linux_dsm_epyc7002/include/linux/netdevice.h

4874 lines
149 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0-or-later */
/*
* 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>
*
* Moved to /usr/include/linux for NET3
*/
#ifndef _LINUX_NETDEVICE_H
#define _LINUX_NETDEVICE_H
#include <linux/timer.h>
#include <linux/bug.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 06:41:36 +07:00
#include <linux/delay.h>
#include <linux/atomic.h>
#include <linux/prefetch.h>
#include <asm/cache.h>
#include <asm/byteorder.h>
#include <linux/percpu.h>
#include <linux/rculist.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 06:41:36 +07:00
#include <linux/workqueue.h>
#include <linux/dynamic_queue_limits.h>
#include <linux/ethtool.h>
#include <net/net_namespace.h>
#ifdef CONFIG_DCB
#include <net/dcbnl.h>
#endif
#include <net/netprio_cgroup.h>
#include <net/xdp.h>
#include <linux/netdev_features.h>
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
#include <linux/neighbour.h>
#include <uapi/linux/netdevice.h>
#include <uapi/linux/if_bonding.h>
#include <uapi/linux/pkt_cls.h>
#include <linux/hashtable.h>
struct netpoll_info;
struct device;
struct phy_device;
struct dsa_port;
struct sfp_bus;
/* 802.11 specific */
struct wireless_dev;
/* 802.15.4 specific */
struct wpan_dev;
struct mpls_dev;
/* UDP Tunnel offloads */
struct udp_tunnel_info;
struct bpf_prog;
struct xdp_buff;
void netdev_set_default_ethtool_ops(struct net_device *dev,
const struct ethtool_ops *ops);
/* 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(), and 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_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*/
};
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;
}
/*
* Compute the worst-case header length according to the protocols
* used.
*/
#if defined(CONFIG_HYPERV_NET)
# define LL_MAX_HEADER 128
#elif defined(CONFIG_WLAN) || IS_ENABLED(CONFIG_AX25)
# if defined(CONFIG_MAC80211_MESH)
# define LL_MAX_HEADER 128
# else
# define LL_MAX_HEADER 96
# endif
#else
# define LL_MAX_HEADER 32
#endif
#if !IS_ENABLED(CONFIG_NET_IPIP) && !IS_ENABLED(CONFIG_NET_IPGRE) && \
!IS_ENABLED(CONFIG_IPV6_SIT) && !IS_ENABLED(CONFIG_IPV6_TUNNEL)
#define MAX_HEADER LL_MAX_HEADER
#else
#define MAX_HEADER (LL_MAX_HEADER + 48)
#endif
/*
* Old network device statistics. Fields are native words
* (unsigned long) so they can be read and written atomically.
*/
struct net_device_stats {
unsigned long rx_packets;
unsigned long tx_packets;
unsigned long rx_bytes;
unsigned long tx_bytes;
unsigned long rx_errors;
unsigned long tx_errors;
unsigned long rx_dropped;
unsigned long tx_dropped;
unsigned long multicast;
unsigned long collisions;
unsigned long rx_length_errors;
unsigned long rx_over_errors;
unsigned long rx_crc_errors;
unsigned long rx_frame_errors;
unsigned long rx_fifo_errors;
unsigned long rx_missed_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;
unsigned long rx_compressed;
unsigned long tx_compressed;
};
#include <linux/cache.h>
#include <linux/skbuff.h>
#ifdef CONFIG_RPS
static keys: Introduce 'struct static_key', static_key_true()/false() and static_key_slow_[inc|dec]() So here's a boot tested patch on top of Jason's series that does all the cleanups I talked about and turns jump labels into a more intuitive to use facility. It should also address the various misconceptions and confusions that surround jump labels. Typical usage scenarios: #include <linux/static_key.h> struct static_key key = STATIC_KEY_INIT_TRUE; if (static_key_false(&key)) do unlikely code else do likely code Or: if (static_key_true(&key)) do likely code else do unlikely code The static key is modified via: static_key_slow_inc(&key); ... static_key_slow_dec(&key); The 'slow' prefix makes it abundantly clear that this is an expensive operation. I've updated all in-kernel code to use this everywhere. Note that I (intentionally) have not pushed through the rename blindly through to the lowest levels: the actual jump-label patching arch facility should be named like that, so we want to decouple jump labels from the static-key facility a bit. On non-jump-label enabled architectures static keys default to likely()/unlikely() branches. Signed-off-by: Ingo Molnar <mingo@elte.hu> Acked-by: Jason Baron <jbaron@redhat.com> Acked-by: Steven Rostedt <rostedt@goodmis.org> Cc: a.p.zijlstra@chello.nl Cc: mathieu.desnoyers@efficios.com Cc: davem@davemloft.net Cc: ddaney.cavm@gmail.com Cc: Linus Torvalds <torvalds@linux-foundation.org> Link: http://lkml.kernel.org/r/20120222085809.GA26397@elte.hu Signed-off-by: Ingo Molnar <mingo@elte.hu>
2012-02-24 14:31:31 +07:00
#include <linux/static_key.h>
extern struct static_key_false rps_needed;
extern struct static_key_false rfs_needed;
#endif
struct neighbour;
struct neigh_parms;
struct sk_buff;
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
#define NETDEV_HW_ADDR_T_MULTICAST 5
bool global_use;
int sync_cnt;
int refcount;
int synced;
struct rcu_head rcu_head;
};
struct netdev_hw_addr_list {
struct list_head list;
int count;
};
#define netdev_hw_addr_list_count(l) ((l)->count)
#define netdev_hw_addr_list_empty(l) (netdev_hw_addr_list_count(l) == 0)
#define netdev_hw_addr_list_for_each(ha, l) \
list_for_each_entry(ha, &(l)->list, list)
#define netdev_uc_count(dev) netdev_hw_addr_list_count(&(dev)->uc)
#define netdev_uc_empty(dev) netdev_hw_addr_list_empty(&(dev)->uc)
#define netdev_for_each_uc_addr(ha, dev) \
netdev_hw_addr_list_for_each(ha, &(dev)->uc)
#define netdev_mc_count(dev) netdev_hw_addr_list_count(&(dev)->mc)
#define netdev_mc_empty(dev) netdev_hw_addr_list_empty(&(dev)->mc)
#define netdev_for_each_mc_addr(ha, dev) \
netdev_hw_addr_list_for_each(ha, &(dev)->mc)
struct hh_cache {
unsigned int hh_len;
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.
*/
#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)
struct header_ops {
int (*create) (struct sk_buff *skb, struct net_device *dev,
unsigned short type, const void *daddr,
const void *saddr, unsigned int len);
int (*parse)(const struct sk_buff *skb, unsigned char *haddr);
int (*cache)(const struct neighbour *neigh, struct hh_cache *hh, __be16 type);
void (*cache_update)(struct hh_cache *hh,
const struct net_device *dev,
const unsigned char *haddr);
bool (*validate)(const char *ll_header, unsigned int len);
__be16 (*parse_protocol)(const struct sk_buff *skb);
};
/* 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 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
int __init netdev_boot_setup(char *str);
struct gro_list {
struct list_head list;
int count;
};
[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 06:41:36 +07:00
/*
* size of gro hash buckets, must less than bit number of
* napi_struct::gro_bitmask
[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 06:41:36 +07:00
*/
#define GRO_HASH_BUCKETS 8
/*
* Structure for NAPI scheduling similar to tasklet but with weighting
*/
[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 06:41:36 +07:00
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
[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 06:41:36 +07:00
* can remove from the list right before clearing the bit.
*/
struct list_head poll_list;
unsigned long state;
int weight;
unsigned long gro_bitmask;
[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 06:41:36 +07:00
int (*poll)(struct napi_struct *, int);
#ifdef CONFIG_NETPOLL
int poll_owner;
#endif
struct net_device *dev;
struct gro_list gro_hash[GRO_HASH_BUCKETS];
struct sk_buff *skb;
net: gro: add a per device gro flush timer Tuning coalescing parameters on NIC can be really hard. Servers can handle both bulk and RPC like traffic, with conflicting goals : bulk flows want as big GRO packets as possible, RPC want minimal latencies. To reach big GRO packets on 10Gbe NIC, one can use : ethtool -C eth0 rx-usecs 4 rx-frames 44 But this penalizes rpc sessions, with an increase of latencies, up to 50% in some cases, as NICs generally do not force an interrupt when a packet with TCP Push flag is received. Some NICs do not have an absolute timer, only a timer rearmed for every incoming packet. This patch uses a different strategy : Let GRO stack decides what do do, based on traffic pattern. Packets with Push flag wont be delayed. Packets without Push flag might be held in GRO engine, if we keep receiving data. This new mechanism is off by default, and shall be enabled by setting /sys/class/net/ethX/gro_flush_timeout to a value in nanosecond. To fully enable this mechanism, drivers should use napi_complete_done() instead of napi_complete(). Tested: Ran 200 netperf TCP_STREAM from A to B (10Gbe mlx4 link, 8 RX queues) Without this feature, we send back about 305,000 ACK per second. GRO aggregation ratio is low (811/305 = 2.65 segments per GRO packet) Setting a timer of 2000 nsec is enough to increase GRO packet sizes and reduce number of ACK packets. (811/19.2 = 42) Receiver performs less calls to upper stacks, less wakes up. This also reduces cpu usage on the sender, as it receives less ACK packets. Note that reducing number of wakes up increases cpu efficiency, but can decrease QPS, as applications wont have the chance to warmup cpu caches doing a partial read of RPC requests/answers if they fit in one skb. B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811269.80 305732.30 1199462.57 19705.72 0.00 0.00 0.50 B:~# echo 2000 >/sys/class/net/eth0/gro_flush_timeout B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811577.30 19230.80 1199916.51 1239.80 0.00 0.00 0.50 Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-07 12:09:44 +07:00
struct hrtimer timer;
struct list_head dev_list;
struct hlist_node napi_hash_node;
unsigned int napi_id;
[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 06:41:36 +07:00
};
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 06:41:36 +07:00
NAPI_STATE_SCHED, /* Poll is scheduled */
net: solve a NAPI race While playing with mlx4 hardware timestamping of RX packets, I found that some packets were received by TCP stack with a ~200 ms delay... Since the timestamp was provided by the NIC, and my probe was added in tcp_v4_rcv() while in BH handler, I was confident it was not a sender issue, or a drop in the network. This would happen with a very low probability, but hurting RPC workloads. A NAPI driver normally arms the IRQ after the napi_complete_done(), after NAPI_STATE_SCHED is cleared, so that the hard irq handler can grab it. Problem is that if another point in the stack grabs NAPI_STATE_SCHED bit while IRQ are not disabled, we might have later an IRQ firing and finding this bit set, right before napi_complete_done() clears it. This can happen with busy polling users, or if gro_flush_timeout is used. But some other uses of napi_schedule() in drivers can cause this as well. thread 1 thread 2 (could be on same cpu, or not) // busy polling or napi_watchdog() napi_schedule(); ... napi->poll() device polling: read 2 packets from ring buffer Additional 3rd packet is available. device hard irq // does nothing because NAPI_STATE_SCHED bit is owned by thread 1 napi_schedule(); napi_complete_done(napi, 2); rearm_irq(); Note that rearm_irq() will not force the device to send an additional IRQ for the packet it already signaled (3rd packet in my example) This patch adds a new NAPI_STATE_MISSED bit, that napi_schedule_prep() can set if it could not grab NAPI_STATE_SCHED Then napi_complete_done() properly reschedules the napi to make sure we do not miss something. Since we manipulate multiple bits at once, use cmpxchg() like in sk_busy_loop() to provide proper transactions. In v2, I changed napi_watchdog() to use a relaxed variant of napi_schedule_prep() : No need to set NAPI_STATE_MISSED from this point. In v3, I added more details in the changelog and clears NAPI_STATE_MISSED in busy_poll_stop() In v4, I added the ideas given by Alexander Duyck in v3 review Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Alexander Duyck <alexander.duyck@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-01 01:34:50 +07:00
NAPI_STATE_MISSED, /* reschedule a napi */
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 14:22:26 +07:00
NAPI_STATE_NPSVC, /* Netpoll - don't dequeue from poll_list */
NAPI_STATE_HASHED, /* In NAPI hash (busy polling possible) */
NAPI_STATE_NO_BUSY_POLL,/* Do not add in napi_hash, no busy polling */
net: busy-poll: allow preemption in sk_busy_loop() After commit 4cd13c21b207 ("softirq: Let ksoftirqd do its job"), sk_busy_loop() needs a bit of care : softirqs might be delayed since we do not allow preemption yet. This patch adds preemptiom points in sk_busy_loop(), and makes sure no unnecessary cache line dirtying or atomic operations are done while looping. A new flag is added into napi->state : NAPI_STATE_IN_BUSY_POLL This prevents napi_complete_done() from clearing NAPIF_STATE_SCHED, so that sk_busy_loop() does not have to grab it again. Similarly, netpoll_poll_lock() is done one time. This gives about 10 to 20 % improvement in various busy polling tests, especially when many threads are busy polling in configurations with large number of NIC queues. This should allow experimenting with bigger delays without hurting overall latencies. Tested: On a 40Gb mlx4 NIC, 32 RX/TX queues. echo 70 >/proc/sys/net/core/busy_read for i in `seq 1 40`; do echo -n $i: ; ./super_netperf $i -H lpaa24 -t UDP_RR -- -N -n; done Before: After: 1: 90072 92819 2: 157289 184007 3: 235772 213504 4: 344074 357513 5: 394755 458267 6: 461151 487819 7: 549116 625963 8: 544423 716219 9: 720460 738446 10: 794686 837612 11: 915998 923960 12: 937507 925107 13: 1019677 971506 14: 1046831 1113650 15: 1114154 1148902 16: 1105221 1179263 17: 1266552 1299585 18: 1258454 1383817 19: 1341453 1312194 20: 1363557 1488487 21: 1387979 1501004 22: 1417552 1601683 23: 1550049 1642002 24: 1568876 1601915 25: 1560239 1683607 26: 1640207 1745211 27: 1706540 1723574 28: 1638518 1722036 29: 1734309 1757447 30: 1782007 1855436 31: 1724806 1888539 32: 1717716 1944297 33: 1778716 1869118 34: 1805738 1983466 35: 1815694 2020758 36: 1893059 2035632 37: 1843406 2034653 38: 1888830 2086580 39: 1972827 2143567 40: 1877729 2181851 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Willem de Bruijn <willemb@google.com> Cc: Adam Belay <abelay@google.com> Cc: Tariq Toukan <tariqt@mellanox.com> Cc: Yuval Mintz <Yuval.Mintz@cavium.com> Cc: Ariel Elior <ariel.elior@cavium.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-11-16 01:15:11 +07:00
NAPI_STATE_IN_BUSY_POLL,/* sk_busy_loop() owns this NAPI */
};
enum {
net: solve a NAPI race While playing with mlx4 hardware timestamping of RX packets, I found that some packets were received by TCP stack with a ~200 ms delay... Since the timestamp was provided by the NIC, and my probe was added in tcp_v4_rcv() while in BH handler, I was confident it was not a sender issue, or a drop in the network. This would happen with a very low probability, but hurting RPC workloads. A NAPI driver normally arms the IRQ after the napi_complete_done(), after NAPI_STATE_SCHED is cleared, so that the hard irq handler can grab it. Problem is that if another point in the stack grabs NAPI_STATE_SCHED bit while IRQ are not disabled, we might have later an IRQ firing and finding this bit set, right before napi_complete_done() clears it. This can happen with busy polling users, or if gro_flush_timeout is used. But some other uses of napi_schedule() in drivers can cause this as well. thread 1 thread 2 (could be on same cpu, or not) // busy polling or napi_watchdog() napi_schedule(); ... napi->poll() device polling: read 2 packets from ring buffer Additional 3rd packet is available. device hard irq // does nothing because NAPI_STATE_SCHED bit is owned by thread 1 napi_schedule(); napi_complete_done(napi, 2); rearm_irq(); Note that rearm_irq() will not force the device to send an additional IRQ for the packet it already signaled (3rd packet in my example) This patch adds a new NAPI_STATE_MISSED bit, that napi_schedule_prep() can set if it could not grab NAPI_STATE_SCHED Then napi_complete_done() properly reschedules the napi to make sure we do not miss something. Since we manipulate multiple bits at once, use cmpxchg() like in sk_busy_loop() to provide proper transactions. In v2, I changed napi_watchdog() to use a relaxed variant of napi_schedule_prep() : No need to set NAPI_STATE_MISSED from this point. In v3, I added more details in the changelog and clears NAPI_STATE_MISSED in busy_poll_stop() In v4, I added the ideas given by Alexander Duyck in v3 review Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Alexander Duyck <alexander.duyck@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-01 01:34:50 +07:00
NAPIF_STATE_SCHED = BIT(NAPI_STATE_SCHED),
NAPIF_STATE_MISSED = BIT(NAPI_STATE_MISSED),
NAPIF_STATE_DISABLE = BIT(NAPI_STATE_DISABLE),
NAPIF_STATE_NPSVC = BIT(NAPI_STATE_NPSVC),
NAPIF_STATE_HASHED = BIT(NAPI_STATE_HASHED),
NAPIF_STATE_NO_BUSY_POLL = BIT(NAPI_STATE_NO_BUSY_POLL),
NAPIF_STATE_IN_BUSY_POLL = BIT(NAPI_STATE_IN_BUSY_POLL),
[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 06:41:36 +07:00
};
enum gro_result {
GRO_MERGED,
GRO_MERGED_FREE,
GRO_HELD,
GRO_NORMAL,
GRO_DROP,
GRO_CONSUMED,
};
typedef enum gro_result gro_result_t;
/*
* enum rx_handler_result - Possible return values for rx_handlers.
* @RX_HANDLER_CONSUMED: skb was consumed by rx_handler, do not process it
* further.
* @RX_HANDLER_ANOTHER: Do another round in receive path. This is indicated in
* case skb->dev was changed by rx_handler.
* @RX_HANDLER_EXACT: Force exact delivery, no wildcard.
* @RX_HANDLER_PASS: Do nothing, pass the skb as if no rx_handler was called.
*
* rx_handlers are functions called from inside __netif_receive_skb(), to do
* special processing of the skb, prior to delivery to protocol handlers.
*
* Currently, a net_device can only have a single rx_handler registered. Trying
* to register a second rx_handler will return -EBUSY.
*
* To register a rx_handler on a net_device, use netdev_rx_handler_register().
* To unregister a rx_handler on a net_device, use
* netdev_rx_handler_unregister().
*
* Upon return, rx_handler is expected to tell __netif_receive_skb() what to
* do with the skb.
*
* If the rx_handler consumed the skb in some way, it should return
* RX_HANDLER_CONSUMED. This is appropriate when the rx_handler arranged for
* the skb to be delivered in some other way.
*
* If the rx_handler changed skb->dev, to divert the skb to another
* net_device, it should return RX_HANDLER_ANOTHER. The rx_handler for the
* new device will be called if it exists.
*
* If the rx_handler decides the skb should be ignored, it should return
* RX_HANDLER_EXACT. The skb will only be delivered to protocol handlers that
* are registered on exact device (ptype->dev == skb->dev).
*
* If the rx_handler didn't change skb->dev, but wants the skb to be normally
* delivered, it should return RX_HANDLER_PASS.
*
* A device without a registered rx_handler will behave as if rx_handler
* returned RX_HANDLER_PASS.
*/
enum rx_handler_result {
RX_HANDLER_CONSUMED,
RX_HANDLER_ANOTHER,
RX_HANDLER_EXACT,
RX_HANDLER_PASS,
};
typedef enum rx_handler_result rx_handler_result_t;
typedef rx_handler_result_t rx_handler_func_t(struct sk_buff **pskb);
void __napi_schedule(struct napi_struct *n);
void __napi_schedule_irqoff(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 06:41:36 +07:00
static inline bool napi_disable_pending(struct napi_struct *n)
{
return test_bit(NAPI_STATE_DISABLE, &n->state);
}
net: solve a NAPI race While playing with mlx4 hardware timestamping of RX packets, I found that some packets were received by TCP stack with a ~200 ms delay... Since the timestamp was provided by the NIC, and my probe was added in tcp_v4_rcv() while in BH handler, I was confident it was not a sender issue, or a drop in the network. This would happen with a very low probability, but hurting RPC workloads. A NAPI driver normally arms the IRQ after the napi_complete_done(), after NAPI_STATE_SCHED is cleared, so that the hard irq handler can grab it. Problem is that if another point in the stack grabs NAPI_STATE_SCHED bit while IRQ are not disabled, we might have later an IRQ firing and finding this bit set, right before napi_complete_done() clears it. This can happen with busy polling users, or if gro_flush_timeout is used. But some other uses of napi_schedule() in drivers can cause this as well. thread 1 thread 2 (could be on same cpu, or not) // busy polling or napi_watchdog() napi_schedule(); ... napi->poll() device polling: read 2 packets from ring buffer Additional 3rd packet is available. device hard irq // does nothing because NAPI_STATE_SCHED bit is owned by thread 1 napi_schedule(); napi_complete_done(napi, 2); rearm_irq(); Note that rearm_irq() will not force the device to send an additional IRQ for the packet it already signaled (3rd packet in my example) This patch adds a new NAPI_STATE_MISSED bit, that napi_schedule_prep() can set if it could not grab NAPI_STATE_SCHED Then napi_complete_done() properly reschedules the napi to make sure we do not miss something. Since we manipulate multiple bits at once, use cmpxchg() like in sk_busy_loop() to provide proper transactions. In v2, I changed napi_watchdog() to use a relaxed variant of napi_schedule_prep() : No need to set NAPI_STATE_MISSED from this point. In v3, I added more details in the changelog and clears NAPI_STATE_MISSED in busy_poll_stop() In v4, I added the ideas given by Alexander Duyck in v3 review Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Alexander Duyck <alexander.duyck@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-01 01:34:50 +07:00
bool napi_schedule_prep(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 06:41:36 +07:00
/**
* napi_schedule - schedule NAPI poll
* @n: NAPI context
[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 06:41:36 +07:00
*
* 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);
}
/**
* napi_schedule_irqoff - schedule NAPI poll
* @n: NAPI context
*
* Variant of napi_schedule(), assuming hard irqs are masked.
*/
static inline void napi_schedule_irqoff(struct napi_struct *n)
{
if (napi_schedule_prep(n))
__napi_schedule_irqoff(n);
}
/* Try to reschedule poll. Called by dev->poll() after napi_complete(). */
static inline bool napi_reschedule(struct napi_struct *napi)
{
if (napi_schedule_prep(napi)) {
__napi_schedule(napi);
return true;
}
return false;
}
bool napi_complete_done(struct napi_struct *n, int work_done);
[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 06:41:36 +07:00
/**
* napi_complete - NAPI processing complete
* @n: NAPI context
[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 06:41:36 +07:00
*
* Mark NAPI processing as complete.
net: gro: add a per device gro flush timer Tuning coalescing parameters on NIC can be really hard. Servers can handle both bulk and RPC like traffic, with conflicting goals : bulk flows want as big GRO packets as possible, RPC want minimal latencies. To reach big GRO packets on 10Gbe NIC, one can use : ethtool -C eth0 rx-usecs 4 rx-frames 44 But this penalizes rpc sessions, with an increase of latencies, up to 50% in some cases, as NICs generally do not force an interrupt when a packet with TCP Push flag is received. Some NICs do not have an absolute timer, only a timer rearmed for every incoming packet. This patch uses a different strategy : Let GRO stack decides what do do, based on traffic pattern. Packets with Push flag wont be delayed. Packets without Push flag might be held in GRO engine, if we keep receiving data. This new mechanism is off by default, and shall be enabled by setting /sys/class/net/ethX/gro_flush_timeout to a value in nanosecond. To fully enable this mechanism, drivers should use napi_complete_done() instead of napi_complete(). Tested: Ran 200 netperf TCP_STREAM from A to B (10Gbe mlx4 link, 8 RX queues) Without this feature, we send back about 305,000 ACK per second. GRO aggregation ratio is low (811/305 = 2.65 segments per GRO packet) Setting a timer of 2000 nsec is enough to increase GRO packet sizes and reduce number of ACK packets. (811/19.2 = 42) Receiver performs less calls to upper stacks, less wakes up. This also reduces cpu usage on the sender, as it receives less ACK packets. Note that reducing number of wakes up increases cpu efficiency, but can decrease QPS, as applications wont have the chance to warmup cpu caches doing a partial read of RPC requests/answers if they fit in one skb. B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811269.80 305732.30 1199462.57 19705.72 0.00 0.00 0.50 B:~# echo 2000 >/sys/class/net/eth0/gro_flush_timeout B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811577.30 19230.80 1199916.51 1239.80 0.00 0.00 0.50 Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-07 12:09:44 +07:00
* Consider using napi_complete_done() instead.
* Return false if device should avoid rearming interrupts.
[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 06:41:36 +07:00
*/
static inline bool napi_complete(struct napi_struct *n)
net: gro: add a per device gro flush timer Tuning coalescing parameters on NIC can be really hard. Servers can handle both bulk and RPC like traffic, with conflicting goals : bulk flows want as big GRO packets as possible, RPC want minimal latencies. To reach big GRO packets on 10Gbe NIC, one can use : ethtool -C eth0 rx-usecs 4 rx-frames 44 But this penalizes rpc sessions, with an increase of latencies, up to 50% in some cases, as NICs generally do not force an interrupt when a packet with TCP Push flag is received. Some NICs do not have an absolute timer, only a timer rearmed for every incoming packet. This patch uses a different strategy : Let GRO stack decides what do do, based on traffic pattern. Packets with Push flag wont be delayed. Packets without Push flag might be held in GRO engine, if we keep receiving data. This new mechanism is off by default, and shall be enabled by setting /sys/class/net/ethX/gro_flush_timeout to a value in nanosecond. To fully enable this mechanism, drivers should use napi_complete_done() instead of napi_complete(). Tested: Ran 200 netperf TCP_STREAM from A to B (10Gbe mlx4 link, 8 RX queues) Without this feature, we send back about 305,000 ACK per second. GRO aggregation ratio is low (811/305 = 2.65 segments per GRO packet) Setting a timer of 2000 nsec is enough to increase GRO packet sizes and reduce number of ACK packets. (811/19.2 = 42) Receiver performs less calls to upper stacks, less wakes up. This also reduces cpu usage on the sender, as it receives less ACK packets. Note that reducing number of wakes up increases cpu efficiency, but can decrease QPS, as applications wont have the chance to warmup cpu caches doing a partial read of RPC requests/answers if they fit in one skb. B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811269.80 305732.30 1199462.57 19705.72 0.00 0.00 0.50 B:~# echo 2000 >/sys/class/net/eth0/gro_flush_timeout B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811577.30 19230.80 1199916.51 1239.80 0.00 0.00 0.50 Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-07 12:09:44 +07:00
{
return napi_complete_done(n, 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 06:41:36 +07:00
/**
* napi_hash_del - remove a NAPI from global table
* @napi: NAPI context
*
* Warning: caller must observe RCU grace period
* before freeing memory containing @napi, if
* this function returns true.
* Note: core networking stack automatically calls it
* from netif_napi_del().
* Drivers might want to call this helper to combine all
* the needed RCU grace periods into a single one.
*/
bool napi_hash_del(struct napi_struct *napi);
[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 06:41:36 +07:00
/**
* napi_disable - prevent NAPI from scheduling
* @n: NAPI context
[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 06:41:36 +07:00
*
* Stop NAPI from being scheduled on this context.
* Waits till any outstanding processing completes.
*/
net: gro: add a per device gro flush timer Tuning coalescing parameters on NIC can be really hard. Servers can handle both bulk and RPC like traffic, with conflicting goals : bulk flows want as big GRO packets as possible, RPC want minimal latencies. To reach big GRO packets on 10Gbe NIC, one can use : ethtool -C eth0 rx-usecs 4 rx-frames 44 But this penalizes rpc sessions, with an increase of latencies, up to 50% in some cases, as NICs generally do not force an interrupt when a packet with TCP Push flag is received. Some NICs do not have an absolute timer, only a timer rearmed for every incoming packet. This patch uses a different strategy : Let GRO stack decides what do do, based on traffic pattern. Packets with Push flag wont be delayed. Packets without Push flag might be held in GRO engine, if we keep receiving data. This new mechanism is off by default, and shall be enabled by setting /sys/class/net/ethX/gro_flush_timeout to a value in nanosecond. To fully enable this mechanism, drivers should use napi_complete_done() instead of napi_complete(). Tested: Ran 200 netperf TCP_STREAM from A to B (10Gbe mlx4 link, 8 RX queues) Without this feature, we send back about 305,000 ACK per second. GRO aggregation ratio is low (811/305 = 2.65 segments per GRO packet) Setting a timer of 2000 nsec is enough to increase GRO packet sizes and reduce number of ACK packets. (811/19.2 = 42) Receiver performs less calls to upper stacks, less wakes up. This also reduces cpu usage on the sender, as it receives less ACK packets. Note that reducing number of wakes up increases cpu efficiency, but can decrease QPS, as applications wont have the chance to warmup cpu caches doing a partial read of RPC requests/answers if they fit in one skb. B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811269.80 305732.30 1199462.57 19705.72 0.00 0.00 0.50 B:~# echo 2000 >/sys/class/net/eth0/gro_flush_timeout B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811577.30 19230.80 1199916.51 1239.80 0.00 0.00 0.50 Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-07 12:09:44 +07:00
void napi_disable(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 06:41:36 +07:00
/**
* napi_enable - enable NAPI scheduling
* @n: NAPI context
[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 06:41:36 +07:00
*
* 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_atomic();
[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 06:41:36 +07:00
clear_bit(NAPI_STATE_SCHED, &n->state);
netpoll: Close race condition between poll_one_napi and napi_disable Drivers might call napi_disable while not holding the napi instance poll_lock. In those instances, its possible for a race condition to exist between poll_one_napi and napi_disable. That is to say, poll_one_napi only tests the NAPI_STATE_SCHED bit to see if there is work to do during a poll, and as such the following may happen: CPU0 CPU1 ndo_tx_timeout napi_poll_dev napi_disable poll_one_napi test_and_set_bit (ret 0) test_bit (ret 1) reset adapter napi_poll_routine If the adapter gets a tx timeout without a napi instance scheduled, its possible for the adapter to think it has exclusive access to the hardware (as the napi instance is now scheduled via the napi_disable call), while the netpoll code thinks there is simply work to do. The result is parallel hardware access leading to corrupt data structures in the driver, and a crash. Additionaly, there is another, more critical race between netpoll and napi_disable. The disabled napi state is actually identical to the scheduled state for a given napi instance. The implication being that, if a napi instance is disabled, a netconsole instance would see the napi state of the device as having been scheduled, and poll it, likely while the driver was dong something requiring exclusive access. In the case above, its fairly clear that not having the rings in a state ready to be polled will cause any number of crashes. The fix should be pretty easy. netpoll uses its own bit to indicate that that the napi instance is in a state of being serviced by netpoll (NAPI_STATE_NPSVC). We can just gate disabling on that bit as well as the sched bit. That should prevent netpoll from conducting a napi poll if we convert its set bit to a test_and_set_bit operation to provide mutual exclusion Change notes: V2) Remove a trailing whtiespace Resubmit with proper subject prefix V3) Clean up spacing nits Signed-off-by: Neil Horman <nhorman@tuxdriver.com> CC: "David S. Miller" <davem@davemloft.net> CC: jmaxwell@redhat.com Tested-by: jmaxwell@redhat.com Signed-off-by: David S. Miller <davem@davemloft.net>
2015-09-24 01:57:58 +07:00
clear_bit(NAPI_STATE_NPSVC, &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 06:41:36 +07:00
}
/**
* 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)
{
if (IS_ENABLED(CONFIG_SMP))
while (test_bit(NAPI_STATE_SCHED, &n->state))
msleep(1);
else
barrier();
}
/**
* napi_if_scheduled_mark_missed - if napi is running, set the
* NAPIF_STATE_MISSED
* @n: NAPI context
*
* If napi is running, set the NAPIF_STATE_MISSED, and return true if
* NAPI is scheduled.
**/
static inline bool napi_if_scheduled_mark_missed(struct napi_struct *n)
{
unsigned long val, new;
do {
val = READ_ONCE(n->state);
if (val & NAPIF_STATE_DISABLE)
return true;
if (!(val & NAPIF_STATE_SCHED))
return false;
new = val | NAPIF_STATE_MISSED;
} while (cmpxchg(&n->state, val, new) != val);
return true;
}
enum netdev_queue_state_t {
__QUEUE_STATE_DRV_XOFF,
__QUEUE_STATE_STACK_XOFF,
__QUEUE_STATE_FROZEN,
};
#define QUEUE_STATE_DRV_XOFF (1 << __QUEUE_STATE_DRV_XOFF)
#define QUEUE_STATE_STACK_XOFF (1 << __QUEUE_STATE_STACK_XOFF)
#define QUEUE_STATE_FROZEN (1 << __QUEUE_STATE_FROZEN)
#define QUEUE_STATE_ANY_XOFF (QUEUE_STATE_DRV_XOFF | QUEUE_STATE_STACK_XOFF)
#define QUEUE_STATE_ANY_XOFF_OR_FROZEN (QUEUE_STATE_ANY_XOFF | \
QUEUE_STATE_FROZEN)
#define QUEUE_STATE_DRV_XOFF_OR_FROZEN (QUEUE_STATE_DRV_XOFF | \
QUEUE_STATE_FROZEN)
/*
* __QUEUE_STATE_DRV_XOFF is used by drivers to stop the transmit queue. The
* netif_tx_* functions below are used to manipulate this flag. The
* __QUEUE_STATE_STACK_XOFF flag is used by the stack to stop the transmit
* queue independently. The netif_xmit_*stopped functions below are called
* to check if the queue has been stopped by the driver or stack (either
* of the XOFF bits are set in the state). Drivers should not need to call
* netif_xmit*stopped functions, they should only be using netif_tx_*.
*/
struct netdev_queue {
/*
* read-mostly part
*/
struct net_device *dev;
struct Qdisc __rcu *qdisc;
struct Qdisc *qdisc_sleeping;
#ifdef CONFIG_SYSFS
xps: Transmit Packet Steering This patch implements transmit packet steering (XPS) for multiqueue devices. XPS selects a transmit queue during packet transmission based on configuration. This is done by mapping the CPU transmitting the packet to a queue. This is the transmit side analogue to RPS-- where RPS is selecting a CPU based on receive queue, XPS selects a queue based on the CPU (previously there was an XPS patch from Eric Dumazet, but that might more appropriately be called transmit completion steering). Each transmit queue can be associated with a number of CPUs which will use the queue to send packets. This is configured as a CPU mask on a per queue basis in: /sys/class/net/eth<n>/queues/tx-<n>/xps_cpus The mappings are stored per device in an inverted data structure that maps CPUs to queues. In the netdevice structure this is an array of num_possible_cpu structures where each structure holds and array of queue_indexes for queues which that CPU can use. The benefits of XPS are improved locality in the per queue data structures. Also, transmit completions are more likely to be done nearer to the sending thread, so this should promote locality back to the socket on free (e.g. UDP). The benefits of XPS are dependent on cache hierarchy, application load, and other factors. XPS would nominally be configured so that a queue would only be shared by CPUs which are sharing a cache, the degenerative configuration woud be that each CPU has it's own queue. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. bnx2x on 16 core AMD XPS (16 queues, 1 TX queue per CPU) 1234K at 100% CPU No XPS (16 queues) 996K at 100% CPU Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-11-21 20:17:27 +07:00
struct kobject kobj;
#endif
#if defined(CONFIG_XPS) && defined(CONFIG_NUMA)
int numa_node;
#endif
unsigned long tx_maxrate;
/*
* Number of TX timeouts for this queue
* (/sys/class/net/DEV/Q/trans_timeout)
*/
unsigned long trans_timeout;
/* Subordinate device that the queue has been assigned to */
struct net_device *sb_dev;
#ifdef CONFIG_XDP_SOCKETS
struct xdp_umem *umem;
#endif
/*
* write-mostly part
*/
spinlock_t _xmit_lock ____cacheline_aligned_in_smp;
int xmit_lock_owner;
/*
* Time (in jiffies) of last Tx
*/
unsigned long trans_start;
unsigned long state;
#ifdef CONFIG_BQL
struct dql dql;
#endif
} ____cacheline_aligned_in_smp;
extern int sysctl_fb_tunnels_only_for_init_net;
net: introduce a knob to control whether to inherit devconf config There have been many people complaining about the inconsistent behaviors of IPv4 and IPv6 devconf when creating new network namespaces. Currently, for IPv4, we inherit all current settings from init_net, but for IPv6 we reset all setting to default. This patch introduces a new /proc file /proc/sys/net/core/devconf_inherit_init_net to control the behavior of whether to inhert sysctl current settings from init_net. This file itself is only available in init_net. As demonstrated below: Initial setup in init_net: # cat /proc/sys/net/ipv4/conf/all/rp_filter 2 # cat /proc/sys/net/ipv6/conf/all/accept_dad 1 Default value 0 (current behavior): # ip netns del test # ip netns add test # ip netns exec test cat /proc/sys/net/ipv4/conf/all/rp_filter 2 # ip netns exec test cat /proc/sys/net/ipv6/conf/all/accept_dad 0 Set to 1 (inherit from init_net): # echo 1 > /proc/sys/net/core/devconf_inherit_init_net # ip netns del test # ip netns add test # ip netns exec test cat /proc/sys/net/ipv4/conf/all/rp_filter 2 # ip netns exec test cat /proc/sys/net/ipv6/conf/all/accept_dad 1 Set to 2 (reset to default): # echo 2 > /proc/sys/net/core/devconf_inherit_init_net # ip netns del test # ip netns add test # ip netns exec test cat /proc/sys/net/ipv4/conf/all/rp_filter 0 # ip netns exec test cat /proc/sys/net/ipv6/conf/all/accept_dad 0 Set to a value out of range (invalid): # echo 3 > /proc/sys/net/core/devconf_inherit_init_net -bash: echo: write error: Invalid argument # echo -1 > /proc/sys/net/core/devconf_inherit_init_net -bash: echo: write error: Invalid argument Reported-by: Zhu Yanjun <Yanjun.Zhu@windriver.com> Reported-by: Tonghao Zhang <xiangxia.m.yue@gmail.com> Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com> Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com> Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com> Acked-by: Tonghao Zhang <xiangxia.m.yue@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-18 14:27:11 +07:00
extern int sysctl_devconf_inherit_init_net;
static inline bool net_has_fallback_tunnels(const struct net *net)
{
return net == &init_net ||
!IS_ENABLED(CONFIG_SYSCTL) ||
!sysctl_fb_tunnels_only_for_init_net;
}
static inline int netdev_queue_numa_node_read(const struct netdev_queue *q)
{
#if defined(CONFIG_XPS) && defined(CONFIG_NUMA)
return q->numa_node;
#else
return NUMA_NO_NODE;
#endif
}
static inline void netdev_queue_numa_node_write(struct netdev_queue *q, int node)
{
#if defined(CONFIG_XPS) && defined(CONFIG_NUMA)
q->numa_node = node;
#endif
}
#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 15:03:29 +07:00
/*
* 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)))
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 15:03:29 +07:00
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
/*
* The rps_dev_flow structure contains the mapping of a flow to a CPU, the
* tail pointer for that CPU's input queue at the time of last enqueue, and
* a hardware filter index.
