2 * Copyright (c) 2007-2009 Robert N. M. Watson
3 * Copyright (c) 2010-2011 Juniper Networks, Inc.
6 * This software was developed by Robert N. M. Watson under contract
7 * to Juniper Networks, Inc.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #ifndef _NET_NETISR_H_
34 #define _NET_NETISR_H_
37 * The netisr (network interrupt service routine) provides a deferred
38 * execution evironment in which (generally inbound) network processing can
39 * take place. Protocols register handlers which will be executed directly,
40 * or via deferred dispatch, depending on the circumstances.
42 * Historically, this was implemented by the BSD software ISR facility; it is
43 * now implemented via a software ithread (SWI).
47 * Protocol numbers, which are encoded in monitoring applications and kernel
48 * modules. Internally, these are used in bit shift operations so must have
49 * a value 0 < proto < 32; we currently further limit at compile-time to 16
50 * for array-sizing purposes.
53 #define NETISR_IGMP 2 /* IGMPv3 output queue */
54 #define NETISR_ROUTE 3 /* routing socket */
55 #define NETISR_ARP 4 /* same as AF_LINK */
56 #define NETISR_ETHER 5 /* ethernet input */
59 #define NETISR_EPAIR 8 /* if_epair(4) */
60 #define NETISR_IP_DIRECT 9 /* direct-dispatch IPv4 */
61 #define NETISR_IPV6_DIRECT 10 /* direct-dispatch IPv6 */
64 * Protocol ordering and affinity policy constants. See the detailed
65 * discussion of policies later in the file.
67 #define NETISR_POLICY_SOURCE 1 /* Maintain source ordering. */
68 #define NETISR_POLICY_FLOW 2 /* Maintain flow ordering. */
69 #define NETISR_POLICY_CPU 3 /* Protocol determines CPU placement. */
72 * Protocol dispatch policy constants; selects whether and when direct
73 * dispatch is permitted.
75 #define NETISR_DISPATCH_DEFAULT 0 /* Use global default. */
76 #define NETISR_DISPATCH_DEFERRED 1 /* Always defer dispatch. */
77 #define NETISR_DISPATCH_HYBRID 2 /* Allow hybrid dispatch. */
78 #define NETISR_DISPATCH_DIRECT 3 /* Always direct dispatch. */
81 * Monitoring data structures, exported by sysctl(2).
83 * Three sysctls are defined. First, a per-protocol structure exported by
86 #define NETISR_NAMEMAXLEN 32
87 struct sysctl_netisr_proto {
88 u_int snp_version; /* Length of struct. */
89 char snp_name[NETISR_NAMEMAXLEN]; /* nh_name */
90 u_int snp_proto; /* nh_proto */
91 u_int snp_qlimit; /* nh_qlimit */
92 u_int snp_policy; /* nh_policy */
93 u_int snp_flags; /* Various flags. */
94 u_int snp_dispatch; /* Dispatch policy. */
99 * Flags for sysctl_netisr_proto.snp_flags.
101 #define NETISR_SNP_FLAGS_M2FLOW 0x00000001 /* nh_m2flow */
102 #define NETISR_SNP_FLAGS_M2CPUID 0x00000002 /* nh_m2cpuid */
103 #define NETISR_SNP_FLAGS_DRAINEDCPU 0x00000004 /* nh_drainedcpu */
106 * Next, a structure per-workstream, with per-protocol data, exported as
107 * net.isr.workstream.
109 struct sysctl_netisr_workstream {
110 u_int snws_version; /* Length of struct. */
111 u_int snws_flags; /* Various flags. */
112 u_int snws_wsid; /* Workstream ID. */
113 u_int snws_cpu; /* nws_cpu */
114 u_int _snws_ispare[12];
118 * Flags for sysctl_netisr_workstream.snws_flags
120 #define NETISR_SNWS_FLAGS_INTR 0x00000001 /* nws_intr_event */
123 * Finally, a per-workstream-per-protocol structure, exported as
126 struct sysctl_netisr_work {
127 u_int snw_version; /* Length of struct. */
128 u_int snw_wsid; /* Workstream ID. */
129 u_int snw_proto; /* Protocol number. */
130 u_int snw_len; /* nw_len */
131 u_int snw_watermark; /* nw_watermark */
132 u_int _snw_ispare[3];
134 uint64_t snw_dispatched; /* nw_dispatched */
135 uint64_t snw_hybrid_dispatched; /* nw_hybrid_dispatched */
136 uint64_t snw_qdrops; /* nw_qdrops */
137 uint64_t snw_queued; /* nw_queued */
138 uint64_t snw_handled; /* nw_handled */
140 uint64_t _snw_llspare[7];
146 * Protocols express ordering constraints and affinity preferences by
147 * implementing one or neither of nh_m2flow and nh_m2cpuid, which are used by
148 * netisr to determine which per-CPU workstream to assign mbufs to.
