freebsd-dev/sys/netinet
Robert Watson d4b5cae49b Reimplement the netisr framework in order to support parallel netisr
threads:

- Support up to one netisr thread per CPU, each processings its own
  workstream, or set of per-protocol queues.  Threads may be bound
  to specific CPUs, or allowed to migrate, based on a global policy.

  In the future it would be desirable to support topology-centric
  policies, such as "one netisr per package".

- Allow each protocol to advertise an ordering policy, which can
  currently be one of:

  NETISR_POLICY_SOURCE: packets must maintain ordering with respect to
    an implicit or explicit source (such as an interface or socket).

  NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work,
    as well as allowing protocols to provide a flow generation function
    for mbufs without flow identifers (m2flow).  Falls back on
    NETISR_POLICY_SOURCE if now flow ID is available.

  NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for
    each packet handled by netisr (m2cpuid).

- Provide utility functions for querying the number of workstreams
  being used, as well as a mapping function from workstream to CPU ID,
  which protocols may use in work placement decisions.

- Add explicit interfaces to get and set per-protocol queue limits, and
  get and clear drop counters, which query data or apply changes across
  all workstreams.

- Add a more extensible netisr registration interface, in which
  protocols declare 'struct netisr_handler' structures for each
  registered NETISR_ type.  These include name, handler function,
  optional mbuf to flow ID function, optional mbuf to CPU ID function,
  queue limit, and ordering policy.  Padding is present to allow these
  to be expanded in the future.  If no queue limit is declared, then
  a default is used.

- Queue limits are now per-workstream, and raised from the previous
  IFQ_MAXLEN default of 50 to 256.

- All protocols are updated to use the new registration interface, and
  with the exception of netnatm, default queue limits.  Most protocols
  register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use
  NETISR_POLICY_FLOW, and will therefore take advantage of driver-
  generated flow IDs if present.

- Formalize a non-packet based interface between interface polling and
  the netisr, rather than having polling pretend to be two protocols.
  Provide two explicit hooks in the netisr worker for start and end
  events for runs: netisr_poll() and netisr_pollmore(), as well as a
  function, netisr_sched_poll(), to allow the polling code to schedule
  netisr execution.  DEVICE_POLLING still embeds single-netisr
  assumptions in its implementation, so for now if it is compiled into
  the kernel, a single and un-bound netisr thread is enforced
  regardless of tunable configuration.

In the default configuration, the new netisr implementation maintains
the same basic assumptions as the previous implementation: a single,
un-bound worker thread processes all deferred work, and direct dispatch
is enabled by default wherever possible.

Performance measurement shows a marginal performance improvement over
the old implementation due to the use of batched dequeue.

An rmlock is used to synchronize use and registration/unregistration
using the framework; currently, synchronized use is disabled
(replicating current netisr policy) due to a measurable 3%-6% hit in
ping-pong micro-benchmarking.  It will be enabled once further rmlock
optimization has taken place.  However, in practice, netisrs are
rarely registered or unregistered at runtime.

A new man page for netisr will follow, but since one doesn't currently
exist, it hasn't been updated.

This change is not appropriate for MFC, although the polling shutdown
handler should be merged to 7-STABLE.

Bump __FreeBSD_version.

