver uses the primitives __napi_complete/__napi_rx_complete. Protocols will set the gro_receive and gro_complete function pointers in order to participate in this scheme. In addition to the packet, gro_receive will get a list of currently held packets. Each packet in the list has a same_flow field which is non-zero if it is a potential match for the new packet. For each packet that may match, they also have a flush field which is non-zero if the held packet must not be merged with the new packet. Once gro_receive has determined that the new skb matches a held packet, the held packet may be processed immediately if the new skb cannot be merged with it. In this case gro_receive should return the pointer to the existing skb in gro_list. Otherwise the new skb should be merged into the existing packet and NULL should be returned, unless the new skb makes it impossible for any further merges to be made (e.g., FIN packet) where the merged skb should be returned. Whenever the skb is merged into an existing entry, the gro_receive function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb merely matches an existing entry but can't be merged with it, then this shouldn't be set. If gro_receive finds it pointless to hold the new skb for future merging, it should set NAPI_GRO_CB(skb)->flush. Held packets will be flushed by napi_gro_flush which is called by napi_complete and napi_rx_complete. Currently held packets are stored in a singly liked list just like LRO. The list is limited to a maximum of 8 entries. In future, this may be expanded to use a hash table to allow more flows to be held for merging. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 10:38:52 +03:00
struct sk_buff *skb);
extern void napi_reuse_skb(struct napi_struct *napi,
struct sk_buff *skb);
extern struct sk_buff * napi_get_frags(struct napi_struct *napi);
extern gro_result_t napi_frags_finish(struct napi_struct *napi,
struct sk_buff *skb,
gro_result_t ret);
extern struct sk_buff * napi_frags_skb(struct napi_struct *napi);
extern gro_result_t napi_gro_frags(struct napi_struct *napi);
static inline void napi_free_frags(struct napi_struct *napi)
{
kfree_skb(napi->skb);
napi->skb = NULL;
}
extern void netif_nit_deliver(struct sk_buff *skb);
extern int dev_valid_name(const char *name);
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
extern int dev_ioctl(struct net *net, unsigned int cmd, void __user *);
extern int dev_ethtool(struct net *net, struct ifreq *);
extern unsigned dev_get_flags(const struct net_device *);
extern int __dev_change_flags(struct net_device *, unsigned int flags);
extern int dev_change_flags(struct net_device *, unsigned);
extern void __dev_notify_flags(struct net_device *, unsigned int old_flags);
extern int dev_change_name(struct net_device *, const char *);
extern int dev_set_alias(struct net_device *, const char *, size_t);
extern int dev_change_net_namespace(struct net_device *,
struct net *, const char *);
extern int dev_set_mtu(struct net_device *, int);
extern int dev_set_mac_address(struct net_device *,
struct sockaddr *);
extern int dev_hard_start_xmit(struct sk_buff *skb,
struct net_device *dev,
struct netdev_queue *txq);
extern int dev_forward_skb(struct net_device *dev,
struct sk_buff *skb);
extern int netdev_budget;
/* Called by rtnetlink.c:rtnl_unlock() */
extern void netdev_run_todo(void);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* dev_put - release reference to device
* @dev: network device
*
* Release reference to device to allow it to be freed.
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
*/
static inline void dev_put(struct net_device *dev)
{
atomic_dec(&dev->refcnt);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* dev_hold - get reference to device
* @dev: network device
*
* Hold reference to device to keep it from being freed.
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
*/
static inline void dev_hold(struct net_device *dev)
{
atomic_inc(&dev->refcnt);
}
/* Carrier loss detection, dial on demand. The functions netif_carrier_on
* and _off may be called from IRQ context, but it is caller
* who is responsible for serialization of these calls.
*
* The name carrier is inappropriate, these functions should really be
* called netif_lowerlayer_*() because they represent the state of any
* kind of lower layer not just hardware media.
*/
extern void linkwatch_fire_event(struct net_device *dev);
linkwatch: linkwatch_forget_dev() to speedup device dismantle Herbert Xu a écrit : > On Tue, Nov 17, 2009 at 04:26:04AM -0800, David Miller wrote: >> Really, the link watch stuff is just due for a redesign. I don't >> think a simple hack is going to cut it this time, sorry Eric :-) > > I have no objections against any redesigns, but since the only > caller of linkwatch_forget_dev runs in process context with the > RTNL, it could also legally emit those events. Thanks guys, here an updated version then, before linkwatch surgery ? In this version, I force the event to be sent synchronously. [PATCH net-next-2.6] linkwatch: linkwatch_forget_dev() to speedup device dismantle time ip link del eth3.103 ; time ip link del eth3.104 ; time ip link del eth3.105 real 0m0.266s user 0m0.000s sys 0m0.001s real 0m0.770s user 0m0.000s sys 0m0.000s real 0m1.022s user 0m0.000s sys 0m0.000s One problem of current schem in vlan dismantle phase is the holding of device done by following chain : vlan_dev_stop() -> netif_carrier_off(dev) -> linkwatch_fire_event(dev) -> dev_hold() ... And __linkwatch_run_queue() runs up to one second later... A generic fix to this problem is to add a linkwatch_forget_dev() method to unlink the device from the list of watched devices. dev->link_watch_next becomes dev->link_watch_list (and use a bit more memory), to be able to unlink device in O(1). After patch : time ip link del eth3.103 ; time ip link del eth3.104 ; time ip link del eth3.105 real 0m0.024s user 0m0.000s sys 0m0.000s real 0m0.032s user 0m0.000s sys 0m0.001s real 0m0.033s user 0m0.000s sys 0m0.000s Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-11-17 08:59:21 +03:00
extern void linkwatch_forget_dev(struct net_device *dev);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_carrier_ok - test if carrier present
* @dev: network device
*
* Check if carrier is present on device
*/
static inline int netif_carrier_ok(const struct net_device *dev)
{
return !test_bit(__LINK_STATE_NOCARRIER, &dev->state);
}
extern unsigned long dev_trans_start(struct net_device *dev);
extern void __netdev_watchdog_up(struct net_device *dev);
extern void netif_carrier_on(struct net_device *dev);
extern void netif_carrier_off(struct net_device *dev);
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_dormant_on - mark device as dormant.
