2 * Copyright (c) 2001,2002,2003 Jonathan Lemon <jlemon@FreeBSD.org>
3 * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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
32 #include <sys/param.h>
34 #include <sys/rtprio.h>
35 #include <sys/systm.h>
36 #include <sys/interrupt.h>
37 #include <sys/kernel.h>
38 #include <sys/kthread.h>
40 #include <sys/malloc.h>
42 #include <sys/random.h>
43 #include <sys/resourcevar.h>
44 #include <sys/sysctl.h>
45 #include <sys/unistd.h>
46 #include <machine/atomic.h>
47 #include <machine/cpu.h>
48 #include <machine/stdarg.h>
51 #include <sys/socket.h>
54 #include <net/if_types.h>
55 #include <net/if_var.h>
56 #include <net/netisr.h>
59 * debug_mpsafenet controls network subsystem-wide use of the Giant lock,
60 * from system calls down to interrupt handlers. It can be changed only via
61 * a tunable at boot, not at run-time, due to the complexity of unwinding.
62 * The compiled default is set via a kernel option; right now, the default
63 * unless otherwise specified is to run the network stack without Giant.
66 int debug_mpsafenet = 0;
68 int debug_mpsafenet = 1;
70 int debug_mpsafenet_toolatetotwiddle = 0;
72 TUNABLE_INT("debug.mpsafenet", &debug_mpsafenet);
73 SYSCTL_INT(_debug, OID_AUTO, mpsafenet, CTLFLAG_RD, &debug_mpsafenet, 0,
74 "Enable/disable MPSAFE network support");
76 volatile unsigned int netisr; /* scheduling bits for network */
80 struct ifqueue *ni_queue;
87 * Not all network code is currently capable of running MPSAFE; however,
88 * most of it is. Since those sections that are not are generally optional
89 * components not shipped with default kernels, we provide a basic way to
90 * determine whether MPSAFE operation is permitted: based on a default of
91 * yes, we permit non-MPSAFE components to use a registration call to
92 * identify that they require Giant. If the system is early in the boot
93 * process still, then we change the debug_mpsafenet setting to choose a
94 * non-MPSAFE execution mode (degraded). If it's too late for that (since
95 * the setting cannot be changed at run time), we generate a console warning
96 * that the configuration may be unsafe.
98 static int mpsafe_warn_count;
101 * Function call implementing registration of a non-MPSAFE network component.
104 net_warn_not_mpsafe(const char *component)
108 * If we're running with Giant over the network stack, there is no
111 if (!debug_mpsafenet)
115 * If it's not too late to change the MPSAFE setting for the network
116 * stack, do so now. This effectively suppresses warnings by
117 * components registering later.
119 if (!debug_mpsafenet_toolatetotwiddle) {
121 printf("WARNING: debug.mpsafenet forced to 0 as %s requires "
122 "Giant\n", component);
127 * We must run without Giant, so generate a console warning with some
128 * information with what to do about it. The system may be operating
131 printf("WARNING: Network stack Giant-free, but %s requires Giant.\n",
133 if (mpsafe_warn_count == 0)
134 printf(" Consider adding 'options NET_WITH_GIANT' or "
135 "setting debug.mpsafenet=0\n");
140 * This sysinit is run after any pre-loaded or compiled-in components have
141 * announced that they require Giant, but before any modules loaded at
145 net_mpsafe_toolate(void *arg)
148 debug_mpsafenet_toolatetotwiddle = 1;
150 if (!debug_mpsafenet)
151 printf("WARNING: MPSAFE network stack disabled, expect "
152 "reduced performance.\n");
155 SYSINIT(net_mpsafe_toolate, SI_SUB_SETTINGS, SI_ORDER_ANY, net_mpsafe_toolate,
159 legacy_setsoftnet(void)
161 swi_sched(net_ih, 0);
165 netisr_register(int num, netisr_t *handler, struct ifqueue *inq, int flags)
168 KASSERT(!(num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs))),
169 ("bad isr %d", num));
170 netisrs[num].ni_handler = handler;
171 netisrs[num].ni_queue = inq;
172 if ((flags & NETISR_MPSAFE) && !debug_mpsafenet)
173 flags &= ~NETISR_MPSAFE;
174 netisrs[num].ni_flags = flags;
178 netisr_unregister(int num)
182 KASSERT(!(num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs))),
183 ("bad isr %d", num));