Reviewed by:	bz
2009-06-01 10:41:38 +00:00
..
libalias What's the point of adjusting a checksum if we are going to toss the 2009-04-11 15:26:31 +00:00
accf_data.c Add FBSDID to all files in netinet so that people can more 2007-10-07 20:44:24 +00:00
accf_dns.c Correct the sense of a test so that this filter always waits for the full 2009-05-26 20:00:30 +00:00
accf_http.c Add FBSDID to all files in netinet so that people can more 2007-10-07 20:44:24 +00:00
icmp6.h Add MLDv2 prototypes and defines. 2009-04-29 10:20:17 +00:00
icmp_var.h Update stats in struct icmpstat and icmp6stat using four new 2009-04-12 13:22:33 +00:00
if_atm.c This main goals of this project are: 2008-12-15 06:10:57 +00:00
if_atm.h
if_ether.c Reimplement the netisr framework in order to support parallel netisr 2009-06-01 10:41:38 +00:00
if_ether.h remove dependency on net/if.h of this header 2009-02-16 15:07:40 +00:00
igmp_var.h Update stats in struct igmpstat using two new macros: 2009-04-12 13:41:13 +00:00
igmp.c Reimplement the netisr framework in order to support parallel netisr 2009-06-01 10:41:38 +00:00
igmp.h Comment IGMP_PIM as being very historic, as in, don't use. 2009-03-19 01:15:26 +00:00
in_cksum.c Add FBSDID to all files in netinet so that people can more 2007-10-07 20:44:24 +00:00
in_gif.c Update stats in struct ipstat using four new macros, IPSTAT_ADD(), 2009-04-11 23:35:20 +00:00
in_gif.h
in_mcast.c Cut over to KTR_INET for CTR. 2009-04-29 10:14:16 +00:00
in_pcb.c - Rename IP_NONLOCALOK IP socket option to IP_BINDANY, to be more consistent 2009-06-01 10:30:00 +00:00
in_pcb.h - Rename IP_NONLOCALOK IP socket option to IP_BINDANY, to be more consistent 2009-06-01 10:30:00 +00:00
in_proto.c Merge IGMPv3 and Source-Specific Multicast (SSM) to the FreeBSD 2009-03-09 17:53:05 +00:00
in_rmx.c Change the curvnet variable from a global const struct vnet *, 2009-05-05 10:56:12 +00:00
in_systm.h Use uint32_t instead of n_long and n_time, and uint16_t instead of n_short. 2009-02-13 15:14:43 +00:00
in_var.h Remove bogus comment. 2009-05-09 18:50:01 +00:00
in.c If including vnet.h one has to include opt_route.h as well. This is 2009-05-22 23:03:15 +00:00
in.h - Rename IP_NONLOCALOK IP socket option to IP_BINDANY, to be more consistent 2009-06-01 10:30:00 +00:00
ip6.h Start removing IPv6 Type 0 Routing header code. 2009-03-03 13:12:12 +00:00
ip_carp.c Bite the bullet, and make the IPv6 SSM and MLDv2 mega-commit: 2009-04-29 19:19:13 +00:00
ip_carp.h Update stats in struct carpstats using two new macros: CARPSTATS_ADD() 2009-04-12 14:19:37 +00:00
ip_divert.c Reimplement the netisr framework in order to support parallel netisr 2009-06-01 10:41:38 +00:00
ip_divert.h
ip_dummynet.c Add emulation of delay profiles, which lets you model various 2009-04-09 12:46:00 +00:00
ip_dummynet.h Add emulation of delay profiles, which lets you model various 2009-04-09 12:46:00 +00:00
ip_ecn.c Add FBSDID to all files in netinet so that people can more 2007-10-07 20:44:24 +00:00
ip_ecn.h
ip_encap.c Add FBSDID to all files in netinet so that people can more 2007-10-07 20:44:24 +00:00
ip_encap.h
ip_fastfwd.c Change if_output to take a struct route as its fourth argument in order 2009-04-16 20:30:28 +00:00
ip_fw2.c Revert the logical change of r192341. 2009-05-18 22:34:44 +00:00
ip_fw_nat.c Prefer actual field names (if_addrhead, ifa_link) to macros aliasing 2009-04-20 22:40:44 +00:00
ip_fw_pfil.c Permit buiding kernels with options VIMAGE, restricted to only a single 2009-04-30 13:36:26 +00:00
ip_fw.h Make indentation more uniform accross vnet container structs. 2009-05-02 08:16:26 +00:00
ip_gre.c Enqueue de-capsulated packet instead of performing direct dispatch. It's 2008-07-04 21:01:30 +00:00
ip_gre.h