* @dev: network device
*
* Mark device as dormant (as per RFC2863).
*
* The dormant state indicates that the relevant interface is not
* actually in a condition to pass packets (i.e., it is not 'up') but is
* in a "pending" state, waiting for some external event. For "on-
* demand" interfaces, this new state identifies the situation where the
* interface is waiting for events to place it in the up state.
*
*/
static inline void netif_dormant_on(struct net_device *dev)
{
if (!test_and_set_bit(__LINK_STATE_DORMANT, &dev->state))
linkwatch_fire_event(dev);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_dormant_off - set device as not dormant.
* @dev: network device
*
* Device is not in dormant state.
*/
static inline void netif_dormant_off(struct net_device *dev)
{
if (test_and_clear_bit(__LINK_STATE_DORMANT, &dev->state))
linkwatch_fire_event(dev);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_dormant - test if carrier present
* @dev: network device
*
* Check if carrier is present on device
*/
static inline int netif_dormant(const struct net_device *dev)
{
return test_bit(__LINK_STATE_DORMANT, &dev->state);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_oper_up - test if device is operational
* @dev: network device
*
* Check if carrier is operational
*/
static inline int netif_oper_up(const struct net_device *dev)
{
return (dev->operstate == IF_OPER_UP ||
dev->operstate == IF_OPER_UNKNOWN /* backward compat */);
}
[NET]: Make NAPI polling independent of struct net_device objects. Several devices have multiple independant RX queues per net device, and some have a single interrupt doorbell for several queues. In either case, it's easier to support layouts like that if the structure representing the poll is independant from the net device itself. The signature of the ->poll() call back goes from: int foo_poll(struct net_device *dev, int *budget) to int foo_poll(struct napi_struct *napi, int budget) The caller is returned the number of RX packets processed (or the number of "NAPI credits" consumed if you want to get abstract). The callee no longer messes around bumping dev->quota, *budget, etc. because that is all handled in the caller upon return. The napi_struct is to be embedded in the device driver private data structures. Furthermore, it is the driver's responsibility to disable all NAPI instances in it's ->stop() device close handler. Since the napi_struct is privatized into the driver's private data structures, only the driver knows how to get at all of the napi_struct instances it may have per-device. With lots of help and suggestions from Rusty Russell, Roland Dreier, Michael Chan, Jeff Garzik, and Jamal Hadi Salim. Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra, Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan. [ Ported to current tree and all drivers converted. Integrated Stephen's follow-on kerneldoc additions, and restored poll_list handling to the old style to fix mutual exclusion issues. -DaveM ] Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-04 03:41:36 +04:00
/**
* netif_device_present - is device available or removed
* @dev: network device
*
* Check if device has not been removed from system.
*/
static inline int netif_device_present(struct net_device *dev)
{
return test_bit(__LINK_STATE_PRESENT, &dev->state);
}
[NET]: Deinline some larger functions from netdevice.h On a allyesconfig'ured kernel: Size Uses Wasted Name and definition ===== ==== ====== ================================================ 95 162 12075 netif_wake_queue include/linux/netdevice.h 129 86 9265 dev_kfree_skb_any include/linux/netdevice.h 127 56 5885 netif_device_attach include/linux/netdevice.h 73 86 4505 dev_kfree_skb_irq include/linux/netdevice.h 46 60 1534 netif_device_detach include/linux/netdevice.h 119 16 1485 __netif_rx_schedule include/linux/netdevice.h 143 5 492 netif_rx_schedule include/linux/netdevice.h 81 7 366 netif_schedule include/linux/netdevice.h netif_wake_queue is big because __netif_schedule is a big inline: static inline void __netif_schedule(struct net_device *dev) { if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) { unsigned long flags; struct softnet_data *sd; local_irq_save(flags); sd = &__get_cpu_var(softnet_data); dev->next_sched = sd->output_queue; sd->output_queue = dev; raise_softirq_irqoff(NET_TX_SOFTIRQ); local_irq_restore(flags); } } static inline void netif_wake_queue(struct net_device *dev) { #ifdef CONFIG_NETPOLL_TRAP if (netpoll_trap()) return; #endif if (test_and_clear_bit(__LINK_STATE_XOFF, &dev->state)) __netif_schedule(dev); } By de-inlining __netif_schedule we are saving a lot of text at each callsite of netif_wake_queue and netif_schedule. __netif_rx_schedule is also big, and it makes more sense to keep both of them out of line. Patch also deinlines dev_kfree_skb_any. We can deinline dev_kfree_skb_irq instead... oh well. netif_device_attach/detach are not hot paths, we can deinline them too. Signed-off-by: Denis Vlasenko <vda@ilport.com.ua> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-03-30 03:57:29 +04:00
