master
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2007-2009 Robert N. M. Watson
5 * Copyright (c) 2010-2011 Juniper Networks, Inc.
6 * All rights reserved.
7 *
8 * This software was developed by Robert N. M. Watson under contract
9 * to Juniper Networks, Inc.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 */
32
33#ifndef _NET_NETISR_H_
34#define _NET_NETISR_H_
35
36/*
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.
41 *
42 * Historically, this was implemented by the BSD software ISR facility; it is
43 * now implemented via a software ithread (SWI).
44 */
45
46/*
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.
51 */
52#define NETISR_IP 1
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 */
57#define NETISR_IPV6 6
58#define NETISR_IP_DIRECT 9 /* direct-dispatch IPv4 */
59#define NETISR_IPV6_DIRECT 10 /* direct-dispatch IPv6 */
60
61/*
62 * Protocol ordering and affinity policy constants. See the detailed
63 * discussion of policies later in the file.
64 */
65#define NETISR_POLICY_SOURCE 1 /* Maintain source ordering. */
66#define NETISR_POLICY_FLOW 2 /* Maintain flow ordering. */
67#define NETISR_POLICY_CPU 3 /* Protocol determines CPU placement. */
68
69/*
70 * Protocol dispatch policy constants; selects whether and when direct
71 * dispatch is permitted.
72 */
73#define NETISR_DISPATCH_DEFAULT 0 /* Use global default. */
74#define NETISR_DISPATCH_DEFERRED 1 /* Always defer dispatch. */
75#define NETISR_DISPATCH_HYBRID 2 /* Allow hybrid dispatch. */
76#define NETISR_DISPATCH_DIRECT 3 /* Always direct dispatch. */
77
78/*
79 * Monitoring data structures, exported by sysctl(2).
80 *
81 * Three sysctls are defined. First, a per-protocol structure exported by
82 * net.isr.proto.
83 */
84#define NETISR_NAMEMAXLEN 32
85struct sysctl_netisr_proto {
86 u_int snp_version; /* Length of struct. */
87 char snp_name[NETISR_NAMEMAXLEN]; /* nh_name */
88 u_int snp_proto; /* nh_proto */
89 u_int snp_qlimit; /* nh_qlimit */
90 u_int snp_policy; /* nh_policy */
91 u_int snp_flags; /* Various flags. */
92 u_int snp_dispatch; /* Dispatch policy. */
93 u_int _snp_ispare[6];
94};
95
96/*
97 * Flags for sysctl_netisr_proto.snp_flags.
98 */
99#define NETISR_SNP_FLAGS_M2FLOW 0x00000001 /* nh_m2flow */
100#define NETISR_SNP_FLAGS_M2CPUID 0x00000002 /* nh_m2cpuid */
101#define NETISR_SNP_FLAGS_DRAINEDCPU 0x00000004 /* nh_drainedcpu */
102
103/*
104 * Next, a structure per-workstream, with per-protocol data, exported as
105 * net.isr.workstream.
106 */
107struct sysctl_netisr_workstream {
108 u_int snws_version; /* Length of struct. */
109 u_int snws_flags; /* Various flags. */
110 u_int snws_wsid; /* Workstream ID. */
111 u_int snws_cpu; /* nws_cpu */
112 u_int _snws_ispare[12];
113};
114
115/*
116 * Flags for sysctl_netisr_workstream.snws_flags
117 */
118#define NETISR_SNWS_FLAGS_INTR 0x00000001 /* nws_intr_event */
119
120/*
121 * Finally, a per-workstream-per-protocol structure, exported as
122 * net.isr.work.
123 */
124struct sysctl_netisr_work {
125 u_int snw_version; /* Length of struct. */
126 u_int snw_wsid; /* Workstream ID. */
127 u_int snw_proto; /* Protocol number. */
128 u_int snw_len; /* nw_len */
129 u_int snw_watermark; /* nw_watermark */
130 u_int _snw_ispare[3];
131
132 uint64_t snw_dispatched; /* nw_dispatched */
133 uint64_t snw_hybrid_dispatched; /* nw_hybrid_dispatched */
134 uint64_t snw_qdrops; /* nw_qdrops */
135 uint64_t snw_queued; /* nw_queued */
136 uint64_t snw_handled; /* nw_handled */
137
138 uint64_t _snw_llspare[7];
139};
140
141#ifdef _KERNEL
142
143/*-
144 * Protocols express ordering constraints and affinity preferences by
145 * implementing one or neither of nh_m2flow and nh_m2cpuid, which are used by
146 * netisr to determine which per-CPU workstream to assign mbufs to.