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
*/
struct rps_dev_flow {
u16 cpu;
u16 filter;
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
unsigned int last_qtail;
};
#define RPS_NO_FILTER 0xffff
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
/*
* The rps_dev_flow_table structure contains a table of flow mappings.
*/
struct rps_dev_flow_table {
unsigned int mask;
struct rcu_head rcu;
struct rps_dev_flow flows[0];
};
#define RPS_DEV_FLOW_TABLE_SIZE(_num) (sizeof(struct rps_dev_flow_table) + \
((_num) * sizeof(struct rps_dev_flow)))
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
/*
* The rps_sock_flow_table contains mappings of flows to the last CPU
* on which they were processed by the application (set in recvmsg).
* Each entry is a 32bit value. Upper part is the high-order bits
* of flow hash, lower part is CPU number.
* rps_cpu_mask is used to partition the space, depending on number of
* possible CPUs : rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1
* For example, if 64 CPUs are possible, rps_cpu_mask = 0x3f,
* meaning we use 32-6=26 bits for the hash.
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
*/
struct rps_sock_flow_table {
u32 mask;
u32 ents[0] ____cacheline_aligned_in_smp;
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
};
#define RPS_SOCK_FLOW_TABLE_SIZE(_num) (offsetof(struct rps_sock_flow_table, ents[_num]))
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
#define RPS_NO_CPU 0xffff
extern u32 rps_cpu_mask;
extern struct rps_sock_flow_table __rcu *rps_sock_flow_table;
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
static inline void rps_record_sock_flow(struct rps_sock_flow_table *table,
u32 hash)
{
if (table && hash) {
unsigned int index = hash & table->mask;
u32 val = hash & ~rps_cpu_mask;
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
/* We only give a hint, preemption can change CPU under us */
val |= raw_smp_processor_id();
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
if (table->ents[index] != val)
table->ents[index] = val;
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
}
}
#ifdef CONFIG_RFS_ACCEL
bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, u32 flow_id,
u16 filter_id);
#endif
#endif /* CONFIG_RPS */
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 15:03:29 +07:00
/* This structure contains an instance of an RX queue. */
struct netdev_rx_queue {
#ifdef CONFIG_RPS
struct rps_map __rcu *rps_map;
struct rps_dev_flow_table __rcu *rps_flow_table;
#endif
struct kobject kobj;
struct net_device *dev;
struct xdp_rxq_info xdp_rxq;
#ifdef CONFIG_XDP_SOCKETS
struct xdp_umem *umem;
#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 15:03:29 +07:00
} ____cacheline_aligned_in_smp;
/*
* RX queue sysfs structures and functions.
*/
struct rx_queue_attribute {
struct attribute attr;
ssize_t (*show)(struct netdev_rx_queue *queue, char *buf);
ssize_t (*store)(struct netdev_rx_queue *queue,
const char *buf, size_t len);
};
#ifdef CONFIG_XPS
/*
* This structure holds an XPS map which can be of variable length. The
* map is an array of queues.
*/
struct xps_map {
unsigned int len;
unsigned int alloc_len;
struct rcu_head rcu;
u16 queues[0];
};
#define XPS_MAP_SIZE(_num) (sizeof(struct xps_map) + ((_num) * sizeof(u16)))
#define XPS_MIN_MAP_ALLOC ((L1_CACHE_ALIGN(offsetof(struct xps_map, queues[1])) \
- sizeof(struct xps_map)) / sizeof(u16))
/*
* This structure holds all XPS maps for device. Maps are indexed by CPU.
*/
struct xps_dev_maps {
struct rcu_head rcu;
struct xps_map __rcu *attr_map[0]; /* Either CPUs map or RXQs map */
};
#define XPS_CPU_DEV_MAPS_SIZE(_tcs) (sizeof(struct xps_dev_maps) + \
(nr_cpu_ids * (_tcs) * sizeof(struct xps_map *)))
#define XPS_RXQ_DEV_MAPS_SIZE(_tcs, _rxqs) (sizeof(struct xps_dev_maps) +\
(_rxqs * (_tcs) * sizeof(struct xps_map *)))
#endif /* CONFIG_XPS */
net: implement mechanism for HW based QOS This patch provides a mechanism for lower layer devices to steer traffic using skb->priority to tx queues. This allows for hardware based QOS schemes to use the default qdisc without incurring the penalties related to global state and the qdisc lock. While reliably receiving skbs on the correct tx ring to avoid head of line blocking resulting from shuffling in the LLD. Finally, all the goodness from txq caching and xps/rps can still be leveraged. Many drivers and hardware exist with the ability to implement QOS schemes in the hardware but currently these drivers tend to rely on firmware to reroute specific traffic, a driver specific select_queue or the queue_mapping action in the qdisc. By using select_queue for this drivers need to be updated for each and every traffic type and we lose the goodness of much of the upstream work. Firmware solutions are inherently inflexible. And finally if admins are expected to build a qdisc and filter rules to steer traffic this requires knowledge of how the hardware is currently configured. The number of tx queues and the queue offsets may change depending on resources. Also this approach incurs all the overhead of a qdisc with filters. With the mechanism in this patch users can set skb priority using expected methods ie setsockopt() or the stack can set the priority directly. Then the skb will be steered to the correct tx queues aligned with hardware QOS traffic classes. In the normal case with single traffic class and all queues in this class everything works as is until the LLD enables multiple tcs. To steer the skb we mask out the lower 4 bits of the priority and allow the hardware to configure upto 15 distinct classes of traffic. This is expected to be sufficient for most applications at any rate it is more then the 8021Q spec designates and is equal to the number of prio bands currently implemented in the default qdisc. This in conjunction with a userspace application such as lldpad can be used to implement 8021Q transmission selection algorithms one of these algorithms being the extended transmission selection algorithm currently being used for DCB. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-01-17 15:06:04 +07:00
#define TC_MAX_QUEUE 16
#define TC_BITMASK 15
/* HW offloaded queuing disciplines txq count and offset maps */
struct netdev_tc_txq {
u16 count;
u16 offset;
};
#if defined(CONFIG_FCOE) || defined(CONFIG_FCOE_MODULE)
/*
* This structure is to hold information about the device
* configured to run FCoE protocol stack.
*/
struct netdev_fcoe_hbainfo {
char manufacturer[64];
char serial_number[64];
char hardware_version[64];
char driver_version[64];
char optionrom_version[64];
char firmware_version[64];
char model[256];
char model_description[256];
};
#endif
#define MAX_PHYS_ITEM_ID_LEN 32
/* This structure holds a unique identifier to identify some
* physical item (port for example) used by a netdevice.
*/
struct netdev_phys_item_id {
unsigned char id[MAX_PHYS_ITEM_ID_LEN];
unsigned char id_len;
};
static inline bool netdev_phys_item_id_same(struct netdev_phys_item_id *a,
struct netdev_phys_item_id *b)
{
return a->id_len == b->id_len &&
memcmp(a->id, b->id, a->id_len) == 0;
}
typedef u16 (*select_queue_fallback_t)(struct net_device *dev,
struct sk_buff *skb,
struct net_device *sb_dev);
enum tc_setup_type {
TC_SETUP_QDISC_MQPRIO,
TC_SETUP_CLSU32,
TC_SETUP_CLSFLOWER,
TC_SETUP_CLSMATCHALL,
TC_SETUP_CLSBPF,
TC_SETUP_BLOCK,
TC_SETUP_QDISC_CBS,
TC_SETUP_QDISC_RED,
TC_SETUP_QDISC_PRIO,
TC_SETUP_QDISC_MQ,
net/sched: Introduce the ETF Qdisc The ETF (Earliest TxTime First) qdisc uses the information added earlier in this series (the socket option SO_TXTIME and the new role of sk_buff->tstamp) to schedule packets transmission based on absolute time. For some workloads, just bandwidth enforcement is not enough, and precise control of the transmission of packets is necessary. Example: $ tc qdisc replace dev enp2s0 parent root handle 100 mqprio num_tc 3 \ map 2 2 1 0 2 2 2 2 2 2 2 2 2 2 2 2 queues 1@0 1@1 2@2 hw 0 $ tc qdisc add dev enp2s0 parent 100:1 etf delta 100000 \ clockid CLOCK_TAI In this example, the Qdisc will provide SW best-effort for the control of the transmission time to the network adapter, the time stamp in the socket will be in reference to the clockid CLOCK_TAI and packets will leave the qdisc "delta" (100000) nanoseconds before its transmission time. The ETF qdisc will buffer packets sorted by their txtime. It will drop packets on enqueue() if their skbuff clockid does not match the clock reference of the Qdisc. Moreover, on dequeue(), a packet will be dropped if it expires while being enqueued. The qdisc also supports the SO_TXTIME deadline mode. For this mode, it will dequeue a packet as soon as possible and change the skb timestamp to 'now' during etf_dequeue(). Note that both the qdisc's and the SO_TXTIME ABIs allow for a clockid to be configured, but it's been decided that usage of CLOCK_TAI should be enforced until we decide to allow for other clockids to be used. The rationale here is that PTP times are usually in the TAI scale, thus no other clocks should be necessary. For now, the qdisc will return EINVAL if any clocks other than CLOCK_TAI are used. Signed-off-by: Jesus Sanchez-Palencia <jesus.sanchez-palencia@intel.com> Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-04 05:42:53 +07:00
TC_SETUP_QDISC_ETF,
TC_SETUP_ROOT_QDISC,
TC_SETUP_QDISC_GRED,
};
/* These structures hold the attributes of bpf state that are being passed
* to the netdevice through the bpf op.
*/
enum bpf_netdev_command {
/* Set or clear a bpf program used in the earliest stages of packet
* rx. The prog will have been loaded as BPF_PROG_TYPE_XDP. The callee
* is responsible for calling bpf_prog_put on any old progs that are
* stored. In case of error, the callee need not release the new prog
* reference, but on success it takes ownership and must bpf_prog_put
* when it is no longer used.
*/
XDP_SETUP_PROG,
XDP_SETUP_PROG_HW,
XDP_QUERY_PROG,
XDP_QUERY_PROG_HW,
/* BPF program for offload callbacks, invoked at program load time. */
BPF_OFFLOAD_MAP_ALLOC,
BPF_OFFLOAD_MAP_FREE,
XDP_SETUP_XSK_UMEM,
};
struct bpf_prog_offload_ops;
struct netlink_ext_ack;
struct xdp_umem;
struct netdev_bpf {
enum bpf_netdev_command command;
union {
/* XDP_SETUP_PROG */
struct {
u32 flags;
struct bpf_prog *prog;
struct netlink_ext_ack *extack;
};
/* XDP_QUERY_PROG, XDP_QUERY_PROG_HW */
struct {
u32 prog_id;
/* flags with which program was installed */
u32 prog_flags;
};
/* BPF_OFFLOAD_MAP_ALLOC, BPF_OFFLOAD_MAP_FREE */
struct {
struct bpf_offloaded_map *offmap;
};
/* XDP_SETUP_XSK_UMEM */
struct {
struct xdp_umem *umem;
u16 queue_id;
} xsk;
};
};
#ifdef CONFIG_XFRM_OFFLOAD
struct xfrmdev_ops {
int (*xdo_dev_state_add) (struct xfrm_state *x);
void (*xdo_dev_state_delete) (struct xfrm_state *x);
void (*xdo_dev_state_free) (struct xfrm_state *x);
bool (*xdo_dev_offload_ok) (struct sk_buff *skb,
struct xfrm_state *x);
void (*xdo_dev_state_advance_esn) (struct xfrm_state *x);
};
#endif
struct dev_ifalias {
struct rcu_head rcuhead;
char ifalias[];
};
struct devlink;
struct tlsdev_ops;
/*
* 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 a network device is registered.
* The network device can use this for any late stage initialization
* or semantic validation. It can fail with an error code which will
* be propagated 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 a network device transitions to the up
* state.
*
* int (*ndo_stop)(struct net_device *dev);
* This function is called when a network device transitions 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.
* Returns NETDEV_TX_OK. Can return NETDEV_TX_BUSY, but you should stop
* the queue before that can happen; it's for obsolete devices and weird
* corner cases, but the stack really does a non-trivial amount
* of useless work if you return NETDEV_TX_BUSY.
* Required; cannot be NULL.
*
* netdev_features_t (*ndo_features_check)(struct sk_buff *skb,
* struct net_device *dev
* netdev_features_t features);
* Called by core transmit path to determine if device is capable of
* performing offload operations on a given packet. This is to give
* the device an opportunity to implement any restrictions that cannot
* be otherwise expressed by feature flags. The check is called with
* the set of features that the stack has calculated and it returns
* those the driver believes to be appropriate.
*
net: core: explicitly select a txq before doing l2 forwarding Currently, the tx queue were selected implicitly in ndo_dfwd_start_xmit(). The will cause several issues: - NETIF_F_LLTX were removed for macvlan, so txq lock were done for macvlan instead of lower device which misses the necessary txq synchronization for lower device such as txq stopping or frozen required by dev watchdog or control path. - dev_hard_start_xmit() was called with NULL txq which bypasses the net device watchdog. - dev_hard_start_xmit() does not check txq everywhere which will lead a crash when tso is disabled for lower device. Fix this by explicitly introducing a new param for .ndo_select_queue() for just selecting queues in the case of l2 forwarding offload. netdev_pick_tx() was also extended to accept this parameter and dev_queue_xmit_accel() was used to do l2 forwarding transmission. With this fixes, NETIF_F_LLTX could be preserved for macvlan and there's no need to check txq against NULL in dev_hard_start_xmit(). Also there's no need to keep a dedicated ndo_dfwd_start_xmit() and we can just reuse the code of dev_queue_xmit() to do the transmission. In the future, it was also required for macvtap l2 forwarding support since it provides a necessary synchronization method. Cc: John Fastabend <john.r.fastabend@intel.com> Cc: Neil Horman <nhorman@tuxdriver.com> Cc: e1000-devel@lists.sourceforge.net Signed-off-by: Jason Wang <jasowang@redhat.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Acked-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-10 15:18:26 +07:00
* u16 (*ndo_select_queue)(struct net_device *dev, struct sk_buff *skb,
* struct net_device *sb_dev);
* Called to decide which queue to use 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 promiscuous is enabled.
*
* void (*ndo_set_rx_mode)(struct net_device *dev);
* This function is called device changes address list filtering.
* If driver handles unicast address filtering, it should set
* IFF_UNICAST_FLT in its priv_flags.
*
* 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 requests an ioctl which can't be handled by
* the generic interface code. If not defined ioctls 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 reasons; 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.
*
* void (*ndo_tx_timeout)(struct net_device *dev);
* Callback used when the transmitter has not made any progress
* for dev->watchdog ticks.
*
* void (*ndo_get_stats64)(struct net_device *dev,
* struct rtnl_link_stats64 *storage);
* struct net_device_stats* (*ndo_get_stats)(struct net_device *dev);
* Called when a user wants to get the network device usage
* statistics. Drivers must do one of the following:
* 1. Define @ndo_get_stats64 to fill in a zero-initialised
* rtnl_link_stats64 structure passed by the caller.
* 2. Define @ndo_get_stats to update a net_device_stats structure
* (which should normally be dev->stats) and return a pointer to
* it. The structure may be changed asynchronously only if each
* field is written atomically.
* 3. Update dev->stats asynchronously and atomically, and define
* neither operation.
*
* bool (*ndo_has_offload_stats)(const struct net_device *dev, int attr_id)
* Return true if this device supports offload stats of this attr_id.
*
* int (*ndo_get_offload_stats)(int attr_id, const struct net_device *dev,
* void *attr_data)
* Get statistics for offload operations by attr_id. Write it into the
* attr_data pointer.
*
* int (*ndo_vlan_rx_add_vid)(struct net_device *dev, __be16 proto, u16 vid);
* If device supports VLAN filtering this function is called when a
* VLAN id is registered.
*
* int (*ndo_vlan_rx_kill_vid)(struct net_device *dev, __be16 proto, u16 vid);
* If device supports VLAN filtering 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, __be16 proto);
net-next:v4: Add support to configure SR-IOV VF minimum and maximum Tx rate through ip tool. o min_tx_rate puts lower limit on the VF bandwidth. VF is guaranteed to have a bandwidth of at least this value. max_tx_rate puts cap on the VF bandwidth. VF can have a bandwidth of up to this value. o A new handler set_vf_rate for attr IFLA_VF_RATE has been introduced which takes 4 arguments: netdev, VF number, min_tx_rate, max_tx_rate o ndo_set_vf_rate replaces ndo_set_vf_tx_rate handler. o Drivers that currently implement ndo_set_vf_tx_rate should now call ndo_set_vf_rate instead and reject attempt to set a minimum bandwidth greater than 0 for IFLA_VF_TX_RATE when IFLA_VF_RATE is not yet implemented by driver. o If user enters only one of either min_tx_rate or max_tx_rate, then, userland should read back the other value from driver and set both for IFLA_VF_RATE. Drivers that have not yet implemented IFLA_VF_RATE should always return min_tx_rate as 0 when read from ip tool. o If both IFLA_VF_TX_RATE and IFLA_VF_RATE options are specified, then IFLA_VF_RATE should override. o Idea is to have consistent display of rate values to user. o Usage example: - ./ip link set p4p1 vf 0 rate 900 ./ip link show p4p1 32: p4p1: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT qlen 1000 link/ether 00:0e:1e:08:b0:f0 brd ff:ff:ff:ff:ff:ff vf 0 MAC 3e:a0:ca:bd:ae:5a, tx rate 900 (Mbps), max_tx_rate 900Mbps vf 1 MAC f6:c6:7c:3f:3d:6c vf 2 MAC 56:32:43:98:d7:71 vf 3 MAC d6:be:c3:b5:85:ff vf 4 MAC ee:a9:9a:1e:19:14 vf 5 MAC 4a:d0:4c:07:52:18 vf 6 MAC 3a:76:44:93:62:f9 vf 7 MAC 82:e9:e7:e3:15:1a ./ip link set p4p1 vf 0 max_tx_rate 300 min_tx_rate 200 ./ip link show p4p1 32: p4p1: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT qlen 1000 link/ether 00:0e:1e:08:b0:f0 brd ff:ff:ff:ff:ff:ff vf 0 MAC 3e:a0:ca:bd:ae:5a, tx rate 300 (Mbps), max_tx_rate 300Mbps, min_tx_rate 200Mbps vf 1 MAC f6:c6:7c:3f:3d:6c vf 2 MAC 56:32:43:98:d7:71 vf 3 MAC d6:be:c3:b5:85:ff vf 4 MAC ee:a9:9a:1e:19:14 vf 5 MAC 4a:d0:4c:07:52:18 vf 6 MAC 3a:76:44:93:62:f9 vf 7 MAC 82:e9:e7:e3:15:1a ./ip link set p4p1 vf 0 max_tx_rate 600 rate 300 ./ip link show p4p1 32: p4p1: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT qlen 1000 link/ether 00:0e:1e:08:b0:f brd ff:ff:ff:ff:ff:ff vf 0 MAC 3e:a0:ca:bd:ae:5, tx rate 600 (Mbps), max_tx_rate 600Mbps, min_tx_rate 200Mbps vf 1 MAC f6:c6:7c:3f:3d:6c vf 2 MAC 56:32:43:98:d7:71 vf 3 MAC d6:be:c3:b5:85:ff vf 4 MAC ee:a9:9a:1e:19:14 vf 5 MAC 4a:d0:4c:07:52:18 vf 6 MAC 3a:76:44:93:62:f9 vf 7 MAC 82:e9:e7:e3:15:1a Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-22 20:59:05 +07:00
* int (*ndo_set_vf_rate)(struct net_device *dev, int vf, int min_tx_rate,
* int max_tx_rate);
* int (*ndo_set_vf_spoofchk)(struct net_device *dev, int vf, bool setting);
* int (*ndo_set_vf_trust)(struct net_device *dev, int vf, bool setting);
* int (*ndo_get_vf_config)(struct net_device *dev,
* int vf, struct ifla_vf_info *ivf);
* int (*ndo_set_vf_link_state)(struct net_device *dev, int vf, int link_state);
net: Add netlink support for virtual port management (was iovnl) Add new netdev ops ndo_{set|get}_vf_port to allow setting of port-profile on a netdev interface. Extends netlink socket RTM_SETLINK/ RTM_GETLINK with two new sub msgs called IFLA_VF_PORTS and IFLA_PORT_SELF (added to end of IFLA_cmd list). These are both nested atrtibutes using this layout: [IFLA_NUM_VF] [IFLA_VF_PORTS] [IFLA_VF_PORT] [IFLA_PORT_*], ... [IFLA_VF_PORT] [IFLA_PORT_*], ... ... [IFLA_PORT_SELF] [IFLA_PORT_*], ... These attributes are design to be set and get symmetrically. VF_PORTS is a list of VF_PORTs, one for each VF, when dealing with an SR-IOV device. PORT_SELF is for the PF of the SR-IOV device, in case it wants to also have a port-profile, or for the case where the VF==PF, like in enic patch 2/2 of this patch set. A port-profile is used to configure/enable the external switch virtual port backing the netdev interface, not to configure the host-facing side of the netdev. A port-profile is an identifier known to the switch. How port- profiles are installed on the switch or how available port-profiles are made know to the host is outside the scope of this patch. There are two types of port-profiles specs in the netlink msg. The first spec is for 802.1Qbg (pre-)standard, VDP protocol. The second spec is for devices that run a similar protocol as VDP but in firmware, thus hiding the protocol details. In either case, the specs have much in common and makes sense to define the netlink msg as the union of the two specs. For example, both specs have a notition of associating/deassociating a port-profile. And both specs require some information from the hypervisor manager, such as client port instance ID. The general flow is the port-profile is applied to a host netdev interface using RTM_SETLINK, the receiver of the RTM_SETLINK msg communicates with the switch, and the switch virtual port backing the host netdev interface is configured/enabled based on the settings defined by the port-profile. What those settings comprise, and how those settings are managed is again outside the scope of this patch, since this patch only deals with the first step in the flow. Signed-off-by: Scott Feldman <scofeldm@cisco.com> Signed-off-by: Roopa Prabhu <roprabhu@cisco.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-18 12:49:55 +07:00
* int (*ndo_set_vf_port)(struct net_device *dev, int vf,
* struct nlattr *port[]);
*
* Enable or disable the VF ability to query its RSS Redirection Table and
* Hash Key. This is needed since on some devices VF share this information
* with PF and querying it may introduce a theoretical security risk.
* int (*ndo_set_vf_rss_query_en)(struct net_device *dev, int vf, bool setting);
net: Add netlink support for virtual port management (was iovnl) Add new netdev ops ndo_{set|get}_vf_port to allow setting of port-profile on a netdev interface. Extends netlink socket RTM_SETLINK/ RTM_GETLINK with two new sub msgs called IFLA_VF_PORTS and IFLA_PORT_SELF (added to end of IFLA_cmd list). These are both nested atrtibutes using this layout: [IFLA_NUM_VF] [IFLA_VF_PORTS] [IFLA_VF_PORT] [IFLA_PORT_*], ... [IFLA_VF_PORT] [IFLA_PORT_*], ... ... [IFLA_PORT_SELF] [IFLA_PORT_*], ... These attributes are design to be set and get symmetrically. VF_PORTS is a list of VF_PORTs, one for each VF, when dealing with an SR-IOV device. PORT_SELF is for the PF of the SR-IOV device, in case it wants to also have a port-profile, or for the case where the VF==PF, like in enic patch 2/2 of this patch set. A port-profile is used to configure/enable the external switch virtual port backing the netdev interface, not to configure the host-facing side of the netdev. A port-profile is an identifier known to the switch. How port- profiles are installed on the switch or how available port-profiles are made know to the host is outside the scope of this patch. There are two types of port-profiles specs in the netlink msg. The first spec is for 802.1Qbg (pre-)standard, VDP protocol. The second spec is for devices that run a similar protocol as VDP but in firmware, thus hiding the protocol details. In either case, the specs have much in common and makes sense to define the netlink msg as the union of the two specs. For example, both specs have a notition of associating/deassociating a port-profile. And both specs require some information from the hypervisor manager, such as client port instance ID. The general flow is the port-profile is applied to a host netdev interface using RTM_SETLINK, the receiver of the RTM_SETLINK msg communicates with the switch, and the switch virtual port backing the host netdev interface is configured/enabled based on the settings defined by the port-profile. What those settings comprise, and how those settings are managed is again outside the scope of this patch, since this patch only deals with the first step in the flow. Signed-off-by: Scott Feldman <scofeldm@cisco.com> Signed-off-by: Roopa Prabhu <roprabhu@cisco.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-18 12:49:55 +07:00
* int (*ndo_get_vf_port)(struct net_device *dev, int vf, struct sk_buff *skb);
* int (*ndo_setup_tc)(struct net_device *dev, enum tc_setup_type type,
* void *type_data);
* Called to setup any 'tc' scheduler, classifier or action on @dev.
* This is always called from the stack with the rtnl lock held and netif
* tx queues stopped. This allows the netdevice to perform queue
* management safely.
*
* Fiber Channel over Ethernet (FCoE) offload functions.
* int (*ndo_fcoe_enable)(struct net_device *dev);
* Called when the FCoE protocol stack wants to start using LLD for FCoE
* so the underlying device can perform whatever needed configuration or
* initialization to support acceleration of FCoE traffic.
*
* int (*ndo_fcoe_disable)(struct net_device *dev);
* Called when the FCoE protocol stack wants to stop using LLD for FCoE
* so the underlying device can perform whatever needed clean-ups to
* stop supporting acceleration of FCoE traffic.
*
* int (*ndo_fcoe_ddp_setup)(struct net_device *dev, u16 xid,
* struct scatterlist *sgl, unsigned int sgc);
* Called when the FCoE Initiator wants to initialize an I/O that
* is a possible candidate for Direct Data Placement (DDP). The LLD can
* perform necessary setup and returns 1 to indicate the device is set up
* successfully to perform DDP on this I/O, otherwise this returns 0.
*
* int (*ndo_fcoe_ddp_done)(struct net_device *dev, u16 xid);
* Called when the FCoE Initiator/Target is done with the DDPed I/O as
* indicated by the FC exchange id 'xid', so the underlying device can
* clean up and reuse resources for later DDP requests.
*
* int (*ndo_fcoe_ddp_target)(struct net_device *dev, u16 xid,
* struct scatterlist *sgl, unsigned int sgc);
* Called when the FCoE Target wants to initialize an I/O that
* is a possible candidate for Direct Data Placement (DDP). The LLD can
* perform necessary setup and returns 1 to indicate the device is set up
* successfully to perform DDP on this I/O, otherwise this returns 0.
*
* int (*ndo_fcoe_get_hbainfo)(struct net_device *dev,
* struct netdev_fcoe_hbainfo *hbainfo);
* Called when the FCoE Protocol stack wants information on the underlying
* device. This information is utilized by the FCoE protocol stack to
* register attributes with Fiber Channel management service as per the
* FC-GS Fabric Device Management Information(FDMI) specification.
*
* int (*ndo_fcoe_get_wwn)(struct net_device *dev, u64 *wwn, int type);
* Called when the underlying device wants to override default World Wide
* Name (WWN) generation mechanism in FCoE protocol stack to pass its own
* World Wide Port Name (WWPN) or World Wide Node Name (WWNN) to the FCoE
* protocol stack to use.
*
* RFS acceleration.
* int (*ndo_rx_flow_steer)(struct net_device *dev, const struct sk_buff *skb,
* u16 rxq_index, u32 flow_id);
* Set hardware filter for RFS. rxq_index is the target queue index;
* flow_id is a flow ID to be passed to rps_may_expire_flow() later.
* Return the filter ID on success, or a negative error code.
*
* Slave management functions (for bridge, bonding, etc).
* int (*ndo_add_slave)(struct net_device *dev, struct net_device *slave_dev);
* Called to make another netdev an underling.
*
* int (*ndo_del_slave)(struct net_device *dev, struct net_device *slave_dev);
* Called to release previously enslaved netdev.
*
* Feature/offload setting functions.
* netdev_features_t (*ndo_fix_features)(struct net_device *dev,
* netdev_features_t features);
* Adjusts the requested feature flags according to device-specific
* constraints, and returns the resulting flags. Must not modify
* the device state.
*
* int (*ndo_set_features)(struct net_device *dev, netdev_features_t features);
* Called to update device configuration to new features. Passed
* feature set might be less than what was returned by ndo_fix_features()).
* Must return >0 or -errno if it changed dev->features itself.
*
* int (*ndo_fdb_add)(struct ndmsg *ndm, struct nlattr *tb[],
* struct net_device *dev,
* const unsigned char *addr, u16 vid, u16 flags,
* struct netlink_ext_ack *extack);
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
* Adds an FDB entry to dev for addr.
* int (*ndo_fdb_del)(struct ndmsg *ndm, struct nlattr *tb[],
* struct net_device *dev,
* const unsigned char *addr, u16 vid)
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
* Deletes the FDB entry from dev coresponding to addr.
* int (*ndo_fdb_dump)(struct sk_buff *skb, struct netlink_callback *cb,
* struct net_device *dev, struct net_device *filter_dev,
rtnetlink: fdb dump: optimize by saving last interface markers fdb dumps spanning multiple skb's currently restart from the first interface again for every skb. This results in unnecessary iterations on the already visited interfaces and their fdb entries. In large scale setups, we have seen this to slow down fdb dumps considerably. On a system with 30k macs we see fdb dumps spanning across more than 300 skbs. To fix the problem, this patch replaces the existing single fdb marker with three markers: netdev hash entries, netdevs and fdb index to continue where we left off instead of restarting from the first netdev. This is consistent with link dumps. In the process of fixing the performance issue, this patch also re-implements fix done by commit 472681d57a5d ("net: ndo_fdb_dump should report -EMSGSIZE to rtnl_fdb_dump") (with an internal fix from Wilson Kok) in the following ways: - change ndo_fdb_dump handlers to return error code instead of the last fdb index - use cb->args strictly for dump frag markers and not error codes. This is consistent with other dump functions. Below results were taken on a system with 1000 netdevs and 35085 fdb entries: before patch: $time bridge fdb show | wc -l 15065 real 1m11.791s user 0m0.070s sys 1m8.395s (existing code does not return all macs) after patch: $time bridge fdb show | wc -l 35085 real 0m2.017s user 0m0.113s sys 0m1.942s Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com> Signed-off-by: Wilson Kok <wkok@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-31 11:56:45 +07:00
* int *idx)
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
* Used to add FDB entries to dump requests. Implementers should add
* entries to skb and update idx with the number of entries.
*
* int (*ndo_bridge_setlink)(struct net_device *dev, struct nlmsghdr *nlh,
* u16 flags, struct netlink_ext_ack *extack)
* int (*ndo_bridge_getlink)(struct sk_buff *skb, u32 pid, u32 seq,
* struct net_device *dev, u32 filter_mask,
* int nlflags)
* int (*ndo_bridge_dellink)(struct net_device *dev, struct nlmsghdr *nlh,
* u16 flags);
*
* int (*ndo_change_carrier)(struct net_device *dev, bool new_carrier);
* Called to change device carrier. Soft-devices (like dummy, team, etc)
* which do not represent real hardware may define this to allow their
* userspace components to manage their virtual carrier state. Devices
* that determine carrier state from physical hardware properties (eg
* network cables) or protocol-dependent mechanisms (eg
* USB_CDC_NOTIFY_NETWORK_CONNECTION) should NOT implement this function.
*
* int (*ndo_get_phys_port_id)(struct net_device *dev,
* struct netdev_phys_item_id *ppid);
* Called to get ID of physical port of this device. If driver does
* not implement this, it is assumed that the hw is not able to have
* multiple net devices on single physical port.
*
* int (*ndo_get_port_parent_id)(struct net_device *dev,
* struct netdev_phys_item_id *ppid)
* Called to get the parent ID of the physical port of this device.
*
* void (*ndo_udp_tunnel_add)(struct net_device *dev,
* struct udp_tunnel_info *ti);
* Called by UDP tunnel to notify a driver about the UDP port and socket
* address family that a UDP tunnel is listnening to. It is called only
* when a new port starts listening. The operation is protected by the
* RTNL.
*
* void (*ndo_udp_tunnel_del)(struct net_device *dev,
* struct udp_tunnel_info *ti);
* Called by UDP tunnel to notify the driver about a UDP port and socket
* address family that the UDP tunnel is not listening to anymore. The
* operation is protected by the RTNL.
*
* void* (*ndo_dfwd_add_station)(struct net_device *pdev,
* struct net_device *dev)
* Called by upper layer devices to accelerate switching or other
* station functionality into hardware. 'pdev is the lowerdev
* to use for the offload and 'dev' is the net device that will
* back the offload. Returns a pointer to the private structure
* the upper layer will maintain.
* void (*ndo_dfwd_del_station)(struct net_device *pdev, void *priv)
* Called by upper layer device to delete the station created
* by 'ndo_dfwd_add_station'. 'pdev' is the net device backing
* the station and priv is the structure returned by the add
* operation.
* int (*ndo_set_tx_maxrate)(struct net_device *dev,
* int queue_index, u32 maxrate);
* Called when a user wants to set a max-rate limitation of specific
* TX queue.
* int (*ndo_get_iflink)(const struct net_device *dev);
* Called to get the iflink value of this device.
* void (*ndo_change_proto_down)(struct net_device *dev,
* bool proto_down);
* This function is used to pass protocol port error state information
* to the switch driver. The switch driver can react to the proto_down
* by doing a phys down on the associated switch port.
* int (*ndo_fill_metadata_dst)(struct net_device *dev, struct sk_buff *skb);
* This function is used to get egress tunnel information for given skb.
* This is useful for retrieving outer tunnel header parameters while
* sampling packet.
* void (*ndo_set_rx_headroom)(struct net_device *dev, int needed_headroom);
* This function is used to specify the headroom that the skb must
* consider when allocation skb during packet reception. Setting
* appropriate rx headroom value allows avoiding skb head copy on
* forward. Setting a negative value resets the rx headroom to the
* default value.
* int (*ndo_bpf)(struct net_device *dev, struct netdev_bpf *bpf);
* This function is used to set or query state related to XDP on the
* netdevice and manage BPF offload. See definition of
* enum bpf_netdev_command for details.
* int (*ndo_xdp_xmit)(struct net_device *dev, int n, struct xdp_frame **xdp,
* u32 flags);
* This function is used to submit @n XDP packets for transmit on a
* netdevice. Returns number of frames successfully transmitted, frames
* that got dropped are freed/returned via xdp_return_frame().
* Returns negative number, means general error invoking ndo, meaning
* no frames were xmit'ed and core-caller will free all frames.
* struct devlink_port *(*ndo_get_devlink_port)(struct net_device *dev);
* Get devlink port instance associated with a given netdev.
* Called with a reference on the netdevice and devlink locks only,
* rtnl_lock is not held.