extern void netif_device_detach(struct net_device *dev);
[NET]: Deinline some larger functions from netdevice.h On a allyesconfig'ured kernel: Size Uses Wasted Name and definition ===== ==== ====== ================================================ 95 162 12075 netif_wake_queue include/linux/netdevice.h 129 86 9265 dev_kfree_skb_any include/linux/netdevice.h 127 56 5885 netif_device_attach include/linux/netdevice.h 73 86 4505 dev_kfree_skb_irq include/linux/netdevice.h 46 60 1534 netif_device_detach include/linux/netdevice.h 119 16 1485 __netif_rx_schedule include/linux/netdevice.h 143 5 492 netif_rx_schedule include/linux/netdevice.h 81 7 366 netif_schedule include/linux/netdevice.h netif_wake_queue is big because __netif_schedule is a big inline: static inline void __netif_schedule(struct net_device *dev) { if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) { unsigned long flags; struct softnet_data *sd; local_irq_save(flags); sd = &__get_cpu_var(softnet_data); dev->next_sched = sd->output_queue; sd->output_queue = dev; raise_softirq_irqoff(NET_TX_SOFTIRQ); local_irq_restore(flags); } } static inline void netif_wake_queue(struct net_device *dev) { #ifdef CONFIG_NETPOLL_TRAP if (netpoll_trap()) return; #endif if (test_and_clear_bit(__LINK_STATE_XOFF, &dev->state)) __netif_schedule(dev); } By de-inlining __netif_schedule we are saving a lot of text at each callsite of netif_wake_queue and netif_schedule. __netif_rx_schedule is also big, and it makes more sense to keep both of them out of line. Patch also deinlines dev_kfree_skb_any. We can deinline dev_kfree_skb_irq instead... oh well. netif_device_attach/detach are not hot paths, we can deinline them too. Signed-off-by: Denis Vlasenko <vda@ilport.com.ua> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-03-30 03:57:29 +04:00
extern void netif_device_attach(struct net_device *dev);
/*
* Network interface message level settings
*/
#define HAVE_NETIF_MSG 1
enum {
NETIF_MSG_DRV = 0x0001,
NETIF_MSG_PROBE = 0x0002,
NETIF_MSG_LINK = 0x0004,
NETIF_MSG_TIMER = 0x0008,
NETIF_MSG_IFDOWN = 0x0010,
NETIF_MSG_IFUP = 0x0020,
NETIF_MSG_RX_ERR = 0x0040,
NETIF_MSG_TX_ERR = 0x0080,
NETIF_MSG_TX_QUEUED = 0x0100,
NETIF_MSG_INTR = 0x0200,
NETIF_MSG_TX_DONE = 0x0400,
NETIF_MSG_RX_STATUS = 0x0800,
NETIF_MSG_PKTDATA = 0x1000,
NETIF_MSG_HW = 0x2000,
NETIF_MSG_WOL = 0x4000,
};
#define netif_msg_drv(p) ((p)->msg_enable & NETIF_MSG_DRV)
#define netif_msg_probe(p) ((p)->msg_enable & NETIF_MSG_PROBE)
#define netif_msg_link(p) ((p)->msg_enable & NETIF_MSG_LINK)
#define netif_msg_timer(p) ((p)->msg_enable & NETIF_MSG_TIMER)
#define netif_msg_ifdown(p) ((p)->msg_enable & NETIF_MSG_IFDOWN)
#define netif_msg_ifup(p) ((p)->msg_enable & NETIF_MSG_IFUP)
#define netif_msg_rx_err(p) ((p)->msg_enable & NETIF_MSG_RX_ERR)
#define netif_msg_tx_err(p) ((p)->msg_enable & NETIF_MSG_TX_ERR)
#define netif_msg_tx_queued(p) ((p)->msg_enable & NETIF_MSG_TX_QUEUED)
#define netif_msg_intr(p) ((p)->msg_enable & NETIF_MSG_INTR)
#define netif_msg_tx_done(p) ((p)->msg_enable & NETIF_MSG_TX_DONE)
#define netif_msg_rx_status(p) ((p)->msg_enable & NETIF_MSG_RX_STATUS)
#define netif_msg_pktdata(p) ((p)->msg_enable & NETIF_MSG_PKTDATA)
#define netif_msg_hw(p) ((p)->msg_enable & NETIF_MSG_HW)
#define netif_msg_wol(p) ((p)->msg_enable & NETIF_MSG_WOL)
static inline u32 netif_msg_init(int debug_value, int default_msg_enable_bits)
{
/* use default */
if (debug_value < 0 || debug_value >= (sizeof(u32) * 8))
return default_msg_enable_bits;
if (debug_value == 0) /* no output */
return 0;
/* set low N bits */
return (1 << debug_value) - 1;
}
static inline void __netif_tx_lock(struct netdev_queue *txq, int cpu)
[NET]: Add netif_tx_lock Various drivers use xmit_lock internally to synchronise with their transmission routines. They do so without setting xmit_lock_owner. This is fine as long as netpoll is not in use. With netpoll it is possible for deadlocks to occur if xmit_lock_owner isn't set. This is because if a printk occurs while xmit_lock is held and xmit_lock_owner is not set can cause netpoll to attempt to take xmit_lock recursively. While it is possible to resolve this by getting netpoll to use trylock, it is suboptimal because netpoll's sole objective is to maximise the chance of getting the printk out on the wire. So delaying or dropping the message is to be avoided as much as possible. So the only alternative is to always set xmit_lock_owner. The following patch does this by introducing the netif_tx_lock family of functions that take care of setting/unsetting xmit_lock_owner. I renamed xmit_lock to _xmit_lock to indicate that it should not be used directly. I didn't provide irq versions of the netif_tx_lock functions since xmit_lock is meant to be a BH-disabling lock. This is pretty much a straight text substitution except for a small bug fix in winbond. It currently uses netif_stop_queue/spin_unlock_wait to stop transmission. This is unsafe as an IRQ can potentially wake up the queue. So it is safer to use netif_tx_disable. The hamradio bits used spin_lock_irq but it is unnecessary as xmit_lock must never be taken in an IRQ handler. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-06-09 23:20:56 +04:00
