xref: /freebsd/sys/net/netisr.c (revision 7773002178c8dbc52b44e4d705f07706409af8e4)
1 /*-
2  * Copyright (c) 2001,2002,2003 Jonathan Lemon <jlemon@FreeBSD.org>
3  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD$
28  */
29 
30 #include <sys/param.h>
31 #include <sys/bus.h>
32 #include <sys/rtprio.h>
33 #include <sys/systm.h>
34 #include <sys/interrupt.h>
35 #include <sys/kernel.h>
36 #include <sys/kthread.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/proc.h>
40 #include <sys/random.h>
41 #include <sys/resourcevar.h>
42 #include <sys/sysctl.h>
43 #include <sys/unistd.h>
44 #include <machine/atomic.h>
45 #include <machine/cpu.h>
46 #include <machine/stdarg.h>
47 
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 
51 #include <net/if.h>
52 #include <net/if_types.h>
53 #include <net/if_var.h>
54 #include <net/netisr.h>
55 
56 /*
57  * XXX this is a temporary measure to allow folks to
58  * XXX disable Giant locking in the network code without
59  * XXX recompiling--in case of problems.
60  */
61 int	debug_mpsafenet = 0;
62 TUNABLE_INT("debug.mpsafenet", &debug_mpsafenet);
63 SYSCTL_INT(_debug, OID_AUTO, mpsafenet, CTLFLAG_RD, &debug_mpsafenet, 0,
64     "Enable/disable MPSAFE network support");
65 
66 volatile unsigned int	netisr;	/* scheduling bits for network */
67 
68 struct netisr {
69 	netisr_t	*ni_handler;
70 	struct ifqueue	*ni_queue;
71 } netisrs[32];
72 
73 static struct mtx netisr_mtx;
74 static void *net_ih;
75 
76 void
77 legacy_setsoftnet(void)
78 {
79 	swi_sched(net_ih, 0);
80 }
81 
82 void
83 netisr_register(int num, netisr_t *handler, struct ifqueue *inq)
84 {
85 
86 	KASSERT(!(num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs))),
87 	    ("bad isr %d", num));
88 	netisrs[num].ni_handler = handler;
89 	netisrs[num].ni_queue = inq;
90 }
91 
92 void
93 netisr_unregister(int num)
94 {
95 	struct netisr *ni;
96 	int s;
97 
98 	KASSERT(!(num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs))),
99 	    ("bad isr %d", num));
100 	ni = &netisrs[num];
101 	ni->ni_handler = NULL;
102 	if (ni->ni_queue != NULL) {
103 		s = splimp();
104 		IF_DRAIN(ni->ni_queue);
105 		splx(s);
106 	}
107 }
108 
109 struct isrstat {
110 	int	isrs_count;			/* dispatch count */
111 	int	isrs_directed;			/* ...successfully dispatched */
112 	int	isrs_deferred;			/* ...queued instead */
113 	int	isrs_queued;			/* intentionally queueued */
114 	int	isrs_swi_count;			/* swi_net handlers called */
115 };
116 static struct isrstat isrstat;
117 
118 SYSCTL_NODE(_net, OID_AUTO, isr, CTLFLAG_RW, 0, "netisr counters");
119 
120 static int	netisr_enable = 0;
121 SYSCTL_INT(_net_isr, OID_AUTO, enable, CTLFLAG_RW,
122     &netisr_enable, 0, "enable direct dispatch");
123 TUNABLE_INT("net.isr.enable", &netisr_enable);
124 
125 SYSCTL_INT(_net_isr, OID_AUTO, count, CTLFLAG_RD,
126     &isrstat.isrs_count, 0, "");
127 SYSCTL_INT(_net_isr, OID_AUTO, directed, CTLFLAG_RD,
128     &isrstat.isrs_directed, 0, "");
129 SYSCTL_INT(_net_isr, OID_AUTO, deferred, CTLFLAG_RD,
130     &isrstat.isrs_deferred, 0, "");
131 SYSCTL_INT(_net_isr, OID_AUTO, queued, CTLFLAG_RD,
132     &isrstat.isrs_queued, 0, "");
133 SYSCTL_INT(_net_isr, OID_AUTO, swi_count, CTLFLAG_RD,
134     &isrstat.isrs_swi_count, 0, "");
135 
136 /*
137  * Process all packets currently present in a netisr queue.  Used to
138  * drain an existing set of packets waiting for processing when we
139  * begin direct dispatch, to avoid processing packets out of order.
