xref: /freebsd/sys/net/netisr.h (revision c4fbf89fc502c29acacf5f944e450e5563a2c889)
1 /*-
2  * Copyright (c) 2007-2009 Robert N. M. Watson
3  * Copyright (c) 2010 Juniper Networks, Inc.
4  * All rights reserved.
5  *
6  * This software was developed by Robert N. M. Watson under contract
7  * to Juniper Networks, Inc.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
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.
17  *
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
28  * SUCH DAMAGE.
29  *
30  * $FreeBSD$
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_AARP	4		/* Appletalk ARP */
56 #define	NETISR_ATALK2	5		/* Appletalk phase 2 */
57 #define	NETISR_ATALK1	6		/* Appletalk phase 1 */
58 #define	NETISR_ARP	7		/* same as AF_LINK */
59 #define	NETISR_IPX	8		/* same as AF_IPX */
60 #define	NETISR_ETHER	9		/* ethernet input */
61 #define	NETISR_IPV6	10
62 #define	NETISR_NATM	11
63 #define	NETISR_EPAIR	12		/* if_epair(4) */
64 
65 /*
66  * Protocol ordering and affinity policy constants.  See the detailed
67  * discussion of policies later in the file.
68  */
69 #define	NETISR_POLICY_SOURCE	1	/* Maintain source ordering. */
70 #define	NETISR_POLICY_FLOW	2	/* Maintain flow ordering. */
71 #define	NETISR_POLICY_CPU	3	/* Protocol determines CPU placement. */
72 
73 /*
74  * Monitoring data structures, exported by sysctl(2).
75  *
76  * Three sysctls are defined.  First, a per-protocol structure exported by
77  * net.isr.proto.
78  */
79 #define	NETISR_NAMEMAXLEN	32
80 struct sysctl_netisr_proto {
81 	u_int	snp_version;			/* Length of struct. */
82 	char	snp_name[NETISR_NAMEMAXLEN];	/* nh_name */
83 	u_int	snp_proto;			/* nh_proto */
84 	u_int	snp_qlimit;			/* nh_qlimit */
85 	u_int	snp_policy;			/* nh_policy */
86 	u_int	snp_flags;			/* Various flags. */
87 	u_int	_snp_ispare[7];
88 };
89 
90 /*
91  * Flags for sysctl_netisr_proto.snp_flags.
92  */
93 #define	NETISR_SNP_FLAGS_M2FLOW		0x00000001	/* nh_m2flow */
94 #define	NETISR_SNP_FLAGS_M2CPUID	0x00000002	/* nh_m2cpuid */
95 #define	NETISR_SNP_FLAGS_DRAINEDCPU	0x00000004	/* nh_drainedcpu */
96 
97 /*
98  * Next, a structure per-workstream, with per-protocol data, exported as
99  * net.isr.workstream.
100  */
101 struct sysctl_netisr_workstream {
102 	u_int	snws_version;			/* Length of struct. */
103 	u_int	snws_flags;			/* Various flags. */
104 	u_int	snws_wsid;			/* Workstream ID. */
105 	u_int	snws_cpu;			/* nws_cpu */
106 	u_int	_snws_ispare[12];
107 };
108 
109 /*
110  * Flags for sysctl_netisr_workstream.snws_flags
111  */
112 #define	NETISR_SNWS_FLAGS_INTR		0x00000001	/* nws_intr_event */
113 
114 /*
115  * Finally, a per-workstream-per-protocol structure, exported as
116  * net.isr.work.
117  */
118 struct sysctl_netisr_work {
119 	u_int	snw_version;			/* Length of struct. */
120 	u_int	snw_wsid;			/* Workstream ID. */
121 	u_int	snw_proto;			/* Protocol number. */
122 	u_int	snw_len;			/* nw_len */
123 	u_int	snw_watermark;			/* nw_watermark */
124 	u_int	_snw_ispare[3];
125 
126 	uint64_t	snw_dispatched;		/* nw_dispatched */
127 	uint64_t	snw_hybrid_dispatched;	/* nw_hybrid_dispatched */
128 	uint64_t	snw_qdrops;		/* nw_qdrops */
129 	uint64_t	snw_queued;		/* nw_queued */
130 	uint64_t	snw_handled;		/* nw_handled */
131 
132 	uint64_t	_snw_llspare[7];
133 };
134 
135 #ifdef _KERNEL
136 
137 /*-
138  * Protocols express ordering constraints and affinity preferences by
139  * implementing one or neither of nh_m2flow and nh_m2cpuid, which are used by
140  * netisr to determine which per-CPU workstream to assign mbufs to.