*/
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);
netdev_features_t (*ndo_features_check)(struct sk_buff *skb,
struct net_device *dev,
netdev_features_t features);
u16 (*ndo_select_queue)(struct net_device *dev,
net: core: explicitly select a txq before doing l2 forwarding Currently, the tx queue were selected implicitly in ndo_dfwd_start_xmit(). The will cause several issues: - NETIF_F_LLTX were removed for macvlan, so txq lock were done for macvlan instead of lower device which misses the necessary txq synchronization for lower device such as txq stopping or frozen required by dev watchdog or control path. - dev_hard_start_xmit() was called with NULL txq which bypasses the net device watchdog. - dev_hard_start_xmit() does not check txq everywhere which will lead a crash when tso is disabled for lower device. Fix this by explicitly introducing a new param for .ndo_select_queue() for just selecting queues in the case of l2 forwarding offload. netdev_pick_tx() was also extended to accept this parameter and dev_queue_xmit_accel() was used to do l2 forwarding transmission. With this fixes, NETIF_F_LLTX could be preserved for macvlan and there's no need to check txq against NULL in dev_hard_start_xmit(). Also there's no need to keep a dedicated ndo_dfwd_start_xmit() and we can just reuse the code of dev_queue_xmit() to do the transmission. In the future, it was also required for macvtap l2 forwarding support since it provides a necessary synchronization method. Cc: John Fastabend <john.r.fastabend@intel.com> Cc: Neil Horman <nhorman@tuxdriver.com> Cc: e1000-devel@lists.sourceforge.net Signed-off-by: Jason Wang <jasowang@redhat.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Acked-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-10 15:18:26 +07:00
struct sk_buff *skb,
struct net_device *sb_dev);
void (*ndo_change_rx_flags)(struct net_device *dev,
int flags);
void (*ndo_set_rx_mode)(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);
void (*ndo_get_stats64)(struct net_device *dev,
struct rtnl_link_stats64 *storage);
bool (*ndo_has_offload_stats)(const struct net_device *dev, int attr_id);
int (*ndo_get_offload_stats)(int attr_id,
const struct net_device *dev,
void *attr_data);
struct net_device_stats* (*ndo_get_stats)(struct net_device *dev);
int (*ndo_vlan_rx_add_vid)(struct net_device *dev,
__be16 proto, u16 vid);
int (*ndo_vlan_rx_kill_vid)(struct net_device *dev,
__be16 proto, u16 vid);
#ifdef CONFIG_NET_POLL_CONTROLLER
void (*ndo_poll_controller)(struct net_device *dev);
int (*ndo_netpoll_setup)(struct net_device *dev,
netpoll: Remove gfp parameter from __netpoll_setup The gfp parameter was added in: commit 47be03a28cc6c80e3aa2b3e8ed6d960ff0c5c0af Author: Amerigo Wang <amwang@redhat.com> Date: Fri Aug 10 01:24:37 2012 +0000 netpoll: use GFP_ATOMIC in slave_enable_netpoll() and __netpoll_setup() slave_enable_netpoll() and __netpoll_setup() may be called with read_lock() held, so should use GFP_ATOMIC to allocate memory. Eric suggested to pass gfp flags to __netpoll_setup(). Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: "David S. Miller" <davem@davemloft.net> Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Cong Wang <amwang@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> The reason for the gfp parameter was removed in: commit c4cdef9b7183159c23c7302aaf270d64c549f557 Author: dingtianhong <dingtianhong@huawei.com> Date: Tue Jul 23 15:25:27 2013 +0800 bonding: don't call slave_xxx_netpoll under spinlocks The slave_xxx_netpoll will call synchronize_rcu_bh(), so the function may schedule and sleep, it should't be called under spinlocks. bond_netpoll_setup() and bond_netpoll_cleanup() are always protected by rtnl lock, it is no need to take the read lock, as the slave list couldn't be changed outside rtnl lock. Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net> Nothing else that calls __netpoll_setup or ndo_netpoll_setup requires a gfp paramter, so remove the gfp parameter from both of these functions making the code clearer. Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-28 05:36:38 +07:00
struct netpoll_info *info);
void (*ndo_netpoll_cleanup)(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, __be16 proto);
net-next:v4: Add support to configure SR-IOV VF minimum and maximum Tx rate through ip tool. o min_tx_rate puts lower limit on the VF bandwidth. VF is guaranteed to have a bandwidth of at least this value. max_tx_rate puts cap on the VF bandwidth. VF can have a bandwidth of up to this value. o A new handler set_vf_rate for attr IFLA_VF_RATE has been introduced which takes 4 arguments: netdev, VF number, min_tx_rate, max_tx_rate o ndo_set_vf_rate replaces ndo_set_vf_tx_rate handler. o Drivers that currently implement ndo_set_vf_tx_rate should now call ndo_set_vf_rate instead and reject attempt to set a minimum bandwidth greater than 0 for IFLA_VF_TX_RATE when IFLA_VF_RATE is not yet implemented by driver. o If user enters only one of either min_tx_rate or max_tx_rate, then, userland should read back the other value from driver and set both for IFLA_VF_RATE. Drivers that have not yet implemented IFLA_VF_RATE should always return min_tx_rate as 0 when read from ip tool. o If both IFLA_VF_TX_RATE and IFLA_VF_RATE options are specified, then IFLA_VF_RATE should override. o Idea is to have consistent display of rate values to user. o Usage example: - ./ip link set p4p1 vf 0 rate 900 ./ip link show p4p1 32: p4p1: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT qlen 1000 link/ether 00:0e:1e:08:b0:f0 brd ff:ff:ff:ff:ff:ff vf 0 MAC 3e:a0:ca:bd:ae:5a, tx rate 900 (Mbps), max_tx_rate 900Mbps vf 1 MAC f6:c6:7c:3f:3d:6c vf 2 MAC 56:32:43:98:d7:71 vf 3 MAC d6:be:c3:b5:85:ff vf 4 MAC ee:a9:9a:1e:19:14 vf 5 MAC 4a:d0:4c:07:52:18 vf 6 MAC 3a:76:44:93:62:f9 vf 7 MAC 82:e9:e7:e3:15:1a ./ip link set p4p1 vf 0 max_tx_rate 300 min_tx_rate 200 ./ip link show p4p1 32: p4p1: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT qlen 1000 link/ether 00:0e:1e:08:b0:f0 brd ff:ff:ff:ff:ff:ff vf 0 MAC 3e:a0:ca:bd:ae:5a, tx rate 300 (Mbps), max_tx_rate 300Mbps, min_tx_rate 200Mbps vf 1 MAC f6:c6:7c:3f:3d:6c vf 2 MAC 56:32:43:98:d7:71 vf 3 MAC d6:be:c3:b5:85:ff vf 4 MAC ee:a9:9a:1e:19:14 vf 5 MAC 4a:d0:4c:07:52:18 vf 6 MAC 3a:76:44:93:62:f9 vf 7 MAC 82:e9:e7:e3:15:1a ./ip link set p4p1 vf 0 max_tx_rate 600 rate 300 ./ip link show p4p1 32: p4p1: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT qlen 1000 link/ether 00:0e:1e:08:b0:f brd ff:ff:ff:ff:ff:ff vf 0 MAC 3e:a0:ca:bd:ae:5, tx rate 600 (Mbps), max_tx_rate 600Mbps, min_tx_rate 200Mbps vf 1 MAC f6:c6:7c:3f:3d:6c vf 2 MAC 56:32:43:98:d7:71 vf 3 MAC d6:be:c3:b5:85:ff vf 4 MAC ee:a9:9a:1e:19:14 vf 5 MAC 4a:d0:4c:07:52:18 vf 6 MAC 3a:76:44:93:62:f9 vf 7 MAC 82:e9:e7:e3:15:1a Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-22 20:59:05 +07:00
int (*ndo_set_vf_rate)(struct net_device *dev,
int vf, int min_tx_rate,
int max_tx_rate);
int (*ndo_set_vf_spoofchk)(struct net_device *dev,
int vf, bool setting);
int (*ndo_set_vf_trust)(struct net_device *dev,
int vf, bool setting);
int (*ndo_get_vf_config)(struct net_device *dev,
int vf,
struct ifla_vf_info *ivf);
int (*ndo_set_vf_link_state)(struct net_device *dev,
int vf, int link_state);
int (*ndo_get_vf_stats)(struct net_device *dev,
int vf,
struct ifla_vf_stats
*vf_stats);
net: Add netlink support for virtual port management (was iovnl) Add new netdev ops ndo_{set|get}_vf_port to allow setting of port-profile on a netdev interface. Extends netlink socket RTM_SETLINK/ RTM_GETLINK with two new sub msgs called IFLA_VF_PORTS and IFLA_PORT_SELF (added to end of IFLA_cmd list). These are both nested atrtibutes using this layout: [IFLA_NUM_VF] [IFLA_VF_PORTS] [IFLA_VF_PORT] [IFLA_PORT_*], ... [IFLA_VF_PORT] [IFLA_PORT_*], ... ... [IFLA_PORT_SELF] [IFLA_PORT_*], ... These attributes are design to be set and get symmetrically. VF_PORTS is a list of VF_PORTs, one for each VF, when dealing with an SR-IOV device. PORT_SELF is for the PF of the SR-IOV device, in case it wants to also have a port-profile, or for the case where the VF==PF, like in enic patch 2/2 of this patch set. A port-profile is used to configure/enable the external switch virtual port backing the netdev interface, not to configure the host-facing side of the netdev. A port-profile is an identifier known to the switch. How port- profiles are installed on the switch or how available port-profiles are made know to the host is outside the scope of this patch. There are two types of port-profiles specs in the netlink msg. The first spec is for 802.1Qbg (pre-)standard, VDP protocol. The second spec is for devices that run a similar protocol as VDP but in firmware, thus hiding the protocol details. In either case, the specs have much in common and makes sense to define the netlink msg as the union of the two specs. For example, both specs have a notition of associating/deassociating a port-profile. And both specs require some information from the hypervisor manager, such as client port instance ID. The general flow is the port-profile is applied to a host netdev interface using RTM_SETLINK, the receiver of the RTM_SETLINK msg communicates with the switch, and the switch virtual port backing the host netdev interface is configured/enabled based on the settings defined by the port-profile. What those settings comprise, and how those settings are managed is again outside the scope of this patch, since this patch only deals with the first step in the flow. Signed-off-by: Scott Feldman <scofeldm@cisco.com> Signed-off-by: Roopa Prabhu <roprabhu@cisco.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-18 12:49:55 +07:00
int (*ndo_set_vf_port)(struct net_device *dev,
int vf,
struct nlattr *port[]);
int (*ndo_get_vf_port)(struct net_device *dev,
int vf, struct sk_buff *skb);
int (*ndo_set_vf_guid)(struct net_device *dev,
int vf, u64 guid,
int guid_type);
int (*ndo_set_vf_rss_query_en)(
struct net_device *dev,
int vf, bool setting);
int (*ndo_setup_tc)(struct net_device *dev,
enum tc_setup_type type,
void *type_data);
#if IS_ENABLED(CONFIG_FCOE)
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);
int (*ndo_fcoe_ddp_target)(struct net_device *dev,
u16 xid,
struct scatterlist *sgl,
unsigned int sgc);
int (*ndo_fcoe_get_hbainfo)(struct net_device *dev,
struct netdev_fcoe_hbainfo *hbainfo);
#endif
#if IS_ENABLED(CONFIG_LIBFCOE)
#define NETDEV_FCOE_WWNN 0
#define NETDEV_FCOE_WWPN 1
int (*ndo_fcoe_get_wwn)(struct net_device *dev,
u64 *wwn, int type);
#endif
#ifdef CONFIG_RFS_ACCEL
int (*ndo_rx_flow_steer)(struct net_device *dev,
const struct sk_buff *skb,
u16 rxq_index,
u32 flow_id);
#endif
int (*ndo_add_slave)(struct net_device *dev,
struct net_device *slave_dev,
struct netlink_ext_ack *extack);
int (*ndo_del_slave)(struct net_device *dev,
struct net_device *slave_dev);
netdev_features_t (*ndo_fix_features)(struct net_device *dev,
netdev_features_t features);
int (*ndo_set_features)(struct net_device *dev,
netdev_features_t features);
int (*ndo_neigh_construct)(struct net_device *dev,
struct neighbour *n);
void (*ndo_neigh_destroy)(struct net_device *dev,
struct neighbour *n);
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
int (*ndo_fdb_add)(struct ndmsg *ndm,
struct nlattr *tb[],
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
struct net_device *dev,
const unsigned char *addr,
u16 vid,
u16 flags,
struct netlink_ext_ack *extack);
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
int (*ndo_fdb_del)(struct ndmsg *ndm,
struct nlattr *tb[],
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
struct net_device *dev,
const unsigned char *addr,
u16 vid);
net: add generic PF_BRIDGE:RTM_ FDB hooks This adds two new flags NTF_MASTER and NTF_SELF that can now be used to specify where PF_BRIDGE netlink commands should be sent. NTF_MASTER sends the commands to the 'dev->master' device for parsing. Typically this will be the linux net/bridge, or open-vswitch devices. Also without any flags set the command will be handled by the master device as well so that current user space tools continue to work as expected. The NTF_SELF flag will push the PF_BRIDGE commands to the device. In the basic example below the commands are then parsed and programmed in the embedded bridge. Note if both NTF_SELF and NTF_MASTER bits are set then the command will be sent to both 'dev->master' and 'dev' this allows user space to easily keep the embedded bridge and software bridge in sync. There is a slight complication in the case with both flags set when an error occurs. To resolve this the rtnl handler clears the NTF_ flag in the netlink ack to indicate which sets completed successfully. The add/del handlers will abort as soon as any error occurs. To support this new net device ops were added to call into the device and the existing bridging code was refactored to use these. There should be no required changes in user space to support the current bridge behavior. A basic setup with a SR-IOV enabled NIC looks like this, veth0 veth2 | | ------------ | bridge0 | <---- software bridging ------------ / / ethx.y ethx VF PF \ \ <---- propagate FDB entries to HW \ \ -------------------- | Embedded Bridge | <---- hardware offloaded switching -------------------- In this case the embedded bridge must be managed to allow 'veth0' to communicate with 'ethx.y' correctly. At present drivers managing the embedded bridge either send frames onto the network which then get dropped by the switch OR the embedded bridge will flood these frames. With this patch we have a mechanism to manage the embedded bridge correctly from user space. This example is specific to SR-IOV but replacing the VF with another PF or dropping this into the DSA framework generates similar management issues. Examples session using the 'br'[1] tool to add, dump and then delete a mac address with a new "embedded" option and enabled ixgbe driver: # br fdb add 22:35:19:ac:60:59 dev eth3 # br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static #br fdb add 22:35:19:ac:60:59 embedded dev eth3 #br fdb port mac addr flags veth0 22:35:19:ac:60:58 static veth0 9a:5f:81:f7:f6:ec local eth3 00:1b:21:55:23:59 local eth3 22:35:19:ac:60:59 static veth0 22:35:19:ac:60:57 static eth3 22:35:19:ac:60:59 local embedded #br fdb del 22:35:19:ac:60:59 embedded dev eth3 I added a couple lines to 'br' to set the flags correctly is all. It is my opinion that the merit of this patch is now embedded and SW bridges can both be modeled correctly in user space using very nearly the same message passing. [1] 'br' tool was published as an RFC here and will be renamed 'bridge' http://patchwork.ozlabs.org/patch/117664/ Thanks to Jamal Hadi Salim, Stephen Hemminger and Ben Hutchings for valuable feedback, suggestions, and review. v2: fixed api descriptions and error case with both NTF_SELF and NTF_MASTER set plus updated patch description. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-15 13:43:56 +07:00
int (*ndo_fdb_dump)(struct sk_buff *skb,
struct netlink_callback *cb,
struct net_device *dev,
struct net_device *filter_dev,
rtnetlink: fdb dump: optimize by saving last interface markers fdb dumps spanning multiple skb's currently restart from the first interface again for every skb. This results in unnecessary iterations on the already visited interfaces and their fdb entries. In large scale setups, we have seen this to slow down fdb dumps considerably. On a system with 30k macs we see fdb dumps spanning across more than 300 skbs. To fix the problem, this patch replaces the existing single fdb marker with three markers: netdev hash entries, netdevs and fdb index to continue where we left off instead of restarting from the first netdev. This is consistent with link dumps. In the process of fixing the performance issue, this patch also re-implements fix done by commit 472681d57a5d ("net: ndo_fdb_dump should report -EMSGSIZE to rtnl_fdb_dump") (with an internal fix from Wilson Kok) in the following ways: - change ndo_fdb_dump handlers to return error code instead of the last fdb index - use cb->args strictly for dump frag markers and not error codes. This is consistent with other dump functions. Below results were taken on a system with 1000 netdevs and 35085 fdb entries: before patch: $time bridge fdb show | wc -l 15065 real 1m11.791s user 0m0.070s sys 1m8.395s (existing code does not return all macs) after patch: $time bridge fdb show | wc -l 35085 real 0m2.017s user 0m0.113s sys 0m1.942s Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com> Signed-off-by: Wilson Kok <wkok@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-31 11:56:45 +07:00
int *idx);
int (*ndo_fdb_get)(struct sk_buff *skb,
struct nlattr *tb[],
struct net_device *dev,
const unsigned char *addr,
u16 vid, u32 portid, u32 seq,
struct netlink_ext_ack *extack);
int (*ndo_bridge_setlink)(struct net_device *dev,
struct nlmsghdr *nlh,
u16 flags,
struct netlink_ext_ack *extack);
int (*ndo_bridge_getlink)(struct sk_buff *skb,
u32 pid, u32 seq,
struct net_device *dev,
u32 filter_mask,
int nlflags);
int (*ndo_bridge_dellink)(struct net_device *dev,
struct nlmsghdr *nlh,
u16 flags);
int (*ndo_change_carrier)(struct net_device *dev,
bool new_carrier);
int (*ndo_get_phys_port_id)(struct net_device *dev,
struct netdev_phys_item_id *ppid);
int (*ndo_get_port_parent_id)(struct net_device *dev,
struct netdev_phys_item_id *ppid);
int (*ndo_get_phys_port_name)(struct net_device *dev,
char *name, size_t len);
void (*ndo_udp_tunnel_add)(struct net_device *dev,
struct udp_tunnel_info *ti);
void (*ndo_udp_tunnel_del)(struct net_device *dev,
struct udp_tunnel_info *ti);
void* (*ndo_dfwd_add_station)(struct net_device *pdev,
struct net_device *dev);
void (*ndo_dfwd_del_station)(struct net_device *pdev,
void *priv);
int (*ndo_get_lock_subclass)(struct net_device *dev);
int (*ndo_set_tx_maxrate)(struct net_device *dev,
int queue_index,
u32 maxrate);
int (*ndo_get_iflink)(const struct net_device *dev);
int (*ndo_change_proto_down)(struct net_device *dev,
bool proto_down);
int (*ndo_fill_metadata_dst)(struct net_device *dev,
struct sk_buff *skb);
void (*ndo_set_rx_headroom)(struct net_device *dev,
int needed_headroom);
int (*ndo_bpf)(struct net_device *dev,
struct netdev_bpf *bpf);
int (*ndo_xdp_xmit)(struct net_device *dev, int n,
struct xdp_frame **xdp,
u32 flags);
int (*ndo_xsk_async_xmit)(struct net_device *dev,
u32 queue_id);
struct devlink_port * (*ndo_get_devlink_port)(struct net_device *dev);
};
/**
* enum net_device_priv_flags - &struct net_device priv_flags
*
* These are the &struct net_device, they are only set internally
* by drivers and used in the kernel. These flags are invisible to
* userspace; this means that the order of these flags can change
* during any kernel release.
*
* You should have a pretty good reason to be extending these flags.
*
* @IFF_802_1Q_VLAN: 802.1Q VLAN device
* @IFF_EBRIDGE: Ethernet bridging device
* @IFF_BONDING: bonding master or slave
* @IFF_ISATAP: ISATAP interface (RFC4214)
* @IFF_WAN_HDLC: WAN HDLC device
* @IFF_XMIT_DST_RELEASE: dev_hard_start_xmit() is allowed to
* release skb->dst
* @IFF_DONT_BRIDGE: disallow bridging this ether dev
* @IFF_DISABLE_NETPOLL: disable netpoll at run-time
* @IFF_MACVLAN_PORT: device used as macvlan port
* @IFF_BRIDGE_PORT: device used as bridge port
* @IFF_OVS_DATAPATH: device used as Open vSwitch datapath port
* @IFF_TX_SKB_SHARING: The interface supports sharing skbs on transmit
* @IFF_UNICAST_FLT: Supports unicast filtering
* @IFF_TEAM_PORT: device used as team port
* @IFF_SUPP_NOFCS: device supports sending custom FCS
* @IFF_LIVE_ADDR_CHANGE: device supports hardware address
* change when it's running
* @IFF_MACVLAN: Macvlan device
* @IFF_XMIT_DST_RELEASE_PERM: IFF_XMIT_DST_RELEASE not taking into account
* underlying stacked devices
* @IFF_L3MDEV_MASTER: device is an L3 master device
* @IFF_NO_QUEUE: device can run without qdisc attached
* @IFF_OPENVSWITCH: device is a Open vSwitch master
* @IFF_L3MDEV_SLAVE: device is enslaved to an L3 master device
* @IFF_TEAM: device is a team device
ethtool: correctly ensure {GS}CHANNELS doesn't conflict with GS{RXFH} Ethernet drivers implementing both {GS}RXFH and {GS}CHANNELS ethtool ops incorrectly allow SCHANNELS when it would conflict with the settings from SRXFH. This occurs because it is not possible for drivers to understand whether their Rx flow indirection table has been configured or is in the default state. In addition, drivers currently behave in various ways when increasing the number of Rx channels. Some drivers will always destroy the Rx flow indirection table when this occurs, whether it has been set by the user or not. Other drivers will attempt to preserve the table even if the user has never modified it from the default driver settings. Neither of these situation is desirable because it leads to unexpected behavior or loss of user configuration. The correct behavior is to simply return -EINVAL when SCHANNELS would conflict with the current Rx flow table settings. However, it should only do so if the current settings were modified by the user. If we required that the new settings never conflict with the current (default) Rx flow settings, we would force users to first reduce their Rx flow settings and then reduce the number of Rx channels. This patch proposes a solution implemented in net/core/ethtool.c which ensures that all drivers behave correctly. It checks whether the RXFH table has been configured to non-default settings, and stores this information in a private netdev flag. When the number of channels is requested to change, it first ensures that the current Rx flow table is not going to assign flows to now disabled channels. Signed-off-by: Jacob Keller <jacob.e.keller@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-09 07:05:03 +07:00
* @IFF_RXFH_CONFIGURED: device has had Rx Flow indirection table configured
* @IFF_PHONY_HEADROOM: the headroom value is controlled by an external
* entity (i.e. the master device for bridged veth)
* @IFF_MACSEC: device is a MACsec device
* @IFF_NO_RX_HANDLER: device doesn't support the rx_handler hook
* @IFF_FAILOVER: device is a failover master device
* @IFF_FAILOVER_SLAVE: device is lower dev of a failover master device
ipvlan, l3mdev: fix broken l3s mode wrt local routes While implementing ipvlan l3 and l3s mode for kubernetes CNI plugin, I ran into the issue that while l3 mode is working fine, l3s mode does not have any connectivity to kube-apiserver and hence all pods end up in Error state as well. The ipvlan master device sits on top of a bond device and hostns traffic to kube-apiserver (also running in hostns) is DNATed from 10.152.183.1:443 to 139.178.29.207:37573 where the latter is the address of the bond0. While in l3 mode, a curl to https://10.152.183.1:443 or to https://139.178.29.207:37573 works fine from hostns, neither of them do in case of l3s. In the latter only a curl to https://127.0.0.1:37573 appeared to work where for local addresses of bond0 I saw kernel suddenly starting to emit ARP requests to query HW address of bond0 which remained unanswered and neighbor entries in INCOMPLETE state. These ARP requests only happen while in l3s. Debugging this further, I found the issue is that l3s mode is piggy- backing on l3 master device, and in this case local routes are using l3mdev_master_dev_rcu(dev) instead of net->loopback_dev as per commit f5a0aab84b74 ("net: ipv4: dst for local input routes should use l3mdev if relevant") and 5f02ce24c269 ("net: l3mdev: Allow the l3mdev to be a loopback"). I found that reverting them back into using the net->loopback_dev fixed ipvlan l3s connectivity and got everything working for the CNI. Now judging from 4fbae7d83c98 ("ipvlan: Introduce l3s mode") and the l3mdev paper in [0] the only sole reason why ipvlan l3s is relying on l3 master device is to get the l3mdev_ip_rcv() receive hook for setting the dst entry of the input route without adding its own ipvlan specific hacks into the receive path, however, any l3 domain semantics beyond just that are breaking l3s operation. Note that ipvlan also has the ability to dynamically switch its internal operation from l3 to l3s for all ports via ipvlan_set_port_mode() at runtime. In any case, l3 vs l3s soley distinguishes itself by 'de-confusing' netfilter through switching skb->dev to ipvlan slave device late in NF_INET_LOCAL_IN before handing the skb to L4. Minimal fix taken here is to add a IFF_L3MDEV_RX_HANDLER flag which, if set from ipvlan setup, gets us only the wanted l3mdev_l3_rcv() hook without any additional l3mdev semantics on top. This should also have minimal impact since dev->priv_flags is already hot in cache. With this set, l3s mode is working fine and I also get things like masquerading pod traffic on the ipvlan master properly working. [0] https://netdevconf.org/1.2/papers/ahern-what-is-l3mdev-paper.pdf Fixes: f5a0aab84b74 ("net: ipv4: dst for local input routes should use l3mdev if relevant") Fixes: 5f02ce24c269 ("net: l3mdev: Allow the l3mdev to be a loopback") Fixes: 4fbae7d83c98 ("ipvlan: Introduce l3s mode") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Cc: Mahesh Bandewar <maheshb@google.com> Cc: David Ahern <dsa@cumulusnetworks.com> Cc: Florian Westphal <fw@strlen.de> Cc: Martynas Pumputis <m@lambda.lt> Acked-by: David Ahern <dsa@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-30 18:49:48 +07:00
* @IFF_L3MDEV_RX_HANDLER: only invoke the rx handler of L3 master device
failover: allow name change on IFF_UP slave interfaces When a netdev appears through hot plug then gets enslaved by a failover master that is already up and running, the slave will be opened right away after getting enslaved. Today there's a race that userspace (udev) may fail to rename the slave if the kernel (net_failover) opens the slave earlier than when the userspace rename happens. Unlike bond or team, the primary slave of failover can't be renamed by userspace ahead of time, since the kernel initiated auto-enslavement is unable to, or rather, is never meant to be synchronized with the rename request from userspace. As the failover slave interfaces are not designed to be operated directly by userspace apps: IP configuration, filter rules with regard to network traffic passing and etc., should all be done on master interface. In general, userspace apps only care about the name of master interface, while slave names are less important as long as admin users can see reliable names that may carry other information describing the netdev. For e.g., they can infer that "ens3nsby" is a standby slave of "ens3", while for a name like "eth0" they can't tell which master it belongs to. Historically the name of IFF_UP interface can't be changed because there might be admin script or management software that is already relying on such behavior and assumes that the slave name can't be changed once UP. But failover is special: with the in-kernel auto-enslavement mechanism, the userspace expectation for device enumeration and bring-up order is already broken. Previously initramfs and various userspace config tools were modified to bypass failover slaves because of auto-enslavement and duplicate MAC address. Similarly, in case that users care about seeing reliable slave name, the new type of failover slaves needs to be taken care of specifically in userspace anyway. It's less risky to lift up the rename restriction on failover slave which is already UP. Although it's possible this change may potentially break userspace component (most likely configuration scripts or management software) that assumes slave name can't be changed while UP, it's relatively a limited and controllable set among all userspace components, which can be fixed specifically to listen for the rename events on failover slaves. Userspace component interacting with slaves is expected to be changed to operate on failover master interface instead, as the failover slave is dynamic in nature which may come and go at any point. The goal is to make the role of failover slaves less relevant, and userspace components should only deal with failover master in the long run. Fixes: 30c8bd5aa8b2 ("net: Introduce generic failover module") Signed-off-by: Si-Wei Liu <si-wei.liu@oracle.com> Reviewed-by: Liran Alon <liran.alon@oracle.com> Acked-by: Sridhar Samudrala <sridhar.samudrala@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-09 06:45:27 +07:00
* @IFF_LIVE_RENAME_OK: rename is allowed while device is up and running
*/
enum netdev_priv_flags {
IFF_802_1Q_VLAN = 1<<0,
IFF_EBRIDGE = 1<<1,
IFF_BONDING = 1<<2,
IFF_ISATAP = 1<<3,
IFF_WAN_HDLC = 1<<4,
IFF_XMIT_DST_RELEASE = 1<<5,
IFF_DONT_BRIDGE = 1<<6,
IFF_DISABLE_NETPOLL = 1<<7,
IFF_MACVLAN_PORT = 1<<8,
IFF_BRIDGE_PORT = 1<<9,
IFF_OVS_DATAPATH = 1<<10,
IFF_TX_SKB_SHARING = 1<<11,
IFF_UNICAST_FLT = 1<<12,
IFF_TEAM_PORT = 1<<13,
IFF_SUPP_NOFCS = 1<<14,
IFF_LIVE_ADDR_CHANGE = 1<<15,
IFF_MACVLAN = 1<<16,
IFF_XMIT_DST_RELEASE_PERM = 1<<17,
IFF_L3MDEV_MASTER = 1<<18,
IFF_NO_QUEUE = 1<<19,
IFF_OPENVSWITCH = 1<<20,
IFF_L3MDEV_SLAVE = 1<<21,
IFF_TEAM = 1<<22,
IFF_RXFH_CONFIGURED = 1<<23,
IFF_PHONY_HEADROOM = 1<<24,
IFF_MACSEC = 1<<25,
IFF_NO_RX_HANDLER = 1<<26,
IFF_FAILOVER = 1<<27,
IFF_FAILOVER_SLAVE = 1<<28,
ipvlan, l3mdev: fix broken l3s mode wrt local routes While implementing ipvlan l3 and l3s mode for kubernetes CNI plugin, I ran into the issue that while l3 mode is working fine, l3s mode does not have any connectivity to kube-apiserver and hence all pods end up in Error state as well. The ipvlan master device sits on top of a bond device and hostns traffic to kube-apiserver (also running in hostns) is DNATed from 10.152.183.1:443 to 139.178.29.207:37573 where the latter is the address of the bond0. While in l3 mode, a curl to https://10.152.183.1:443 or to https://139.178.29.207:37573 works fine from hostns, neither of them do in case of l3s. In the latter only a curl to https://127.0.0.1:37573 appeared to work where for local addresses of bond0 I saw kernel suddenly starting to emit ARP requests to query HW address of bond0 which remained unanswered and neighbor entries in INCOMPLETE state. These ARP requests only happen while in l3s. Debugging this further, I found the issue is that l3s mode is piggy- backing on l3 master device, and in this case local routes are using l3mdev_master_dev_rcu(dev) instead of net->loopback_dev as per commit f5a0aab84b74 ("net: ipv4: dst for local input routes should use l3mdev if relevant") and 5f02ce24c269 ("net: l3mdev: Allow the l3mdev to be a loopback"). I found that reverting them back into using the net->loopback_dev fixed ipvlan l3s connectivity and got everything working for the CNI. Now judging from 4fbae7d83c98 ("ipvlan: Introduce l3s mode") and the l3mdev paper in [0] the only sole reason why ipvlan l3s is relying on l3 master device is to get the l3mdev_ip_rcv() receive hook for setting the dst entry of the input route without adding its own ipvlan specific hacks into the receive path, however, any l3 domain semantics beyond just that are breaking l3s operation. Note that ipvlan also has the ability to dynamically switch its internal operation from l3 to l3s for all ports via ipvlan_set_port_mode() at runtime. In any case, l3 vs l3s soley distinguishes itself by 'de-confusing' netfilter through switching skb->dev to ipvlan slave device late in NF_INET_LOCAL_IN before handing the skb to L4. Minimal fix taken here is to add a IFF_L3MDEV_RX_HANDLER flag which, if set from ipvlan setup, gets us only the wanted l3mdev_l3_rcv() hook without any additional l3mdev semantics on top. This should also have minimal impact since dev->priv_flags is already hot in cache. With this set, l3s mode is working fine and I also get things like masquerading pod traffic on the ipvlan master properly working. [0] https://netdevconf.org/1.2/papers/ahern-what-is-l3mdev-paper.pdf Fixes: f5a0aab84b74 ("net: ipv4: dst for local input routes should use l3mdev if relevant") Fixes: 5f02ce24c269 ("net: l3mdev: Allow the l3mdev to be a loopback") Fixes: 4fbae7d83c98 ("ipvlan: Introduce l3s mode") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Cc: Mahesh Bandewar <maheshb@google.com> Cc: David Ahern <dsa@cumulusnetworks.com> Cc: Florian Westphal <fw@strlen.de> Cc: Martynas Pumputis <m@lambda.lt> Acked-by: David Ahern <dsa@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-30 18:49:48 +07:00
IFF_L3MDEV_RX_HANDLER = 1<<29,
failover: allow name change on IFF_UP slave interfaces When a netdev appears through hot plug then gets enslaved by a failover master that is already up and running, the slave will be opened right away after getting enslaved. Today there's a race that userspace (udev) may fail to rename the slave if the kernel (net_failover) opens the slave earlier than when the userspace rename happens. Unlike bond or team, the primary slave of failover can't be renamed by userspace ahead of time, since the kernel initiated auto-enslavement is unable to, or rather, is never meant to be synchronized with the rename request from userspace. As the failover slave interfaces are not designed to be operated directly by userspace apps: IP configuration, filter rules with regard to network traffic passing and etc., should all be done on master interface. In general, userspace apps only care about the name of master interface, while slave names are less important as long as admin users can see reliable names that may carry other information describing the netdev. For e.g., they can infer that "ens3nsby" is a standby slave of "ens3", while for a name like "eth0" they can't tell which master it belongs to. Historically the name of IFF_UP interface can't be changed because there might be admin script or management software that is already relying on such behavior and assumes that the slave name can't be changed once UP. But failover is special: with the in-kernel auto-enslavement mechanism, the userspace expectation for device enumeration and bring-up order is already broken. Previously initramfs and various userspace config tools were modified to bypass failover slaves because of auto-enslavement and duplicate MAC address. Similarly, in case that users care about seeing reliable slave name, the new type of failover slaves needs to be taken care of specifically in userspace anyway. It's less risky to lift up the rename restriction on failover slave which is already UP. Although it's possible this change may potentially break userspace component (most likely configuration scripts or management software) that assumes slave name can't be changed while UP, it's relatively a limited and controllable set among all userspace components, which can be fixed specifically to listen for the rename events on failover slaves. Userspace component interacting with slaves is expected to be changed to operate on failover master interface instead, as the failover slave is dynamic in nature which may come and go at any point. The goal is to make the role of failover slaves less relevant, and userspace components should only deal with failover master in the long run. Fixes: 30c8bd5aa8b2 ("net: Introduce generic failover module") Signed-off-by: Si-Wei Liu <si-wei.liu@oracle.com> Reviewed-by: Liran Alon <liran.alon@oracle.com> Acked-by: Sridhar Samudrala <sridhar.samudrala@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-09 06:45:27 +07:00
IFF_LIVE_RENAME_OK = 1<<30,
};
#define IFF_802_1Q_VLAN IFF_802_1Q_VLAN
#define IFF_EBRIDGE IFF_EBRIDGE
#define IFF_BONDING IFF_BONDING
#define IFF_ISATAP IFF_ISATAP
#define IFF_WAN_HDLC IFF_WAN_HDLC
#define IFF_XMIT_DST_RELEASE IFF_XMIT_DST_RELEASE
#define IFF_DONT_BRIDGE IFF_DONT_BRIDGE
#define IFF_DISABLE_NETPOLL IFF_DISABLE_NETPOLL
#define IFF_MACVLAN_PORT IFF_MACVLAN_PORT
#define IFF_BRIDGE_PORT IFF_BRIDGE_PORT
#define IFF_OVS_DATAPATH IFF_OVS_DATAPATH
#define IFF_TX_SKB_SHARING IFF_TX_SKB_SHARING
#define IFF_UNICAST_FLT IFF_UNICAST_FLT
#define IFF_TEAM_PORT IFF_TEAM_PORT
#define IFF_SUPP_NOFCS IFF_SUPP_NOFCS
#define IFF_LIVE_ADDR_CHANGE IFF_LIVE_ADDR_CHANGE
#define IFF_MACVLAN IFF_MACVLAN
#define IFF_XMIT_DST_RELEASE_PERM IFF_XMIT_DST_RELEASE_PERM
#define IFF_L3MDEV_MASTER IFF_L3MDEV_MASTER
#define IFF_NO_QUEUE IFF_NO_QUEUE
#define IFF_OPENVSWITCH IFF_OPENVSWITCH
#define IFF_L3MDEV_SLAVE IFF_L3MDEV_SLAVE
#define IFF_TEAM IFF_TEAM
ethtool: correctly ensure {GS}CHANNELS doesn't conflict with GS{RXFH} Ethernet drivers implementing both {GS}RXFH and {GS}CHANNELS ethtool ops incorrectly allow SCHANNELS when it would conflict with the settings from SRXFH. This occurs because it is not possible for drivers to understand whether their Rx flow indirection table has been configured or is in the default state. In addition, drivers currently behave in various ways when increasing the number of Rx channels. Some drivers will always destroy the Rx flow indirection table when this occurs, whether it has been set by the user or not. Other drivers will attempt to preserve the table even if the user has never modified it from the default driver settings. Neither of these situation is desirable because it leads to unexpected behavior or loss of user configuration. The correct behavior is to simply return -EINVAL when SCHANNELS would conflict with the current Rx flow table settings. However, it should only do so if the current settings were modified by the user. If we required that the new settings never conflict with the current (default) Rx flow settings, we would force users to first reduce their Rx flow settings and then reduce the number of Rx channels. This patch proposes a solution implemented in net/core/ethtool.c which ensures that all drivers behave correctly. It checks whether the RXFH table has been configured to non-default settings, and stores this information in a private netdev flag. When the number of channels is requested to change, it first ensures that the current Rx flow table is not going to assign flows to now disabled channels. Signed-off-by: Jacob Keller <jacob.e.keller@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-09 07:05:03 +07:00
#define IFF_RXFH_CONFIGURED IFF_RXFH_CONFIGURED
#define IFF_MACSEC IFF_MACSEC
#define IFF_NO_RX_HANDLER IFF_NO_RX_HANDLER
#define IFF_FAILOVER IFF_FAILOVER
#define IFF_FAILOVER_SLAVE IFF_FAILOVER_SLAVE
ipvlan, l3mdev: fix broken l3s mode wrt local routes While implementing ipvlan l3 and l3s mode for kubernetes CNI plugin, I ran into the issue that while l3 mode is working fine, l3s mode does not have any connectivity to kube-apiserver and hence all pods end up in Error state as well. The ipvlan master device sits on top of a bond device and hostns traffic to kube-apiserver (also running in hostns) is DNATed from 10.152.183.1:443 to 139.178.29.207:37573 where the latter is the address of the bond0. While in l3 mode, a curl to https://10.152.183.1:443 or to https://139.178.29.207:37573 works fine from hostns, neither of them do in case of l3s. In the latter only a curl to https://127.0.0.1:37573 appeared to work where for local addresses of bond0 I saw kernel suddenly starting to emit ARP requests to query HW address of bond0 which remained unanswered and neighbor entries in INCOMPLETE state. These ARP requests only happen while in l3s. Debugging this further, I found the issue is that l3s mode is piggy- backing on l3 master device, and in this case local routes are using l3mdev_master_dev_rcu(dev) instead of net->loopback_dev as per commit f5a0aab84b74 ("net: ipv4: dst for local input routes should use l3mdev if relevant") and 5f02ce24c269 ("net: l3mdev: Allow the l3mdev to be a loopback"). I found that reverting them back into using the net->loopback_dev fixed ipvlan l3s connectivity and got everything working for the CNI. Now judging from 4fbae7d83c98 ("ipvlan: Introduce l3s mode") and the l3mdev paper in [0] the only sole reason why ipvlan l3s is relying on l3 master device is to get the l3mdev_ip_rcv() receive hook for setting the dst entry of the input route without adding its own ipvlan specific hacks into the receive path, however, any l3 domain semantics beyond just that are breaking l3s operation. Note that ipvlan also has the ability to dynamically switch its internal operation from l3 to l3s for all ports via ipvlan_set_port_mode() at runtime. In any case, l3 vs l3s soley distinguishes itself by 'de-confusing' netfilter through switching skb->dev to ipvlan slave device late in NF_INET_LOCAL_IN before handing the skb to L4. Minimal fix taken here is to add a IFF_L3MDEV_RX_HANDLER flag which, if set from ipvlan setup, gets us only the wanted l3mdev_l3_rcv() hook without any additional l3mdev semantics on top. This should also have minimal impact since dev->priv_flags is already hot in cache. With this set, l3s mode is working fine and I also get things like masquerading pod traffic on the ipvlan master properly working. [0] https://netdevconf.org/1.2/papers/ahern-what-is-l3mdev-paper.pdf Fixes: f5a0aab84b74 ("net: ipv4: dst for local input routes should use l3mdev if relevant") Fixes: 5f02ce24c269 ("net: l3mdev: Allow the l3mdev to be a loopback") Fixes: 4fbae7d83c98 ("ipvlan: Introduce l3s mode") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Cc: Mahesh Bandewar <maheshb@google.com> Cc: David Ahern <dsa@cumulusnetworks.com> Cc: Florian Westphal <fw@strlen.de> Cc: Martynas Pumputis <m@lambda.lt> Acked-by: David Ahern <dsa@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-30 18:49:48 +07:00
#define IFF_L3MDEV_RX_HANDLER IFF_L3MDEV_RX_HANDLER
failover: allow name change on IFF_UP slave interfaces When a netdev appears through hot plug then gets enslaved by a failover master that is already up and running, the slave will be opened right away after getting enslaved. Today there's a race that userspace (udev) may fail to rename the slave if the kernel (net_failover) opens the slave earlier than when the userspace rename happens. Unlike bond or team, the primary slave of failover can't be renamed by userspace ahead of time, since the kernel initiated auto-enslavement is unable to, or rather, is never meant to be synchronized with the rename request from userspace. As the failover slave interfaces are not designed to be operated directly by userspace apps: IP configuration, filter rules with regard to network traffic passing and etc., should all be done on master interface. In general, userspace apps only care about the name of master interface, while slave names are less important as long as admin users can see reliable names that may carry other information describing the netdev. For e.g., they can infer that "ens3nsby" is a standby slave of "ens3", while for a name like "eth0" they can't tell which master it belongs to. Historically the name of IFF_UP interface can't be changed because there might be admin script or management software that is already relying on such behavior and assumes that the slave name can't be changed once UP. But failover is special: with the in-kernel auto-enslavement mechanism, the userspace expectation for device enumeration and bring-up order is already broken. Previously initramfs and various userspace config tools were modified to bypass failover slaves because of auto-enslavement and duplicate MAC address. Similarly, in case that users care about seeing reliable slave name, the new type of failover slaves needs to be taken care of specifically in userspace anyway. It's less risky to lift up the rename restriction on failover slave which is already UP. Although it's possible this change may potentially break userspace component (most likely configuration scripts or management software) that assumes slave name can't be changed while UP, it's relatively a limited and controllable set among all userspace components, which can be fixed specifically to listen for the rename events on failover slaves. Userspace component interacting with slaves is expected to be changed to operate on failover master interface instead, as the failover slave is dynamic in nature which may come and go at any point. The goal is to make the role of failover slaves less relevant, and userspace components should only deal with failover master in the long run. Fixes: 30c8bd5aa8b2 ("net: Introduce generic failover module") Signed-off-by: Si-Wei Liu <si-wei.liu@oracle.com> Reviewed-by: Liran Alon <liran.alon@oracle.com> Acked-by: Sridhar Samudrala <sridhar.samudrala@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-09 06:45:27 +07:00
#define IFF_LIVE_RENAME_OK IFF_LIVE_RENAME_OK
/**
* struct net_device - 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.