140  */
141 static void
142 netisr_processqueue(struct netisr *ni)
143 {
144 	struct mbuf *m;
145 
146 	for (;;) {
147 		IF_DEQUEUE(ni->ni_queue, m);
148 		if (m == NULL)
149 			break;
150 		ni->ni_handler(m);
151 	}
152 }
153 
154 /*
155  * Call the netisr directly instead of queueing the packet, if possible.
156  *
157  * Ideally, the permissibility of calling the routine would be determined
158  * by checking if splnet() was asserted at the time the device interrupt
159  * occurred; if so, this indicates that someone is in the network stack.
160  *
161  * However, bus_setup_intr uses INTR_TYPE_NET, which sets splnet before
162  * calling the interrupt handler, so the previous mask is unavailable.
163  * Approximate this by checking intr_nesting_level instead; if any SWI
164  * handlers are running, the packet is queued instead.
165  */
166 void
167 netisr_dispatch(int num, struct mbuf *m)
168 {
169 	struct netisr *ni;
170 
171 	isrstat.isrs_count++;
172 	KASSERT(!(num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs))),
173 	    ("bad isr %d", num));
174 	ni = &netisrs[num];
175 	if (ni->ni_queue == NULL) {
176 		m_freem(m);
177 		return;
178 	}
179 	if (netisr_enable && mtx_trylock(&netisr_mtx)) {
180 		isrstat.isrs_directed++;
181 		/*
182 		 * One slight problem here is that packets might bypass
183 		 * each other in the stack, if an earlier one happened
184 		 * to get stuck in the queue.
185 		 *
186 		 * we can either:
187 		 *	a. drain the queue before handling this packet,
188 		 *	b. fallback to queueing the packet,
189 		 *	c. sweep the issue under the rug and ignore it.
190 		 *
191 		 * Currently, we do a).  Previously, we did c).
192 		 */
193 		netisr_processqueue(ni);
194 		ni->ni_handler(m);
195 		mtx_unlock(&netisr_mtx);
196 	} else {
197 		isrstat.isrs_deferred++;
198 		if (IF_HANDOFF(ni->ni_queue, m, NULL))
199 			schednetisr(num);
200 	}
201 }
202 
203 /*
204  * Same as above, but always queue.
205  * This is either used in places where we are not confident that
206  * direct dispatch is possible, or where queueing is required.
207  */
208 int
209 netisr_queue(int num, struct mbuf *m)
210 {
211 	struct netisr *ni;
212 
213 	KASSERT(!(num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs))),
214 	    ("bad isr %d", num));
215 	ni = &netisrs[num];
216 	if (ni->ni_queue == NULL) {
217 		m_freem(m);
218 		return (1);
219 	}
220 	isrstat.isrs_queued++;
221 	if (!IF_HANDOFF(ni->ni_queue, m, NULL))
222 		return (0);
223 	schednetisr(num);
224 	return (1);
225 }
226 
227 static void
228 swi_net(void *dummy)
229 {
230 	struct netisr *ni;
231 	u_int bits;
232 	int i;
233 #ifdef DEVICE_POLLING
234 	const int polling = 1;
235 #else
236 	const int polling = 0;
237 #endif
238 
239 	mtx_lock(&netisr_mtx);
240 	do {
241 		bits = atomic_readandclear_int(&netisr);
242 		if (bits == 0)
243 			break;
244 		while ((i = ffs(bits)) != 0) {
245 			isrstat.isrs_swi_count++;
246 			i--;
247 			bits &= ~(1 << i);
248 			ni = &netisrs[i];
249 			if (ni->ni_handler == NULL) {
250 				printf("swi_net: unregistered isr %d.\n", i);
251 				continue;
252 			}
253 			if (ni->ni_queue == NULL)
254 				ni->ni_handler(NULL);
255 			else
256 				netisr_processqueue(ni);
257 		}
258 	} while (polling);
259 	mtx_unlock(&netisr_mtx);
260 }
261 
262 static void
263 start_netisr(void *dummy)
264 {
265 
266 	mtx_init(&netisr_mtx, "netisr lock", NULL, MTX_DEF);
267 	if (swi_add(NULL, "net", swi_net, NULL, SWI_NET, 0, &net_ih))
268 		panic("start_netisr");
269 }
270 SYSINIT(start_netisr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_netisr, NULL)
271