141  *
142  * The following policies may be used by protocols:
143  *
144  * NETISR_POLICY_SOURCE - netisr should maintain source ordering without
145  *                        advice from the protocol.  netisr will ignore any
146  *                        flow IDs present on the mbuf for the purposes of
147  *                        work placement.
148  *
149  * NETISR_POLICY_FLOW - netisr should maintain flow ordering as defined by
150  *                      the mbuf header flow ID field.  If the protocol
151  *                      implements nh_m2flow, then netisr will query the
152  *                      protocol in the event that the mbuf doesn't have a
153  *                      flow ID, falling back on source ordering.
154  *
155  * NETISR_POLICY_CPU - netisr will delegate all work placement decisions to
156  *                     the protocol, querying nh_m2cpuid for each packet.
157  *
158  * Protocols might make decisions about work placement based on an existing
159  * calculated flow ID on the mbuf, such as one provided in hardware, the
160  * receive interface pointed to by the mbuf (if any), the optional source
161  * identifier passed at some dispatch points, or even parse packet headers to
162  * calculate a flow.  Both protocol handlers may return a new mbuf pointer
163  * for the chain, or NULL if the packet proves invalid or m_pullup() fails.
164  *
165  * XXXRW: If we eventually support dynamic reconfiguration, there should be
166  * protocol handlers to notify them of CPU configuration changes so that they
167  * can rebalance work.
168  */
169 struct mbuf;
170 typedef void		 netisr_handler_t (struct mbuf *m);
171 typedef struct mbuf	*netisr_m2cpuid_t(struct mbuf *m, uintptr_t source,
172 			 u_int *cpuid);
173 typedef	struct mbuf	*netisr_m2flow_t(struct mbuf *m, uintptr_t source);
174 typedef void		 netisr_drainedcpu_t(u_int cpuid);
175 
176 /*
177  * Data structure describing a protocol handler.
178  */
179 struct netisr_handler {
180 	const char	*nh_name;	/* Character string protocol name. */
181 	netisr_handler_t *nh_handler;	/* Protocol handler. */
182 	netisr_m2flow_t	*nh_m2flow;	/* Query flow for untagged packet. */
183 	netisr_m2cpuid_t *nh_m2cpuid;	/* Query CPU to process mbuf on. */
184 	netisr_drainedcpu_t *nh_drainedcpu; /* Callback when drained a queue. */
185 	u_int		 nh_proto;	/* Integer protocol ID. */
186 	u_int		 nh_qlimit;	/* Maximum per-CPU queue depth. */
187 	u_int		 nh_policy;	/* Work placement policy. */
188 	u_int		 nh_ispare[5];	/* For future use. */
189 	void		*nh_pspare[4];	/* For future use. */
190 };
191 
192 /*
193  * Register, unregister, and other netisr handler management functions.
194  */
195 void	netisr_clearqdrops(const struct netisr_handler *nhp);
196 void	netisr_getqdrops(const struct netisr_handler *nhp,
197 	    u_int64_t *qdropsp);
198 void	netisr_getqlimit(const struct netisr_handler *nhp, u_int *qlimitp);
199 void	netisr_register(const struct netisr_handler *nhp);
200 int	netisr_setqlimit(const struct netisr_handler *nhp, u_int qlimit);
201 void	netisr_unregister(const struct netisr_handler *nhp);
202 
203 /*
204  * Process a packet destined for a protocol, and attempt direct dispatch.
205  * Supplemental source ordering information can be passed using the _src
206  * variant.
207  */
208 int	netisr_dispatch(u_int proto, struct mbuf *m);
209 int	netisr_dispatch_src(u_int proto, uintptr_t source, struct mbuf *m);
210 int	netisr_queue(u_int proto, struct mbuf *m);
211 int	netisr_queue_src(u_int proto, uintptr_t source, struct mbuf *m);
212 
213 /*
214  * Provide a default implementation of "map an ID to a CPU ID".
215  */
216 u_int	netisr_default_flow2cpu(u_int flowid);
217 
218 /*
219  * Utility routines to return the number of CPUs participting in netisr, and
220  * to return a mapping from a number to a CPU ID that can be used with the
221  * scheduler.
222  */
223 u_int	netisr_get_cpucount(void);
224 u_int	netisr_get_cpuid(u_int cpunumber);
225 
226 /*
227  * Interfaces between DEVICE_POLLING and netisr.
228  */
229 void	netisr_sched_poll(void);
230 void	netisr_poll(void);
231 void	netisr_pollmore(void);
232 
233 #endif /* !_KERNEL */
234 #endif /* !_NET_NETISR_H_ */
235