*
* @name: 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
* of the interface.
*
* @name_hlist: Device name hash chain, please keep it close to name[]
* @ifalias: SNMP alias
* @mem_end: Shared memory end
* @mem_start: Shared memory start
* @base_addr: Device I/O address
* @irq: Device IRQ number
*
* @state: Generic network queuing layer state, see netdev_state_t
* @dev_list: The global list of network devices
* @napi_list: List entry used for polling NAPI devices
* @unreg_list: List entry when we are unregistering the
* device; see the function unregister_netdev
* @close_list: List entry used when we are closing the device
* @ptype_all: Device-specific packet handlers for all protocols
* @ptype_specific: Device-specific, protocol-specific packet handlers
*
* @adj_list: Directly linked devices, like slaves for bonding
* @features: Currently active device features
* @hw_features: User-changeable features
*
* @wanted_features: User-requested features
* @vlan_features: Mask of features inheritable by VLAN devices
*
* @hw_enc_features: Mask of features inherited by encapsulating devices
* This field indicates what encapsulation
* offloads the hardware is capable of doing,
* and drivers will need to set them appropriately.
*
* @mpls_features: Mask of features inheritable by MPLS
*
* @ifindex: interface index
* @group: The group the device belongs to
*
* @stats: Statistics struct, which was left as a legacy, use
* rtnl_link_stats64 instead
*
* @rx_dropped: Dropped packets by core network,
* do not use this in drivers
* @tx_dropped: Dropped packets by core network,
* do not use this in drivers
* @rx_nohandler: nohandler dropped packets by core network on
* inactive devices, do not use this in drivers
* @carrier_up_count: Number of times the carrier has been up
* @carrier_down_count: Number of times the carrier has been down
*
* @wireless_handlers: List of functions to handle Wireless Extensions,
* instead of ioctl,
* see <net/iw_handler.h> for details.
* @wireless_data: Instance data managed by the core of wireless extensions
*
* @netdev_ops: Includes several pointers to callbacks,
* if one wants to override the ndo_*() functions
* @ethtool_ops: Management operations
* @ndisc_ops: Includes callbacks for different IPv6 neighbour
* discovery handling. Necessary for e.g. 6LoWPAN.
* @header_ops: Includes callbacks for creating,parsing,caching,etc
* of Layer 2 headers.
*
* @flags: Interface flags (a la BSD)
* @priv_flags: Like 'flags' but invisible to userspace,
* see if.h for the definitions
* @gflags: Global flags ( kept as legacy )
* @padded: How much padding added by alloc_netdev()
* @operstate: RFC2863 operstate
* @link_mode: Mapping policy to operstate
* @if_port: Selectable AUI, TP, ...
* @dma: DMA channel
* @mtu: Interface MTU value
* @min_mtu: Interface Minimum MTU value
* @max_mtu: Interface Maximum MTU value
* @type: Interface hardware type
* @hard_header_len: Maximum hardware header length.
* @min_header_len: Minimum hardware header length
*
* @needed_headroom: Extra headroom the hardware may need, but not in all
* cases can this be guaranteed
* @needed_tailroom: Extra tailroom the hardware may need, but not in all
* cases can this be guaranteed. Some cases also use
* LL_MAX_HEADER instead to allocate the skb
*
* interface address info:
*
* @perm_addr: Permanent hw address
* @addr_assign_type: Hw address assignment type
* @addr_len: Hardware address length
* @neigh_priv_len: Used in neigh_alloc()
* @dev_id: Used to differentiate devices that share
* the same link layer address
* @dev_port: Used to differentiate devices that share
* the same function
* @addr_list_lock: XXX: need comments on this one
* @uc_promisc: Counter that indicates promiscuous mode
* has been enabled due to the need to listen to
* additional unicast addresses in a device that
* does not implement ndo_set_rx_mode()
* @uc: unicast mac addresses
* @mc: multicast mac addresses
* @dev_addrs: list of device hw addresses
* @queues_kset: Group of all Kobjects in the Tx and RX queues
* @promiscuity: Number of times the NIC is told to work in
* promiscuous mode; if it becomes 0 the NIC will
* exit promiscuous mode
* @allmulti: Counter, enables or disables allmulticast mode
*
* @vlan_info: VLAN info
* @dsa_ptr: dsa specific data
* @tipc_ptr: TIPC specific data
* @atalk_ptr: AppleTalk link
* @ip_ptr: IPv4 specific data
* @dn_ptr: DECnet specific data
* @ip6_ptr: IPv6 specific data
* @ax25_ptr: AX.25 specific data
* @ieee80211_ptr: IEEE 802.11 specific data, assign before registering
*
* @dev_addr: Hw address (before bcast,
* because most packets are unicast)
*
* @_rx: Array of RX queues
* @num_rx_queues: Number of RX queues
* allocated at register_netdev() time
* @real_num_rx_queues: Number of RX queues currently active in device
*
* @rx_handler: handler for received packets
* @rx_handler_data: XXX: need comments on this one
* @miniq_ingress: ingress/clsact qdisc specific data for
* ingress processing
* @ingress_queue: XXX: need comments on this one
* @broadcast: hw bcast address
*
* @rx_cpu_rmap: CPU reverse-mapping for RX completion interrupts,
* indexed by RX queue number. Assigned by driver.
* This must only be set if the ndo_rx_flow_steer
* operation is defined
* @index_hlist: Device index hash chain
*
* @_tx: Array of TX queues
* @num_tx_queues: Number of TX queues allocated at alloc_netdev_mq() time
* @real_num_tx_queues: Number of TX queues currently active in device
* @qdisc: Root qdisc from userspace point of view
* @tx_queue_len: Max frames per queue allowed
* @tx_global_lock: XXX: need comments on this one
*
* @xps_maps: XXX: need comments on this one
* @miniq_egress: clsact qdisc specific data for
* egress processing
* @watchdog_timeo: Represents the timeout that is used by
* the watchdog (see dev_watchdog())
* @watchdog_timer: List of timers
*
* @pcpu_refcnt: Number of references to this device
* @todo_list: Delayed register/unregister
* @link_watch_list: XXX: need comments on this one
*
* @reg_state: Register/unregister state machine
* @dismantle: Device is going to be freed
* @rtnl_link_state: This enum represents the phases of creating
* a new link
*
net: Fix inconsistent teardown and release of private netdev state. Network devices can allocate reasources and private memory using netdev_ops->ndo_init(). However, the release of these resources can occur in one of two different places. Either netdev_ops->ndo_uninit() or netdev->destructor(). The decision of which operation frees the resources depends upon whether it is necessary for all netdev refs to be released before it is safe to perform the freeing. netdev_ops->ndo_uninit() presumably can occur right after the NETDEV_UNREGISTER notifier completes and the unicast and multicast address lists are flushed. netdev->destructor(), on the other hand, does not run until the netdev references all go away. Further complicating the situation is that netdev->destructor() almost universally does also a free_netdev(). This creates a problem for the logic in register_netdevice(). Because all callers of register_netdevice() manage the freeing of the netdev, and invoke free_netdev(dev) if register_netdevice() fails. If netdev_ops->ndo_init() succeeds, but something else fails inside of register_netdevice(), it does call ndo_ops->ndo_uninit(). But it is not able to invoke netdev->destructor(). This is because netdev->destructor() will do a free_netdev() and then the caller of register_netdevice() will do the same. However, this means that the resources that would normally be released by netdev->destructor() will not be. Over the years drivers have added local hacks to deal with this, by invoking their destructor parts by hand when register_netdevice() fails. Many drivers do not try to deal with this, and instead we have leaks. Let's close this hole by formalizing the distinction between what private things need to be freed up by netdev->destructor() and whether the driver needs unregister_netdevice() to perform the free_netdev(). netdev->priv_destructor() performs all actions to free up the private resources that used to be freed by netdev->destructor(), except for free_netdev(). netdev->needs_free_netdev is a boolean that indicates whether free_netdev() should be done at the end of unregister_netdevice(). Now, register_netdevice() can sanely release all resources after ndo_ops->ndo_init() succeeds, by invoking both ndo_ops->ndo_uninit() and netdev->priv_destructor(). And at the end of unregister_netdevice(), we invoke netdev->priv_destructor() and optionally call free_netdev(). Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-08 23:52:56 +07:00
* @needs_free_netdev: Should unregister perform free_netdev?
* @priv_destructor: Called from unregister
* @npinfo: XXX: need comments on this one
* @nd_net: Network namespace this network device is inside
*
* @ml_priv: Mid-layer private
* @lstats: Loopback statistics
* @tstats: Tunnel statistics
* @dstats: Dummy statistics
* @vstats: Virtual ethernet statistics
*
* @garp_port: GARP
* @mrp_port: MRP
*
* @dev: Class/net/name entry
* @sysfs_groups: Space for optional device, statistics and wireless
* sysfs groups
*
* @sysfs_rx_queue_group: Space for optional per-rx queue attributes
* @rtnl_link_ops: Rtnl_link_ops
*
* @gso_max_size: Maximum size of generic segmentation offload
* @gso_max_segs: Maximum number of segments that can be passed to the
* NIC for GSO
*
* @dcbnl_ops: Data Center Bridging netlink ops
* @num_tc: Number of traffic classes in the net device
* @tc_to_txq: XXX: need comments on this one
* @prio_tc_map: XXX: need comments on this one
*
* @fcoe_ddp_xid: Max exchange id for FCoE LRO by ddp
*
* @priomap: XXX: need comments on this one
* @phydev: Physical device may attach itself
* for hardware timestamping
* @sfp_bus: attached &struct sfp_bus structure.
*
* @qdisc_tx_busylock: lockdep class annotating Qdisc->busylock spinlock
* @qdisc_running_key: lockdep class annotating Qdisc->running seqcount
*
* @proto_down: protocol port state information can be sent to the
* switch driver and used to set the phys state of the
* switch port.
*
* @wol_enabled: Wake-on-LAN is enabled
*
* FIXME: cleanup struct net_device such that network protocol info
* moves out.
*/
struct net_device {
char name[IFNAMSIZ];
struct hlist_node name_hlist;
struct dev_ifalias __rcu *ifalias;
/*
* I/O specific fields
* FIXME: Merge these and struct ifmap into one
*/
unsigned long mem_end;
unsigned long mem_start;
unsigned long base_addr;
int irq;
/*
* Some hardware also needs these fields (state,dev_list,
* napi_list,unreg_list,close_list) but they are not
* part of the usual set specified in Space.c.
*/
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 06:41:36 +07:00
struct list_head napi_list;
struct list_head unreg_list;
struct list_head close_list;
struct list_head ptype_all;
struct list_head ptype_specific;
net: add adj_list to save only neighbours Currently, we distinguish neighbours (first-level linked devices) from non-neighbours by the neighbour bool in the netdev_adjacent. This could be quite time-consuming in case we would like to traverse *only* through neighbours - cause we'd have to traverse through all devices and check for this flag, and in a (quite common) scenario where we have lots of vlans on top of bridge, which is on top of a bond - the bonding would have to go through all those vlans to get its upper neighbour linked devices. This situation is really unpleasant, cause there are already a lot of cases when a device with slaves needs to go through them in hot path. To fix this, introduce a new upper/lower device lists structure - adj_list, which contains only the neighbours. It works always in pair with the all_adj_list structure (renamed from upper/lower_dev_list), i.e. both of them contain the same links, only that all_adj_list contains also non-neighbour device links. It's really a small change visible, currently, only for __netdev_adjacent_dev_insert/remove(), and doesn't change the main linked logic at all. Also, add some comments a fix a name collision in netdev_for_each_upper_dev_rcu() and rework the naming by the following rules: netdev_(all_)(upper|lower)_* If "all_" is present, then we work with the whole list of upper/lower devices, otherwise - only with direct neighbours. Uninline functions - to get better stack traces. CC: "David S. Miller" <davem@davemloft.net> CC: Eric Dumazet <edumazet@google.com> CC: Jiri Pirko <jiri@resnulli.us> CC: Alexander Duyck <alexander.h.duyck@intel.com> CC: Cong Wang <amwang@redhat.com> Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-09-25 14:20:07 +07:00
struct {
struct list_head upper;
struct list_head lower;
} adj_list;
netdev_features_t features;
netdev_features_t hw_features;
netdev_features_t wanted_features;
netdev_features_t vlan_features;
netdev_features_t hw_enc_features;
MPLS: Add limited GSO support In the case where a non-MPLS packet is received and an MPLS stack is added it may well be the case that the original skb is GSO but the NIC used for transmit does not support GSO of MPLS packets. The aim of this code is to provide GSO in software for MPLS packets whose skbs are GSO. SKB Usage: When an implementation adds an MPLS stack to a non-MPLS packet it should do the following to skb metadata: * Set skb->inner_protocol to the old non-MPLS ethertype of the packet. skb->inner_protocol is added by this patch. * Set skb->protocol to the new MPLS ethertype of the packet. * Set skb->network_header to correspond to the end of the L3 header, including the MPLS label stack. I have posted a patch, "[PATCH v3.29] datapath: Add basic MPLS support to kernel" which adds MPLS support to the kernel datapath of Open vSwtich. That patch sets the above requirements in datapath/actions.c:push_mpls() and was used to exercise this code. The datapath patch is against the Open vSwtich tree but it is intended that it be added to the Open vSwtich code present in the mainline Linux kernel at some point. Features: I believe that the approach that I have taken is at least partially consistent with the handling of other protocols. Jesse, I understand that you have some ideas here. I am more than happy to change my implementation. This patch adds dev->mpls_features which may be used by devices to advertise features supported for MPLS packets. A new NETIF_F_MPLS_GSO feature is added for devices which support hardware MPLS GSO offload. Currently no devices support this and MPLS GSO always falls back to software. Alternate Implementation: One possible alternate implementation is to teach netif_skb_features() and skb_network_protocol() about MPLS, in a similar way to their understanding of VLANs. I believe this would avoid the need for net/mpls/mpls_gso.c and in particular the calls to __skb_push() and __skb_push() in mpls_gso_segment(). I have decided on the implementation in this patch as it should not introduce any overhead in the case where mpls_gso is not compiled into the kernel or inserted as a module. MPLS GSO suggested by Jesse Gross. Based in part on "v4 GRE: Add TCP segmentation offload for GRE" by Pravin B Shelar. Cc: Jesse Gross <jesse@nicira.com> Cc: Pravin B Shelar <pshelar@nicira.com> Signed-off-by: Simon Horman <horms@verge.net.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-05-24 04:02:52 +07:00
netdev_features_t mpls_features;
netdev_features_t gso_partial_features;
int ifindex;
int group;
struct net_device_stats stats;
atomic_long_t rx_dropped;
atomic_long_t tx_dropped;
atomic_long_t rx_nohandler;
/* Stats to monitor link on/off, flapping */
atomic_t carrier_up_count;
atomic_t carrier_down_count;
#ifdef CONFIG_WIRELESS_EXT
const struct iw_handler_def *wireless_handlers;
struct iw_public_data *wireless_data;
#endif
const struct net_device_ops *netdev_ops;
const struct ethtool_ops *ethtool_ops;
#ifdef CONFIG_NET_L3_MASTER_DEV
const struct l3mdev_ops *l3mdev_ops;
#endif
#if IS_ENABLED(CONFIG_IPV6)
const struct ndisc_ops *ndisc_ops;
#endif
#ifdef CONFIG_XFRM_OFFLOAD
const struct xfrmdev_ops *xfrmdev_ops;
#endif
#if IS_ENABLED(CONFIG_TLS_DEVICE)
const struct tlsdev_ops *tlsdev_ops;
#endif
const struct header_ops *header_ops;
unsigned int flags;
unsigned int priv_flags;
unsigned short gflags;
unsigned short padded;
unsigned char operstate;
unsigned char link_mode;
unsigned char if_port;
unsigned char dma;
unsigned int mtu;
unsigned int min_mtu;
unsigned int max_mtu;
unsigned short type;
unsigned short hard_header_len;
unsigned char min_header_len;
unsigned short needed_headroom;
unsigned short needed_tailroom;
/* Interface address info. */
unsigned char perm_addr[MAX_ADDR_LEN];
unsigned char addr_assign_type;
unsigned char addr_len;
unsigned short neigh_priv_len;
unsigned short dev_id;
unsigned short dev_port;
spinlock_t addr_list_lock;
unsigned char name_assign_type;
bool uc_promisc;
struct netdev_hw_addr_list uc;
struct netdev_hw_addr_list mc;
struct netdev_hw_addr_list dev_addrs;
#ifdef CONFIG_SYSFS
struct kset *queues_kset;
#endif
unsigned int promiscuity;
unsigned int allmulti;
/* Protocol-specific pointers */
#if IS_ENABLED(CONFIG_VLAN_8021Q)
struct vlan_info __rcu *vlan_info;
#endif
#if IS_ENABLED(CONFIG_NET_DSA)
struct dsa_port *dsa_ptr;
#endif
#if IS_ENABLED(CONFIG_TIPC)
struct tipc_bearer __rcu *tipc_ptr;
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 20:44:02 +07:00
#endif
#if IS_ENABLED(CONFIG_IRDA) || IS_ENABLED(CONFIG_ATALK)
void *atalk_ptr;
#endif
struct in_device __rcu *ip_ptr;
#if IS_ENABLED(CONFIG_DECNET)
struct dn_dev __rcu *dn_ptr;
#endif
struct inet6_dev __rcu *ip6_ptr;
#if IS_ENABLED(CONFIG_AX25)
void *ax25_ptr;
#endif
struct wireless_dev *ieee80211_ptr;
struct wpan_dev *ieee802154_ptr;
#if IS_ENABLED(CONFIG_MPLS_ROUTING)
struct mpls_dev __rcu *mpls_ptr;
#endif
/*
* Cache lines mostly used on receive path (including eth_type_trans())
*/
/* Interface address info used in eth_type_trans() */
unsigned char *dev_addr;
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 15:03:29 +07:00
struct netdev_rx_queue *_rx;
unsigned int num_rx_queues;
unsigned int real_num_rx_queues;
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 15:03:29 +07:00
struct bpf_prog __rcu *xdp_prog;
net: gro: add a per device gro flush timer Tuning coalescing parameters on NIC can be really hard. Servers can handle both bulk and RPC like traffic, with conflicting goals : bulk flows want as big GRO packets as possible, RPC want minimal latencies. To reach big GRO packets on 10Gbe NIC, one can use : ethtool -C eth0 rx-usecs 4 rx-frames 44 But this penalizes rpc sessions, with an increase of latencies, up to 50% in some cases, as NICs generally do not force an interrupt when a packet with TCP Push flag is received. Some NICs do not have an absolute timer, only a timer rearmed for every incoming packet. This patch uses a different strategy : Let GRO stack decides what do do, based on traffic pattern. Packets with Push flag wont be delayed. Packets without Push flag might be held in GRO engine, if we keep receiving data. This new mechanism is off by default, and shall be enabled by setting /sys/class/net/ethX/gro_flush_timeout to a value in nanosecond. To fully enable this mechanism, drivers should use napi_complete_done() instead of napi_complete(). Tested: Ran 200 netperf TCP_STREAM from A to B (10Gbe mlx4 link, 8 RX queues) Without this feature, we send back about 305,000 ACK per second. GRO aggregation ratio is low (811/305 = 2.65 segments per GRO packet) Setting a timer of 2000 nsec is enough to increase GRO packet sizes and reduce number of ACK packets. (811/19.2 = 42) Receiver performs less calls to upper stacks, less wakes up. This also reduces cpu usage on the sender, as it receives less ACK packets. Note that reducing number of wakes up increases cpu efficiency, but can decrease QPS, as applications wont have the chance to warmup cpu caches doing a partial read of RPC requests/answers if they fit in one skb. B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811269.80 305732.30 1199462.57 19705.72 0.00 0.00 0.50 B:~# echo 2000 >/sys/class/net/eth0/gro_flush_timeout B:~# sar -n DEV 1 10 | grep eth0 | tail -1 Average: eth0 811577.30 19230.80 1199916.51 1239.80 0.00 0.00 0.50 Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-07 12:09:44 +07:00
unsigned long gro_flush_timeout;
rx_handler_func_t __rcu *rx_handler;
void __rcu *rx_handler_data;
#ifdef CONFIG_NET_CLS_ACT
struct mini_Qdisc __rcu *miniq_ingress;
net: sched: further simplify handle_ing Ingress qdisc has no other purpose than calling into tc_classify() that executes attached classifier(s) and action(s). It has a 1:1 relationship to dev->ingress_queue. After having commit 087c1a601ad7 ("net: sched: run ingress qdisc without locks") removed the central ingress lock, one major contention point is gone. The extra indirection layers however, are not necessary for calling into ingress qdisc. pktgen calling locally into netif_receive_skb() with a dummy u32, single CPU result on a Supermicro X10SLM-F, Xeon E3-1240: before ~21,1 Mpps, after patch ~22,9 Mpps. We can redirect the private classifier list to the netdev directly, without changing any classifier API bits (!) and execute on that from handle_ing() side. The __QDISC_STATE_DEACTIVATE test can be removed, ingress qdisc doesn't have a queue and thus dev_deactivate_queue() is also not applicable, ingress_cl_list provides similar behaviour. In other words, ingress qdisc acts like TCQ_F_BUILTIN qdisc. One next possible step is the removal of the dev's ingress (dummy) netdev_queue, and to only have the list member in the netdevice itself. Note, the filter chain is RCU protected and individual filter elements are being kfree'd by sched subsystem after RCU grace period. RCU read lock is being held by __netif_receive_skb_core(). Joint work with Alexei Starovoitov. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: Alexei Starovoitov <ast@plumgrid.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-05-10 03:51:32 +07:00
#endif
struct netdev_queue __rcu *ingress_queue;
netfilter: add netfilter ingress hook after handle_ing() under unique static key This patch adds the Netfilter ingress hook just after the existing tc ingress hook, that seems to be the consensus solution for this. Note that the Netfilter hook resides under the global static key that enables ingress filtering. Nonetheless, Netfilter still also has its own static key for minimal impact on the existing handle_ing(). * Without this patch: Result: OK: 6216490(c6216338+d152) usec, 100000000 (60byte,0frags) 16086246pps 7721Mb/sec (7721398080bps) errors: 100000000 42.46% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core 25.92% kpktgend_0 [kernel.kallsyms] [k] kfree_skb 7.81% kpktgend_0 [pktgen] [k] pktgen_thread_worker 5.62% kpktgend_0 [kernel.kallsyms] [k] ip_rcv 2.70% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal 2.34% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk 1.44% kpktgend_0 [kernel.kallsyms] [k] __build_skb * With this patch: Result: OK: 6214833(c6214731+d101) usec, 100000000 (60byte,0frags) 16090536pps 7723Mb/sec (7723457280bps) errors: 100000000 41.23% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core 26.57% kpktgend_0 [kernel.kallsyms] [k] kfree_skb 7.72% kpktgend_0 [pktgen] [k] pktgen_thread_worker 5.55% kpktgend_0 [kernel.kallsyms] [k] ip_rcv 2.78% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal 2.06% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk 1.43% kpktgend_0 [kernel.kallsyms] [k] __build_skb * Without this patch + tc ingress: tc filter add dev eth4 parent ffff: protocol ip prio 1 \ u32 match ip dst 4.3.2.1/32 Result: OK: 9269001(c9268821+d179) usec, 100000000 (60byte,0frags) 10788648pps 5178Mb/sec (5178551040bps) errors: 100000000 40.99% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core 17.50% kpktgend_0 [kernel.kallsyms] [k] kfree_skb 11.77% kpktgend_0 [cls_u32] [k] u32_classify 5.62% kpktgend_0 [kernel.kallsyms] [k] tc_classify_compat 5.18% kpktgend_0 [pktgen] [k] pktgen_thread_worker 3.23% kpktgend_0 [kernel.kallsyms] [k] tc_classify 2.97% kpktgend_0 [kernel.kallsyms] [k] ip_rcv 1.83% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal 1.50% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk 0.99% kpktgend_0 [kernel.kallsyms] [k] __build_skb * With this patch + tc ingress: tc filter add dev eth4 parent ffff: protocol ip prio 1 \ u32 match ip dst 4.3.2.1/32 Result: OK: 9308218(c9308091+d126) usec, 100000000 (60byte,0frags) 10743194pps 5156Mb/sec (5156733120bps) errors: 100000000 42.01% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core 17.78% kpktgend_0 [kernel.kallsyms] [k] kfree_skb 11.70% kpktgend_0 [cls_u32] [k] u32_classify 5.46% kpktgend_0 [kernel.kallsyms] [k] tc_classify_compat 5.16% kpktgend_0 [pktgen] [k] pktgen_thread_worker 2.98% kpktgend_0 [kernel.kallsyms] [k] ip_rcv 2.84% kpktgend_0 [kernel.kallsyms] [k] tc_classify 1.96% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal 1.57% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk Note that the results are very similar before and after. I can see gcc gets the code under the ingress static key out of the hot path. Then, on that cold branch, it generates the code to accomodate the netfilter ingress static key. My explanation for this is that this reduces the pressure on the instruction cache for non-users as the new code is out of the hot path, and it comes with minimal impact for tc ingress users. Using gcc version 4.8.4 on: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 8 [...] L1d cache: 16K L1i cache: 64K L2 cache: 2048K L3 cache: 8192K Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Acked-by: Alexei Starovoitov <ast@plumgrid.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-05-13 23:19:38 +07:00
#ifdef CONFIG_NETFILTER_INGRESS
struct nf_hook_entries __rcu *nf_hooks_ingress;
netfilter: add netfilter ingress hook after handle_ing() under unique static key This patch adds the Netfilter ingress hook just after the existing tc ingress hook, that seems to be the consensus solution for this. Note that the Netfilter hook resides under the global static key that enables ingress filtering. Nonetheless, Netfilter still also has its own static key for minimal impact on the existing handle_ing(). * Without this patch: Result: OK: 6216490(c6216338+d152) usec, 100000000 (60byte,0frags) 16086246pps 7721Mb/sec (7721398080bps) errors: 100000000 42.46% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core 25.92% kpktgend_0 [kernel.kallsyms] [k] kfree_skb 7.81% kpktgend_0 [pktgen] [k] pktgen_thread_worker 5.62% kpktgend_0 [kernel.kallsyms] [k] ip_rcv 2.70% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal 2.34% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk 1.44% kpktgend_0 [kernel.kallsyms] [k] __build_skb * With this patch: Result: OK: 6214833(c6214731+d101) usec, 100000000 (60byte,0frags) 16090536pps 7723Mb/sec (7723457280bps) errors: 100000000 41.23% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core 26.57% kpktgend_0 [kernel.kallsyms] [k] kfree_skb 7.72% kpktgend_0 [pktgen] [k] pktgen_thread_worker 5.55% kpktgend_0 [kernel.kallsyms] [k] ip_rcv 2.78% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal 2.06% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk 1.43% kpktgend_0 [kernel.kallsyms] [k] __build_skb * Without this patch + tc ingress: tc filter add dev eth4 parent ffff: protocol ip prio 1 \ u32 match ip dst 4.3.2.1/32 Result: OK: 9269001(c9268821+d179) usec, 100000000 (60byte,0frags) 10788648pps 5178Mb/sec (5178551040bps) errors: 100000000 40.99% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core 17.50% kpktgend_0 [kernel.kallsyms] [k] kfree_skb 11.77% kpktgend_0 [cls_u32] [k] u32_classify 5.62% kpktgend_0 [kernel.kallsyms] [k] tc_classify_compat 5.18% kpktgend_0 [pktgen] [k] pktgen_thread_worker 3.23% kpktgend_0 [kernel.kallsyms] [k] tc_classify 2.97% kpktgend_0 [kernel.kallsyms] [k] ip_rcv 1.83% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal 1.50% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk 0.99% kpktgend_0 [kernel.kallsyms] [k] __build_skb * With this patch + tc ingress: tc filter add dev eth4 parent ffff: protocol ip prio 1 \ u32 match ip dst 4.3.2.1/32 Result: OK: 9308218(c9308091+d126) usec, 100000000 (60byte,0frags) 10743194pps 5156Mb/sec (5156733120bps) errors: 100000000 42.01% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core 17.78% kpktgend_0 [kernel.kallsyms] [k] kfree_skb 11.70% kpktgend_0 [cls_u32] [k] u32_classify 5.46% kpktgend_0 [kernel.kallsyms] [k] tc_classify_compat 5.16% kpktgend_0 [pktgen] [k] pktgen_thread_worker 2.98% kpktgend_0 [kernel.kallsyms] [k] ip_rcv 2.84% kpktgend_0 [kernel.kallsyms] [k] tc_classify 1.96% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal 1.57% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk Note that the results are very similar before and after. I can see gcc gets the code under the ingress static key out of the hot path. Then, on that cold branch, it generates the code to accomodate the netfilter ingress static key. My explanation for this is that this reduces the pressure on the instruction cache for non-users as the new code is out of the hot path, and it comes with minimal impact for tc ingress users. Using gcc version 4.8.4 on: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 8 [...] L1d cache: 16K L1i cache: 64K L2 cache: 2048K L3 cache: 8192K Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Acked-by: Alexei Starovoitov <ast@plumgrid.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-05-13 23:19:38 +07:00
#endif
net: sched: further simplify handle_ing Ingress qdisc has no other purpose than calling into tc_classify() that executes attached classifier(s) and action(s). It has a 1:1 relationship to dev->ingress_queue. After having commit 087c1a601ad7 ("net: sched: run ingress qdisc without locks") removed the central ingress lock, one major contention point is gone. The extra indirection layers however, are not necessary for calling into ingress qdisc. pktgen calling locally into netif_receive_skb() with a dummy u32, single CPU result on a Supermicro X10SLM-F, Xeon E3-1240: before ~21,1 Mpps, after patch ~22,9 Mpps. We can redirect the private classifier list to the netdev directly, without changing any classifier API bits (!) and execute on that from handle_ing() side. The __QDISC_STATE_DEACTIVATE test can be removed, ingress qdisc doesn't have a queue and thus dev_deactivate_queue() is also not applicable, ingress_cl_list provides similar behaviour. In other words, ingress qdisc acts like TCQ_F_BUILTIN qdisc. One next possible step is the removal of the dev's ingress (dummy) netdev_queue, and to only have the list member in the netdevice itself. Note, the filter chain is RCU protected and individual filter elements are being kfree'd by sched subsystem after RCU grace period. RCU read lock is being held by __netif_receive_skb_core(). Joint work with Alexei Starovoitov. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: Alexei Starovoitov <ast@plumgrid.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-05-10 03:51:32 +07:00
unsigned char broadcast[MAX_ADDR_LEN];
#ifdef CONFIG_RFS_ACCEL
struct cpu_rmap *rx_cpu_rmap;
#endif
struct hlist_node index_hlist;
/*
* Cache lines mostly used on transmit path
*/
struct netdev_queue *_tx ____cacheline_aligned_in_smp;
unsigned int num_tx_queues;
unsigned int real_num_tx_queues;
struct Qdisc *qdisc;
#ifdef CONFIG_NET_SCHED
DECLARE_HASHTABLE (qdisc_hash, 4);
#endif
unsigned int tx_queue_len;
spinlock_t tx_global_lock;
int watchdog_timeo;
#ifdef CONFIG_XPS
struct xps_dev_maps __rcu *xps_cpus_map;
struct xps_dev_maps __rcu *xps_rxqs_map;
#endif
net, sched: add clsact qdisc This work adds a generalization of the ingress qdisc as a qdisc holding only classifiers. The clsact qdisc works on ingress, but also on egress. In both cases, it's execution happens without taking the qdisc lock, and the main difference for the egress part compared to prior version of [1] is that this can be applied with _any_ underlying real egress qdisc (also classless ones). Besides solving the use-case of [1], that is, allowing for more programmability on assigning skb->priority for the mqprio case that is supported by most popular 10G+ NICs, it also opens up a lot more flexibility for other tc applications. The main work on classification can already be done at clsact egress time if the use-case allows and state stored for later retrieval f.e. again in skb->priority with major/minors (which is checked by most classful qdiscs before consulting tc_classify()) and/or in other skb fields like skb->tc_index for some light-weight post-processing to get to the eventual classid in case of a classful qdisc. Another use case is that the clsact egress part allows to have a central egress counterpart to the ingress classifiers, so that classifiers can easily share state (e.g. in cls_bpf via eBPF maps) for ingress and egress. Currently, default setups like mq + pfifo_fast would require for this to use, for example, prio qdisc instead (to get a tc_classify() run) and to duplicate the egress classifier for each queue. With clsact, it allows for leaving the setup as is, it can additionally assign skb->priority to put the skb in one of pfifo_fast's bands and it can share state with maps. Moreover, we can access the skb's dst entry (f.e. to retrieve tclassid) w/o the need to perform a skb_dst_force() to hold on to it any longer. In lwt case, we can also use this facility to setup dst metadata via cls_bpf (bpf_skb_set_tunnel_key()) without needing a real egress qdisc just for that (case of IFF_NO_QUEUE devices, for example). The realization can be done without any changes to the scheduler core framework. All it takes is that we have two a-priori defined minors/child classes, where we can mux between ingress and egress classifier list (dev->ingress_cl_list and dev->egress_cl_list, latter stored close to dev->_tx to avoid extra cacheline miss for moderate loads). The egress part is a bit similar modelled to handle_ing() and patched to a noop in case the functionality is not used. Both handlers are now called sch_handle_ingress() and sch_handle_egress(), code sharing among the two doesn't seem practical as there are various minor differences in both paths, so that making them conditional in a single handler would rather slow things down. Full compatibility to ingress qdisc is provided as well. Since both piggyback on TC_H_CLSACT, only one of them (ingress/clsact) can exist per netdevice, and thus ingress qdisc specific behaviour can be retained for user space. This means, either a user does 'tc qdisc add dev foo ingress' and configures ingress qdisc as usual, or the 'tc qdisc add dev foo clsact' alternative, where both, ingress and egress classifier can be configured as in the below example. ingress qdisc supports attaching classifier to any minor number whereas clsact has two fixed minors for muxing between the lists, therefore to not break user space setups, they are better done as two separate qdiscs. I decided to extend the sch_ingress module with clsact functionality so that commonly used code can be reused, the module is being aliased with sch_clsact so that it can be auto-loaded properly. Alternative would have been to add a flag when initializing ingress to alter its behaviour plus aliasing to a different name (as it's more than just ingress). However, the first would end up, based on the flag, choosing the new/old behaviour by calling different function implementations to handle each anyway, the latter would require to register ingress qdisc once again under different alias. So, this really begs to provide a minimal, cleaner approach to have Qdisc_ops and Qdisc_class_ops by its own that share callbacks used by both. Example, adding qdisc: # tc qdisc add dev foo clsact # tc qdisc show dev foo qdisc mq 0: root qdisc pfifo_fast 0: parent :1 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc pfifo_fast 0: parent :2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc pfifo_fast 0: parent :3 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc pfifo_fast 0: parent :4 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc clsact ffff: parent ffff:fff1 Adding filters (deleting, etc works analogous by specifying ingress/egress): # tc filter add dev foo ingress bpf da obj bar.o sec ingress # tc filter add dev foo egress bpf da obj bar.o sec egress # tc filter show dev foo ingress filter protocol all pref 49152 bpf filter protocol all pref 49152 bpf handle 0x1 bar.o:[ingress] direct-action # tc filter show dev foo egress filter protocol all pref 49152 bpf filter protocol all pref 49152 bpf handle 0x1 bar.o:[egress] direct-action A 'tc filter show dev foo' or 'tc filter show dev foo parent ffff:' will show an empty list for clsact. Either using the parent names (ingress/egress) or specifying the full major/minor will then show the related filter lists. Prior work on a mqprio prequeue() facility [1] was done mainly by John Fastabend. [1] http://patchwork.ozlabs.org/patch/512949/ Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-01-08 04:29:47 +07:00