ip_icmp.c In icmp_reflect(), acquire the inteface address list lock when 2009-04-20 13:45:39 +00:00
ip_icmp.h Use uint32_t instead of n_long and n_time, and uint16_t instead of n_short. 2009-02-13 15:14:43 +00:00
ip_id.c Rather than using hidden includes (with cicular dependencies), 2008-12-02 21:37:28 +00:00
ip_input.c Reimplement the netisr framework in order to support parallel netisr 2009-06-01 10:41:38 +00:00
ip_ipsec.c Add sysctls to toggle the behaviour of the (former) IPSEC_FILTERTUNNEL 2009-05-23 16:42:38 +00:00
ip_ipsec.h Take the route mtu into account, if available, when sending an 2008-04-09 05:17:18 +00:00
ip_mroute.c Use KTR_INET for MROUTING CTRs. 2009-04-29 10:17:08 +00:00
ip_mroute.h Bracket struct mfc and struct rtdetq with #ifdef _KERNEL. 2009-04-21 12:47:09 +00:00
ip_options.c Update stats in struct ipstat using four new macros, IPSTAT_ADD(), 2009-04-11 23:35:20 +00:00
ip_options.h Add function ip_checkrouteralert(), which will be used 2009-03-04 02:51:22 +00:00
ip_output.c - Rename IP_NONLOCALOK IP socket option to IP_BINDANY, to be more consistent 2009-06-01 10:30:00 +00:00
ip_var.h Update stats in struct ipstat using four new macros, IPSTAT_ADD(), 2009-04-11 23:35:20 +00:00
ip.h Use uint32_t instead of n_long and n_time, and uint16_t instead of n_short. 2009-02-13 15:14:43 +00:00
pim_var.h Update stats in struct pimstat using two new macros: PIMSTAT_ADD() 2009-04-12 14:06:26 +00:00
pim.h
raw_ip.c - Rename IP_NONLOCALOK IP socket option to IP_BINDANY, to be more consistent 2009-06-01 10:30:00 +00:00
sctp_asconf.c Code from the hack-session known as the IETF (and a 2008-12-06 13:19:54 +00:00
sctp_asconf.h Code from the hack-session known as the IETF (and a 2008-12-06 13:19:54 +00:00
sctp_auth.c - Cleanup checksum code. 2009-02-03 11:04:03 +00:00
sctp_auth.h Code from the hack-session known as the IETF (and a 2008-12-06 13:19:54 +00:00
sctp_bsd_addr.c Add missing address lock when we look at the ifa list 2009-04-14 19:20:27 +00:00
sctp_bsd_addr.h 1) Adds the rest of the VIMAGE change macros 2008-07-09 16:45:30 +00:00
sctp_cc_functions.c - Macro-izes the packed declaration in all headers. 2008-06-14 07:58:05 +00:00
sctp_cc_functions.h 1) Adds the rest of the VIMAGE change macros 2008-07-09 16:45:30 +00:00
sctp_constants.h Fixes several PR-SCTP releated bugs. 2009-03-14 13:42:13 +00:00
sctp_crc32.c repository sync to multi-OS repo ... spaceing change 2009-05-07 16:43:49 +00:00
sctp_crc32.h This commit fixes the issue with alias_sctp.c. No 2009-02-14 11:34:57 +00:00
sctp_header.h Add the add-stream capability. Still needs more 2009-02-20 15:03:54 +00:00
sctp_indata.c Move the flight size reduction to right after 2009-04-14 07:50:29 +00:00
sctp_indata.h Code from the hack-session known as the IETF (and a 2008-12-06 13:19:54 +00:00
sctp_input.c Many bug fixes (from the IETF hack-fest): 2009-04-04 11:43:32 +00:00
sctp_input.h 1) Adds the rest of the VIMAGE change macros 2008-07-09 16:45:30 +00:00
sctp_lock_bsd.h - Make strict-sacks be the default. 2008-08-28 09:44:07 +00:00
sctp_os_bsd.h Unbreak options VIMAGE kernel builds. 2009-05-06 08:49:39 +00:00
sctp_os.h - Fix address add handling to clear cached routes and source addresses 2007-08-24 00:53:53 +00:00
sctp_output.c Many bug fixes (from the IETF hack-fest): 2009-04-04 11:43:32 +00:00
sctp_output.h Fix the add stream feature of strm-reset to really work: 2009-02-27 20:54:45 +00:00
sctp_pcb.c Adds missing sysctl to manage the vtag_time_wait time. This will 2009-05-30 11:14:41 +00:00
sctp_pcb.h - Cleanup checksum code. 2009-02-03 11:04:03 +00:00
sctp_peeloff.c - Macro-izes the packed declaration in all headers. 2008-06-14 07:58:05 +00:00
sctp_peeloff.h - Copyright change, cisco's silly tool wants it to say: 2007-05-08 17:01:12 +00:00