{
spin_lock(&txq->_xmit_lock);
txq->xmit_lock_owner = cpu;
}
static inline void __netif_tx_lock_bh(struct netdev_queue *txq)
{
spin_lock_bh(&txq->_xmit_lock);
txq->xmit_lock_owner = smp_processor_id();
}
static inline int __netif_tx_trylock(struct netdev_queue *txq)
{
int ok = spin_trylock(&txq->_xmit_lock);
if (likely(ok))
txq->xmit_lock_owner = smp_processor_id();
return ok;
}
static inline void __netif_tx_unlock(struct netdev_queue *txq)
{
txq->xmit_lock_owner = -1;
spin_unlock(&txq->_xmit_lock);
}
static inline void __netif_tx_unlock_bh(struct netdev_queue *txq)
{
txq->xmit_lock_owner = -1;
spin_unlock_bh(&txq->_xmit_lock);
}
static inline void txq_trans_update(struct netdev_queue *txq)
{
if (txq->xmit_lock_owner != -1)
txq->trans_start = jiffies;
}
/**
* netif_tx_lock - grab network device transmit lock
* @dev: network device
*
* Get network device transmit lock
*/
static inline void netif_tx_lock(struct net_device *dev)
{
unsigned int i;
int cpu;
spin_lock(&dev->tx_global_lock);
cpu = smp_processor_id();
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
/* We are the only thread of execution doing a
* freeze, but we have to grab the _xmit_lock in
* order to synchronize with threads which are in
* the ->hard_start_xmit() handler and already
* checked the frozen bit.
*/
__netif_tx_lock(txq, cpu);
set_bit(__QUEUE_STATE_FROZEN, &txq->state);
__netif_tx_unlock(txq);
}
[NET]: Add netif_tx_lock Various drivers use xmit_lock internally to synchronise with their transmission routines. They do so without setting xmit_lock_owner. This is fine as long as netpoll is not in use. With netpoll it is possible for deadlocks to occur if xmit_lock_owner isn't set. This is because if a printk occurs while xmit_lock is held and xmit_lock_owner is not set can cause netpoll to attempt to take xmit_lock recursively. While it is possible to resolve this by getting netpoll to use trylock, it is suboptimal because netpoll's sole objective is to maximise the chance of getting the printk out on the wire. So delaying or dropping the message is to be avoided as much as possible. So the only alternative is to always set xmit_lock_owner. The following patch does this by introducing the netif_tx_lock family of functions that take care of setting/unsetting xmit_lock_owner. I renamed xmit_lock to _xmit_lock to indicate that it should not be used directly. I didn't provide irq versions of the netif_tx_lock functions since xmit_lock is meant to be a BH-disabling lock. This is pretty much a straight text substitution except for a small bug fix in winbond. It currently uses netif_stop_queue/spin_unlock_wait to stop transmission. This is unsafe as an IRQ can potentially wake up the queue. So it is safer to use netif_tx_disable. The hamradio bits used spin_lock_irq but it is unnecessary as xmit_lock must never be taken in an IRQ handler. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-06-09 23:20:56 +04:00
}
static inline void netif_tx_lock_bh(struct net_device *dev)
{
local_bh_disable();
netif_tx_lock(dev);
[NET]: Add netif_tx_lock Various drivers use xmit_lock internally to synchronise with their transmission routines. They do so without setting xmit_lock_owner. This is fine as long as netpoll is not in use. With netpoll it is possible for deadlocks to occur if xmit_lock_owner isn't set. This is because if a printk occurs while xmit_lock is held and xmit_lock_owner is not set can cause netpoll to attempt to take xmit_lock recursively. While it is possible to resolve this by getting netpoll to use trylock, it is suboptimal because netpoll's sole objective is to maximise the chance of getting the printk out on the wire. So delaying or dropping the message is to be avoided as much as possible. So the only alternative is to always set xmit_lock_owner. The following patch does this by introducing the netif_tx_lock family of functions that take care of setting/unsetting xmit_lock_owner. I renamed xmit_lock to _xmit_lock to indicate that it should not be used directly. I didn't provide irq versions of the netif_tx_lock functions since xmit_lock is meant to be a BH-disabling lock. This is pretty much a straight text substitution except for a small bug fix in winbond. It currently uses netif_stop_queue/spin_unlock_wait to stop transmission. This is unsafe as an IRQ can potentially wake up the queue. So it is safer to use netif_tx_disable. The hamradio bits used spin_lock_irq but it is unnecessary as xmit_lock must never be taken in an IRQ handler. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-06-09 23:20:56 +04:00
}
static inline void netif_tx_unlock(struct net_device *dev)
{
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
/* No need to grab the _xmit_lock here. If the
* queue is not stopped for another reason, we
* force a schedule.