#ifdef CONFIG_NET_CLS_ACT
struct mini_Qdisc __rcu *miniq_egress;
net, sched: add clsact qdisc This work adds a generalization of the ingress qdisc as a qdisc holding only classifiers. The clsact qdisc works on ingress, but also on egress. In both cases, it's execution happens without taking the qdisc lock, and the main difference for the egress part compared to prior version of [1] is that this can be applied with _any_ underlying real egress qdisc (also classless ones). Besides solving the use-case of [1], that is, allowing for more programmability on assigning skb->priority for the mqprio case that is supported by most popular 10G+ NICs, it also opens up a lot more flexibility for other tc applications. The main work on classification can already be done at clsact egress time if the use-case allows and state stored for later retrieval f.e. again in skb->priority with major/minors (which is checked by most classful qdiscs before consulting tc_classify()) and/or in other skb fields like skb->tc_index for some light-weight post-processing to get to the eventual classid in case of a classful qdisc. Another use case is that the clsact egress part allows to have a central egress counterpart to the ingress classifiers, so that classifiers can easily share state (e.g. in cls_bpf via eBPF maps) for ingress and egress. Currently, default setups like mq + pfifo_fast would require for this to use, for example, prio qdisc instead (to get a tc_classify() run) and to duplicate the egress classifier for each queue. With clsact, it allows for leaving the setup as is, it can additionally assign skb->priority to put the skb in one of pfifo_fast's bands and it can share state with maps. Moreover, we can access the skb's dst entry (f.e. to retrieve tclassid) w/o the need to perform a skb_dst_force() to hold on to it any longer. In lwt case, we can also use this facility to setup dst metadata via cls_bpf (bpf_skb_set_tunnel_key()) without needing a real egress qdisc just for that (case of IFF_NO_QUEUE devices, for example). The realization can be done without any changes to the scheduler core framework. All it takes is that we have two a-priori defined minors/child classes, where we can mux between ingress and egress classifier list (dev->ingress_cl_list and dev->egress_cl_list, latter stored close to dev->_tx to avoid extra cacheline miss for moderate loads). The egress part is a bit similar modelled to handle_ing() and patched to a noop in case the functionality is not used. Both handlers are now called sch_handle_ingress() and sch_handle_egress(), code sharing among the two doesn't seem practical as there are various minor differences in both paths, so that making them conditional in a single handler would rather slow things down. Full compatibility to ingress qdisc is provided as well. Since both piggyback on TC_H_CLSACT, only one of them (ingress/clsact) can exist per netdevice, and thus ingress qdisc specific behaviour can be retained for user space. This means, either a user does 'tc qdisc add dev foo ingress' and configures ingress qdisc as usual, or the 'tc qdisc add dev foo clsact' alternative, where both, ingress and egress classifier can be configured as in the below example. ingress qdisc supports attaching classifier to any minor number whereas clsact has two fixed minors for muxing between the lists, therefore to not break user space setups, they are better done as two separate qdiscs. I decided to extend the sch_ingress module with clsact functionality so that commonly used code can be reused, the module is being aliased with sch_clsact so that it can be auto-loaded properly. Alternative would have been to add a flag when initializing ingress to alter its behaviour plus aliasing to a different name (as it's more than just ingress). However, the first would end up, based on the flag, choosing the new/old behaviour by calling different function implementations to handle each anyway, the latter would require to register ingress qdisc once again under different alias. So, this really begs to provide a minimal, cleaner approach to have Qdisc_ops and Qdisc_class_ops by its own that share callbacks used by both. Example, adding qdisc: # tc qdisc add dev foo clsact # tc qdisc show dev foo qdisc mq 0: root qdisc pfifo_fast 0: parent :1 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc pfifo_fast 0: parent :2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc pfifo_fast 0: parent :3 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc pfifo_fast 0: parent :4 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc clsact ffff: parent ffff:fff1 Adding filters (deleting, etc works analogous by specifying ingress/egress): # tc filter add dev foo ingress bpf da obj bar.o sec ingress # tc filter add dev foo egress bpf da obj bar.o sec egress # tc filter show dev foo ingress filter protocol all pref 49152 bpf filter protocol all pref 49152 bpf handle 0x1 bar.o:[ingress] direct-action # tc filter show dev foo egress filter protocol all pref 49152 bpf filter protocol all pref 49152 bpf handle 0x1 bar.o:[egress] direct-action A 'tc filter show dev foo' or 'tc filter show dev foo parent ffff:' will show an empty list for clsact. Either using the parent names (ingress/egress) or specifying the full major/minor will then show the related filter lists. Prior work on a mqprio prequeue() facility [1] was done mainly by John Fastabend. [1] http://patchwork.ozlabs.org/patch/512949/ Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-01-08 04:29:47 +07:00
#endif
/* These may be needed for future network-power-down code. */
struct timer_list watchdog_timer;
int __percpu *pcpu_refcnt;
struct list_head todo_list;
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 12:59:21 +07:00
struct list_head link_watch_list;
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:8;
bool dismantle;
enum {
RTNL_LINK_INITIALIZED,
RTNL_LINK_INITIALIZING,
} rtnl_link_state:16;
net: Fix inconsistent teardown and release of private netdev state. Network devices can allocate reasources and private memory using netdev_ops->ndo_init(). However, the release of these resources can occur in one of two different places. Either netdev_ops->ndo_uninit() or netdev->destructor(). The decision of which operation frees the resources depends upon whether it is necessary for all netdev refs to be released before it is safe to perform the freeing. netdev_ops->ndo_uninit() presumably can occur right after the NETDEV_UNREGISTER notifier completes and the unicast and multicast address lists are flushed. netdev->destructor(), on the other hand, does not run until the netdev references all go away. Further complicating the situation is that netdev->destructor() almost universally does also a free_netdev(). This creates a problem for the logic in register_netdevice(). Because all callers of register_netdevice() manage the freeing of the netdev, and invoke free_netdev(dev) if register_netdevice() fails. If netdev_ops->ndo_init() succeeds, but something else fails inside of register_netdevice(), it does call ndo_ops->ndo_uninit(). But it is not able to invoke netdev->destructor(). This is because netdev->destructor() will do a free_netdev() and then the caller of register_netdevice() will do the same. However, this means that the resources that would normally be released by netdev->destructor() will not be. Over the years drivers have added local hacks to deal with this, by invoking their destructor parts by hand when register_netdevice() fails. Many drivers do not try to deal with this, and instead we have leaks. Let's close this hole by formalizing the distinction between what private things need to be freed up by netdev->destructor() and whether the driver needs unregister_netdevice() to perform the free_netdev(). netdev->priv_destructor() performs all actions to free up the private resources that used to be freed by netdev->destructor(), except for free_netdev(). netdev->needs_free_netdev is a boolean that indicates whether free_netdev() should be done at the end of unregister_netdevice(). Now, register_netdevice() can sanely release all resources after ndo_ops->ndo_init() succeeds, by invoking both ndo_ops->ndo_uninit() and netdev->priv_destructor(). And at the end of unregister_netdevice(), we invoke netdev->priv_destructor() and optionally call free_netdev(). Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-08 23:52:56 +07:00
bool needs_free_netdev;
void (*priv_destructor)(struct net_device *dev);
#ifdef CONFIG_NETPOLL
struct netpoll_info __rcu *npinfo;
#endif
possible_net_t nd_net;
/* mid-layer private */
union {
void *ml_priv;
struct pcpu_lstats __percpu *lstats;
struct pcpu_sw_netstats __percpu *tstats;
struct pcpu_dstats __percpu *dstats;
};
#if IS_ENABLED(CONFIG_GARP)
struct garp_port __rcu *garp_port;
#endif
#if IS_ENABLED(CONFIG_MRP)
struct mrp_port __rcu *mrp_port;
#endif
struct device dev;
const struct attribute_group *sysfs_groups[4];
const struct attribute_group *sysfs_rx_queue_group;
const struct rtnl_link_ops *rtnl_link_ops;
[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 17:43:19 +07:00
/* for setting kernel sock attribute on TCP connection setup */
#define GSO_MAX_SIZE 65536
unsigned int gso_max_size;
#define GSO_MAX_SEGS 65535
u16 gso_max_segs;
#ifdef CONFIG_DCB
const struct dcbnl_rtnl_ops *dcbnl_ops;
#endif
s16 num_tc;
struct netdev_tc_txq tc_to_txq[TC_MAX_QUEUE];
u8 prio_tc_map[TC_BITMASK + 1];
#if IS_ENABLED(CONFIG_FCOE)
unsigned int fcoe_ddp_xid;
#endif
#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
struct netprio_map __rcu *priomap;
#endif
struct phy_device *phydev;
struct sfp_bus *sfp_bus;
struct lock_class_key *qdisc_tx_busylock;
struct lock_class_key *qdisc_running_key;
bool proto_down;
unsigned wol_enabled:1;
};
#define to_net_dev(d) container_of(d, struct net_device, dev)
net: Generic XDP This provides a generic SKB based non-optimized XDP path which is used if either the driver lacks a specific XDP implementation, or the user requests it via a new IFLA_XDP_FLAGS value named XDP_FLAGS_SKB_MODE. It is arguable that perhaps I should have required something like this as part of the initial XDP feature merge. I believe this is critical for two reasons: 1) Accessibility. More people can play with XDP with less dependencies. Yes I know we have XDP support in virtio_net, but that just creates another depedency for learning how to use this facility. I wrote this to make life easier for the XDP newbies. 2) As a model for what the expected semantics are. If there is a pure generic core implementation, it serves as a semantic example for driver folks adding XDP support. One thing I have not tried to address here is the issue of XDP_PACKET_HEADROOM, thanks to Daniel for spotting that. It seems incredibly expensive to do a skb_cow(skb, XDP_PACKET_HEADROOM) or whatever even if the XDP program doesn't try to push headers at all. I think we really need the verifier to somehow propagate whether certain XDP helpers are used or not. v5: - Handle both negative and positive offset after running prog - Fix mac length in XDP_TX case (Alexei) - Use rcu_dereference_protected() in free_netdev (kbuild test robot) v4: - Fix MAC header adjustmnet before calling prog (David Ahern) - Disable LRO when generic XDP is installed (Michael Chan) - Bypass qdisc et al. on XDP_TX and record the event (Alexei) - Do not perform generic XDP on reinjected packets (DaveM) v3: - Make sure XDP program sees packet at MAC header, push back MAC header if we do XDP_TX. (Alexei) - Elide GRO when generic XDP is in use. (Alexei) - Add XDP_FLAG_SKB_MODE flag which the user can use to request generic XDP even if the driver has an XDP implementation. (Alexei) - Report whether SKB mode is in use in rtnl_xdp_fill() via XDP_FLAGS attribute. (Daniel) v2: - Add some "fall through" comments in switch statements based upon feedback from Andrew Lunn - Use RCU for generic xdp_prog, thanks to Johannes Berg. Tested-by: Andy Gospodarek <andy@greyhouse.net> Tested-by: Jesper Dangaard Brouer <brouer@redhat.com> Tested-by: David Ahern <dsa@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-19 02:36:58 +07:00
static inline bool netif_elide_gro(const struct net_device *dev)
{
if (!(dev->features & NETIF_F_GRO) || dev->xdp_prog)
return true;
return false;
}
#define NETDEV_ALIGN 32
net: implement mechanism for HW based QOS This patch provides a mechanism for lower layer devices to steer traffic using skb->priority to tx queues. This allows for hardware based QOS schemes to use the default qdisc without incurring the penalties related to global state and the qdisc lock. While reliably receiving skbs on the correct tx ring to avoid head of line blocking resulting from shuffling in the LLD. Finally, all the goodness from txq caching and xps/rps can still be leveraged. Many drivers and hardware exist with the ability to implement QOS schemes in the hardware but currently these drivers tend to rely on firmware to reroute specific traffic, a driver specific select_queue or the queue_mapping action in the qdisc. By using select_queue for this drivers need to be updated for each and every traffic type and we lose the goodness of much of the upstream work. Firmware solutions are inherently inflexible. And finally if admins are expected to build a qdisc and filter rules to steer traffic this requires knowledge of how the hardware is currently configured. The number of tx queues and the queue offsets may change depending on resources. Also this approach incurs all the overhead of a qdisc with filters. With the mechanism in this patch users can set skb priority using expected methods ie setsockopt() or the stack can set the priority directly. Then the skb will be steered to the correct tx queues aligned with hardware QOS traffic classes. In the normal case with single traffic class and all queues in this class everything works as is until the LLD enables multiple tcs. To steer the skb we mask out the lower 4 bits of the priority and allow the hardware to configure upto 15 distinct classes of traffic. This is expected to be sufficient for most applications at any rate it is more then the 8021Q spec designates and is equal to the number of prio bands currently implemented in the default qdisc. This in conjunction with a userspace application such as lldpad can be used to implement 8021Q transmission selection algorithms one of these algorithms being the extended transmission selection algorithm currently being used for DCB. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-01-17 15:06:04 +07:00
static inline
int netdev_get_prio_tc_map(const struct net_device *dev, u32 prio)
{
return dev->prio_tc_map[prio & TC_BITMASK];
}
static inline
int netdev_set_prio_tc_map(struct net_device *dev, u8 prio, u8 tc)
{
if (tc >= dev->num_tc)
return -EINVAL;
dev->prio_tc_map[prio & TC_BITMASK] = tc & TC_BITMASK;
return 0;
}
int netdev_txq_to_tc(struct net_device *dev, unsigned int txq);
void netdev_reset_tc(struct net_device *dev);
int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset);
int netdev_set_num_tc(struct net_device *dev, u8 num_tc);
net: implement mechanism for HW based QOS This patch provides a mechanism for lower layer devices to steer traffic using skb->priority to tx queues. This allows for hardware based QOS schemes to use the default qdisc without incurring the penalties related to global state and the qdisc lock. While reliably receiving skbs on the correct tx ring to avoid head of line blocking resulting from shuffling in the LLD. Finally, all the goodness from txq caching and xps/rps can still be leveraged. Many drivers and hardware exist with the ability to implement QOS schemes in the hardware but currently these drivers tend to rely on firmware to reroute specific traffic, a driver specific select_queue or the queue_mapping action in the qdisc. By using select_queue for this drivers need to be updated for each and every traffic type and we lose the goodness of much of the upstream work. Firmware solutions are inherently inflexible. And finally if admins are expected to build a qdisc and filter rules to steer traffic this requires knowledge of how the hardware is currently configured. The number of tx queues and the queue offsets may change depending on resources. Also this approach incurs all the overhead of a qdisc with filters. With the mechanism in this patch users can set skb priority using expected methods ie setsockopt() or the stack can set the priority directly. Then the skb will be steered to the correct tx queues aligned with hardware QOS traffic classes. In the normal case with single traffic class and all queues in this class everything works as is until the LLD enables multiple tcs. To steer the skb we mask out the lower 4 bits of the priority and allow the hardware to configure upto 15 distinct classes of traffic. This is expected to be sufficient for most applications at any rate it is more then the 8021Q spec designates and is equal to the number of prio bands currently implemented in the default qdisc. This in conjunction with a userspace application such as lldpad can be used to implement 8021Q transmission selection algorithms one of these algorithms being the extended transmission selection algorithm currently being used for DCB. Signed-off-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-01-17 15:06:04 +07:00
static inline
int netdev_get_num_tc(struct net_device *dev)
{
return dev->num_tc;
}
void netdev_unbind_sb_channel(struct net_device *dev,
struct net_device *sb_dev);
int netdev_bind_sb_channel_queue(struct net_device *dev,
struct net_device *sb_dev,
u8 tc, u16 count, u16 offset);
int netdev_set_sb_channel(struct net_device *dev, u16 channel);
static inline int netdev_get_sb_channel(struct net_device *dev)
{
return max_t(int, -dev->num_tc, 0);
}
static inline
struct netdev_queue *netdev_get_tx_queue(const struct net_device *dev,
unsigned int index)
{
return &dev->_tx[index];
}
static inline struct netdev_queue *skb_get_tx_queue(const struct net_device *dev,
const struct sk_buff *skb)
{
return netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
}
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);
}
#define netdev_lockdep_set_classes(dev) \
{ \
static struct lock_class_key qdisc_tx_busylock_key; \
static struct lock_class_key qdisc_running_key; \
static struct lock_class_key qdisc_xmit_lock_key; \
static struct lock_class_key dev_addr_list_lock_key; \
unsigned int i; \
\
(dev)->qdisc_tx_busylock = &qdisc_tx_busylock_key; \
(dev)->qdisc_running_key = &qdisc_running_key; \
lockdep_set_class(&(dev)->addr_list_lock, \
&dev_addr_list_lock_key); \
for (i = 0; i < (dev)->num_tx_queues; i++) \
lockdep_set_class(&(dev)->_tx[i]._xmit_lock, \
&qdisc_xmit_lock_key); \
}
u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
struct net_device *sb_dev);
struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
struct sk_buff *skb,
struct net_device *sb_dev);
/* returns the headroom that the master device needs to take in account
* when forwarding to this dev
*/
static inline unsigned netdev_get_fwd_headroom(struct net_device *dev)
{
return dev->priv_flags & IFF_PHONY_HEADROOM ? 0 : dev->needed_headroom;
}
static inline void netdev_set_rx_headroom(struct net_device *dev, int new_hr)
{
if (dev->netdev_ops->ndo_set_rx_headroom)
dev->netdev_ops->ndo_set_rx_headroom(dev, new_hr);
}
/* set the device rx headroom to the dev's default */
static inline void netdev_reset_rx_headroom(struct net_device *dev)
{
netdev_set_rx_headroom(dev, -1);
}
/*
* Net namespace inlines
*/
static inline
struct net *dev_net(const struct net_device *dev)
{
return read_pnet(&dev->nd_net);
}
static inline
void dev_net_set(struct net_device *dev, struct net *net)
{
write_pnet(&dev->nd_net, net);
}
[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 06:41:36 +07:00
/**
* 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
* fine-grained identification of different network device types. For
* example Ethernet, Wireless LAN, Bluetooth, WiMAX etc.
*/
#define SET_NETDEV_DEVTYPE(net, devtype) ((net)->dev.type = (devtype))
/* Default NAPI poll() weight
* Device drivers are strongly advised to not use bigger value
*/
#define NAPI_POLL_WEIGHT 64
/**
* 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 14:38:52 +07:00
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 06:41:36 +07:00
/**
* netif_tx_napi_add - initialize a NAPI context
* @dev: network device
* @napi: NAPI context
* @poll: polling function
* @weight: default weight
*
* This variant of netif_napi_add() should be used from drivers using NAPI
* to exclusively poll a TX queue.
* This will avoid we add it into napi_hash[], thus polluting this hash table.
*/
static inline void netif_tx_napi_add(struct net_device *dev,
struct napi_struct *napi,
int (*poll)(struct napi_struct *, int),
int weight)
{
set_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state);
netif_napi_add(dev, napi, poll, weight);
}
/**
* 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 14:38:52 +07:00
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 14:38:52 +07:00
/* This is non-zero if the packet cannot be merged with the new skb. */
net-gre-gro: Add GRE support to the GRO stack This patch built on top of Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") to add the support of the standard GRE (RFC1701/RFC2784/RFC2890) to the GRO stack. It also serves as an example for supporting other encapsulation protocols in the GRO stack in the future. The patch supports version 0 and all the flags (key, csum, seq#) but will flush any pkt with the S (seq#) flag. This is because the S flag is not support by GSO, and a GRO pkt may end up in the forwarding path, thus requiring GSO support to break it up correctly. Currently the "packet_offload" structure only contains L3 (ETH_P_IP/ ETH_P_IPV6) GRO offload support so the encapped pkts are limited to IP pkts (i.e., w/o L2 hdr). But support for other protocol type can be easily added, so is the support for GRE variations like NVGRE. The patch also support csum offload. Specifically if the csum flag is on and the h/w is capable of checksumming the payload (CHECKSUM_COMPLETE), the code will take advantage of the csum computed by the h/w when validating the GRE csum. Note that commit 60769a5dcd8755715c7143b4571d5c44f01796f1 "ipv4: gre: add GRO capability" already introduces GRO capability to IPv4 GRE tunnels, using the gro_cells infrastructure. But GRO is done after GRE hdr has been removed (i.e., decapped). The following patch applies GRO when pkts first come in (before hitting the GRE tunnel code). There is some performance advantage for applying GRO as early as possible. Also this approach is transparent to other subsystem like Open vSwitch where GRE decap is handled outside of the IP stack hence making it harder for the gro_cells stuff to apply. On the other hand, some NICs are still not capable of hashing on the inner hdr of a GRE pkt (RSS). In that case the GRO processing of pkts from the same remote host will all happen on the same CPU and the performance may be suboptimal. I'm including some rough preliminary performance numbers below. Note that the performance will be highly dependent on traffic load, mix as usual. Moreover it also depends on NIC offload features hence the following is by no means a comprehesive study. Local testing and tuning will be needed to decide the best setting. All tests spawned 50 copies of netperf TCP_STREAM and ran for 30 secs. (super_netperf 50 -H 192.168.1.18 -l 30) An IP GRE tunnel with only the key flag on (e.g., ip tunnel add gre1 mode gre local 10.246.17.18 remote 10.246.17.17 ttl 255 key 123) is configured. The GRO support for pkts AFTER decap are controlled through the device feature of the GRE device (e.g., ethtool -K gre1 gro on/off). 1.1 ethtool -K gre1 gro off; ethtool -K eth0 gro off thruput: 9.16Gbps CPU utilization: 19% 1.2 ethtool -K gre1 gro on; ethtool -K eth0 gro off thruput: 5.9Gbps CPU utilization: 15% 1.3 ethtool -K gre1 gro off; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 12-13% 1.4 ethtool -K gre1 gro on; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 10% The following tests were performed on a different NIC that is capable of csum offload. I.e., the h/w is capable of computing IP payload csum (CHECKSUM_COMPLETE). 2.1 ethtool -K gre1 gro on (hence will use gro_cells) 2.1.1 ethtool -K eth0 gro off; csum offload disabled thruput: 8.53Gbps CPU utilization: 9% 2.1.2 ethtool -K eth0 gro off; csum offload enabled thruput: 8.97Gbps CPU utilization: 7-8% 2.1.3 ethtool -K eth0 gro on; csum offload disabled thruput: 8.83Gbps CPU utilization: 5-6% 2.1.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.98Gbps CPU utilization: 5% 2.2 ethtool -K gre1 gro off 2.2.1 ethtool -K eth0 gro off; csum offload disabled thruput: 5.93Gbps CPU utilization: 9% 2.2.2 ethtool -K eth0 gro off; csum offload enabled thruput: 5.62Gbps CPU utilization: 8% 2.2.3 ethtool -K eth0 gro on; csum offload disabled thruput: 7.69Gbps CPU utilization: 8% 2.2.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.96Gbps CPU utilization: 5-6% Signed-off-by: H.K. Jerry Chu <hkchu@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-08 01:23:19 +07:00
u16 flush;
/* Save the IP ID here and check when we get to the transport layer */
u16 flush_id;
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 14:38:52 +07:00
/* Number of segments aggregated. */
u16 count;
/* Start offset for remote checksum offload */
u16 gro_remcsum_start;
/* jiffies when first packet was created/queued */
unsigned long age;
/* Used in ipv6_gro_receive() and foo-over-udp */
u16 proto;
/* This is non-zero if the packet may be of the same flow. */
u8 same_flow:1;
/* Used in tunnel GRO receive */
u8 encap_mark:1;
/* GRO checksum is valid */
u8 csum_valid:1;
/* Number of checksums via CHECKSUM_UNNECESSARY */
u8 csum_cnt:3;
/* Free the skb? */
u8 free:2;
#define NAPI_GRO_FREE 1
#define NAPI_GRO_FREE_STOLEN_HEAD 2
/* Used in foo-over-udp, set in udp[46]_gro_receive */
u8 is_ipv6:1;
/* Used in GRE, set in fou/gue_gro_receive */
u8 is_fou:1;
/* Used to determine if flush_id can be ignored */
u8 is_atomic:1;
/* Number of gro_receive callbacks this packet already went through */
u8 recursion_counter:4;
/* 1 bit hole */
net-gre-gro: Add GRE support to the GRO stack This patch built on top of Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") to add the support of the standard GRE (RFC1701/RFC2784/RFC2890) to the GRO stack. It also serves as an example for supporting other encapsulation protocols in the GRO stack in the future. The patch supports version 0 and all the flags (key, csum, seq#) but will flush any pkt with the S (seq#) flag. This is because the S flag is not support by GSO, and a GRO pkt may end up in the forwarding path, thus requiring GSO support to break it up correctly. Currently the "packet_offload" structure only contains L3 (ETH_P_IP/ ETH_P_IPV6) GRO offload support so the encapped pkts are limited to IP pkts (i.e., w/o L2 hdr). But support for other protocol type can be easily added, so is the support for GRE variations like NVGRE. The patch also support csum offload. Specifically if the csum flag is on and the h/w is capable of checksumming the payload (CHECKSUM_COMPLETE), the code will take advantage of the csum computed by the h/w when validating the GRE csum. Note that commit 60769a5dcd8755715c7143b4571d5c44f01796f1 "ipv4: gre: add GRO capability" already introduces GRO capability to IPv4 GRE tunnels, using the gro_cells infrastructure. But GRO is done after GRE hdr has been removed (i.e., decapped). The following patch applies GRO when pkts first come in (before hitting the GRE tunnel code). There is some performance advantage for applying GRO as early as possible. Also this approach is transparent to other subsystem like Open vSwitch where GRE decap is handled outside of the IP stack hence making it harder for the gro_cells stuff to apply. On the other hand, some NICs are still not capable of hashing on the inner hdr of a GRE pkt (RSS). In that case the GRO processing of pkts from the same remote host will all happen on the same CPU and the performance may be suboptimal. I'm including some rough preliminary performance numbers below. Note that the performance will be highly dependent on traffic load, mix as usual. Moreover it also depends on NIC offload features hence the following is by no means a comprehesive study. Local testing and tuning will be needed to decide the best setting. All tests spawned 50 copies of netperf TCP_STREAM and ran for 30 secs. (super_netperf 50 -H 192.168.1.18 -l 30) An IP GRE tunnel with only the key flag on (e.g., ip tunnel add gre1 mode gre local 10.246.17.18 remote 10.246.17.17 ttl 255 key 123) is configured. The GRO support for pkts AFTER decap are controlled through the device feature of the GRE device (e.g., ethtool -K gre1 gro on/off). 1.1 ethtool -K gre1 gro off; ethtool -K eth0 gro off thruput: 9.16Gbps CPU utilization: 19% 1.2 ethtool -K gre1 gro on; ethtool -K eth0 gro off thruput: 5.9Gbps CPU utilization: 15% 1.3 ethtool -K gre1 gro off; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 12-13% 1.4 ethtool -K gre1 gro on; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 10% The following tests were performed on a different NIC that is capable of csum offload. I.e., the h/w is capable of computing IP payload csum (CHECKSUM_COMPLETE). 2.1 ethtool -K gre1 gro on (hence will use gro_cells) 2.1.1 ethtool -K eth0 gro off; csum offload disabled thruput: 8.53Gbps CPU utilization: 9% 2.1.2 ethtool -K eth0 gro off; csum offload enabled thruput: 8.97Gbps CPU utilization: 7-8% 2.1.3 ethtool -K eth0 gro on; csum offload disabled thruput: 8.83Gbps CPU utilization: 5-6% 2.1.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.98Gbps CPU utilization: 5% 2.2 ethtool -K gre1 gro off 2.2.1 ethtool -K eth0 gro off; csum offload disabled thruput: 5.93Gbps CPU utilization: 9% 2.2.2 ethtool -K eth0 gro off; csum offload enabled thruput: 5.62Gbps CPU utilization: 8% 2.2.3 ethtool -K eth0 gro on; csum offload disabled thruput: 7.69Gbps CPU utilization: 8% 2.2.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.96Gbps CPU utilization: 5-6% Signed-off-by: H.K. Jerry Chu <hkchu@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-08 01:23:19 +07:00
/* used to support CHECKSUM_COMPLETE for tunneling protocols */
__wsum csum;
/* used in skb_gro_receive() slow path */
struct sk_buff *last;
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 14:38:52 +07:00
};
#define NAPI_GRO_CB(skb) ((struct napi_gro_cb *)(skb)->cb)
#define GRO_RECURSION_LIMIT 15
static inline int gro_recursion_inc_test(struct sk_buff *skb)
{
return ++NAPI_GRO_CB(skb)->recursion_counter == GRO_RECURSION_LIMIT;
}
typedef struct sk_buff *(*gro_receive_t)(struct list_head *, struct sk_buff *);
static inline struct sk_buff *call_gro_receive(gro_receive_t cb,
struct list_head *head,
struct sk_buff *skb)
{
if (unlikely(gro_recursion_inc_test(skb))) {
NAPI_GRO_CB(skb)->flush |= 1;
return NULL;
}
return cb(head, skb);
}
typedef struct sk_buff *(*gro_receive_sk_t)(struct sock *, struct list_head *,
struct sk_buff *);
static inline struct sk_buff *call_gro_receive_sk(gro_receive_sk_t cb,
struct sock *sk,
struct list_head *head,
struct sk_buff *skb)
{
if (unlikely(gro_recursion_inc_test(skb))) {
NAPI_GRO_CB(skb)->flush |= 1;
return NULL;
}
return cb(sk, head, skb);
}
struct packet_type {
__be16 type; /* This is really htons(ether_type). */
bool ignore_outgoing;
struct net_device *dev; /* NULL is wildcarded here */
int (*func) (struct sk_buff *,
struct net_device *,
struct packet_type *,
struct net_device *);
void (*list_func) (struct list_head *,
struct packet_type *,
struct net_device *);
bool (*id_match)(struct packet_type *ptype,
struct sock *sk);
void *af_packet_priv;
struct list_head list;
};
struct offload_callbacks {
struct sk_buff *(*gso_segment)(struct sk_buff *skb,
netdev_features_t features);
struct sk_buff *(*gro_receive)(struct list_head *head,
struct sk_buff *skb);
int (*gro_complete)(struct sk_buff *skb, int nhoff);
};
struct packet_offload {
__be16 type; /* This is really htons(ether_type). */
u16 priority;
struct offload_callbacks callbacks;
struct list_head list;
};
/* often modified stats are per-CPU, other are shared (netdev->stats) */
struct pcpu_sw_netstats {
u64 rx_packets;
u64 rx_bytes;
u64 tx_packets;
u64 tx_bytes;
struct u64_stats_sync syncp;
} __aligned(4 * sizeof(u64));
struct pcpu_lstats {
u64 packets;
u64 bytes;
struct u64_stats_sync syncp;
} __aligned(2 * sizeof(u64));
#define __netdev_alloc_pcpu_stats(type, gfp) \
({ \
typeof(type) __percpu *pcpu_stats = alloc_percpu_gfp(type, gfp);\
if (pcpu_stats) { \
int __cpu; \
for_each_possible_cpu(__cpu) { \
typeof(type) *stat; \
stat = per_cpu_ptr(pcpu_stats, __cpu); \
u64_stats_init(&stat->syncp); \
} \
} \
pcpu_stats; \
})
#define netdev_alloc_pcpu_stats(type) \
__netdev_alloc_pcpu_stats(type, GFP_KERNEL)
enum netdev_lag_tx_type {
NETDEV_LAG_TX_TYPE_UNKNOWN,
NETDEV_LAG_TX_TYPE_RANDOM,
NETDEV_LAG_TX_TYPE_BROADCAST,
NETDEV_LAG_TX_TYPE_ROUNDROBIN,
NETDEV_LAG_TX_TYPE_ACTIVEBACKUP,
NETDEV_LAG_TX_TYPE_HASH,
};
enum netdev_lag_hash {
NETDEV_LAG_HASH_NONE,
NETDEV_LAG_HASH_L2,
NETDEV_LAG_HASH_L34,
NETDEV_LAG_HASH_L23,
NETDEV_LAG_HASH_E23,
NETDEV_LAG_HASH_E34,
NETDEV_LAG_HASH_UNKNOWN,
};
struct netdev_lag_upper_info {
enum netdev_lag_tx_type tx_type;
enum netdev_lag_hash hash_type;
};
struct netdev_lag_lower_state_info {
u8 link_up : 1,
tx_enabled : 1;
};
#include <linux/notifier.h>
/* netdevice notifier chain. Please remember to update netdev_cmd_to_name()
* and the rtnetlink notification exclusion list in rtnetlink_event() when
* adding new types.