sctp_structs.h add an llentry to struct route{_in6} to allow it to be passed around with 2009-04-15 20:34:19 +00:00
sctp_sysctl.c Adds missing sysctl to manage the vtag_time_wait time. This will 2009-05-30 11:14:41 +00:00
sctp_sysctl.h Adds missing sysctl to manage the vtag_time_wait time. This will 2009-05-30 11:14:41 +00:00
sctp_timer.c Fixes several PR-SCTP releated bugs. 2009-03-14 13:42:13 +00:00
sctp_timer.h Invariants changes that make more sense. 2008-10-27 13:53:31 +00:00
sctp_uio.h Make sctp_uio user to kernel structure match the 2009-05-30 10:50:40 +00:00
sctp_usrreq.c Many bug fixes (from the IETF hack-fest): 2009-04-04 11:43:32 +00:00
sctp_var.h Fixes several PR-SCTP releated bugs. 2009-03-14 13:42:13 +00:00
sctp.h Fixes several PR-SCTP releated bugs. 2009-03-14 13:42:13 +00:00
sctputil.c Fix a small memory leak from the nr-sack code - the mapping array 2009-05-30 10:56:27 +00:00
sctputil.h Fixes several PR-SCTP releated bugs. 2009-03-14 13:42:13 +00:00
tcp_debug.c To my reading there are no real consumers of ip6_plen (IPv6 2008-09-07 20:44:45 +00:00
tcp_debug.h Use uint32_t instead of n_long and n_time, and uint16_t instead of n_short. 2009-02-13 15:14:43 +00:00
tcp_fsm.h Make tcpstates[] static, and make sure TCPSTATES is defined before 2007-07-30 11:06:42 +00:00
tcp_hostcache.c A NOP change: style / whitespace cleanup of the noise that slipped 2009-05-08 14:34:25 +00:00
tcp_hostcache.h Add an essential .h file that skipped from the last commit (r185419). 2008-11-28 23:39:25 +00:00
tcp_input.c Correct handling of SYN packets that are to the left of the current window of an ESTABLISHED connection. 2009-05-27 17:02:10 +00:00
tcp_lro.c Don't calculate checksum if it has already been validated 2008-08-24 02:31:09 +00:00
tcp_lro.h Add generic TCP LOR into netinet 2008-06-11 22:12:50 +00:00
tcp_offload.c Rather than using hidden includes (with cicular dependencies), 2008-12-02 21:37:28 +00:00
tcp_offload.h - Switch to INP_WLOCK macro from inp_wlock 2008-07-21 21:22:56 +00:00
tcp_output.c Update stats in struct tcpstat using two new macros, TCPSTAT_ADD() and 2009-04-11 22:07:19 +00:00
tcp_reass.c Remove comment about moving tcp_reass() to its own file named tcp_reass.c, 2009-05-25 14:51:47 +00:00
tcp_sack.c Update stats in struct tcpstat using two new macros, TCPSTAT_ADD() and 2009-04-11 22:07:19 +00:00
tcp_seq.h
tcp_subr.c For UDP with introducing the UDP control block, the uma zone had to 2009-05-23 17:02:30 +00:00
tcp_syncache.c Permit buiding kernels with options VIMAGE, restricted to only a single 2009-04-30 13:36:26 +00:00
tcp_syncache.h Permit buiding kernels with options VIMAGE, restricted to only a single 2009-04-30 13:36:26 +00:00
tcp_timer.c Update stats in struct tcpstat using two new macros, TCPSTAT_ADD() and 2009-04-11 22:07:19 +00:00
tcp_timer.h Two changes: 2007-09-24 05:26:24 +00:00
tcp_timewait.c Unbreak options VIMAGE + nooptions INVARIANTS kernel builds. 2009-05-02 05:02:28 +00:00
tcp_usrreq.c Update stats in struct tcpstat using two new macros, TCPSTAT_ADD() and 2009-04-11 22:07:19 +00:00
tcp_var.h Permit buiding kernels with options VIMAGE, restricted to only a single 2009-04-30 13:36:26 +00:00
tcp.h add rcv_nxt, snd_nxt, and toe offload id to FreeBSD-specific 2008-05-05 20:13:31 +00:00
tcpip.h
toedev.h Don't check if an interface can do tcp offload if there are no offload devices registered on the system. 2008-09-01 05:30:22 +00:00
udp_usrreq.c Add hierarchical jails. A jail may further virtualize its environment 2009-05-27 14:11:23 +00:00
udp_var.h Implement UDP control block support. 2009-05-23 16:51:13 +00:00
udp.h Gratuitous UDP restyling toward style(9) in 7.x. 2007-02-20 10:13:11 +00:00
vinet.h Implement UDP control block support. 2009-05-23 16:51:13 +00:00