*/
clear_bit(__QUEUE_STATE_FROZEN, &txq->state);
netif_schedule_queue(txq);
}
spin_unlock(&dev->tx_global_lock);
[NET]: Add netif_tx_lock Various drivers use xmit_lock internally to synchronise with their transmission routines. They do so without setting xmit_lock_owner. This is fine as long as netpoll is not in use. With netpoll it is possible for deadlocks to occur if xmit_lock_owner isn't set. This is because if a printk occurs while xmit_lock is held and xmit_lock_owner is not set can cause netpoll to attempt to take xmit_lock recursively. While it is possible to resolve this by getting netpoll to use trylock, it is suboptimal because netpoll's sole objective is to maximise the chance of getting the printk out on the wire. So delaying or dropping the message is to be avoided as much as possible. So the only alternative is to always set xmit_lock_owner. The following patch does this by introducing the netif_tx_lock family of functions that take care of setting/unsetting xmit_lock_owner. I renamed xmit_lock to _xmit_lock to indicate that it should not be used directly. I didn't provide irq versions of the netif_tx_lock functions since xmit_lock is meant to be a BH-disabling lock. This is pretty much a straight text substitution except for a small bug fix in winbond. It currently uses netif_stop_queue/spin_unlock_wait to stop transmission. This is unsafe as an IRQ can potentially wake up the queue. So it is safer to use netif_tx_disable. The hamradio bits used spin_lock_irq but it is unnecessary as xmit_lock must never be taken in an IRQ handler. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-06-09 23:20:56 +04:00
}
static inline void netif_tx_unlock_bh(struct net_device *dev)
{
netif_tx_unlock(dev);
local_bh_enable();
[NET]: Add netif_tx_lock Various drivers use xmit_lock internally to synchronise with their transmission routines. They do so without setting xmit_lock_owner. This is fine as long as netpoll is not in use. With netpoll it is possible for deadlocks to occur if xmit_lock_owner isn't set. This is because if a printk occurs while xmit_lock is held and xmit_lock_owner is not set can cause netpoll to attempt to take xmit_lock recursively. While it is possible to resolve this by getting netpoll to use trylock, it is suboptimal because netpoll's sole objective is to maximise the chance of getting the printk out on the wire. So delaying or dropping the message is to be avoided as much as possible. So the only alternative is to always set xmit_lock_owner. The following patch does this by introducing the netif_tx_lock family of functions that take care of setting/unsetting xmit_lock_owner. I renamed xmit_lock to _xmit_lock to indicate that it should not be used directly. I didn't provide irq versions of the netif_tx_lock functions since xmit_lock is meant to be a BH-disabling lock. This is pretty much a straight text substitution except for a small bug fix in winbond. It currently uses netif_stop_queue/spin_unlock_wait to stop transmission. This is unsafe as an IRQ can potentially wake up the queue. So it is safer to use netif_tx_disable. The hamradio bits used spin_lock_irq but it is unnecessary as xmit_lock must never be taken in an IRQ handler. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2006-06-09 23:20:56 +04:00
}
#define HARD_TX_LOCK(dev, txq, cpu) { \
if ((dev->features & NETIF_F_LLTX) == 0) { \
__netif_tx_lock(txq, cpu); \
} \
}
#define HARD_TX_UNLOCK(dev, txq) { \
if ((dev->features & NETIF_F_LLTX) == 0) { \
__netif_tx_unlock(txq); \
} \
}
static inline void netif_tx_disable(struct net_device *dev)
{
unsigned int i;
int cpu;
local_bh_disable();
cpu = smp_processor_id();
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
__netif_tx_lock(txq, cpu);
netif_tx_stop_queue(txq);
__netif_tx_unlock(txq);
}
local_bh_enable();
}
static inline void netif_addr_lock(struct net_device *dev)
{
spin_lock(&dev->addr_list_lock);
}
static inline void netif_addr_lock_bh(struct net_device *dev)
{
spin_lock_bh(&dev->addr_list_lock);
}
static inline void netif_addr_unlock(struct net_device *dev)
{
spin_unlock(&dev->addr_list_lock);
}
static inline void netif_addr_unlock_bh(struct net_device *dev)
{
spin_unlock_bh(&dev->addr_list_lock);
}
/*
* dev_addrs walker. Should be used only for read access. Call with
* rcu_read_lock held.