*/
enum netdev_cmd {
NETDEV_UP = 1, /* For now you can't veto a device up/down */
NETDEV_DOWN,
NETDEV_REBOOT, /* Tell a protocol stack a network interface
detected a hardware crash and restarted
- we can use this eg to kick tcp sessions
once done */
NETDEV_CHANGE, /* Notify device state change */
NETDEV_REGISTER,
NETDEV_UNREGISTER,
NETDEV_CHANGEMTU, /* notify after mtu change happened */
NETDEV_CHANGEADDR, /* notify after the address change */
NETDEV_PRE_CHANGEADDR, /* notify before the address change */
NETDEV_GOING_DOWN,
NETDEV_CHANGENAME,
NETDEV_FEAT_CHANGE,
NETDEV_BONDING_FAILOVER,
NETDEV_PRE_UP,
NETDEV_PRE_TYPE_CHANGE,
NETDEV_POST_TYPE_CHANGE,
NETDEV_POST_INIT,
NETDEV_RELEASE,
NETDEV_NOTIFY_PEERS,
NETDEV_JOIN,
NETDEV_CHANGEUPPER,
NETDEV_RESEND_IGMP,
NETDEV_PRECHANGEMTU, /* notify before mtu change happened */
NETDEV_CHANGEINFODATA,
NETDEV_BONDING_INFO,
NETDEV_PRECHANGEUPPER,
NETDEV_CHANGELOWERSTATE,
NETDEV_UDP_TUNNEL_PUSH_INFO,
NETDEV_UDP_TUNNEL_DROP_INFO,
NETDEV_CHANGE_TX_QUEUE_LEN,
NETDEV_CVLAN_FILTER_PUSH_INFO,
NETDEV_CVLAN_FILTER_DROP_INFO,
NETDEV_SVLAN_FILTER_PUSH_INFO,
NETDEV_SVLAN_FILTER_DROP_INFO,
};
const char *netdev_cmd_to_name(enum netdev_cmd cmd);
int register_netdevice_notifier(struct notifier_block *nb);
int unregister_netdevice_notifier(struct notifier_block *nb);
struct netdev_notifier_info {
struct net_device *dev;
struct netlink_ext_ack *extack;
};
2018-10-09 22:48:14 +07:00
struct netdev_notifier_info_ext {
struct netdev_notifier_info info; /* must be first */
union {
u32 mtu;
} ext;
};
struct netdev_notifier_change_info {
struct netdev_notifier_info info; /* must be first */
unsigned int flags_changed;
};
struct netdev_notifier_changeupper_info {
struct netdev_notifier_info info; /* must be first */
struct net_device *upper_dev; /* new upper dev */
bool master; /* is upper dev master */
bool linking; /* is the notification for link or unlink */
void *upper_info; /* upper dev info */
};
struct netdev_notifier_changelowerstate_info {
struct netdev_notifier_info info; /* must be first */
void *lower_state_info; /* is lower dev state */
};
struct netdev_notifier_pre_changeaddr_info {
struct netdev_notifier_info info; /* must be first */
const unsigned char *dev_addr;
};
static inline void netdev_notifier_info_init(struct netdev_notifier_info *info,
struct net_device *dev)
{
info->dev = dev;
info->extack = NULL;
}
static inline struct net_device *
netdev_notifier_info_to_dev(const struct netdev_notifier_info *info)
{
return info->dev;
}
static inline struct netlink_ext_ack *
netdev_notifier_info_to_extack(const struct netdev_notifier_info *info)
{
return info->extack;
}
int call_netdevice_notifiers(unsigned long val, struct net_device *dev);
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>
2007-09-18 01:56:21 +07:00
#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>
2007-09-18 01:56:21 +07:00
#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)
#define for_each_netdev_in_bond_rcu(bond, slave) \
for_each_netdev_rcu(&init_net, slave) \
if (netdev_master_upper_dev_get_rcu(slave) == (bond))
[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>
2007-09-18 01:56:21 +07:00
#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(list_next_rcu(&dev->dev_list));
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);
}
static inline struct net_device *first_net_device_rcu(struct net *net)
{
struct list_head *lh = rcu_dereference(list_next_rcu(&net->dev_base_head));
return lh == &net->dev_base_head ? NULL : net_device_entry(lh);
}
int netdev_boot_setup_check(struct net_device *dev);
unsigned long netdev_boot_base(const char *prefix, int unit);
struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
const char *hwaddr);
struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type);
struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type);
void dev_add_pack(struct packet_type *pt);
void dev_remove_pack(struct packet_type *pt);
void __dev_remove_pack(struct packet_type *pt);
void dev_add_offload(struct packet_offload *po);
void dev_remove_offload(struct packet_offload *po);
int dev_get_iflink(const struct net_device *dev);
int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb);
struct net_device *__dev_get_by_flags(struct net *net, unsigned short flags,
unsigned short mask);
struct net_device *dev_get_by_name(struct net *net, const char *name);
struct net_device *dev_get_by_name_rcu(struct net *net, const char *name);
struct net_device *__dev_get_by_name(struct net *net, const char *name);
int dev_alloc_name(struct net_device *dev, const char *name);
int dev_open(struct net_device *dev, struct netlink_ext_ack *extack);
void dev_close(struct net_device *dev);
void dev_close_many(struct list_head *head, bool unlink);
void dev_disable_lro(struct net_device *dev);
int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *newskb);
u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
struct net_device *sb_dev);
u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
struct net_device *sb_dev);
int dev_queue_xmit(struct sk_buff *skb);
int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev);
int dev_direct_xmit(struct sk_buff *skb, u16 queue_id);
int register_netdevice(struct net_device *dev);
void unregister_netdevice_queue(struct net_device *dev, struct list_head *head);
void unregister_netdevice_many(struct list_head *head);
static inline void unregister_netdevice(struct net_device *dev)
{
unregister_netdevice_queue(dev, NULL);
}
int netdev_refcnt_read(const struct net_device *dev);
void free_netdev(struct net_device *dev);
void netdev_freemem(struct net_device *dev);
void synchronize_net(void);
int init_dummy_netdev(struct net_device *dev);
struct net_device *dev_get_by_index(struct net *net, int ifindex);
struct net_device *__dev_get_by_index(struct net *net, int ifindex);
struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex);
struct net_device *dev_get_by_napi_id(unsigned int napi_id);
int netdev_get_name(struct net *net, char *name, int ifindex);
int dev_restart(struct net_device *dev);
int skb_gro_receive(struct sk_buff *p, 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_frag0_invalidate(struct sk_buff *skb)
{
NAPI_GRO_CB(skb)->frag0 = NULL;
NAPI_GRO_CB(skb)->frag0_len = 0;
}
static inline void *skb_gro_header_slow(struct sk_buff *skb, unsigned int hlen,
unsigned int offset)
{
if (!pskb_may_pull(skb, hlen))
return NULL;
skb_gro_frag0_invalidate(skb);
return skb->data + offset;
}
static inline void *skb_gro_network_header(struct sk_buff *skb)
{
return (NAPI_GRO_CB(skb)->frag0 ?: skb->data) +
skb_network_offset(skb);
}
net-gre-gro: Add GRE support to the GRO stack This patch built on top of Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") to add the support of the standard GRE (RFC1701/RFC2784/RFC2890) to the GRO stack. It also serves as an example for supporting other encapsulation protocols in the GRO stack in the future. The patch supports version 0 and all the flags (key, csum, seq#) but will flush any pkt with the S (seq#) flag. This is because the S flag is not support by GSO, and a GRO pkt may end up in the forwarding path, thus requiring GSO support to break it up correctly. Currently the "packet_offload" structure only contains L3 (ETH_P_IP/ ETH_P_IPV6) GRO offload support so the encapped pkts are limited to IP pkts (i.e., w/o L2 hdr). But support for other protocol type can be easily added, so is the support for GRE variations like NVGRE. The patch also support csum offload. Specifically if the csum flag is on and the h/w is capable of checksumming the payload (CHECKSUM_COMPLETE), the code will take advantage of the csum computed by the h/w when validating the GRE csum. Note that commit 60769a5dcd8755715c7143b4571d5c44f01796f1 "ipv4: gre: add GRO capability" already introduces GRO capability to IPv4 GRE tunnels, using the gro_cells infrastructure. But GRO is done after GRE hdr has been removed (i.e., decapped). The following patch applies GRO when pkts first come in (before hitting the GRE tunnel code). There is some performance advantage for applying GRO as early as possible. Also this approach is transparent to other subsystem like Open vSwitch where GRE decap is handled outside of the IP stack hence making it harder for the gro_cells stuff to apply. On the other hand, some NICs are still not capable of hashing on the inner hdr of a GRE pkt (RSS). In that case the GRO processing of pkts from the same remote host will all happen on the same CPU and the performance may be suboptimal. I'm including some rough preliminary performance numbers below. Note that the performance will be highly dependent on traffic load, mix as usual. Moreover it also depends on NIC offload features hence the following is by no means a comprehesive study. Local testing and tuning will be needed to decide the best setting. All tests spawned 50 copies of netperf TCP_STREAM and ran for 30 secs. (super_netperf 50 -H 192.168.1.18 -l 30) An IP GRE tunnel with only the key flag on (e.g., ip tunnel add gre1 mode gre local 10.246.17.18 remote 10.246.17.17 ttl 255 key 123) is configured. The GRO support for pkts AFTER decap are controlled through the device feature of the GRE device (e.g., ethtool -K gre1 gro on/off). 1.1 ethtool -K gre1 gro off; ethtool -K eth0 gro off thruput: 9.16Gbps CPU utilization: 19% 1.2 ethtool -K gre1 gro on; ethtool -K eth0 gro off thruput: 5.9Gbps CPU utilization: 15% 1.3 ethtool -K gre1 gro off; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 12-13% 1.4 ethtool -K gre1 gro on; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 10% The following tests were performed on a different NIC that is capable of csum offload. I.e., the h/w is capable of computing IP payload csum (CHECKSUM_COMPLETE). 2.1 ethtool -K gre1 gro on (hence will use gro_cells) 2.1.1 ethtool -K eth0 gro off; csum offload disabled thruput: 8.53Gbps CPU utilization: 9% 2.1.2 ethtool -K eth0 gro off; csum offload enabled thruput: 8.97Gbps CPU utilization: 7-8% 2.1.3 ethtool -K eth0 gro on; csum offload disabled thruput: 8.83Gbps CPU utilization: 5-6% 2.1.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.98Gbps CPU utilization: 5% 2.2 ethtool -K gre1 gro off 2.2.1 ethtool -K eth0 gro off; csum offload disabled thruput: 5.93Gbps CPU utilization: 9% 2.2.2 ethtool -K eth0 gro off; csum offload enabled thruput: 5.62Gbps CPU utilization: 8% 2.2.3 ethtool -K eth0 gro on; csum offload disabled thruput: 7.69Gbps CPU utilization: 8% 2.2.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.96Gbps CPU utilization: 5-6% Signed-off-by: H.K. Jerry Chu <hkchu@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-08 01:23:19 +07:00
static inline void skb_gro_postpull_rcsum(struct sk_buff *skb,
const void *start, unsigned int len)
{
if (NAPI_GRO_CB(skb)->csum_valid)
net-gre-gro: Add GRE support to the GRO stack This patch built on top of Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") to add the support of the standard GRE (RFC1701/RFC2784/RFC2890) to the GRO stack. It also serves as an example for supporting other encapsulation protocols in the GRO stack in the future. The patch supports version 0 and all the flags (key, csum, seq#) but will flush any pkt with the S (seq#) flag. This is because the S flag is not support by GSO, and a GRO pkt may end up in the forwarding path, thus requiring GSO support to break it up correctly. Currently the "packet_offload" structure only contains L3 (ETH_P_IP/ ETH_P_IPV6) GRO offload support so the encapped pkts are limited to IP pkts (i.e., w/o L2 hdr). But support for other protocol type can be easily added, so is the support for GRE variations like NVGRE. The patch also support csum offload. Specifically if the csum flag is on and the h/w is capable of checksumming the payload (CHECKSUM_COMPLETE), the code will take advantage of the csum computed by the h/w when validating the GRE csum. Note that commit 60769a5dcd8755715c7143b4571d5c44f01796f1 "ipv4: gre: add GRO capability" already introduces GRO capability to IPv4 GRE tunnels, using the gro_cells infrastructure. But GRO is done after GRE hdr has been removed (i.e., decapped). The following patch applies GRO when pkts first come in (before hitting the GRE tunnel code). There is some performance advantage for applying GRO as early as possible. Also this approach is transparent to other subsystem like Open vSwitch where GRE decap is handled outside of the IP stack hence making it harder for the gro_cells stuff to apply. On the other hand, some NICs are still not capable of hashing on the inner hdr of a GRE pkt (RSS). In that case the GRO processing of pkts from the same remote host will all happen on the same CPU and the performance may be suboptimal. I'm including some rough preliminary performance numbers below. Note that the performance will be highly dependent on traffic load, mix as usual. Moreover it also depends on NIC offload features hence the following is by no means a comprehesive study. Local testing and tuning will be needed to decide the best setting. All tests spawned 50 copies of netperf TCP_STREAM and ran for 30 secs. (super_netperf 50 -H 192.168.1.18 -l 30) An IP GRE tunnel with only the key flag on (e.g., ip tunnel add gre1 mode gre local 10.246.17.18 remote 10.246.17.17 ttl 255 key 123) is configured. The GRO support for pkts AFTER decap are controlled through the device feature of the GRE device (e.g., ethtool -K gre1 gro on/off). 1.1 ethtool -K gre1 gro off; ethtool -K eth0 gro off thruput: 9.16Gbps CPU utilization: 19% 1.2 ethtool -K gre1 gro on; ethtool -K eth0 gro off thruput: 5.9Gbps CPU utilization: 15% 1.3 ethtool -K gre1 gro off; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 12-13% 1.4 ethtool -K gre1 gro on; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 10% The following tests were performed on a different NIC that is capable of csum offload. I.e., the h/w is capable of computing IP payload csum (CHECKSUM_COMPLETE). 2.1 ethtool -K gre1 gro on (hence will use gro_cells) 2.1.1 ethtool -K eth0 gro off; csum offload disabled thruput: 8.53Gbps CPU utilization: 9% 2.1.2 ethtool -K eth0 gro off; csum offload enabled thruput: 8.97Gbps CPU utilization: 7-8% 2.1.3 ethtool -K eth0 gro on; csum offload disabled thruput: 8.83Gbps CPU utilization: 5-6% 2.1.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.98Gbps CPU utilization: 5% 2.2 ethtool -K gre1 gro off 2.2.1 ethtool -K eth0 gro off; csum offload disabled thruput: 5.93Gbps CPU utilization: 9% 2.2.2 ethtool -K eth0 gro off; csum offload enabled thruput: 5.62Gbps CPU utilization: 8% 2.2.3 ethtool -K eth0 gro on; csum offload disabled thruput: 7.69Gbps CPU utilization: 8% 2.2.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.96Gbps CPU utilization: 5-6% Signed-off-by: H.K. Jerry Chu <hkchu@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-08 01:23:19 +07:00
NAPI_GRO_CB(skb)->csum = csum_sub(NAPI_GRO_CB(skb)->csum,
csum_partial(start, len, 0));
}
/* GRO checksum functions. These are logical equivalents of the normal
* checksum functions (in skbuff.h) except that they operate on the GRO
* offsets and fields in sk_buff.
*/
__sum16 __skb_gro_checksum_complete(struct sk_buff *skb);
static inline bool skb_at_gro_remcsum_start(struct sk_buff *skb)
{
return (NAPI_GRO_CB(skb)->gro_remcsum_start == skb_gro_offset(skb));
}
static inline bool __skb_gro_checksum_validate_needed(struct sk_buff *skb,
bool zero_okay,
__sum16 check)
{
return ((skb->ip_summed != CHECKSUM_PARTIAL ||
skb_checksum_start_offset(skb) <
skb_gro_offset(skb)) &&
!skb_at_gro_remcsum_start(skb) &&
NAPI_GRO_CB(skb)->csum_cnt == 0 &&
(!zero_okay || check));
}
static inline __sum16 __skb_gro_checksum_validate_complete(struct sk_buff *skb,
__wsum psum)
{
if (NAPI_GRO_CB(skb)->csum_valid &&
!csum_fold(csum_add(psum, NAPI_GRO_CB(skb)->csum)))
return 0;
NAPI_GRO_CB(skb)->csum = psum;
return __skb_gro_checksum_complete(skb);
}
static inline void skb_gro_incr_csum_unnecessary(struct sk_buff *skb)
{
if (NAPI_GRO_CB(skb)->csum_cnt > 0) {
/* Consume a checksum from CHECKSUM_UNNECESSARY */
NAPI_GRO_CB(skb)->csum_cnt--;
} else {
/* Update skb for CHECKSUM_UNNECESSARY and csum_level when we
* verified a new top level checksum or an encapsulated one
* during GRO. This saves work if we fallback to normal path.
*/
__skb_incr_checksum_unnecessary(skb);
}
}
#define __skb_gro_checksum_validate(skb, proto, zero_okay, check, \
compute_pseudo) \
({ \
__sum16 __ret = 0; \
if (__skb_gro_checksum_validate_needed(skb, zero_okay, check)) \
__ret = __skb_gro_checksum_validate_complete(skb, \
compute_pseudo(skb, proto)); \
if (!__ret) \
skb_gro_incr_csum_unnecessary(skb); \
__ret; \
})
#define skb_gro_checksum_validate(skb, proto, compute_pseudo) \
__skb_gro_checksum_validate(skb, proto, false, 0, compute_pseudo)
#define skb_gro_checksum_validate_zero_check(skb, proto, check, \
compute_pseudo) \
__skb_gro_checksum_validate(skb, proto, true, check, compute_pseudo)
#define skb_gro_checksum_simple_validate(skb) \
__skb_gro_checksum_validate(skb, 0, false, 0, null_compute_pseudo)
static inline bool __skb_gro_checksum_convert_check(struct sk_buff *skb)
{
return (NAPI_GRO_CB(skb)->csum_cnt == 0 &&
!NAPI_GRO_CB(skb)->csum_valid);
}
static inline void __skb_gro_checksum_convert(struct sk_buff *skb,
__sum16 check, __wsum pseudo)
{
NAPI_GRO_CB(skb)->csum = ~pseudo;
NAPI_GRO_CB(skb)->csum_valid = 1;
}
#define skb_gro_checksum_try_convert(skb, proto, check, compute_pseudo) \
do { \
if (__skb_gro_checksum_convert_check(skb)) \
__skb_gro_checksum_convert(skb, check, \
compute_pseudo(skb, proto)); \
} while (0)
struct gro_remcsum {
int offset;
__wsum delta;
};
static inline void skb_gro_remcsum_init(struct gro_remcsum *grc)
{
grc->offset = 0;
grc->delta = 0;
}
static inline void *skb_gro_remcsum_process(struct sk_buff *skb, void *ptr,
unsigned int off, size_t hdrlen,
int start, int offset,
struct gro_remcsum *grc,
bool nopartial)
{
__wsum delta;
size_t plen = hdrlen + max_t(size_t, offset + sizeof(u16), start);
BUG_ON(!NAPI_GRO_CB(skb)->csum_valid);
if (!nopartial) {
NAPI_GRO_CB(skb)->gro_remcsum_start = off + hdrlen + start;
return ptr;
}
ptr = skb_gro_header_fast(skb, off);
if (skb_gro_header_hard(skb, off + plen)) {
ptr = skb_gro_header_slow(skb, off + plen, off);
if (!ptr)
return NULL;
}
delta = remcsum_adjust(ptr + hdrlen, NAPI_GRO_CB(skb)->csum,
start, offset);
/* Adjust skb->csum since we changed the packet */
NAPI_GRO_CB(skb)->csum = csum_add(NAPI_GRO_CB(skb)->csum, delta);
grc->offset = off + hdrlen + offset;
grc->delta = delta;
return ptr;
}
static inline void skb_gro_remcsum_cleanup(struct sk_buff *skb,
struct gro_remcsum *grc)
{
void *ptr;
size_t plen = grc->offset + sizeof(u16);
if (!grc->delta)
return;
ptr = skb_gro_header_fast(skb, grc->offset);
if (skb_gro_header_hard(skb, grc->offset + sizeof(u16))) {
ptr = skb_gro_header_slow(skb, plen, grc->offset);
if (!ptr)
return;
}
remcsum_unadjust((__sum16 *)ptr, grc->delta);
}
#ifdef CONFIG_XFRM_OFFLOAD
static inline void skb_gro_flush_final(struct sk_buff *skb, struct sk_buff *pp, int flush)
{
if (PTR_ERR(pp) != -EINPROGRESS)
NAPI_GRO_CB(skb)->flush |= flush;
}
static inline void skb_gro_flush_final_remcsum(struct sk_buff *skb,
struct sk_buff *pp,
int flush,
struct gro_remcsum *grc)
{
if (PTR_ERR(pp) != -EINPROGRESS) {
NAPI_GRO_CB(skb)->flush |= flush;
skb_gro_remcsum_cleanup(skb, grc);
skb->remcsum_offload = 0;
}
}
#else
static inline void skb_gro_flush_final(struct sk_buff *skb, struct sk_buff *pp, int flush)
{
NAPI_GRO_CB(skb)->flush |= flush;
}
static inline void skb_gro_flush_final_remcsum(struct sk_buff *skb,
struct sk_buff *pp,
int flush,
struct gro_remcsum *grc)
{
NAPI_GRO_CB(skb)->flush |= flush;
skb_gro_remcsum_cleanup(skb, grc);
skb->remcsum_offload = 0;
}
#endif
static inline int dev_hard_header(struct sk_buff *skb, struct net_device *dev,
unsigned short type,
const void *daddr, const void *saddr,
unsigned int 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);
}
static inline __be16 dev_parse_header_protocol(const struct sk_buff *skb)
{
const struct net_device *dev = skb->dev;
if (!dev->header_ops || !dev->header_ops->parse_protocol)
return 0;
return dev->header_ops->parse_protocol(skb);
}
/* ll_header must have at least hard_header_len allocated */
static inline bool dev_validate_header(const struct net_device *dev,
char *ll_header, int len)
{
if (likely(len >= dev->hard_header_len))
return true;
if (len < dev->min_header_len)
return false;
if (capable(CAP_SYS_RAWIO)) {
memset(ll_header + len, 0, dev->hard_header_len - len);
return true;
}
if (dev->header_ops && dev->header_ops->validate)
return dev->header_ops->validate(ll_header, len);
return false;
}
typedef int gifconf_func_t(struct net_device * dev, char __user * bufptr,
int len, int size);
int register_gifconf(unsigned int family, gifconf_func_t *gifconf);
static inline int unregister_gifconf(unsigned int family)
{
return register_gifconf(family, NULL);
}
#ifdef CONFIG_NET_FLOW_LIMIT
#define FLOW_LIMIT_HISTORY (1 << 7) /* must be ^2 and !overflow buckets */
struct sd_flow_limit {
u64 count;
unsigned int num_buckets;
unsigned int history_head;
u16 history[FLOW_LIMIT_HISTORY];
u8 buckets[];
};
extern int netdev_flow_limit_table_len;
#endif /* CONFIG_NET_FLOW_LIMIT */
/*
* Incoming packets are placed on per-CPU queues
*/
struct softnet_data {
struct list_head poll_list;
struct sk_buff_head process_queue;
/* stats */
unsigned int processed;
unsigned int time_squeeze;
unsigned int received_rps;
#ifdef CONFIG_RPS
struct softnet_data *rps_ipi_list;
#endif
#ifdef CONFIG_NET_FLOW_LIMIT
struct sd_flow_limit __rcu *flow_limit;
#endif
struct Qdisc *output_queue;
struct Qdisc **output_queue_tailp;
struct sk_buff *completion_queue;
#ifdef CONFIG_XFRM_OFFLOAD
struct sk_buff_head xfrm_backlog;
#endif
/* written and read only by owning cpu: */
struct {
u16 recursion;
u8 more;
} xmit;
#ifdef CONFIG_RPS
/* input_queue_head should be written by cpu owning this struct,
* and only read by other cpus. Worth using a cache line.
*/
unsigned int input_queue_head ____cacheline_aligned_in_smp;
/* Elements below can be accessed between CPUs for RPS/RFS */
smp: Avoid using two cache lines for struct call_single_data struct call_single_data is used in IPIs to transfer information between CPUs. Its size is bigger than sizeof(unsigned long) and less than cache line size. Currently it is not allocated with any explicit alignment requirements. This makes it possible for allocated call_single_data to cross two cache lines, which results in double the number of the cache lines that need to be transferred among CPUs. This can be fixed by requiring call_single_data to be aligned with the size of call_single_data. Currently the size of call_single_data is the power of 2. If we add new fields to call_single_data, we may need to add padding to make sure the size of new definition is the power of 2 as well. Fortunately, this is enforced by GCC, which will report bad sizes. To set alignment requirements of call_single_data to the size of call_single_data, a struct definition and a typedef is used. To test the effect of the patch, I used the vm-scalability multiple thread swap test case (swap-w-seq-mt). The test will create multiple threads and each thread will eat memory until all RAM and part of swap is used, so that huge number of IPIs are triggered when unmapping memory. In the test, the throughput of memory writing improves ~5% compared with misaligned call_single_data, because of faster IPIs. Suggested-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Huang, Ying <ying.huang@intel.com> [ Add call_single_data_t and align with size of call_single_data. ] Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Juergen Gross <jgross@suse.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/87bmnqd6lz.fsf@yhuang-mobile.sh.intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-08 11:30:00 +07:00
call_single_data_t csd ____cacheline_aligned_in_smp;
struct softnet_data *rps_ipi_next;
unsigned int cpu;
unsigned int input_queue_tail;
#endif
unsigned int dropped;
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 15:03:29 +07:00
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 06:41:36 +07:00
struct napi_struct backlog;
};
static inline void input_queue_head_incr(struct softnet_data *sd)
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
{
#ifdef CONFIG_RPS
sd->input_queue_head++;
#endif
}
static inline void input_queue_tail_incr_save(struct softnet_data *sd,
unsigned int *qtail)
{
#ifdef CONFIG_RPS
*qtail = ++sd->input_queue_tail;
rfs: Receive Flow Steering This patch implements receive flow steering (RFS). RFS steers received packets for layer 3 and 4 processing to the CPU where the application for the corresponding flow is running. RFS is an extension of Receive Packet Steering (RPS). The basic idea of RFS is that when an application calls recvmsg (or sendmsg) the application's running CPU is stored in a hash table that is indexed by the connection's rxhash which is stored in the socket structure. The rxhash is passed in skb's received on the connection from netif_receive_skb. For each received packet, the associated rxhash is used to look up the CPU in the hash table, if a valid CPU is set then the packet is steered to that CPU using the RPS mechanisms. The convolution of the simple approach is that it would potentially allow OOO packets. If threads are thrashing around CPUs or multiple threads are trying to read from the same sockets, a quickly changing CPU value in the hash table could cause rampant OOO packets-- we consider this a non-starter. To avoid OOO packets, this solution implements two types of hash tables: rps_sock_flow_table and rps_dev_flow_table. rps_sock_table is a global hash table. Each entry is just a CPU number and it is populated in recvmsg and sendmsg as described above. This table contains the "desired" CPUs for flows. rps_dev_flow_table is specific to each device queue. Each entry contains a CPU and a tail queue counter. The CPU is the "current" CPU for a matching flow. The tail queue counter holds the value of a tail queue counter for the associated CPU's backlog queue at the time of last enqueue for a flow matching the entry. Each backlog queue has a queue head counter which is incremented on dequeue, and so a queue tail counter is computed as queue head count + queue length. When a packet is enqueued on a backlog queue, the current value of the queue tail counter is saved in the hash entry of the rps_dev_flow_table. And now the trick: when selecting the CPU for RPS (get_rps_cpu) the rps_sock_flow table and the rps_dev_flow table for the RX queue are consulted. When the desired CPU for the flow (found in the rps_sock_flow table) does not match the current CPU (found in the rps_dev_flow table), the current CPU is changed to the desired CPU if one of the following is true: - The current CPU is unset (equal to RPS_NO_CPU) - Current CPU is offline - The current CPU's queue head counter >= queue tail counter in the rps_dev_flow table. This checks if the queue tail has advanced beyond the last packet that was enqueued using this table entry. This guarantees that all packets queued using this entry have been dequeued, thus preserving in order delivery. Making each queue have its own rps_dev_flow table has two advantages: 1) the tail queue counters will be written on each receive, so keeping the table local to interrupting CPU s good for locality. 2) this allows lockless access to the table-- the CPU number and queue tail counter need to be accessed together under mutual exclusion from netif_receive_skb, we assume that this is only called from device napi_poll which is non-reentrant. This patch implements RFS for TCP and connected UDP sockets. It should be usable for other flow oriented protocols. There are two configuration parameters for RFS. The "rps_flow_entries" kernel init parameter sets the number of entries in the rps_sock_flow_table, the per rxqueue sysfs entry "rps_flow_cnt" contains the number of entries in the rps_dev_flow table for the rxqueue. Both are rounded to power of two. The obvious benefit of RFS (over just RPS) is that it achieves CPU locality between the receive processing for a flow and the applications processing; this can result in increased performance (higher pps, lower latency). The benefits of RFS are dependent on cache hierarchy, application load, and other factors. On simple benchmarks, we don't necessarily see improvement and sometimes see degradation. However, for more complex benchmarks and for applications where cache pressure is much higher this technique seems to perform very well. Below are some benchmark results which show the potential benfit of this patch. The netperf test has 500 instances of netperf TCP_RR test with 1 byte req. and resp. The RPC test is an request/response test similar in structure to netperf RR test ith 100 threads on each host, but does more work in userspace that netperf. e1000e on 8 core Intel No RFS or RPS 104K tps at 30% CPU No RFS (best RPS config): 290K tps at 63% CPU RFS 303K tps at 61% CPU RPC test tps CPU% 50/90/99% usec latency Latency StdDev No RFS/RPS 103K 48% 757/900/3185 4472.35 RPS only: 174K 73% 415/993/2468 491.66 RFS 223K 73% 379/651/1382 315.61 Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-17 06:01:27 +07:00
#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 15:03:29 +07:00
DECLARE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
static inline int dev_recursion_level(void)
{
return this_cpu_read(softnet_data.xmit.recursion);
}
#define XMIT_RECURSION_LIMIT 10
static inline bool dev_xmit_recursion(void)
{
return unlikely(__this_cpu_read(softnet_data.xmit.recursion) >
XMIT_RECURSION_LIMIT);
}
static inline void dev_xmit_recursion_inc(void)
{
__this_cpu_inc(softnet_data.xmit.recursion);
}
static inline void dev_xmit_recursion_dec(void)
{
__this_cpu_dec(softnet_data.xmit.recursion);
}
void __netif_schedule(struct Qdisc *q);
void netif_schedule_queue(struct netdev_queue *txq);
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));
}
net: force inlining of netif_tx_start/stop_queue, sock_hold, __sock_put Sometimes gcc mysteriously doesn't inline very small functions we expect to be inlined. See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66122 Arguably, gcc should do better, but gcc people aren't willing to invest time into it, asking to use __always_inline instead. With this .config: http://busybox.net/~vda/kernel_config_OPTIMIZE_INLINING_and_Os, the following functions get deinlined many times. netif_tx_stop_queue: 207 copies, 590 calls: 55 push %rbp 48 89 e5 mov %rsp,%rbp f0 80 8f e0 01 00 00 01 lock orb $0x1,0x1e0(%rdi) 5d pop %rbp c3 retq netif_tx_start_queue: 47 copies, 111 calls 55 push %rbp 48 89 e5 mov %rsp,%rbp f0 80 a7 e0 01 00 00 fe lock andb $0xfe,0x1e0(%rdi) 5d pop %rbp c3 retq sock_hold: 39 copies, 124 calls 55 push %rbp 48 89 e5 mov %rsp,%rbp f0 ff 87 80 00 00 00 lock incl 0x80(%rdi) 5d pop %rbp c3 retq __sock_put: 6 copies, 13 calls 55 push %rbp 48 89 e5 mov %rsp,%rbp f0 ff 8f 80 00 00 00 lock decl 0x80(%rdi) 5d pop %rbp c3 retq This patch fixes this via s/inline/__always_inline/. Code size decrease after the patch is ~2.5k: text data bss dec hex filename 56719876 56364551 36196352 149280779 8e5d80b vmlinux_before 56717440 56364551 36196352 149278343 8e5ce87 vmlinux Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com> CC: David S. Miller <davem@davemloft.net> CC: linux-kernel@vger.kernel.org CC: netdev@vger.kernel.org CC: netfilter-devel@vger.kernel.org Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-08 22:51:54 +07:00
static __always_inline void netif_tx_start_queue(struct netdev_queue *dev_queue)
{
clear_bit(__QUEUE_STATE_DRV_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 06:41:36 +07:00
/**
* 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);
}
}
void netif_tx_wake_queue(struct netdev_queue *dev_queue);
/**
* 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);
}
}
net: force inlining of netif_tx_start/stop_queue, sock_hold, __sock_put Sometimes gcc mysteriously doesn't inline very small functions we expect to be inlined. See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66122 Arguably, gcc should do better, but gcc people aren't willing to invest time into it, asking to use __always_inline instead. With this .config: http://busybox.net/~vda/kernel_config_OPTIMIZE_INLINING_and_Os, the following functions get deinlined many times. netif_tx_stop_queue: 207 copies, 590 calls: 55 push %rbp 48 89 e5 mov %rsp,%rbp f0 80 8f e0 01 00 00 01 lock orb $0x1,0x1e0(%rdi) 5d pop %rbp c3 retq netif_tx_start_queue: 47 copies, 111 calls 55 push %rbp 48 89 e5 mov %rsp,%rbp f0 80 a7 e0 01 00 00 fe lock andb $0xfe,0x1e0(%rdi) 5d pop %rbp c3 retq sock_hold: 39 copies, 124 calls 55 push %rbp 48 89 e5 mov %rsp,%rbp f0 ff 87 80 00 00 00 lock incl 0x80(%rdi) 5d pop %rbp c3 retq __sock_put: 6 copies, 13 calls 55 push %rbp 48 89 e5 mov %rsp,%rbp f0 ff 8f 80 00 00 00 lock decl 0x80(%rdi) 5d pop %rbp c3 retq This patch fixes this via s/inline/__always_inline/. Code size decrease after the patch is ~2.5k: text data bss dec hex filename 56719876 56364551 36196352 149280779 8e5d80b vmlinux_before 56717440 56364551 36196352 149278343 8e5ce87 vmlinux Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com> CC: David S. Miller <davem@davemloft.net> CC: linux-kernel@vger.kernel.org CC: netdev@vger.kernel.org CC: netfilter-devel@vger.kernel.org Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-08 22:51:54 +07:00
static __always_inline void netif_tx_stop_queue(struct netdev_queue *dev_queue)
{
set_bit(__QUEUE_STATE_DRV_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 06:41:36 +07:00
/**
* 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));
}
2015-05-12 02:17:53 +07:00
void netif_tx_stop_all_queues(struct net_device *dev);
static inline bool netif_tx_queue_stopped(const struct netdev_queue *dev_queue)
{
return test_bit(__QUEUE_STATE_DRV_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 06:41:36 +07:00
/**
* 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 bool netif_queue_stopped(const struct net_device *dev)
{
return netif_tx_queue_stopped(netdev_get_tx_queue(dev, 0));
}
static inline bool netif_xmit_stopped(const struct netdev_queue *dev_queue)
{
return dev_queue->state & QUEUE_STATE_ANY_XOFF;
}
static inline bool
netif_xmit_frozen_or_stopped(const struct netdev_queue *dev_queue)
{
return dev_queue->state & QUEUE_STATE_ANY_XOFF_OR_FROZEN;
}
static inline bool
netif_xmit_frozen_or_drv_stopped(const struct netdev_queue *dev_queue)
{
return dev_queue->state & QUEUE_STATE_DRV_XOFF_OR_FROZEN;
}
/**
* netdev_txq_bql_enqueue_prefetchw - prefetch bql data for write
* @dev_queue: pointer to transmit queue
*
* BQL enabled drivers might use this helper in their ndo_start_xmit(),
* to give appropriate hint to the CPU.
*/
static inline void netdev_txq_bql_enqueue_prefetchw(struct netdev_queue *dev_queue)
{
#ifdef CONFIG_BQL
prefetchw(&dev_queue->dql.num_queued);
#endif
}
/**
* netdev_txq_bql_complete_prefetchw - prefetch bql data for write
* @dev_queue: pointer to transmit queue
*
* BQL enabled drivers might use this helper in their TX completion path,
* to give appropriate hint to the CPU.
*/
static inline void netdev_txq_bql_complete_prefetchw(struct netdev_queue *dev_queue)
{
#ifdef CONFIG_BQL
prefetchw(&dev_queue->dql.limit);
#endif
}
static inline void netdev_tx_sent_queue(struct netdev_queue *dev_queue,
unsigned int bytes)
{
#ifdef CONFIG_BQL
dql_queued(&dev_queue->dql, bytes);
if (likely(dql_avail(&dev_queue->dql) >= 0))
return;
set_bit(__QUEUE_STATE_STACK_XOFF, &dev_queue->state);
/*
* The XOFF flag must be set before checking the dql_avail below,
* because in netdev_tx_completed_queue we update the dql_completed
* before checking the XOFF flag.
*/
smp_mb();
/* check again in case another CPU has just made room avail */
if (unlikely(dql_avail(&dev_queue->dql) >= 0))
clear_bit(__QUEUE_STATE_STACK_XOFF, &dev_queue->state);
#endif
}
net: bql: add __netdev_tx_sent_queue() When qdisc_run() tries to use BQL budget to bulk-dequeue a batch of packets, GSO can later transform this list in another list of skbs, and each skb is sent to device ndo_start_xmit(), one at a time, with skb->xmit_more being set to one but for last skb. Problem is that very often, BQL limit is hit in the middle of the packet train, forcing dev_hard_start_xmit() to stop the bulk send and requeue the end of the list. BQL role is to avoid head of line blocking, making sure a qdisc can deliver high priority packets before low priority ones. But there is no way requeued packets can be bypassed by fresh packets in the qdisc. Aborting the bulk send increases TX softirqs, and hot cache lines (after skb_segment()) are wasted. Note that for TSO packets, we never split a packet in the middle because of BQL limit being hit. Drivers should be able to update BQL counters without flipping/caring about BQL status, if the current skb has xmit_more set. Upper layers are ultimately responsible to stop sending another packet train when BQL limit is hit. Code template in a driver might look like the following : send_doorbell = __netdev_tx_sent_queue(tx_queue, nr_bytes, skb->xmit_more); Note that __netdev_tx_sent_queue() use is not mandatory, since following patch will change dev_hard_start_xmit() to not care about BQL status. But it is highly recommended so that xmit_more full benefits can be reached (less doorbells sent, and less atomic operations as well) Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-10-31 22:39:12 +07:00
/* Variant of netdev_tx_sent_queue() for drivers that are aware
* that they should not test BQL status themselves.
* We do want to change __QUEUE_STATE_STACK_XOFF only for the last
* skb of a batch.
* Returns true if the doorbell must be used to kick the NIC.
*/
static inline bool __netdev_tx_sent_queue(struct netdev_queue *dev_queue,
unsigned int bytes,
bool xmit_more)
{
if (xmit_more) {
#ifdef CONFIG_BQL
dql_queued(&dev_queue->dql, bytes);
#endif
return netif_tx_queue_stopped(dev_queue);
}
netdev_tx_sent_queue(dev_queue, bytes);
return true;
}
/**
* netdev_sent_queue - report the number of bytes queued to hardware
* @dev: network device
* @bytes: number of bytes queued to the hardware device queue
*
* Report the number of bytes queued for sending/completion to the network
* device hardware queue. @bytes should be a good approximation and should
* exactly match netdev_completed_queue() @bytes
*/
static inline void netdev_sent_queue(struct net_device *dev, unsigned int bytes)
{
netdev_tx_sent_queue(netdev_get_tx_queue(dev, 0), bytes);
}
static inline bool __netdev_sent_queue(struct net_device *dev,
unsigned int bytes,
bool xmit_more)
{
return __netdev_tx_sent_queue(netdev_get_tx_queue(dev, 0), bytes,
xmit_more);
}
static inline void netdev_tx_completed_queue(struct netdev_queue *dev_queue,
unsigned int pkts, unsigned int bytes)
{
#ifdef CONFIG_BQL
if (unlikely(!bytes))
return;
dql_completed(&dev_queue->dql, bytes);
/*
* Without the memory barrier there is a small possiblity that
* netdev_tx_sent_queue will miss the update and cause the queue to
* be stopped forever
*/
smp_mb();
if (dql_avail(&dev_queue->dql) < 0)
return;
if (test_and_clear_bit(__QUEUE_STATE_STACK_XOFF, &dev_queue->state))
netif_schedule_queue(dev_queue);
#endif
}
/**
* netdev_completed_queue - report bytes and packets completed by device
* @dev: network device
* @pkts: actual number of packets sent over the medium
* @bytes: actual number of bytes sent over the medium
*
* Report the number of bytes and packets transmitted by the network device
* hardware queue over the physical medium, @bytes must exactly match the
* @bytes amount passed to netdev_sent_queue()
*/
static inline void netdev_completed_queue(struct net_device *dev,
unsigned int pkts, unsigned int bytes)
{
netdev_tx_completed_queue(netdev_get_tx_queue(dev, 0), pkts, bytes);
}
static inline void netdev_tx_reset_queue(struct netdev_queue *q)
{
#ifdef CONFIG_BQL
clear_bit(__QUEUE_STATE_STACK_XOFF, &q->state);
dql_reset(&q->dql);
#endif
}
/**
* netdev_reset_queue - reset the packets and bytes count of a network device
* @dev_queue: network device
*
* Reset the bytes and packet count of a network device and clear the
* software flow control OFF bit for this network device
*/
static inline void netdev_reset_queue(struct net_device *dev_queue)
{
netdev_tx_reset_queue(netdev_get_tx_queue(dev_queue, 0));
}
/**
* netdev_cap_txqueue - check if selected tx queue exceeds device queues
* @dev: network device
* @queue_index: given tx queue index
*
* Returns 0 if given tx queue index >= number of device tx queues,
* otherwise returns the originally passed tx queue index.
*/
static inline u16 netdev_cap_txqueue(struct net_device *dev, u16 queue_index)
{
if (unlikely(queue_index >= dev->real_num_tx_queues)) {
net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
dev->name, queue_index,
dev->real_num_tx_queues);
return 0;
}
return queue_index;
}
[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 06:41:36 +07:00
/**
* netif_running - test if up
* @dev: network device
*
* Test if the device has been brought up.
*/
static inline bool 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 06:41:36 +07:00
/**
* 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 06:41:36 +07:00
/**
* 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);
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 06:41:36 +07:00
/**
* 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 bool __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 bool 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 06:41:36 +07:00
/**
* 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);
netif_tx_wake_queue(txq);
}
#ifdef CONFIG_XPS
int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
u16 index);
int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
u16 index, bool is_rxqs_map);
/**
* netif_attr_test_mask - Test a CPU or Rx queue set in a mask
* @j: CPU/Rx queue index
* @mask: bitmask of all cpus/rx queues
* @nr_bits: number of bits in the bitmask
*
* Test if a CPU or Rx queue index is set in a mask of all CPU/Rx queues.
*/
static inline bool netif_attr_test_mask(unsigned long j,
const unsigned long *mask,
unsigned int nr_bits)
{
cpu_max_bits_warn(j, nr_bits);
return test_bit(j, mask);
}
/**
* netif_attr_test_online - Test for online CPU/Rx queue
* @j: CPU/Rx queue index
* @online_mask: bitmask for CPUs/Rx queues that are online
* @nr_bits: number of bits in the bitmask
*
* Returns true if a CPU/Rx queue is online.
*/
static inline bool netif_attr_test_online(unsigned long j,
const unsigned long *online_mask,
unsigned int nr_bits)
{
cpu_max_bits_warn(j, nr_bits);
if (online_mask)
return test_bit(j, online_mask);
return (j < nr_bits);
}
/**
* netif_attrmask_next - get the next CPU/Rx queue in a cpu/Rx queues mask
* @n: CPU/Rx queue index
* @srcp: the cpumask/Rx queue mask pointer
* @nr_bits: number of bits in the bitmask
*
* Returns >= nr_bits if no further CPUs/Rx queues set.
*/
static inline unsigned int netif_attrmask_next(int n, const unsigned long *srcp,
unsigned int nr_bits)
{
/* -1 is a legal arg here. */
if (n != -1)
cpu_max_bits_warn(n, nr_bits);
if (srcp)
return find_next_bit(srcp, nr_bits, n + 1);
return n + 1;
}
/**
* netif_attrmask_next_and - get the next CPU/Rx queue in *src1p & *src2p
* @n: CPU/Rx queue index
* @src1p: the first CPUs/Rx queues mask pointer
* @src2p: the second CPUs/Rx queues mask pointer
* @nr_bits: number of bits in the bitmask
*
* Returns >= nr_bits if no further CPUs/Rx queues set in both.
*/
static inline int netif_attrmask_next_and(int n, const unsigned long *src1p,
const unsigned long *src2p,
unsigned int nr_bits)
{
/* -1 is a legal arg here. */
if (n != -1)
cpu_max_bits_warn(n, nr_bits);
if (src1p && src2p)
return find_next_and_bit(src1p, src2p, nr_bits, n + 1);
else if (src1p)
return find_next_bit(src1p, nr_bits, n + 1);
else if (src2p)
return find_next_bit(src2p, nr_bits, n + 1);
return n + 1;
}
#else
static inline int netif_set_xps_queue(struct net_device *dev,
const struct cpumask *mask,
u16 index)
{
return 0;
}
static inline int __netif_set_xps_queue(struct net_device *dev,
const unsigned long *mask,
u16 index, bool is_rxqs_map)
{
return 0;
}
#endif
[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 06:41:36 +07:00
/**
* netif_is_multiqueue - test if device has multiple transmit queues
* @dev: network device
*
* Check if device has multiple transmit queues
*/
static inline bool netif_is_multiqueue(const struct net_device *dev)
{
return dev->num_tx_queues > 1;
}
int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq);
#ifdef CONFIG_SYSFS
int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq);
#else
static inline int netif_set_real_num_rx_queues(struct net_device *dev,
unsigned int rxqs)
{
dev->real_num_rx_queues = rxqs;
return 0;
}
#endif
static inline struct netdev_rx_queue *
__netif_get_rx_queue(struct net_device *dev, unsigned int rxq)
{
return dev->_rx + rxq;
}
#ifdef CONFIG_SYSFS
static inline unsigned int get_netdev_rx_queue_index(
struct netdev_rx_queue *queue)
{
struct net_device *dev = queue->dev;
int index = queue - dev->_rx;
BUG_ON(index >= dev->num_rx_queues);
return index;
}
#endif
#define DEFAULT_MAX_NUM_RSS_QUEUES (8)
int netif_get_num_default_rss_queues(void);
enum skb_free_reason {
SKB_REASON_CONSUMED,
SKB_REASON_DROPPED,
};
void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason);
void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason);
/*
* It is not allowed to call kfree_skb() or consume_skb() from hardware
* interrupt context or with hardware interrupts being disabled.