185 ni->ni_handler = NULL;
186 if (ni->ni_queue != NULL)
187 IF_DRAIN(ni->ni_queue);
192 int isrs_count; /* dispatch count */
193 int isrs_directed; /* ...directly dispatched */
194 int isrs_deferred; /* ...queued instead */
195 int isrs_queued; /* intentionally queueued */
196 int isrs_drop; /* dropped 'cuz no handler */
197 int isrs_swi_count; /* swi_net handlers called */
199 static struct isrstat isrstat;
201 SYSCTL_NODE(_net, OID_AUTO, isr, CTLFLAG_RW, 0, "netisr counters");
203 static int netisr_enable = 0;
204 SYSCTL_INT(_net_isr, OID_AUTO, enable, CTLFLAG_RW,
205 &netisr_enable, 0, "enable direct dispatch");
206 TUNABLE_INT("net.isr.enable", &netisr_enable);
208 SYSCTL_INT(_net_isr, OID_AUTO, count, CTLFLAG_RD,
209 &isrstat.isrs_count, 0, "");
210 SYSCTL_INT(_net_isr, OID_AUTO, directed, CTLFLAG_RD,
211 &isrstat.isrs_directed, 0, "");
212 SYSCTL_INT(_net_isr, OID_AUTO, deferred, CTLFLAG_RD,
213 &isrstat.isrs_deferred, 0, "");
214 SYSCTL_INT(_net_isr, OID_AUTO, queued, CTLFLAG_RD,
215 &isrstat.isrs_queued, 0, "");
216 SYSCTL_INT(_net_isr, OID_AUTO, drop, CTLFLAG_RD,
217 &isrstat.isrs_drop, 0, "");
218 SYSCTL_INT(_net_isr, OID_AUTO, swi_count, CTLFLAG_RD,
219 &isrstat.isrs_swi_count, 0, "");
222 * Process all packets currently present in a netisr queue. Used to
223 * drain an existing set of packets waiting for processing when we
224 * begin direct dispatch, to avoid processing packets out of order.
227 netisr_processqueue(struct netisr *ni)
232 IF_DEQUEUE(ni->ni_queue, m);
240 * Call the netisr directly instead of queueing the packet, if possible.
243 netisr_dispatch(int num, struct mbuf *m)
247 isrstat.isrs_count++; /* XXX redundant */
248 KASSERT(!(num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs))),
249 ("bad isr %d", num));
251 if (ni->ni_queue == NULL) {
257 * Do direct dispatch only for MPSAFE netisrs (and
258 * only when enabled). Note that when a netisr is
259 * marked MPSAFE we permit multiple concurrent instances
260 * to run. We guarantee only the order in which
261 * packets are processed for each "dispatch point" in
262 * the system (i.e. call to netisr_dispatch or
263 * netisr_queue). This insures ordering of packets
264 * from an interface but does not guarantee ordering
265 * between multiple places in the system (e.g. IP
266 * dispatched from interfaces vs. IP queued from IPSec).
268 if (netisr_enable && (ni->ni_flags & NETISR_MPSAFE)) {
269 isrstat.isrs_directed++;
271 * NB: We used to drain the queue before handling
272 * the packet but now do not. Doing so here will
273 * not preserve ordering so instead we fallback to
274 * guaranteeing order only from dispatch points
275 * in the system (see above).
279 isrstat.isrs_deferred++;
280 if (IF_HANDOFF(ni->ni_queue, m, NULL))
286 * Same as above, but always queue.
287 * This is either used in places where we are not confident that
288 * direct dispatch is possible, or where queueing is required.
289 * It returns (0) on success and ERRNO on failure. On failure the
290 * mbuf has been free'd.
293 netisr_queue(int num, struct mbuf *m)
297 KASSERT(!(num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs))),
298 ("bad isr %d", num));
300 if (ni->ni_queue == NULL) {
305 isrstat.isrs_queued++;
306 if (!IF_HANDOFF(ni->ni_queue, m, NULL))
307 return (ENOBUFS); /* IF_HANDOFF has free'd the mbuf */
318 #ifdef DEVICE_POLLING
319 const int polling = 1;
321 const int polling = 0;
325 bits = atomic_readandclear_int(&netisr);
328 while ((i = ffs(bits)) != 0) {
329 isrstat.isrs_swi_count++;
333 if (ni->ni_handler == NULL) {
334 printf("swi_net: unregistered isr %d.\n", i);
337 if ((ni->ni_flags & NETISR_MPSAFE) == 0) {
339 if (ni->ni_queue == NULL)
340 ni->ni_handler(NULL);
342 netisr_processqueue(ni);
345 if (ni->ni_queue == NULL)
346 ni->ni_handler(NULL);
348 netisr_processqueue(ni);
355 start_netisr(void *dummy)
358 if (swi_add(NULL, "net", swi_net, NULL, SWI_NET, INTR_MPSAFE, &net_ih))
359 panic("start_netisr");
361 SYSINIT(start_netisr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_netisr, NULL)