*/
#define for_each_dev_addr(dev, ha) \
list_for_each_entry_rcu(ha, &dev->dev_addrs.list, list)
/* These functions live elsewhere (drivers/net/net_init.c, but related) */
extern void ether_setup(struct net_device *dev);
/* Support for loadable net-drivers */
extern struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
void (*setup)(struct net_device *),
unsigned int queue_count);
#define alloc_netdev(sizeof_priv, name, setup) \
alloc_netdev_mq(sizeof_priv, name, setup, 1)
extern int register_netdev(struct net_device *dev);
extern void unregister_netdev(struct net_device *dev);
/* Functions used for device addresses handling */
extern int dev_addr_add(struct net_device *dev, unsigned char *addr,
unsigned char addr_type);
extern int dev_addr_del(struct net_device *dev, unsigned char *addr,
unsigned char addr_type);
extern int dev_addr_add_multiple(struct net_device *to_dev,
struct net_device *from_dev,
unsigned char addr_type);
extern int dev_addr_del_multiple(struct net_device *to_dev,
struct net_device *from_dev,
unsigned char addr_type);
extern void dev_addr_flush(struct net_device *dev);
extern int dev_addr_init(struct net_device *dev);
/* Functions used for unicast addresses handling */
extern int dev_uc_add(struct net_device *dev, unsigned char *addr);
extern int dev_uc_del(struct net_device *dev, unsigned char *addr);
extern int dev_uc_sync(struct net_device *to, struct net_device *from);
extern void dev_uc_unsync(struct net_device *to, struct net_device *from);
extern void dev_uc_flush(struct net_device *dev);
extern void dev_uc_init(struct net_device *dev);
/* Functions used for secondary unicast and multicast support */
extern void dev_set_rx_mode(struct net_device *dev);
extern void __dev_set_rx_mode(struct net_device *dev);
extern int dev_mc_delete(struct net_device *dev, void *addr, int alen, int all);
extern int dev_mc_add(struct net_device *dev, void *addr, int alen, int newonly);
extern int dev_mc_sync(struct net_device *to, struct net_device *from);
extern void dev_mc_unsync(struct net_device *to, struct net_device *from);
extern void dev_addr_discard(struct net_device *dev);
extern int __dev_addr_delete(struct dev_addr_list **list, int *count, void *addr, int alen, int all);
extern int __dev_addr_add(struct dev_addr_list **list, int *count, void *addr, int alen, int newonly);
extern int __dev_addr_sync(struct dev_addr_list **to, int *to_count, struct dev_addr_list **from, int *from_count);
extern void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, struct dev_addr_list **from, int *from_count);
extern int dev_set_promiscuity(struct net_device *dev, int inc);
extern int dev_set_allmulti(struct net_device *dev, int inc);
extern void netdev_state_change(struct net_device *dev);
extern int netdev_bonding_change(struct net_device *dev,
unsigned long event);
extern void netdev_features_change(struct net_device *dev);
/* Load a device via the kmod */
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 22:56:21 +04:00
extern void dev_load(struct net *net, const char *name);
extern void dev_mcast_init(void);
extern const struct net_device_stats *dev_get_stats(struct net_device *dev);
extern void dev_txq_stats_fold(const struct net_device *dev, struct net_device_stats *stats);
extern int netdev_max_backlog;
extern int weight_p;
extern int netdev_set_master(struct net_device *dev, struct net_device *master);
extern int skb_checksum_help(struct sk_buff *skb);
extern struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features);
#ifdef CONFIG_BUG
extern void netdev_rx_csum_fault(struct net_device *dev);
#else
static inline void netdev_rx_csum_fault(struct net_device *dev)
{
}
#endif
/* rx skb timestamps */
extern void net_enable_timestamp(void);
extern void net_disable_timestamp(void);
#ifdef CONFIG_PROC_FS
extern void *dev_seq_start(struct seq_file *seq, loff_t *pos);
extern void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos);
extern void dev_seq_stop(struct seq_file *seq, void *v);
#endif
extern int netdev_class_create_file(struct class_attribute *class_attr);
extern void netdev_class_remove_file(struct class_attribute *class_attr);
extern char *netdev_drivername(const struct net_device *dev, char *buffer, int len);
extern void linkwatch_run_queue(void);
unsigned long netdev_increment_features(unsigned long all, unsigned long one,
unsigned long mask);
unsigned long netdev_fix_features(unsigned long features, const char *name);
void netif_stacked_transfer_operstate(const struct net_device *rootdev,
struct net_device *dev);
static inline int net_gso_ok(int features, int gso_type)
{
int feature = gso_type << NETIF_F_GSO_SHIFT;
return (features & feature) == feature;
}
static inline int skb_gso_ok(struct sk_buff *skb, int features)
{
return net_gso_ok(features, skb_shinfo(skb)->gso_type) &&
(!skb_has_frags(skb) || (features & NETIF_F_FRAGLIST));
}
static inline int netif_needs_gso(struct net_device *dev, struct sk_buff *skb)
{
return skb_is_gso(skb) &&
(!skb_gso_ok(skb, dev->features) ||
unlikely(skb->ip_summed != CHECKSUM_PARTIAL));
}
[NET]: Add per-connection option to set max TSO frame size Update: My mailer ate one of Jarek's feedback mails... Fixed the parameter in netif_set_gso_max_size() to be u32, not u16. Fixed the whitespace issue due to a patch import botch. Changed the types from u32 to unsigned int to be more consistent with other variables in the area. Also brought the patch up to the latest net-2.6.26 tree. Update: Made gso_max_size container 32 bits, not 16. Moved the location of gso_max_size within netdev to be less hotpath. Made more consistent names between the sock and netdev layers, and added a define for the max GSO size. Update: Respun for net-2.6.26 tree. Update: changed max_gso_frame_size and sk_gso_max_size from signed to unsigned - thanks Stephen! This patch adds the ability for device drivers to control the size of the TSO frames being sent to them, per TCP connection. By setting the netdevice's gso_max_size value, the socket layer will set the GSO frame size based on that value. This will propogate into the TCP layer, and send TSO's of that size to the hardware. This can be desirable to help tune the bursty nature of TSO on a per-adapter basis, where one may have 1 GbE and 10 GbE devices coexisting in a system, one running multiqueue and the other not, etc. This can also be desirable for devices that cannot support full 64 KB TSO's, but still want to benefit from some level of segmentation offloading. Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-03-21 13:43:19 +03:00