* (in_irq() || irqs_disabled())
*
* We provide four helpers that can be used in following contexts :
*
* dev_kfree_skb_irq(skb) when caller drops a packet from irq context,
* replacing kfree_skb(skb)
*
* dev_consume_skb_irq(skb) when caller consumes a packet from irq context.
* Typically used in place of consume_skb(skb) in TX completion path
*
* dev_kfree_skb_any(skb) when caller doesn't know its current irq context,
* replacing kfree_skb(skb)
*
* dev_consume_skb_any(skb) when caller doesn't know its current irq context,
* and consumed a packet. Used in place of consume_skb(skb)
*/
static inline void dev_kfree_skb_irq(struct sk_buff *skb)
{
__dev_kfree_skb_irq(skb, SKB_REASON_DROPPED);
}
static inline void dev_consume_skb_irq(struct sk_buff *skb)
{
__dev_kfree_skb_irq(skb, SKB_REASON_CONSUMED);
}
static inline void dev_kfree_skb_any(struct sk_buff *skb)
{
__dev_kfree_skb_any(skb, SKB_REASON_DROPPED);
}
static inline void dev_consume_skb_any(struct sk_buff *skb)
{
__dev_kfree_skb_any(skb, SKB_REASON_CONSUMED);
}
void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog);
int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb);
int netif_rx(struct sk_buff *skb);
int netif_rx_ni(struct sk_buff *skb);
int netif_receive_skb(struct sk_buff *skb);
int netif_receive_skb_core(struct sk_buff *skb);
void netif_receive_skb_list(struct list_head *head);
gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb);
void napi_gro_flush(struct napi_struct *napi, bool flush_old);
struct sk_buff *napi_get_frags(struct napi_struct *napi);
gro_result_t napi_gro_frags(struct napi_struct *napi);
net-gre-gro: Add GRE support to the GRO stack This patch built on top of Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") to add the support of the standard GRE (RFC1701/RFC2784/RFC2890) to the GRO stack. It also serves as an example for supporting other encapsulation protocols in the GRO stack in the future. The patch supports version 0 and all the flags (key, csum, seq#) but will flush any pkt with the S (seq#) flag. This is because the S flag is not support by GSO, and a GRO pkt may end up in the forwarding path, thus requiring GSO support to break it up correctly. Currently the "packet_offload" structure only contains L3 (ETH_P_IP/ ETH_P_IPV6) GRO offload support so the encapped pkts are limited to IP pkts (i.e., w/o L2 hdr). But support for other protocol type can be easily added, so is the support for GRE variations like NVGRE. The patch also support csum offload. Specifically if the csum flag is on and the h/w is capable of checksumming the payload (CHECKSUM_COMPLETE), the code will take advantage of the csum computed by the h/w when validating the GRE csum. Note that commit 60769a5dcd8755715c7143b4571d5c44f01796f1 "ipv4: gre: add GRO capability" already introduces GRO capability to IPv4 GRE tunnels, using the gro_cells infrastructure. But GRO is done after GRE hdr has been removed (i.e., decapped). The following patch applies GRO when pkts first come in (before hitting the GRE tunnel code). There is some performance advantage for applying GRO as early as possible. Also this approach is transparent to other subsystem like Open vSwitch where GRE decap is handled outside of the IP stack hence making it harder for the gro_cells stuff to apply. On the other hand, some NICs are still not capable of hashing on the inner hdr of a GRE pkt (RSS). In that case the GRO processing of pkts from the same remote host will all happen on the same CPU and the performance may be suboptimal. I'm including some rough preliminary performance numbers below. Note that the performance will be highly dependent on traffic load, mix as usual. Moreover it also depends on NIC offload features hence the following is by no means a comprehesive study. Local testing and tuning will be needed to decide the best setting. All tests spawned 50 copies of netperf TCP_STREAM and ran for 30 secs. (super_netperf 50 -H 192.168.1.18 -l 30) An IP GRE tunnel with only the key flag on (e.g., ip tunnel add gre1 mode gre local 10.246.17.18 remote 10.246.17.17 ttl 255 key 123) is configured. The GRO support for pkts AFTER decap are controlled through the device feature of the GRE device (e.g., ethtool -K gre1 gro on/off). 1.1 ethtool -K gre1 gro off; ethtool -K eth0 gro off thruput: 9.16Gbps CPU utilization: 19% 1.2 ethtool -K gre1 gro on; ethtool -K eth0 gro off thruput: 5.9Gbps CPU utilization: 15% 1.3 ethtool -K gre1 gro off; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 12-13% 1.4 ethtool -K gre1 gro on; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 10% The following tests were performed on a different NIC that is capable of csum offload. I.e., the h/w is capable of computing IP payload csum (CHECKSUM_COMPLETE). 2.1 ethtool -K gre1 gro on (hence will use gro_cells) 2.1.1 ethtool -K eth0 gro off; csum offload disabled thruput: 8.53Gbps CPU utilization: 9% 2.1.2 ethtool -K eth0 gro off; csum offload enabled thruput: 8.97Gbps CPU utilization: 7-8% 2.1.3 ethtool -K eth0 gro on; csum offload disabled thruput: 8.83Gbps CPU utilization: 5-6% 2.1.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.98Gbps CPU utilization: 5% 2.2 ethtool -K gre1 gro off 2.2.1 ethtool -K eth0 gro off; csum offload disabled thruput: 5.93Gbps CPU utilization: 9% 2.2.2 ethtool -K eth0 gro off; csum offload enabled thruput: 5.62Gbps CPU utilization: 8% 2.2.3 ethtool -K eth0 gro on; csum offload disabled thruput: 7.69Gbps CPU utilization: 8% 2.2.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.96Gbps CPU utilization: 5-6% Signed-off-by: H.K. Jerry Chu <hkchu@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-08 01:23:19 +07:00
struct packet_offload *gro_find_receive_by_type(__be16 type);
struct packet_offload *gro_find_complete_by_type(__be16 type);
static inline void napi_free_frags(struct napi_struct *napi)
{
kfree_skb(napi->skb);
napi->skb = NULL;
}
bool netdev_is_rx_handler_busy(struct net_device *dev);
int netdev_rx_handler_register(struct net_device *dev,
rx_handler_func_t *rx_handler,
void *rx_handler_data);
void netdev_rx_handler_unregister(struct net_device *dev);
bool dev_valid_name(const char *name);
int dev_ioctl(struct net *net, unsigned int cmd, struct ifreq *ifr,
bool *need_copyout);
int dev_ifconf(struct net *net, struct ifconf *, int);
int dev_ethtool(struct net *net, struct ifreq *);
unsigned int dev_get_flags(const struct net_device *);
int __dev_change_flags(struct net_device *dev, unsigned int flags,
struct netlink_ext_ack *extack);
int dev_change_flags(struct net_device *dev, unsigned int flags,
struct netlink_ext_ack *extack);
void __dev_notify_flags(struct net_device *, unsigned int old_flags,
unsigned int gchanges);
int dev_change_name(struct net_device *, const char *);
int dev_set_alias(struct net_device *, const char *, size_t);
int dev_get_alias(const struct net_device *, char *, size_t);
int dev_change_net_namespace(struct net_device *, struct net *, const char *);
int __dev_set_mtu(struct net_device *, int);
int dev_set_mtu_ext(struct net_device *dev, int mtu,
struct netlink_ext_ack *extack);
int dev_set_mtu(struct net_device *, int);
int dev_change_tx_queue_len(struct net_device *, unsigned long);
void dev_set_group(struct net_device *, int);
int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
struct netlink_ext_ack *extack);
int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
struct netlink_ext_ack *extack);
int dev_change_carrier(struct net_device *, bool new_carrier);
int dev_get_phys_port_id(struct net_device *dev,
struct netdev_phys_item_id *ppid);
int dev_get_phys_port_name(struct net_device *dev,
char *name, size_t len);
int dev_get_port_parent_id(struct net_device *dev,
struct netdev_phys_item_id *ppid, bool recurse);
bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b);
int dev_change_proto_down(struct net_device *dev, bool proto_down);
int dev_change_proto_down_generic(struct net_device *dev, bool proto_down);
struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again);
struct sk_buff *dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
struct netdev_queue *txq, int *ret);
xdp: refine xdp api with regards to generic xdp While working on the iproute2 generic XDP frontend, I noticed that as of right now it's possible to have native *and* generic XDP programs loaded both at the same time for the case when a driver supports native XDP. The intended model for generic XDP from b5cdae3291f7 ("net: Generic XDP") is, however, that only one out of the two can be present at once which is also indicated as such in the XDP netlink dump part. The main rationale for generic XDP is to ease accessibility (in case a driver does not yet have XDP support) and to generically provide a semantical model as an example for driver developers wanting to add XDP support. The generic XDP option for an XDP aware driver can still be useful for comparing and testing both implementations. However, it is not intended to have a second XDP processing stage or layer with exactly the same functionality of the first native stage. Only reason could be to have a partial fallback for future XDP features that are not supported yet in the native implementation and we probably also shouldn't strive for such fallback and instead encourage native feature support in the first place. Given there's currently no such fallback issue or use case, lets not go there yet if we don't need to. Therefore, change semantics for loading XDP and bail out if the user tries to load a generic XDP program when a native one is present and vice versa. Another alternative to bailing out would be to handle the transition from one flavor to another gracefully, but that would require to bring the device down, exchange both types of programs, and bring it up again in order to avoid a tiny window where a packet could hit both hooks. Given this complicates the logic for just a debugging feature in the native case, I went with the simpler variant. For the dump, remove IFLA_XDP_FLAGS that was added with b5cdae3291f7 and reuse IFLA_XDP_ATTACHED for indicating the mode. Dumping all or just a subset of flags that were used for loading the XDP prog is suboptimal in the long run since not all flags are useful for dumping and if we start to reuse the same flag definitions for load and dump, then we'll waste bit space. What we really just want is to dump the mode for now. Current IFLA_XDP_ATTACHED semantics are: nothing was installed (0), a program is running at the native driver layer (1). Thus, add a mode that says that a program is running at generic XDP layer (2). Applications will handle this fine in that older binaries will just indicate that something is attached at XDP layer, effectively this is similar to IFLA_XDP_FLAGS attr that we would have had modulo the redundancy. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-12 06:04:46 +07:00
typedef int (*bpf_op_t)(struct net_device *dev, struct netdev_bpf *bpf);
xdp: refine xdp api with regards to generic xdp While working on the iproute2 generic XDP frontend, I noticed that as of right now it's possible to have native *and* generic XDP programs loaded both at the same time for the case when a driver supports native XDP. The intended model for generic XDP from b5cdae3291f7 ("net: Generic XDP") is, however, that only one out of the two can be present at once which is also indicated as such in the XDP netlink dump part. The main rationale for generic XDP is to ease accessibility (in case a driver does not yet have XDP support) and to generically provide a semantical model as an example for driver developers wanting to add XDP support. The generic XDP option for an XDP aware driver can still be useful for comparing and testing both implementations. However, it is not intended to have a second XDP processing stage or layer with exactly the same functionality of the first native stage. Only reason could be to have a partial fallback for future XDP features that are not supported yet in the native implementation and we probably also shouldn't strive for such fallback and instead encourage native feature support in the first place. Given there's currently no such fallback issue or use case, lets not go there yet if we don't need to. Therefore, change semantics for loading XDP and bail out if the user tries to load a generic XDP program when a native one is present and vice versa. Another alternative to bailing out would be to handle the transition from one flavor to another gracefully, but that would require to bring the device down, exchange both types of programs, and bring it up again in order to avoid a tiny window where a packet could hit both hooks. Given this complicates the logic for just a debugging feature in the native case, I went with the simpler variant. For the dump, remove IFLA_XDP_FLAGS that was added with b5cdae3291f7 and reuse IFLA_XDP_ATTACHED for indicating the mode. Dumping all or just a subset of flags that were used for loading the XDP prog is suboptimal in the long run since not all flags are useful for dumping and if we start to reuse the same flag definitions for load and dump, then we'll waste bit space. What we really just want is to dump the mode for now. Current IFLA_XDP_ATTACHED semantics are: nothing was installed (0), a program is running at the native driver layer (1). Thus, add a mode that says that a program is running at generic XDP layer (2). Applications will handle this fine in that older binaries will just indicate that something is attached at XDP layer, effectively this is similar to IFLA_XDP_FLAGS attr that we would have had modulo the redundancy. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-12 06:04:46 +07:00
int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
int fd, u32 flags);
u32 __dev_xdp_query(struct net_device *dev, bpf_op_t xdp_op,
enum bpf_netdev_command cmd);
int xdp_umem_query(struct net_device *dev, u16 queue_id);
xdp: refine xdp api with regards to generic xdp While working on the iproute2 generic XDP frontend, I noticed that as of right now it's possible to have native *and* generic XDP programs loaded both at the same time for the case when a driver supports native XDP. The intended model for generic XDP from b5cdae3291f7 ("net: Generic XDP") is, however, that only one out of the two can be present at once which is also indicated as such in the XDP netlink dump part. The main rationale for generic XDP is to ease accessibility (in case a driver does not yet have XDP support) and to generically provide a semantical model as an example for driver developers wanting to add XDP support. The generic XDP option for an XDP aware driver can still be useful for comparing and testing both implementations. However, it is not intended to have a second XDP processing stage or layer with exactly the same functionality of the first native stage. Only reason could be to have a partial fallback for future XDP features that are not supported yet in the native implementation and we probably also shouldn't strive for such fallback and instead encourage native feature support in the first place. Given there's currently no such fallback issue or use case, lets not go there yet if we don't need to. Therefore, change semantics for loading XDP and bail out if the user tries to load a generic XDP program when a native one is present and vice versa. Another alternative to bailing out would be to handle the transition from one flavor to another gracefully, but that would require to bring the device down, exchange both types of programs, and bring it up again in order to avoid a tiny window where a packet could hit both hooks. Given this complicates the logic for just a debugging feature in the native case, I went with the simpler variant. For the dump, remove IFLA_XDP_FLAGS that was added with b5cdae3291f7 and reuse IFLA_XDP_ATTACHED for indicating the mode. Dumping all or just a subset of flags that were used for loading the XDP prog is suboptimal in the long run since not all flags are useful for dumping and if we start to reuse the same flag definitions for load and dump, then we'll waste bit space. What we really just want is to dump the mode for now. Current IFLA_XDP_ATTACHED semantics are: nothing was installed (0), a program is running at the native driver layer (1). Thus, add a mode that says that a program is running at generic XDP layer (2). Applications will handle this fine in that older binaries will just indicate that something is attached at XDP layer, effectively this is similar to IFLA_XDP_FLAGS attr that we would have had modulo the redundancy. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-12 06:04:46 +07:00
int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb);
int dev_forward_skb(struct net_device *dev, struct sk_buff *skb);
bool is_skb_forwardable(const struct net_device *dev,
const struct sk_buff *skb);
static __always_inline int ____dev_forward_skb(struct net_device *dev,
struct sk_buff *skb)
{
if (skb_orphan_frags(skb, GFP_ATOMIC) ||
unlikely(!is_skb_forwardable(dev, skb))) {
atomic_long_inc(&dev->rx_dropped);
kfree_skb(skb);
return NET_RX_DROP;
}
skb_scrub_packet(skb, true);
skb->priority = 0;
return 0;
}
bool dev_nit_active(struct net_device *dev);
void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev);
extern int netdev_budget;
extern unsigned int netdev_budget_usecs;
/* Called by rtnetlink.c:rtnl_unlock() */
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 06:41:36 +07:00
/**
* 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 06:41:36 +07:00
*/
static inline void dev_put(struct net_device *dev)
{
this_cpu_dec(*dev->pcpu_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 06:41:36 +07:00
/**
* 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 06:41:36 +07:00
*/
static inline void dev_hold(struct net_device *dev)
{
this_cpu_inc(*dev->pcpu_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.
*/
void linkwatch_init_dev(struct net_device *dev);
void linkwatch_fire_event(struct net_device *dev);
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 06:41:36 +07:00
/**
* netif_carrier_ok - test if carrier present
* @dev: network device
*
* Check if carrier is present on device
*/
static inline bool netif_carrier_ok(const struct net_device *dev)
{
return !test_bit(__LINK_STATE_NOCARRIER, &dev->state);
}
unsigned long dev_trans_start(struct net_device *dev);
void __netdev_watchdog_up(struct net_device *dev);
void netif_carrier_on(struct net_device *dev);
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 06:41:36 +07:00
/**
* 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 06:41:36 +07:00
/**
* 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 06:41:36 +07:00
/**
* netif_dormant - test if device is dormant
[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 06:41:36 +07:00
* @dev: network device
*
* Check if device is dormant.
[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 06:41:36 +07:00
*/
static inline bool 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 06:41:36 +07:00
/**
* netif_oper_up - test if device is operational
* @dev: network device
*
* Check if carrier is operational
*/
static inline bool 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 06:41:36 +07:00
/**
* netif_device_present - is device available or removed
* @dev: network device
*
* Check if device has not been removed from system.
*/
static inline bool netif_device_present(struct net_device *dev)
{
return test_bit(__LINK_STATE_PRESENT, &dev->state);
}
void netif_device_detach(struct net_device *dev);
void netif_device_attach(struct net_device *dev);
/*
* Network interface message level settings
*/
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 (1U << 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>
2006-06-10 02:20:56 +07:00
{
spin_lock(&txq->_xmit_lock);
txq->xmit_lock_owner = cpu;
}
static inline bool __netif_tx_acquire(struct netdev_queue *txq)
{
__acquire(&txq->_xmit_lock);
return true;
}
static inline void __netif_tx_release(struct netdev_queue *txq)
{
__release(&txq->_xmit_lock);
}
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 bool __netif_tx_trylock(struct netdev_queue *txq)
{
bool 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;
}
/* legacy drivers only, netdev_start_xmit() sets txq->trans_start */
static inline void netif_trans_update(struct net_device *dev)
{
struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
if (txq->trans_start != jiffies)
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>
2006-06-10 02:20:56 +07:00
}
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>
2006-06-10 02:20:56 +07:00
}
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>
2006-06-10 02:20:56 +07:00
}
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>
2006-06-10 02:20:56 +07:00
}
#define HARD_TX_LOCK(dev, txq, cpu) { \
if ((dev->features & NETIF_F_LLTX) == 0) { \
__netif_tx_lock(txq, cpu); \
} else { \
__netif_tx_acquire(txq); \
} \
}
#define HARD_TX_TRYLOCK(dev, txq) \
(((dev->features & NETIF_F_LLTX) == 0) ? \
__netif_tx_trylock(txq) : \
__netif_tx_acquire(txq))
#define HARD_TX_UNLOCK(dev, txq) { \
if ((dev->features & NETIF_F_LLTX) == 0) { \
__netif_tx_unlock(txq); \
} else { \
__netif_tx_release(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_nested(struct net_device *dev)
{
int subclass = SINGLE_DEPTH_NESTING;
if (dev->netdev_ops->ndo_get_lock_subclass)
subclass = dev->netdev_ops->ndo_get_lock_subclass(dev);
spin_lock_nested(&dev->addr_list_lock, subclass);
}
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) */
void ether_setup(struct net_device *dev);
/* Support for loadable net-drivers */
struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
unsigned char name_assign_type,
void (*setup)(struct net_device *),
unsigned int txqs, unsigned int rxqs);
int dev_get_valid_name(struct net *net, struct net_device *dev,
const char *name);
#define alloc_netdev(sizeof_priv, name, name_assign_type, setup) \
alloc_netdev_mqs(sizeof_priv, name, name_assign_type, setup, 1, 1)
#define alloc_netdev_mq(sizeof_priv, name, name_assign_type, setup, count) \
alloc_netdev_mqs(sizeof_priv, name, name_assign_type, setup, count, \
count)
int register_netdev(struct net_device *dev);
void unregister_netdev(struct net_device *dev);
/* General hardware address lists handling functions */
int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
struct netdev_hw_addr_list *from_list, int addr_len);
void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
struct netdev_hw_addr_list *from_list, int addr_len);
int __hw_addr_sync_dev(struct netdev_hw_addr_list *list,
struct net_device *dev,
int (*sync)(struct net_device *, const unsigned char *),
int (*unsync)(struct net_device *,
const unsigned char *));
int __hw_addr_ref_sync_dev(struct netdev_hw_addr_list *list,
struct net_device *dev,
int (*sync)(struct net_device *,
const unsigned char *, int),
int (*unsync)(struct net_device *,
const unsigned char *, int));
void __hw_addr_ref_unsync_dev(struct netdev_hw_addr_list *list,
struct net_device *dev,
int (*unsync)(struct net_device *,
const unsigned char *, int));
void __hw_addr_unsync_dev(struct netdev_hw_addr_list *list,
struct net_device *dev,
int (*unsync)(struct net_device *,
const unsigned char *));
void __hw_addr_init(struct netdev_hw_addr_list *list);
/* Functions used for device addresses handling */
int dev_addr_add(struct net_device *dev, const unsigned char *addr,
unsigned char addr_type);
int dev_addr_del(struct net_device *dev, const unsigned char *addr,
unsigned char addr_type);
void dev_addr_flush(struct net_device *dev);
int dev_addr_init(struct net_device *dev);
/* Functions used for unicast addresses handling */
int dev_uc_add(struct net_device *dev, const unsigned char *addr);
int dev_uc_add_excl(struct net_device *dev, const unsigned char *addr);
int dev_uc_del(struct net_device *dev, const unsigned char *addr);
int dev_uc_sync(struct net_device *to, struct net_device *from);
int dev_uc_sync_multiple(struct net_device *to, struct net_device *from);
void dev_uc_unsync(struct net_device *to, struct net_device *from);
void dev_uc_flush(struct net_device *dev);
void dev_uc_init(struct net_device *dev);
/**
* __dev_uc_sync - Synchonize device's unicast list
* @dev: device to sync
* @sync: function to call if address should be added
* @unsync: function to call if address should be removed
*
* Add newly added addresses to the interface, and release
* addresses that have been deleted.
*/
static inline int __dev_uc_sync(struct net_device *dev,
int (*sync)(struct net_device *,
const unsigned char *),
int (*unsync)(struct net_device *,
const unsigned char *))
{
return __hw_addr_sync_dev(&dev->uc, dev, sync, unsync);
}
/**
* __dev_uc_unsync - Remove synchronized addresses from device
* @dev: device to sync
* @unsync: function to call if address should be removed
*
* Remove all addresses that were added to the device by dev_uc_sync().
*/
static inline void __dev_uc_unsync(struct net_device *dev,
int (*unsync)(struct net_device *,
const unsigned char *))
{
__hw_addr_unsync_dev(&dev->uc, dev, unsync);
}
/* Functions used for multicast addresses handling */
int dev_mc_add(struct net_device *dev, const unsigned char *addr);
int dev_mc_add_global(struct net_device *dev, const unsigned char *addr);
int dev_mc_add_excl(struct net_device *dev, const unsigned char *addr);
int dev_mc_del(struct net_device *dev, const unsigned char *addr);
int dev_mc_del_global(struct net_device *dev, const unsigned char *addr);
int dev_mc_sync(struct net_device *to, struct net_device *from);
int dev_mc_sync_multiple(struct net_device *to, struct net_device *from);
void dev_mc_unsync(struct net_device *to, struct net_device *from);
void dev_mc_flush(struct net_device *dev);
void dev_mc_init(struct net_device *dev);
/**
* __dev_mc_sync - Synchonize device's multicast list
* @dev: device to sync
* @sync: function to call if address should be added
* @unsync: function to call if address should be removed
*
* Add newly added addresses to the interface, and release
* addresses that have been deleted.
*/
static inline int __dev_mc_sync(struct net_device *dev,
int (*sync)(struct net_device *,
const unsigned char *),
int (*unsync)(struct net_device *,
const unsigned char *))
{
return __hw_addr_sync_dev(&dev->mc, dev, sync, unsync);
}
/**
* __dev_mc_unsync - Remove synchronized addresses from device
* @dev: device to sync
* @unsync: function to call if address should be removed
*
* Remove all addresses that were added to the device by dev_mc_sync().
*/
static inline void __dev_mc_unsync(struct net_device *dev,
int (*unsync)(struct net_device *,
const unsigned char *))
{
__hw_addr_unsync_dev(&dev->mc, dev, unsync);
}
/* Functions used for secondary unicast and multicast support */
void dev_set_rx_mode(struct net_device *dev);
void __dev_set_rx_mode(struct net_device *dev);
int dev_set_promiscuity(struct net_device *dev, int inc);
int dev_set_allmulti(struct net_device *dev, int inc);
void netdev_state_change(struct net_device *dev);
void netdev_notify_peers(struct net_device *dev);
void netdev_features_change(struct net_device *dev);
/* Load a device via the kmod */
void dev_load(struct net *net, const char *name);
struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
struct rtnl_link_stats64 *storage);
void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
const struct net_device_stats *netdev_stats);
extern int netdev_max_backlog;
extern int netdev_tstamp_prequeue;
extern int weight_p;
extern int dev_weight_rx_bias;
extern int dev_weight_tx_bias;
extern int dev_rx_weight;
extern int dev_tx_weight;
bool netdev_has_upper_dev(struct net_device *dev, struct net_device *upper_dev);
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
struct list_head **iter);
struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
struct list_head **iter);
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
/* iterate through upper list, must be called under RCU read lock */
#define netdev_for_each_upper_dev_rcu(dev, updev, iter) \
for (iter = &(dev)->adj_list.upper, \
updev = netdev_upper_get_next_dev_rcu(dev, &(iter)); \
updev; \
updev = netdev_upper_get_next_dev_rcu(dev, &(iter)))
int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
int (*fn)(struct net_device *upper_dev,
void *data),
void *data);
bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
struct net_device *upper_dev);
bool netdev_has_any_upper_dev(struct net_device *dev);
void *netdev_lower_get_next_private(struct net_device *dev,
struct list_head **iter);
void *netdev_lower_get_next_private_rcu(struct net_device *dev,
struct list_head **iter);
#define netdev_for_each_lower_private(dev, priv, iter) \
for (iter = (dev)->adj_list.lower.next, \
priv = netdev_lower_get_next_private(dev, &(iter)); \
priv; \
priv = netdev_lower_get_next_private(dev, &(iter)))
#define netdev_for_each_lower_private_rcu(dev, priv, iter) \
for (iter = &(dev)->adj_list.lower, \
priv = netdev_lower_get_next_private_rcu(dev, &(iter)); \
priv; \
priv = netdev_lower_get_next_private_rcu(dev, &(iter)))
void *netdev_lower_get_next(struct net_device *dev,
struct list_head **iter);
#define netdev_for_each_lower_dev(dev, ldev, iter) \
net: make netdev_for_each_lower_dev safe for device removal When I used netdev_for_each_lower_dev in commit bad531623253 ("vrf: remove slave queue and private slave struct") I thought that it acts like netdev_for_each_lower_private and can be used to remove the current device from the list while walking, but unfortunately it acts more like netdev_for_each_lower_private_rcu and doesn't allow it. The difference is where the "iter" points to, right now it points to the current element and that makes it impossible to remove it. Change the logic to be similar to netdev_for_each_lower_private and make it point to the "next" element so we can safely delete the current one. VRF is the only such user right now, there's no change for the read-only users. Here's what can happen now: [98423.249858] general protection fault: 0000 [#1] SMP [98423.250175] Modules linked in: vrf bridge(O) stp llc nfsd auth_rpcgss oid_registry nfs_acl nfs lockd grace sunrpc crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel jitterentropy_rng sha256_generic hmac drbg ppdev aesni_intel aes_x86_64 glue_helper lrw gf128mul ablk_helper cryptd evdev serio_raw pcspkr virtio_balloon parport_pc parport i2c_piix4 i2c_core virtio_console acpi_cpufreq button 9pnet_virtio 9p 9pnet fscache ipv6 autofs4 ext4 crc16 mbcache jbd2 sg virtio_blk virtio_net sr_mod cdrom e1000 ata_generic ehci_pci uhci_hcd ehci_hcd usbcore usb_common virtio_pci ata_piix libata floppy virtio_ring virtio scsi_mod [last unloaded: bridge] [98423.255040] CPU: 1 PID: 14173 Comm: ip Tainted: G O 4.5.0-rc2+ #81 [98423.255386] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 [98423.255777] task: ffff8800547f5540 ti: ffff88003428c000 task.ti: ffff88003428c000 [98423.256123] RIP: 0010:[<ffffffff81514f3e>] [<ffffffff81514f3e>] netdev_lower_get_next+0x1e/0x30 [98423.256534] RSP: 0018:ffff88003428f940 EFLAGS: 00010207 [98423.256766] RAX: 0002000100000004 RBX: ffff880054ff9000 RCX: 0000000000000000 [98423.257039] RDX: ffff88003428f8b8 RSI: ffff88003428f950 RDI: ffff880054ff90c0 [98423.257287] RBP: ffff88003428f940 R08: 0000000000000000 R09: 0000000000000000 [98423.257537] R10: 0000000000000001 R11: 0000000000000000 R12: ffff88003428f9e0 [98423.257802] R13: ffff880054a5fd00 R14: ffff88003428f970 R15: 0000000000000001 [98423.258055] FS: 00007f3d76881700(0000) GS:ffff88005d000000(0000) knlGS:0000000000000000 [98423.258418] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [98423.258650] CR2: 00007ffe5951ffa8 CR3: 0000000052077000 CR4: 00000000000406e0 [98423.258902] Stack: [98423.259075] ffff88003428f960 ffffffffa0442636 0002000100000004 ffff880054ff9000 [98423.259647] ffff88003428f9b0 ffffffff81518205 ffff880054ff9000 ffff88003428f978 [98423.260208] ffff88003428f978 ffff88003428f9e0 ffff88003428f9e0 ffff880035b35f00 [98423.260739] Call Trace: [98423.260920] [<ffffffffa0442636>] vrf_dev_uninit+0x76/0xa0 [vrf] [98423.261156] [<ffffffff81518205>] rollback_registered_many+0x205/0x390 [98423.261401] [<ffffffff815183ec>] unregister_netdevice_many+0x1c/0x70 [98423.261641] [<ffffffff8153223c>] rtnl_delete_link+0x3c/0x50 [98423.271557] [<ffffffff815335bb>] rtnl_dellink+0xcb/0x1d0 [98423.271800] [<ffffffff811cd7da>] ? __inc_zone_state+0x4a/0x90 [98423.272049] [<ffffffff815337b4>] rtnetlink_rcv_msg+0x84/0x200 [98423.272279] [<ffffffff810cfe7d>] ? trace_hardirqs_on+0xd/0x10 [98423.272513] [<ffffffff8153370b>] ? rtnetlink_rcv+0x1b/0x40 [98423.272755] [<ffffffff81533730>] ? rtnetlink_rcv+0x40/0x40 [98423.272983] [<ffffffff8155d6e7>] netlink_rcv_skb+0x97/0xb0 [98423.273209] [<ffffffff8153371a>] rtnetlink_rcv+0x2a/0x40 [98423.273476] [<ffffffff8155ce8b>] netlink_unicast+0x11b/0x1a0 [98423.273710] [<ffffffff8155d2f1>] netlink_sendmsg+0x3e1/0x610 [98423.273947] [<ffffffff814fbc98>] sock_sendmsg+0x38/0x70 [98423.274175] [<ffffffff814fc253>] ___sys_sendmsg+0x2e3/0x2f0 [98423.274416] [<ffffffff810d841e>] ? do_raw_spin_unlock+0xbe/0x140 [98423.274658] [<ffffffff811e1bec>] ? handle_mm_fault+0x26c/0x2210 [98423.274894] [<ffffffff811e19cd>] ? handle_mm_fault+0x4d/0x2210 [98423.275130] [<ffffffff81269611>] ? __fget_light+0x91/0xb0 [98423.275365] [<ffffffff814fcd42>] __sys_sendmsg+0x42/0x80 [98423.275595] [<ffffffff814fcd92>] SyS_sendmsg+0x12/0x20 [98423.275827] [<ffffffff81611bb6>] entry_SYSCALL_64_fastpath+0x16/0x7a [98423.276073] Code: c3 31 c0 5d c3 0f 1f 84 00 00 00 00 00 66 66 66 66 90 48 8b 06 55 48 81 c7 c0 00 00 00 48 89 e5 48 8b 00 48 39 f8 74 09 48 89 06 <48> 8b 40 e8 5d c3 31 c0 5d c3 0f 1f 84 00 00 00 00 00 66 66 66 [98423.279639] RIP [<ffffffff81514f3e>] netdev_lower_get_next+0x1e/0x30 [98423.279920] RSP <ffff88003428f940> CC: David Ahern <dsa@cumulusnetworks.com> CC: David S. Miller <davem@davemloft.net> CC: Roopa Prabhu <roopa@cumulusnetworks.com> CC: Vlad Yasevich <vyasevic@redhat.com> Fixes: bad531623253 ("vrf: remove slave queue and private slave struct") Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Reviewed-by: David Ahern <dsa@cumulusnetworks.com> Tested-by: David Ahern <dsa@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-18 00:00:31 +07:00
for (iter = (dev)->adj_list.lower.next, \
ldev = netdev_lower_get_next(dev, &(iter)); \
ldev; \
ldev = netdev_lower_get_next(dev, &(iter)))
struct net_device *netdev_all_lower_get_next(struct net_device *dev,
struct list_head **iter);
struct net_device *netdev_all_lower_get_next_rcu(struct net_device *dev,
struct list_head **iter);
int netdev_walk_all_lower_dev(struct net_device *dev,
int (*fn)(struct net_device *lower_dev,
void *data),
void *data);
int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
int (*fn)(struct net_device *lower_dev,
void *data),
void *data);
void *netdev_adjacent_get_private(struct list_head *adj_list);
void *netdev_lower_get_first_private_rcu(struct net_device *dev);
struct net_device *netdev_master_upper_dev_get(struct net_device *dev);
struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev);
int netdev_upper_dev_link(struct net_device *dev, struct net_device *upper_dev,
struct netlink_ext_ack *extack);
int netdev_master_upper_dev_link(struct net_device *dev,
struct net_device *upper_dev,
void *upper_priv, void *upper_info,
struct netlink_ext_ack *extack);
void netdev_upper_dev_unlink(struct net_device *dev,
struct net_device *upper_dev);
void netdev_adjacent_rename_links(struct net_device *dev, char *oldname);
void *netdev_lower_dev_get_private(struct net_device *dev,
struct net_device *lower_dev);
void netdev_lower_state_changed(struct net_device *lower_dev,
void *lower_state_info);
/* RSS keys are 40 or 52 bytes long */
#define NETDEV_RSS_KEY_LEN 52
extern u8 netdev_rss_key[NETDEV_RSS_KEY_LEN] __read_mostly;
void netdev_rss_key_fill(void *buffer, size_t len);
int dev_get_nest_level(struct net_device *dev);
int skb_checksum_help(struct sk_buff *skb);
int skb_crc32c_csum_help(struct sk_buff *skb);
int skb_csum_hwoffload_help(struct sk_buff *skb,
const netdev_features_t features);
struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
netdev_features_t features, bool tx_path);
struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
netdev_features_t features);
struct netdev_bonding_info {
ifslave slave;
ifbond master;
};
struct netdev_notifier_bonding_info {
struct netdev_notifier_info info; /* must be first */
struct netdev_bonding_info bonding_info;
};
void netdev_bonding_info_change(struct net_device *dev,
struct netdev_bonding_info *bonding_info);
static inline
struct sk_buff *skb_gso_segment(struct sk_buff *skb, netdev_features_t features)
{
return __skb_gso_segment(skb, features, true);
}
__be16 skb_network_protocol(struct sk_buff *skb, int *depth);
static inline bool can_checksum_protocol(netdev_features_t features,
__be16 protocol)
{
if (protocol == htons(ETH_P_FCOE))
return !!(features & NETIF_F_FCOE_CRC);
/* Assume this is an IP checksum (not SCTP CRC) */
if (features & NETIF_F_HW_CSUM) {
/* Can checksum everything */
return true;
}
switch (protocol) {
case htons(ETH_P_IP):
return !!(features & NETIF_F_IP_CSUM);
case htons(ETH_P_IPV6):
return !!(features & NETIF_F_IPV6_CSUM);
default:
return false;
}
}
#ifdef CONFIG_BUG
void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb);
#else
static inline void netdev_rx_csum_fault(struct net_device *dev,
struct sk_buff *skb)
{
}
#endif
/* rx skb timestamps */
void net_enable_timestamp(void);
void net_disable_timestamp(void);
#ifdef CONFIG_PROC_FS
int __init dev_proc_init(void);
#else
#define dev_proc_init() 0
#endif
static inline netdev_tx_t __netdev_start_xmit(const struct net_device_ops *ops,
struct sk_buff *skb, struct net_device *dev,
bool more)
{
__this_cpu_write(softnet_data.xmit.more, more);
return ops->ndo_start_xmit(skb, dev);
}
static inline bool netdev_xmit_more(void)
{
return __this_cpu_read(softnet_data.xmit.more);
}
static inline netdev_tx_t netdev_start_xmit(struct sk_buff *skb, struct net_device *dev,
struct netdev_queue *txq, bool more)
{
const struct net_device_ops *ops = dev->netdev_ops;
netdev_tx_t rc;
rc = __netdev_start_xmit(ops, skb, dev, more);
if (rc == NETDEV_TX_OK)
txq_trans_update(txq);
return rc;
}
int netdev_class_create_file_ns(const struct class_attribute *class_attr,
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next Pull networking updates from David Miller: 1) The addition of nftables. No longer will we need protocol aware firewall filtering modules, it can all live in userspace. At the core of nftables is a, for lack of a better term, virtual machine that executes byte codes to inspect packet or metadata (arriving interface index, etc.) and make verdict decisions. Besides support for loading packet contents and comparing them, the interpreter supports lookups in various datastructures as fundamental operations. For example sets are supports, and therefore one could create a set of whitelist IP address entries which have ACCEPT verdicts attached to them, and use the appropriate byte codes to do such lookups. Since the interpreted code is composed in userspace, userspace can do things like optimize things before giving it to the kernel. Another major improvement is the capability of atomically updating portions of the ruleset. In the existing netfilter implementation, one has to update the entire rule set in order to make a change and this is very expensive. Userspace tools exist to create nftables rules using existing netfilter rule sets, but both kernel implementations will need to co-exist for quite some time as we transition from the old to the new stuff. Kudos to Patrick McHardy, Pablo Neira Ayuso, and others who have worked so hard on this. 2) Daniel Borkmann and Hannes Frederic Sowa made several improvements to our pseudo-random number generator, mostly used for things like UDP port randomization and netfitler, amongst other things. In particular the taus88 generater is updated to taus113, and test cases are added. 3) Support 64-bit rates in HTB and TBF schedulers, from Eric Dumazet and Yang Yingliang. 4) Add support for new 577xx tigon3 chips to tg3 driver, from Nithin Sujir. 5) Fix two fatal flaws in TCP dynamic right sizing, from Eric Dumazet, Neal Cardwell, and Yuchung Cheng. 6) Allow IP_TOS and IP_TTL to be specified in sendmsg() ancillary control message data, much like other socket option attributes. From Francesco Fusco. 7) Allow applications to specify a cap on the rate computed automatically by the kernel for pacing flows, via a new SO_MAX_PACING_RATE socket option. From Eric Dumazet. 8) Make the initial autotuned send buffer sizing in TCP more closely reflect actual needs, from Eric Dumazet. 9) Currently early socket demux only happens for TCP sockets, but we can do it for connected UDP sockets too. Implementation from Shawn Bohrer. 10) Refactor inet socket demux with the goal of improving hash demux performance for listening sockets. With the main goals being able to use RCU lookups on even request sockets, and eliminating the listening lock contention. From Eric Dumazet. 11) The bonding layer has many demuxes in it's fast path, and an RCU conversion was started back in 3.11, several changes here extend the RCU usage to even more locations. From Ding Tianhong and Wang Yufen, based upon suggestions by Nikolay Aleksandrov and Veaceslav Falico. 12) Allow stackability of segmentation offloads to, in particular, allow segmentation offloading over tunnels. From Eric Dumazet. 13) Significantly improve the handling of secret keys we input into the various hash functions in the inet hashtables, TCP fast open, as well as syncookies. From Hannes Frederic Sowa. The key fundamental operation is "net_get_random_once()" which uses static keys. Hannes even extended this to ipv4/ipv6 fragmentation handling and our generic flow dissector. 14) The generic driver layer takes care now to set the driver data to NULL on device removal, so it's no longer necessary for drivers to explicitly set it to NULL any more. Many drivers have been cleaned up in this way, from Jingoo Han. 15) Add a BPF based packet scheduler classifier, from Daniel Borkmann. 16) Improve CRC32 interfaces and generic SKB checksum iterators so that SCTP's checksumming can more cleanly be handled. Also from Daniel Borkmann. 17) Add a new PMTU discovery mode, IP_PMTUDISC_INTERFACE, which forces using the interface MTU value. This helps avoid PMTU attacks, particularly on DNS servers. From Hannes Frederic Sowa. 18) Use generic XPS for transmit queue steering rather than internal (re-)implementation in virtio-net. From Jason Wang. * git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1622 commits) random32: add test cases for taus113 implementation random32: upgrade taus88 generator to taus113 from errata paper random32: move rnd_state to linux/random.h random32: add prandom_reseed_late() and call when nonblocking pool becomes initialized random32: add periodic reseeding random32: fix off-by-one in seeding requirement PHY: Add RTL8201CP phy_driver to realtek xtsonic: add missing platform_set_drvdata() in xtsonic_probe() macmace: add missing platform_set_drvdata() in mace_probe() ethernet/arc/arc_emac: add missing platform_set_drvdata() in arc_emac_probe() ipv6: protect for_each_sk_fl_rcu in mem_check with rcu_read_lock_bh vlan: Implement vlan_dev_get_egress_qos_mask as an inline. ixgbe: add warning when max_vfs is out of range. igb: Update link modes display in ethtool netfilter: push reasm skb through instead of original frag skbs ip6_output: fragment outgoing reassembled skb properly MAINTAINERS: mv643xx_eth: take over maintainership from Lennart net_sched: tbf: support of 64bit rates ixgbe: deleting dfwd stations out of order can cause null ptr deref ixgbe: fix build err, num_rx_queues is only available with CONFIG_RPS ...