150 * The following policies may be used by protocols:
152 * NETISR_POLICY_SOURCE - netisr should maintain source ordering without
153 * advice from the protocol. netisr will ignore any
154 * flow IDs present on the mbuf for the purposes of
157 * NETISR_POLICY_FLOW - netisr should maintain flow ordering as defined by
158 * the mbuf header flow ID field. If the protocol
159 * implements nh_m2flow, then netisr will query the
160 * protocol in the event that the mbuf doesn't have a
161 * flow ID, falling back on source ordering.
163 * NETISR_POLICY_CPU - netisr will delegate all work placement decisions to
164 * the protocol, querying nh_m2cpuid for each packet.
166 * Protocols might make decisions about work placement based on an existing
167 * calculated flow ID on the mbuf, such as one provided in hardware, the
168 * receive interface pointed to by the mbuf (if any), the optional source
169 * identifier passed at some dispatch points, or even parse packet headers to
170 * calculate a flow. Both protocol handlers may return a new mbuf pointer
171 * for the chain, or NULL if the packet proves invalid or m_pullup() fails.
173 * XXXRW: If we eventually support dynamic reconfiguration, there should be
174 * protocol handlers to notify them of CPU configuration changes so that they
175 * can rebalance work.
178 typedef void netisr_handler_t(struct mbuf *m);
179 typedef struct mbuf *netisr_m2cpuid_t(struct mbuf *m, uintptr_t source,
181 typedef struct mbuf *netisr_m2flow_t(struct mbuf *m, uintptr_t source);
182 typedef void netisr_drainedcpu_t(u_int cpuid);
184 #define NETISR_CPUID_NONE ((u_int)-1) /* No affinity returned. */
187 * Data structure describing a protocol handler.
189 struct netisr_handler {
190 const char *nh_name; /* Character string protocol name. */
191 netisr_handler_t *nh_handler; /* Protocol handler. */
192 netisr_m2flow_t *nh_m2flow; /* Query flow for untagged packet. */
193 netisr_m2cpuid_t *nh_m2cpuid; /* Query CPU to process mbuf on. */
194 netisr_drainedcpu_t *nh_drainedcpu; /* Callback when drained a queue. */
195 u_int nh_proto; /* Integer protocol ID. */
196 u_int nh_qlimit; /* Maximum per-CPU queue depth. */
197 u_int nh_policy; /* Work placement policy. */
198 u_int nh_dispatch; /* Dispatch policy. */
199 u_int nh_ispare[4]; /* For future use. */
200 void *nh_pspare[4]; /* For future use. */
204 * Register, unregister, and other netisr handler management functions.
206 void netisr_clearqdrops(const struct netisr_handler *nhp);
207 void netisr_getqdrops(const struct netisr_handler *nhp,
209 void netisr_getqlimit(const struct netisr_handler *nhp, u_int *qlimitp);
210 void netisr_register(const struct netisr_handler *nhp);
211 int netisr_setqlimit(const struct netisr_handler *nhp, u_int qlimit);
212 void netisr_unregister(const struct netisr_handler *nhp);
215 * Process a packet destined for a protocol, and attempt direct dispatch.
216 * Supplemental source ordering information can be passed using the _src
219 int netisr_dispatch(u_int proto, struct mbuf *m);
220 int netisr_dispatch_src(u_int proto, uintptr_t source, struct mbuf *m);
221 int netisr_queue(u_int proto, struct mbuf *m);
222 int netisr_queue_src(u_int proto, uintptr_t source, struct mbuf *m);
225 * Provide a default implementation of "map an ID to a CPU ID".
227 u_int netisr_default_flow2cpu(u_int flowid);
230 * Utility routines to return the number of CPUs participting in netisr, and
231 * to return a mapping from a number to a CPU ID that can be used with the
234 u_int netisr_get_cpucount(void);
235 u_int netisr_get_cpuid(u_int cpunumber);
238 * Interfaces between DEVICE_POLLING and netisr.
240 void netisr_sched_poll(void);
241 void netisr_poll(void);
242 void netisr_pollmore(void);
244 #endif /* !_KERNEL */
245 #endif /* !_NET_NETISR_H_ */