static inline void netif_set_gso_max_size(struct net_device *dev,
unsigned int size)
{
dev->gso_max_size = size;
}
bonding: allow bond in mode balance-alb to work properly in bridge -try4.3 [PATCH net-next] bonding: allow bond in mode balance-alb to work properly in bridge -try4.3 (updated) changes v4.2 -> v4.3 - memcpy the address always, not just in case it differs from master->dev_addr - compare_ether_addr_64bits() is not used so there is no direct need to make new header file (I think it would be good to have bond stuff in separate file anyway). changes v4.1 -> v4.2 - use skb->pkt_type == PACKET_HOST compare rather then comparing skb dest addr against skb->dev->dev_addr The problem is described in following bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=487763 Basically here's what's going on. In every mode, bonding interface uses the same mac address for all enslaved devices (except fail_over_mac). Only balance-alb will simultaneously use multiple MAC addresses across different slaves. When you put this kind of bond device into a bridge it will only add one of mac adresses into a hash list of mac addresses, say X. This mac address is marked as local. But this bonding interface also has mac address Y. Now then packet arrives with destination address Y, this address is not marked as local and the packed looks like it needs to be forwarded. This packet is then lost which is wrong. Notice that interfaces can be added and removed from bond while it is in bridge. *** When the multiple addresses for bridge port approach failed to solve this issue due to STP I started to think other way to solve this. I returned to previous solution but tweaked one. This patch solves the situation in the bonding without touching bridge code. For every incoming frame to bonding the destination address is compared to current address of the slave device from which tha packet came. If these two match destination address is replaced by mac address of the master. This address is known by bridge so it is delivered properly. Note that the comparsion is not made directly, it's used skb->pkt_type == PACKET_HOST instead. This is "set" previously in eth_type_trans(). I experimentally tried that this works as good as searching through the slave list (v4 of this patch). Jirka Signed-off-by: Jiri Pirko <jpirko@redhat.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-05-28 05:05:00 +04:00
static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
struct net_device *master)
{
if (skb->pkt_type == PACKET_HOST) {
u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
memcpy(dest, master->dev_addr, ETH_ALEN);
}
}
/* On bonding slaves other than the currently active slave, suppress
* duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
* ARP on active-backup slaves with arp_validate enabled.
*/
static inline int skb_bond_should_drop(struct sk_buff *skb,
struct net_device *master)
{
if (master) {
struct net_device *dev = skb->dev;
if (master->priv_flags & IFF_MASTER_ARPMON)
dev->last_rx = jiffies;
bonding: allow bond in mode balance-alb to work properly in bridge -try4.3 [PATCH net-next] bonding: allow bond in mode balance-alb to work properly in bridge -try4.3 (updated) changes v4.2 -> v4.3 - memcpy the address always, not just in case it differs from master->dev_addr - compare_ether_addr_64bits() is not used so there is no direct need to make new header file (I think it would be good to have bond stuff in separate file anyway). changes v4.1 -> v4.2 - use skb->pkt_type == PACKET_HOST compare rather then comparing skb dest addr against skb->dev->dev_addr The problem is described in following bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=487763 Basically here's what's going on. In every mode, bonding interface uses the same mac address for all enslaved devices (except fail_over_mac). Only balance-alb will simultaneously use multiple MAC addresses across different slaves. When you put this kind of bond device into a bridge it will only add one of mac adresses into a hash list of mac addresses, say X. This mac address is marked as local. But this bonding interface also has mac address Y. Now then packet arrives with destination address Y, this address is not marked as local and the packed looks like it needs to be forwarded. This packet is then lost which is wrong. Notice that interfaces can be added and removed from bond while it is in bridge. *** When the multiple addresses for bridge port approach failed to solve this issue due to STP I started to think other way to solve this. I returned to previous solution but tweaked one. This patch solves the situation in the bonding without touching bridge code. For every incoming frame to bonding the destination address is compared to current address of the slave device from which tha packet came. If these two match destination address is replaced by mac address of the master. This address is known by bridge so it is delivered properly. Note that the comparsion is not made directly, it's used skb->pkt_type == PACKET_HOST instead. This is "set" previously in eth_type_trans(). I experimentally tried that this works as good as searching through the slave list (v4 of this patch). Jirka Signed-off-by: Jiri Pirko <jpirko@redhat.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-05-28 05:05:00 +04:00
if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
/* Do address unmangle. The local destination address
* will be always the one master has. Provides the right
* functionality in a bridge.