2013-11-13 15:40:34 +07:00
const void *ns);
void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next Pull networking updates from David Miller: 1) The addition of nftables. No longer will we need protocol aware firewall filtering modules, it can all live in userspace. At the core of nftables is a, for lack of a better term, virtual machine that executes byte codes to inspect packet or metadata (arriving interface index, etc.) and make verdict decisions. Besides support for loading packet contents and comparing them, the interpreter supports lookups in various datastructures as fundamental operations. For example sets are supports, and therefore one could create a set of whitelist IP address entries which have ACCEPT verdicts attached to them, and use the appropriate byte codes to do such lookups. Since the interpreted code is composed in userspace, userspace can do things like optimize things before giving it to the kernel. Another major improvement is the capability of atomically updating portions of the ruleset. In the existing netfilter implementation, one has to update the entire rule set in order to make a change and this is very expensive. Userspace tools exist to create nftables rules using existing netfilter rule sets, but both kernel implementations will need to co-exist for quite some time as we transition from the old to the new stuff. Kudos to Patrick McHardy, Pablo Neira Ayuso, and others who have worked so hard on this. 2) Daniel Borkmann and Hannes Frederic Sowa made several improvements to our pseudo-random number generator, mostly used for things like UDP port randomization and netfitler, amongst other things. In particular the taus88 generater is updated to taus113, and test cases are added. 3) Support 64-bit rates in HTB and TBF schedulers, from Eric Dumazet and Yang Yingliang. 4) Add support for new 577xx tigon3 chips to tg3 driver, from Nithin Sujir. 5) Fix two fatal flaws in TCP dynamic right sizing, from Eric Dumazet, Neal Cardwell, and Yuchung Cheng. 6) Allow IP_TOS and IP_TTL to be specified in sendmsg() ancillary control message data, much like other socket option attributes. From Francesco Fusco. 7) Allow applications to specify a cap on the rate computed automatically by the kernel for pacing flows, via a new SO_MAX_PACING_RATE socket option. From Eric Dumazet. 8) Make the initial autotuned send buffer sizing in TCP more closely reflect actual needs, from Eric Dumazet. 9) Currently early socket demux only happens for TCP sockets, but we can do it for connected UDP sockets too. Implementation from Shawn Bohrer. 10) Refactor inet socket demux with the goal of improving hash demux performance for listening sockets. With the main goals being able to use RCU lookups on even request sockets, and eliminating the listening lock contention. From Eric Dumazet. 11) The bonding layer has many demuxes in it's fast path, and an RCU conversion was started back in 3.11, several changes here extend the RCU usage to even more locations. From Ding Tianhong and Wang Yufen, based upon suggestions by Nikolay Aleksandrov and Veaceslav Falico. 12) Allow stackability of segmentation offloads to, in particular, allow segmentation offloading over tunnels. From Eric Dumazet. 13) Significantly improve the handling of secret keys we input into the various hash functions in the inet hashtables, TCP fast open, as well as syncookies. From Hannes Frederic Sowa. The key fundamental operation is "net_get_random_once()" which uses static keys. Hannes even extended this to ipv4/ipv6 fragmentation handling and our generic flow dissector. 14) The generic driver layer takes care now to set the driver data to NULL on device removal, so it's no longer necessary for drivers to explicitly set it to NULL any more. Many drivers have been cleaned up in this way, from Jingoo Han. 15) Add a BPF based packet scheduler classifier, from Daniel Borkmann. 16) Improve CRC32 interfaces and generic SKB checksum iterators so that SCTP's checksumming can more cleanly be handled. Also from Daniel Borkmann. 17) Add a new PMTU discovery mode, IP_PMTUDISC_INTERFACE, which forces using the interface MTU value. This helps avoid PMTU attacks, particularly on DNS servers. From Hannes Frederic Sowa. 18) Use generic XPS for transmit queue steering rather than internal (re-)implementation in virtio-net. From Jason Wang. * git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1622 commits) random32: add test cases for taus113 implementation random32: upgrade taus88 generator to taus113 from errata paper random32: move rnd_state to linux/random.h random32: add prandom_reseed_late() and call when nonblocking pool becomes initialized random32: add periodic reseeding random32: fix off-by-one in seeding requirement PHY: Add RTL8201CP phy_driver to realtek xtsonic: add missing platform_set_drvdata() in xtsonic_probe() macmace: add missing platform_set_drvdata() in mace_probe() ethernet/arc/arc_emac: add missing platform_set_drvdata() in arc_emac_probe() ipv6: protect for_each_sk_fl_rcu in mem_check with rcu_read_lock_bh vlan: Implement vlan_dev_get_egress_qos_mask as an inline. ixgbe: add warning when max_vfs is out of range. igb: Update link modes display in ethtool netfilter: push reasm skb through instead of original frag skbs ip6_output: fragment outgoing reassembled skb properly MAINTAINERS: mv643xx_eth: take over maintainership from Lennart net_sched: tbf: support of 64bit rates ixgbe: deleting dfwd stations out of order can cause null ptr deref ixgbe: fix build err, num_rx_queues is only available with CONFIG_RPS ...
2013-11-13 15:40:34 +07:00
const void *ns);
sysfs: make attr namespace interface less convoluted sysfs ns (namespace) implementation became more convoluted than necessary while trying to hide ns information from visible interface. The relatively recent attr ns support is a good example. * attr ns tag is determined by sysfs_ops->namespace() callback while dir tag is determined by kobj_type->namespace(). The placement is arbitrary. * Instead of performing operations with explicit ns tag, the namespace callback is routed through sysfs_attr_ns(), sysfs_ops->namespace(), class_attr_namespace(), class_attr->namespace(). It's not simpler in any sense. The only thing this convolution does is traversing the whole stack backwards. The namespace callbacks are unncessary because the operations involved are inherently synchronous. The information can be provided in in straight-forward top-down direction and reversing that direction is unnecessary and against basic design principles. This backward interface is unnecessarily convoluted and hinders properly separating out sysfs from driver model / kobject for proper layering. This patch updates attr ns support such that * sysfs_ops->namespace() and class_attr->namespace() are dropped. * sysfs_{create|remove}_file_ns(), which take explicit @ns param, are added and sysfs_{create|remove}_file() are now simple wrappers around the ns aware functions. * ns handling is dropped from sysfs_chmod_file(). Nobody uses it at this point. sysfs_chmod_file_ns() can be added later if necessary. * Explicit @ns is propagated through class_{create|remove}_file_ns() and netdev_class_{create|remove}_file_ns(). * driver/net/bonding which is currently the only user of attr namespace is updated to use netdev_class_{create|remove}_file_ns() with @bh->net as the ns tag instead of using the namespace callback. This patch should be an equivalent conversion without any functional difference. It makes the code easier to follow, reduces lines of code a bit and helps proper separation and layering. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kay Sievers <kay@vrfy.org> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-09-12 09:29:04 +07:00
static inline int netdev_class_create_file(const struct class_attribute *class_attr)
sysfs: make attr namespace interface less convoluted sysfs ns (namespace) implementation became more convoluted than necessary while trying to hide ns information from visible interface. The relatively recent attr ns support is a good example. * attr ns tag is determined by sysfs_ops->namespace() callback while dir tag is determined by kobj_type->namespace(). The placement is arbitrary. * Instead of performing operations with explicit ns tag, the namespace callback is routed through sysfs_attr_ns(), sysfs_ops->namespace(), class_attr_namespace(), class_attr->namespace(). It's not simpler in any sense. The only thing this convolution does is traversing the whole stack backwards. The namespace callbacks are unncessary because the operations involved are inherently synchronous. The information can be provided in in straight-forward top-down direction and reversing that direction is unnecessary and against basic design principles. This backward interface is unnecessarily convoluted and hinders properly separating out sysfs from driver model / kobject for proper layering. This patch updates attr ns support such that * sysfs_ops->namespace() and class_attr->namespace() are dropped. * sysfs_{create|remove}_file_ns(), which take explicit @ns param, are added and sysfs_{create|remove}_file() are now simple wrappers around the ns aware functions. * ns handling is dropped from sysfs_chmod_file(). Nobody uses it at this point. sysfs_chmod_file_ns() can be added later if necessary. * Explicit @ns is propagated through class_{create|remove}_file_ns() and netdev_class_{create|remove}_file_ns(). * driver/net/bonding which is currently the only user of attr namespace is updated to use netdev_class_{create|remove}_file_ns() with @bh->net as the ns tag instead of using the namespace callback. This patch should be an equivalent conversion without any functional difference. It makes the code easier to follow, reduces lines of code a bit and helps proper separation and layering. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kay Sievers <kay@vrfy.org> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-09-12 09:29:04 +07:00
{
return netdev_class_create_file_ns(class_attr, NULL);
}
static inline void netdev_class_remove_file(const struct class_attribute *class_attr)
sysfs: make attr namespace interface less convoluted sysfs ns (namespace) implementation became more convoluted than necessary while trying to hide ns information from visible interface. The relatively recent attr ns support is a good example. * attr ns tag is determined by sysfs_ops->namespace() callback while dir tag is determined by kobj_type->namespace(). The placement is arbitrary. * Instead of performing operations with explicit ns tag, the namespace callback is routed through sysfs_attr_ns(), sysfs_ops->namespace(), class_attr_namespace(), class_attr->namespace(). It's not simpler in any sense. The only thing this convolution does is traversing the whole stack backwards. The namespace callbacks are unncessary because the operations involved are inherently synchronous. The information can be provided in in straight-forward top-down direction and reversing that direction is unnecessary and against basic design principles. This backward interface is unnecessarily convoluted and hinders properly separating out sysfs from driver model / kobject for proper layering. This patch updates attr ns support such that * sysfs_ops->namespace() and class_attr->namespace() are dropped. * sysfs_{create|remove}_file_ns(), which take explicit @ns param, are added and sysfs_{create|remove}_file() are now simple wrappers around the ns aware functions. * ns handling is dropped from sysfs_chmod_file(). Nobody uses it at this point. sysfs_chmod_file_ns() can be added later if necessary. * Explicit @ns is propagated through class_{create|remove}_file_ns() and netdev_class_{create|remove}_file_ns(). * driver/net/bonding which is currently the only user of attr namespace is updated to use netdev_class_{create|remove}_file_ns() with @bh->net as the ns tag instead of using the namespace callback. This patch should be an equivalent conversion without any functional difference. It makes the code easier to follow, reduces lines of code a bit and helps proper separation and layering. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kay Sievers <kay@vrfy.org> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-09-12 09:29:04 +07:00
{
netdev_class_remove_file_ns(class_attr, NULL);
}
extern const struct kobj_ns_type_operations net_ns_type_operations;
const char *netdev_drivername(const struct net_device *dev);
void linkwatch_run_queue(void);
static inline netdev_features_t netdev_intersect_features(netdev_features_t f1,
netdev_features_t f2)
{
if ((f1 ^ f2) & NETIF_F_HW_CSUM) {
if (f1 & NETIF_F_HW_CSUM)
f1 |= (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
else
f2 |= (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
}
return f1 & f2;
}
static inline netdev_features_t netdev_get_wanted_features(
struct net_device *dev)
{
return (dev->features & ~dev->hw_features) | dev->wanted_features;
}
netdev_features_t netdev_increment_features(netdev_features_t all,
netdev_features_t one, netdev_features_t mask);
/* Allow TSO being used on stacked device :
* Performing the GSO segmentation before last device
* is a performance improvement.
*/
static inline netdev_features_t netdev_add_tso_features(netdev_features_t features,
netdev_features_t mask)
{
return netdev_increment_features(features, NETIF_F_ALL_TSO, mask);
}
int __netdev_update_features(struct net_device *dev);
void netdev_update_features(struct net_device *dev);
void netdev_change_features(struct net_device *dev);
void netif_stacked_transfer_operstate(const struct net_device *rootdev,
struct net_device *dev);
netdev_features_t passthru_features_check(struct sk_buff *skb,
struct net_device *dev,
netdev_features_t features);
netdev_features_t netif_skb_features(struct sk_buff *skb);
static inline bool net_gso_ok(netdev_features_t features, int gso_type)
{
netdev_features_t feature = (netdev_features_t)gso_type << NETIF_F_GSO_SHIFT;
/* check flags correspondence */
BUILD_BUG_ON(SKB_GSO_TCPV4 != (NETIF_F_TSO >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_DODGY != (NETIF_F_GSO_ROBUST >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_TCP_ECN != (NETIF_F_TSO_ECN >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_TCP_FIXEDID != (NETIF_F_TSO_MANGLEID >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_TCPV6 != (NETIF_F_TSO6 >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_FCOE != (NETIF_F_FSO >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_GRE != (NETIF_F_GSO_GRE >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_GRE_CSUM != (NETIF_F_GSO_GRE_CSUM >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_IPXIP4 != (NETIF_F_GSO_IPXIP4 >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_IPXIP6 != (NETIF_F_GSO_IPXIP6 >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_UDP_TUNNEL != (NETIF_F_GSO_UDP_TUNNEL >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_UDP_TUNNEL_CSUM != (NETIF_F_GSO_UDP_TUNNEL_CSUM >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_PARTIAL != (NETIF_F_GSO_PARTIAL >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_TUNNEL_REMCSUM != (NETIF_F_GSO_TUNNEL_REMCSUM >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_SCTP != (NETIF_F_GSO_SCTP >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_ESP != (NETIF_F_GSO_ESP >> NETIF_F_GSO_SHIFT));
net: accept UFO datagrams from tuntap and packet Tuntap and similar devices can inject GSO packets. Accept type VIRTIO_NET_HDR_GSO_UDP, even though not generating UFO natively. Processes are expected to use feature negotiation such as TUNSETOFFLOAD to detect supported offload types and refrain from injecting other packets. This process breaks down with live migration: guest kernels do not renegotiate flags, so destination hosts need to expose all features that the source host does. Partially revert the UFO removal from 182e0b6b5846~1..d9d30adf5677. This patch introduces nearly(*) no new code to simplify verification. It brings back verbatim tuntap UFO negotiation, VIRTIO_NET_HDR_GSO_UDP insertion and software UFO segmentation. It does not reinstate protocol stack support, hardware offload (NETIF_F_UFO), SKB_GSO_UDP tunneling in SKB_GSO_SOFTWARE or reception of VIRTIO_NET_HDR_GSO_UDP packets in tuntap. To support SKB_GSO_UDP reappearing in the stack, also reinstate logic in act_csum and openvswitch. Achieve equivalence with v4.13 HEAD by squashing in commit 939912216fa8 ("net: skb_needs_check() removes CHECKSUM_UNNECESSARY check for tx.") and reverting commit 8d63bee643f1 ("net: avoid skb_warn_bad_offload false positives on UFO"). (*) To avoid having to bring back skb_shinfo(skb)->ip6_frag_id, ipv6_proxy_select_ident is changed to return a __be32 and this is assigned directly to the frag_hdr. Also, SKB_GSO_UDP is inserted at the end of the enum to minimize code churn. Tested Booted a v4.13 guest kernel with QEMU. On a host kernel before this patch `ethtool -k eth0` shows UFO disabled. After the patch, it is enabled, same as on a v4.13 host kernel. A UFO packet sent from the guest appears on the tap device: host: nc -l -p -u 8000 & tcpdump -n -i tap0 guest: dd if=/dev/zero of=payload.txt bs=1 count=2000 nc -u 192.16.1.1 8000 < payload.txt Direct tap to tap transmission of VIRTIO_NET_HDR_GSO_UDP succeeds, packets arriving fragmented: ./with_tap_pair.sh ./tap_send_ufo tap0 tap1 (from https://github.com/wdebruij/kerneltools/tree/master/tests) Changes v1 -> v2 - simplified set_offload change (review comment) - documented test procedure Link: http://lkml.kernel.org/r/<CAF=yD-LuUeDuL9YWPJD9ykOZ0QCjNeznPDr6whqZ9NGMNF12Mw@mail.gmail.com> Fixes: fb652fdfe837 ("macvlan/macvtap: Remove NETIF_F_UFO advertisement.") Reported-by: Michal Kubecek <mkubecek@suse.cz> Signed-off-by: Willem de Bruijn <willemb@google.com> Acked-by: Jason Wang <jasowang@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-11-21 22:22:25 +07:00
BUILD_BUG_ON(SKB_GSO_UDP != (NETIF_F_GSO_UDP >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_UDP_L4 != (NETIF_F_GSO_UDP_L4 >> NETIF_F_GSO_SHIFT));
return (features & feature) == feature;
}
static inline bool skb_gso_ok(struct sk_buff *skb, netdev_features_t features)
{
return net_gso_ok(features, skb_shinfo(skb)->gso_type) &&
(!skb_has_frag_list(skb) || (features & NETIF_F_FRAGLIST));
}
static inline bool netif_needs_gso(struct sk_buff *skb,
netdev_features_t features)
{
return skb_is_gso(skb) && (!skb_gso_ok(skb, features) ||
unlikely((skb->ip_summed != CHECKSUM_PARTIAL) &&
(skb->ip_summed != CHECKSUM_UNNECESSARY)));
}
[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 17:43:19 +07:00
static inline void netif_set_gso_max_size(struct net_device *dev,
unsigned int size)
{
dev->gso_max_size = size;
}
static inline void skb_gso_error_unwind(struct sk_buff *skb, __be16 protocol,
int pulled_hlen, u16 mac_offset,
int mac_len)
{
skb->protocol = protocol;
skb->encapsulation = 1;
skb_push(skb, pulled_hlen);
skb_reset_transport_header(skb);
skb->mac_header = mac_offset;
skb->network_header = skb->mac_header + mac_len;
skb->mac_len = mac_len;
}
static inline bool netif_is_macsec(const struct net_device *dev)
{
return dev->priv_flags & IFF_MACSEC;
}
static inline bool netif_is_macvlan(const struct net_device *dev)
{
return dev->priv_flags & IFF_MACVLAN;
}
static inline bool netif_is_macvlan_port(const struct net_device *dev)
{
return dev->priv_flags & IFF_MACVLAN_PORT;
}
static inline bool netif_is_bond_master(const struct net_device *dev)
{
return dev->flags & IFF_MASTER && dev->priv_flags & IFF_BONDING;
}
static inline bool netif_is_bond_slave(const struct net_device *dev)
{
return dev->flags & IFF_SLAVE && dev->priv_flags & IFF_BONDING;
}
static inline bool netif_supports_nofcs(struct net_device *dev)
{
return dev->priv_flags & IFF_SUPP_NOFCS;
}
ipvlan, l3mdev: fix broken l3s mode wrt local routes While implementing ipvlan l3 and l3s mode for kubernetes CNI plugin, I ran into the issue that while l3 mode is working fine, l3s mode does not have any connectivity to kube-apiserver and hence all pods end up in Error state as well. The ipvlan master device sits on top of a bond device and hostns traffic to kube-apiserver (also running in hostns) is DNATed from 10.152.183.1:443 to 139.178.29.207:37573 where the latter is the address of the bond0. While in l3 mode, a curl to https://10.152.183.1:443 or to https://139.178.29.207:37573 works fine from hostns, neither of them do in case of l3s. In the latter only a curl to https://127.0.0.1:37573 appeared to work where for local addresses of bond0 I saw kernel suddenly starting to emit ARP requests to query HW address of bond0 which remained unanswered and neighbor entries in INCOMPLETE state. These ARP requests only happen while in l3s. Debugging this further, I found the issue is that l3s mode is piggy- backing on l3 master device, and in this case local routes are using l3mdev_master_dev_rcu(dev) instead of net->loopback_dev as per commit f5a0aab84b74 ("net: ipv4: dst for local input routes should use l3mdev if relevant") and 5f02ce24c269 ("net: l3mdev: Allow the l3mdev to be a loopback"). I found that reverting them back into using the net->loopback_dev fixed ipvlan l3s connectivity and got everything working for the CNI. Now judging from 4fbae7d83c98 ("ipvlan: Introduce l3s mode") and the l3mdev paper in [0] the only sole reason why ipvlan l3s is relying on l3 master device is to get the l3mdev_ip_rcv() receive hook for setting the dst entry of the input route without adding its own ipvlan specific hacks into the receive path, however, any l3 domain semantics beyond just that are breaking l3s operation. Note that ipvlan also has the ability to dynamically switch its internal operation from l3 to l3s for all ports via ipvlan_set_port_mode() at runtime. In any case, l3 vs l3s soley distinguishes itself by 'de-confusing' netfilter through switching skb->dev to ipvlan slave device late in NF_INET_LOCAL_IN before handing the skb to L4. Minimal fix taken here is to add a IFF_L3MDEV_RX_HANDLER flag which, if set from ipvlan setup, gets us only the wanted l3mdev_l3_rcv() hook without any additional l3mdev semantics on top. This should also have minimal impact since dev->priv_flags is already hot in cache. With this set, l3s mode is working fine and I also get things like masquerading pod traffic on the ipvlan master properly working. [0] https://netdevconf.org/1.2/papers/ahern-what-is-l3mdev-paper.pdf Fixes: f5a0aab84b74 ("net: ipv4: dst for local input routes should use l3mdev if relevant") Fixes: 5f02ce24c269 ("net: l3mdev: Allow the l3mdev to be a loopback") Fixes: 4fbae7d83c98 ("ipvlan: Introduce l3s mode") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Cc: Mahesh Bandewar <maheshb@google.com> Cc: David Ahern <dsa@cumulusnetworks.com> Cc: Florian Westphal <fw@strlen.de> Cc: Martynas Pumputis <m@lambda.lt> Acked-by: David Ahern <dsa@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-30 18:49:48 +07:00
static inline bool netif_has_l3_rx_handler(const struct net_device *dev)
{
return dev->priv_flags & IFF_L3MDEV_RX_HANDLER;
}
static inline bool netif_is_l3_master(const struct net_device *dev)
{
return dev->priv_flags & IFF_L3MDEV_MASTER;
}
static inline bool netif_is_l3_slave(const struct net_device *dev)
{
return dev->priv_flags & IFF_L3MDEV_SLAVE;
}
static inline bool netif_is_bridge_master(const struct net_device *dev)
{
return dev->priv_flags & IFF_EBRIDGE;
}
static inline bool netif_is_bridge_port(const struct net_device *dev)
{
return dev->priv_flags & IFF_BRIDGE_PORT;
}
static inline bool netif_is_ovs_master(const struct net_device *dev)
{
return dev->priv_flags & IFF_OPENVSWITCH;
}
static inline bool netif_is_ovs_port(const struct net_device *dev)
{
return dev->priv_flags & IFF_OVS_DATAPATH;
}
static inline bool netif_is_team_master(const struct net_device *dev)
{
return dev->priv_flags & IFF_TEAM;
}
static inline bool netif_is_team_port(const struct net_device *dev)
{
return dev->priv_flags & IFF_TEAM_PORT;
}
static inline bool netif_is_lag_master(const struct net_device *dev)
{
return netif_is_bond_master(dev) || netif_is_team_master(dev);
}
static inline bool netif_is_lag_port(const struct net_device *dev)
{
return netif_is_bond_slave(dev) || netif_is_team_port(dev);
}
ethtool: correctly ensure {GS}CHANNELS doesn't conflict with GS{RXFH} Ethernet drivers implementing both {GS}RXFH and {GS}CHANNELS ethtool ops incorrectly allow SCHANNELS when it would conflict with the settings from SRXFH. This occurs because it is not possible for drivers to understand whether their Rx flow indirection table has been configured or is in the default state. In addition, drivers currently behave in various ways when increasing the number of Rx channels. Some drivers will always destroy the Rx flow indirection table when this occurs, whether it has been set by the user or not. Other drivers will attempt to preserve the table even if the user has never modified it from the default driver settings. Neither of these situation is desirable because it leads to unexpected behavior or loss of user configuration. The correct behavior is to simply return -EINVAL when SCHANNELS would conflict with the current Rx flow table settings. However, it should only do so if the current settings were modified by the user. If we required that the new settings never conflict with the current (default) Rx flow settings, we would force users to first reduce their Rx flow settings and then reduce the number of Rx channels. This patch proposes a solution implemented in net/core/ethtool.c which ensures that all drivers behave correctly. It checks whether the RXFH table has been configured to non-default settings, and stores this information in a private netdev flag. When the number of channels is requested to change, it first ensures that the current Rx flow table is not going to assign flows to now disabled channels. Signed-off-by: Jacob Keller <jacob.e.keller@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-09 07:05:03 +07:00
static inline bool netif_is_rxfh_configured(const struct net_device *dev)
{
return dev->priv_flags & IFF_RXFH_CONFIGURED;
}
static inline bool netif_is_failover(const struct net_device *dev)
{
return dev->priv_flags & IFF_FAILOVER;
}
static inline bool netif_is_failover_slave(const struct net_device *dev)
{
return dev->priv_flags & IFF_FAILOVER_SLAVE;
}
/* This device needs to keep skb dst for qdisc enqueue or ndo_start_xmit() */
static inline void netif_keep_dst(struct net_device *dev)
{
dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM);
}
/* return true if dev can't cope with mtu frames that need vlan tag insertion */
static inline bool netif_reduces_vlan_mtu(struct net_device *dev)
{
/* TODO: reserve and use an additional IFF bit, if we get more users */
return dev->priv_flags & IFF_MACSEC;
}
extern struct pernet_operations __net_initdata loopback_net_ops;
/* 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->name[0] || strchr(dev->name, '%'))
return "(unnamed net_device)";
return dev->name;
}
static inline bool netdev_unregistering(const struct net_device *dev)
{
return dev->reg_state == NETREG_UNREGISTERING;
}
static inline const char *netdev_reg_state(const struct net_device *dev)
{
switch (dev->reg_state) {
case NETREG_UNINITIALIZED: return " (uninitialized)";
case NETREG_REGISTERED: return "";
case NETREG_UNREGISTERING: return " (unregistering)";
case NETREG_UNREGISTERED: return " (unregistered)";
case NETREG_RELEASED: return " (released)";
case NETREG_DUMMY: return " (dummy)";
}
WARN_ONCE(1, "%s: unknown reg_state %d\n", dev->name, dev->reg_state);
return " (unknown)";
}
__printf(3, 4) __cold
void netdev_printk(const char *level, const struct net_device *dev,
const char *format, ...);
__printf(2, 3) __cold
void netdev_emerg(const struct net_device *dev, const char *format, ...);
__printf(2, 3) __cold
void netdev_alert(const struct net_device *dev, const char *format, ...);
__printf(2, 3) __cold
void netdev_crit(const struct net_device *dev, const char *format, ...);
__printf(2, 3) __cold
void netdev_err(const struct net_device *dev, const char *format, ...);
__printf(2, 3) __cold
void netdev_warn(const struct net_device *dev, const char *format, ...);
__printf(2, 3) __cold
void netdev_notice(const struct net_device *dev, const char *format, ...);
__printf(2, 3) __cold
void netdev_info(const struct net_device *dev, const char *format, ...);
#define netdev_level_once(level, dev, fmt, ...) \
do { \
static bool __print_once __read_mostly; \
\
if (!__print_once) { \
__print_once = true; \
netdev_printk(level, dev, fmt, ##__VA_ARGS__); \
} \
} while (0)
#define netdev_emerg_once(dev, fmt, ...) \
netdev_level_once(KERN_EMERG, dev, fmt, ##__VA_ARGS__)
#define netdev_alert_once(dev, fmt, ...) \
netdev_level_once(KERN_ALERT, dev, fmt, ##__VA_ARGS__)
#define netdev_crit_once(dev, fmt, ...) \
netdev_level_once(KERN_CRIT, dev, fmt, ##__VA_ARGS__)
#define netdev_err_once(dev, fmt, ...) \
netdev_level_once(KERN_ERR, dev, fmt, ##__VA_ARGS__)
#define netdev_warn_once(dev, fmt, ...) \
netdev_level_once(KERN_WARNING, dev, fmt, ##__VA_ARGS__)
#define netdev_notice_once(dev, fmt, ...) \
netdev_level_once(KERN_NOTICE, dev, fmt, ##__VA_ARGS__)
#define netdev_info_once(dev, fmt, ...) \
netdev_level_once(KERN_INFO, dev, fmt, ##__VA_ARGS__)
net: don't allow CAP_NET_ADMIN to load non-netdev kernel modules Since a8f80e8ff94ecba629542d9b4b5f5a8ee3eb565c any process with CAP_NET_ADMIN may load any module from /lib/modules/. This doesn't mean that CAP_NET_ADMIN is a superset of CAP_SYS_MODULE as modules are limited to /lib/modules/**. However, CAP_NET_ADMIN capability shouldn't allow anybody load any module not related to networking. This patch restricts an ability of autoloading modules to netdev modules with explicit aliases. This fixes CVE-2011-1019. Arnd Bergmann suggested to leave untouched the old pre-v2.6.32 behavior of loading netdev modules by name (without any prefix) for processes with CAP_SYS_MODULE to maintain the compatibility with network scripts that use autoloading netdev modules by aliases like "eth0", "wlan0". Currently there are only three users of the feature in the upstream kernel: ipip, ip_gre and sit. root@albatros:~# capsh --drop=$(seq -s, 0 11),$(seq -s, 13 34) -- root@albatros:~# grep Cap /proc/$$/status CapInh: 0000000000000000 CapPrm: fffffff800001000 CapEff: fffffff800001000 CapBnd: fffffff800001000 root@albatros:~# modprobe xfs FATAL: Error inserting xfs (/lib/modules/2.6.38-rc6-00001-g2bf4ca3/kernel/fs/xfs/xfs.ko): Operation not permitted root@albatros:~# lsmod | grep xfs root@albatros:~# ifconfig xfs xfs: error fetching interface information: Device not found root@albatros:~# lsmod | grep xfs root@albatros:~# lsmod | grep sit root@albatros:~# ifconfig sit sit: error fetching interface information: Device not found root@albatros:~# lsmod | grep sit root@albatros:~# ifconfig sit0 sit0 Link encap:IPv6-in-IPv4 NOARP MTU:1480 Metric:1 root@albatros:~# lsmod | grep sit sit 10457 0 tunnel4 2957 1 sit For CAP_SYS_MODULE module loading is still relaxed: root@albatros:~# grep Cap /proc/$$/status CapInh: 0000000000000000 CapPrm: ffffffffffffffff CapEff: ffffffffffffffff CapBnd: ffffffffffffffff root@albatros:~# ifconfig xfs xfs: error fetching interface information: Device not found root@albatros:~# lsmod | grep xfs xfs 745319 0 Reference: https://lkml.org/lkml/2011/2/24/203 Signed-off-by: Vasiliy Kulikov <segoon@openwall.com> Signed-off-by: Michael Tokarev <mjt@tls.msk.ru> Acked-by: David S. Miller <davem@davemloft.net> Acked-by: Kees Cook <kees.cook@canonical.com> Signed-off-by: James Morris <jmorris@namei.org>
2011-03-02 04:33:13 +07:00
#define MODULE_ALIAS_NETDEV(device) \
MODULE_ALIAS("netdev-" device)
#if defined(CONFIG_DYNAMIC_DEBUG)
#define netdev_dbg(__dev, format, args...) \
do { \
dynamic_netdev_dbg(__dev, format, ##args); \
} while (0)
#elif defined(DEBUG)
#define netdev_dbg(__dev, format, args...) \
netdev_printk(KERN_DEBUG, __dev, format, ##args)
#else
#define netdev_dbg(__dev, format, args...) \
({ \
if (0) \
netdev_printk(KERN_DEBUG, __dev, format, ##args); \
})
#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%s: " format, netdev_name(dev), \
netdev_reg_state(dev), ##args)
#define netdev_WARN_ONCE(dev, format, args...) \
WARN_ONCE(1, "netdevice: %s%s: " format, netdev_name(dev), \
netdev_reg_state(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_level(level, priv, type, dev, fmt, args...) \
do { \
if (netif_msg_##type(priv)) \
netdev_##level(dev, fmt, ##args); \
} while (0)
#define netif_emerg(priv, type, dev, fmt, args...) \
netif_level(emerg, priv, type, dev, fmt, ##args)
#define netif_alert(priv, type, dev, fmt, args...) \
netif_level(alert, priv, type, dev, fmt, ##args)
#define netif_crit(priv, type, dev, fmt, args...) \
netif_level(crit, priv, type, dev, fmt, ##args)
#define netif_err(priv, type, dev, fmt, args...) \
netif_level(err, priv, type, dev, fmt, ##args)
#define netif_warn(priv, type, dev, fmt, args...) \
netif_level(warn, priv, type, dev, fmt, ##args)
#define netif_notice(priv, type, dev, fmt, args...) \
netif_level(notice, priv, type, dev, fmt, ##args)
#define netif_info(priv, type, dev, fmt, args...) \
netif_level(info, priv, type, dev, fmt, ##args)
#if defined(CONFIG_DYNAMIC_DEBUG)
#define netif_dbg(priv, type, netdev, format, args...) \
do { \
if (netif_msg_##type(priv)) \
dynamic_netdev_dbg(netdev, format, ##args); \
} while (0)
#elif defined(DEBUG)
#define netif_dbg(priv, type, dev, format, args...) \
netif_printk(priv, type, KERN_DEBUG, dev, format, ##args)
#else
#define netif_dbg(priv, type, dev, format, args...) \
({ \
if (0) \
netif_printk(priv, type, KERN_DEBUG, dev, format, ##args); \
0; \
})
#endif
/* if @cond then downgrade to debug, else print at @level */
#define netif_cond_dbg(priv, type, netdev, cond, level, fmt, args...) \
do { \
if (cond) \
netif_dbg(priv, type, netdev, fmt, ##args); \
else \
netif_ ## level(priv, type, netdev, fmt, ##args); \
} while (0)
#if defined(VERBOSE_DEBUG)
#define netif_vdbg netif_dbg
#else
#define netif_vdbg(priv, type, dev, format, args...) \
({ \
if (0) \
netif_printk(priv, type, KERN_DEBUG, dev, format, ##args); \
0; \
})
#endif
/*
* The list of packet types we will receive (as opposed to discard)
* and the routines to invoke.
*
* Why 16. Because with 16 the only overlap we get on a hash of the
* low nibble of the protocol value is RARP/SNAP/X.25.
*
* 0800 IP
* 0001 802.3
* 0002 AX.25
* 0004 802.2
* 8035 RARP
* 0005 SNAP
* 0805 X.25
* 0806 ARP
* 8137 IPX
* 0009 Localtalk
* 86DD IPv6
*/
#define PTYPE_HASH_SIZE (16)
#define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
#endif /* _LINUX_NETDEVICE_H */