147 *
148 * The following policies may be used by protocols:
149 *
150 * NETISR_POLICY_SOURCE - netisr should maintain source ordering without
151 * advice from the protocol. netisr will ignore any
152 * flow IDs present on the mbuf for the purposes of
153 * work placement.
154 *
155 * NETISR_POLICY_FLOW - netisr should maintain flow ordering as defined by
156 * the mbuf header flow ID field. If the protocol
157 * implements nh_m2flow, then netisr will query the
158 * protocol in the event that the mbuf doesn't have a
159 * flow ID, falling back on source ordering.
160 *
161 * NETISR_POLICY_CPU - netisr will delegate all work placement decisions to
162 * the protocol, querying nh_m2cpuid for each packet.
163 *
164 * Protocols might make decisions about work placement based on an existing
165 * calculated flow ID on the mbuf, such as one provided in hardware, the
166 * receive interface pointed to by the mbuf (if any), the optional source
167 * identifier passed at some dispatch points, or even parse packet headers to
168 * calculate a flow. Both protocol handlers may return a new mbuf pointer
169 * for the chain, or NULL if the packet proves invalid or m_pullup() fails.
170 *
171 * XXXRW: If we eventually support dynamic reconfiguration, there should be
172 * protocol handlers to notify them of CPU configuration changes so that they
173 * can rebalance work.
174 */
175struct mbuf;
176typedef void netisr_handler_t(struct mbuf *m);
177typedef struct mbuf *netisr_m2cpuid_t(struct mbuf *m, uintptr_t source,
178 u_int *cpuid);
179typedef struct mbuf *netisr_m2flow_t(struct mbuf *m, uintptr_t source);
180typedef void netisr_drainedcpu_t(u_int cpuid);
181
182#define NETISR_CPUID_NONE ((u_int)-1) /* No affinity returned. */
183
184/*
185 * Data structure describing a protocol handler.
186 */
187struct netisr_handler {
188 const char *nh_name; /* Character string protocol name. */
189 netisr_handler_t *nh_handler; /* Protocol handler. */
190 netisr_m2flow_t *nh_m2flow; /* Query flow for untagged packet. */
191 netisr_m2cpuid_t *nh_m2cpuid; /* Query CPU to process mbuf on. */
192 netisr_drainedcpu_t *nh_drainedcpu; /* Callback when drained a queue. */
193 u_int nh_proto; /* Integer protocol ID. */
194 u_int nh_qlimit; /* Maximum per-CPU queue depth. */
195 u_int nh_policy; /* Work placement policy. */
196 u_int nh_dispatch; /* Dispatch policy. */
197 u_int nh_ispare[4]; /* For future use. */
198 void *nh_pspare[4]; /* For future use. */
199};
200
201/*
202 * Register, unregister, and other netisr handler management functions.
203 */
204void netisr_clearqdrops(const struct netisr_handler *nhp);
205void netisr_getqdrops(const struct netisr_handler *nhp,
206 u_int64_t *qdropsp);
207void netisr_getqlimit(const struct netisr_handler *nhp, u_int *qlimitp);
208void netisr_register(const struct netisr_handler *nhp);
209int netisr_setqlimit(const struct netisr_handler *nhp, u_int qlimit);
210void netisr_unregister(const struct netisr_handler *nhp);
211#ifdef VIMAGE
212void netisr_register_vnet(const struct netisr_handler *nhp);
213void netisr_unregister_vnet(const struct netisr_handler *nhp);
214#endif
215
216/*
217 * Process a packet destined for a protocol, and attempt direct dispatch.
218 * Supplemental source ordering information can be passed using the _src
219 * variant.
220 */
221int netisr_dispatch(u_int proto, struct mbuf *m);
222int netisr_dispatch_src(u_int proto, uintptr_t source, struct mbuf *m);
223int netisr_queue(u_int proto, struct mbuf *m);
224int netisr_queue_src(u_int proto, uintptr_t source, struct mbuf *m);
225
226/*
227 * Provide a default implementation of "map an ID to a CPU ID".
228 */
229u_int netisr_default_flow2cpu(u_int flowid);
230
231/*
232 * Utility routines to return the number of CPUs participting in netisr, and
233 * to return a mapping from a number to a CPU ID that can be used with the
234 * scheduler.
235 */
236u_int netisr_get_cpucount(void);
237u_int netisr_get_cpuid(u_int cpunumber);
238
239/*
240 * Interfaces between DEVICE_POLLING and netisr.
241 */
242void netisr_sched_poll(void);
243void netisr_poll(void);
244void netisr_pollmore(void);
245
246#endif /* !_KERNEL */
247#endif /* !_NET_NETISR_H_ */