*/
skb_bond_set_mac_by_master(skb, master);
}
if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
skb->protocol == __cpu_to_be16(ETH_P_ARP))
return 0;
if (master->priv_flags & IFF_MASTER_ALB) {
if (skb->pkt_type != PACKET_BROADCAST &&
skb->pkt_type != PACKET_MULTICAST)
return 0;
}
if (master->priv_flags & IFF_MASTER_8023AD &&
skb->protocol == __cpu_to_be16(ETH_P_SLOW))
return 0;
return 1;
}
}
return 0;
}
extern struct pernet_operations __net_initdata loopback_net_ops;
static inline int dev_ethtool_get_settings(struct net_device *dev,
struct ethtool_cmd *cmd)
{
if (!dev->ethtool_ops || !dev->ethtool_ops->get_settings)
return -EOPNOTSUPP;
return dev->ethtool_ops->get_settings(dev, cmd);
}
static inline u32 dev_ethtool_get_rx_csum(struct net_device *dev)
{
if (!dev->ethtool_ops || !dev->ethtool_ops->get_rx_csum)
return 0;
return dev->ethtool_ops->get_rx_csum(dev);
}
static inline u32 dev_ethtool_get_flags(struct net_device *dev)
{
if (!dev->ethtool_ops || !dev->ethtool_ops->get_flags)
return 0;
return dev->ethtool_ops->get_flags(dev);
}
/* Logging, debugging and troubleshooting/diagnostic helpers. */
/* netdev_printk helpers, similar to dev_printk */
static inline const char *netdev_name(const struct net_device *dev)
{
if (dev->reg_state != NETREG_REGISTERED)
return "(unregistered net_device)";
return dev->name;
}
#define netdev_printk(level, netdev, format, args...) \
dev_printk(level, (netdev)->dev.parent, \
"%s: " format, \
netdev_name(netdev), ##args)
#define netdev_emerg(dev, format, args...) \
netdev_printk(KERN_EMERG, dev, format, ##args)
#define netdev_alert(dev, format, args...) \
netdev_printk(KERN_ALERT, dev, format, ##args)
#define netdev_crit(dev, format, args...) \
netdev_printk(KERN_CRIT, dev, format, ##args)
#define netdev_err(dev, format, args...) \
netdev_printk(KERN_ERR, dev, format, ##args)
#define netdev_warn(dev, format, args...) \
netdev_printk(KERN_WARNING, dev, format, ##args)
#define netdev_notice(dev, format, args...) \
netdev_printk(KERN_NOTICE, dev, format, ##args)
#define netdev_info(dev, format, args...) \
netdev_printk(KERN_INFO, dev, format, ##args)
#if defined(DEBUG)
#define netdev_dbg(__dev, format, args...) \
netdev_printk(KERN_DEBUG, __dev, format, ##args)
#elif defined(CONFIG_DYNAMIC_DEBUG)
#define netdev_dbg(__dev, format, args...) \
do { \
dynamic_dev_dbg((__dev)->dev.parent, "%s: " format, \
netdev_name(__dev), ##args); \
} while (0)
#else
#define netdev_dbg(__dev, format, args...) \
({ \
if (0) \
netdev_printk(KERN_DEBUG, __dev, format, ##args); \
0; \
})
#endif
#if defined(VERBOSE_DEBUG)
#define netdev_vdbg netdev_dbg
#else
#define netdev_vdbg(dev, format, args...) \
({ \
if (0) \
netdev_printk(KERN_DEBUG, dev, format, ##args); \
0; \
})
#endif
/*
* netdev_WARN() acts like dev_printk(), but with the key difference
* of using a WARN/WARN_ON to get the message out, including the
* file/line information and a backtrace.
*/
#define netdev_WARN(dev, format, args...) \
WARN(1, "netdevice: %s\n" format, netdev_name(dev), ##args);
/* netif printk helpers, similar to netdev_printk */
#define netif_printk(priv, type, level, dev, fmt, args...) \
do { \
if (netif_msg_##type(priv)) \
netdev_printk(level, (dev), fmt, ##args); \
} while (0)
#define netif_emerg(priv, type, dev, fmt, args...) \
netif_printk(priv, type, KERN_EMERG, dev, fmt, ##args)
#define netif_alert(priv, type, dev, fmt, args...) \
netif_printk(priv, type, KERN_ALERT, dev, fmt, ##args)
#define netif_crit(priv, type, dev, fmt, args...) \
netif_printk(priv, type, KERN_CRIT, dev, fmt, ##args)
#define netif_err(priv, type, dev, fmt, args...) \
netif_printk(priv, type, KERN_ERR, dev, fmt, ##args)
#define netif_warn(priv, type, dev, fmt, args...) \
netif_printk(priv, type, KERN_WARNING, dev, fmt, ##args)
#define netif_notice(priv, type, dev, fmt, args...) \
netif_printk(priv, type, KERN_NOTICE, dev, fmt, ##args)
#define netif_info(priv, type, dev, fmt, args...) \
netif_printk(priv, type, KERN_INFO, (dev), fmt, ##args)
#if defined(DEBUG)
#define netif_dbg(priv, type, dev, format, args...) \
netif_printk(priv, type, KERN_DEBUG, dev, format, ##args)
#elif defined(CONFIG_DYNAMIC_DEBUG)
#define netif_dbg(priv, type, netdev, format, args...) \
do { \
if (netif_msg_##type(priv)) \
dynamic_dev_dbg((netdev)->dev.parent, \
"%s: " format, \
netdev_name(netdev), ##args); \
} while (0)
#else
#define netif_dbg(priv, type, dev, format, args...) \
({ \
if (0) \
netif_printk(priv, type, KERN_DEBUG, dev, format, ##args); \
0; \
})
#endif
#if defined(VERBOSE_DEBUG)
#define netif_vdbg netdev_dbg
#else
#define netif_vdbg(priv, type, dev, format, args...) \
({ \
if (0) \
netif_printk(KERN_DEBUG, dev, format, ##args); \
0; \
})
#endif
#endif /* __KERNEL__ */
#endif /* _LINUX_NETDEVICE_H */