kern_cpu.c revision 1.21.6.7 1 /* $NetBSD: kern_cpu.c,v 1.21.6.7 2009/01/17 13:29:18 mjf Exp $ */
2
3 /*-
4 * Copyright (c) 2007, 2008 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Andrew Doran.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 /*-
33 * Copyright (c)2007 YAMAMOTO Takashi,
34 * All rights reserved.
35 *
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
38 * are met:
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
44 *
45 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 * SUCH DAMAGE.
56 */
57
58 #include <sys/cdefs.h>
59 __KERNEL_RCSID(0, "$NetBSD: kern_cpu.c,v 1.21.6.7 2009/01/17 13:29:18 mjf Exp $");
60
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/idle.h>
64 #include <sys/sched.h>
65 #include <sys/intr.h>
66 #include <sys/conf.h>
67 #include <sys/cpu.h>
68 #include <sys/cpuio.h>
69 #include <sys/proc.h>
70 #include <sys/percpu.h>
71 #include <sys/kernel.h>
72 #include <sys/kauth.h>
73 #include <sys/xcall.h>
74 #include <sys/pool.h>
75 #include <sys/kmem.h>
76 #include <sys/select.h>
77 #include <sys/namei.h>
78 #include <sys/callout.h>
79
80 #include <uvm/uvm_extern.h>
81
82 void cpuctlattach(int);
83
84 static void cpu_xc_online(struct cpu_info *);
85 static void cpu_xc_offline(struct cpu_info *);
86
87 dev_type_ioctl(cpuctl_ioctl);
88
89 const struct cdevsw cpuctl_cdevsw = {
90 nullopen, nullclose, nullread, nullwrite, cpuctl_ioctl,
91 nullstop, notty, nopoll, nommap, nokqfilter,
92 D_OTHER | D_MPSAFE
93 };
94
95 kmutex_t cpu_lock;
96 int ncpu;
97 int ncpuonline;
98 bool mp_online;
99 struct cpuqueue cpu_queue = CIRCLEQ_HEAD_INITIALIZER(cpu_queue);
100
101 static struct cpu_info *cpu_infos[MAXCPUS];
102
103 int
104 mi_cpu_attach(struct cpu_info *ci)
105 {
106 int error;
107
108 ci->ci_index = ncpu;
109 cpu_infos[cpu_index(ci)] = ci;
110 CIRCLEQ_INSERT_TAIL(&cpu_queue, ci, ci_data.cpu_qchain);
111 TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
112 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
113
114 sched_cpuattach(ci);
115
116 error = create_idle_lwp(ci);
117 if (error != 0) {
118 /* XXX revert sched_cpuattach */
119 return error;
120 }
121
122 if (ci == curcpu())
123 ci->ci_data.cpu_onproc = curlwp;
124 else
125 ci->ci_data.cpu_onproc = ci->ci_data.cpu_idlelwp;
126
127 percpu_init_cpu(ci);
128 softint_init(ci);
129 callout_init_cpu(ci);
130 xc_init_cpu(ci);
131 pool_cache_cpu_init(ci);
132 selsysinit(ci);
133 cache_cpu_init(ci);
134 TAILQ_INIT(&ci->ci_data.cpu_biodone);
135 ncpu++;
136 ncpuonline++;
137
138 return 0;
139 }
140
141 void
142 cpuctlattach(int dummy)
143 {
144 int maj = cdevsw_lookup_major(&cpuctl_cdevsw);
145 device_register_name(makedev(maj, 0), NULL, true, DEV_OTHER, "cpuctl");
146 }
147
148 int
149 cpuctl_ioctl(dev_t dev, u_long cmd, void *data, int flag, lwp_t *l)
150 {
151 CPU_INFO_ITERATOR cii;
152 cpustate_t *cs;
153 struct cpu_info *ci;
154 int error, i;
155 u_int id;
156
157 error = 0;
158
159 mutex_enter(&cpu_lock);
160 switch (cmd) {
161 case IOC_CPU_SETSTATE:
162 cs = data;
163 error = kauth_authorize_system(l->l_cred,
164 KAUTH_SYSTEM_CPU, KAUTH_REQ_SYSTEM_CPU_SETSTATE, cs, NULL,
165 NULL);
166 if (error != 0)
167 break;
168 if (cs->cs_id >= __arraycount(cpu_infos) ||
169 (ci = cpu_lookup(cs->cs_id)) == NULL) {
170 error = ESRCH;
171 break;
172 }
173 if (!cs->cs_intr) {
174 error = EOPNOTSUPP;
175 break;
176 }
177 error = cpu_setstate(ci, cs->cs_online);
178 break;
179
180 case IOC_CPU_GETSTATE:
181 cs = data;
182 id = cs->cs_id;
183 memset(cs, 0, sizeof(*cs));
184 cs->cs_id = id;
185 if (cs->cs_id >= __arraycount(cpu_infos) ||
186 (ci = cpu_lookup(id)) == NULL) {
187 error = ESRCH;
188 break;
189 }
190 if ((ci->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
191 cs->cs_online = false;
192 else
193 cs->cs_online = true;
194 cs->cs_intr = true;
195 cs->cs_lastmod = ci->ci_schedstate.spc_lastmod;
196 break;
197
198 case IOC_CPU_MAPID:
199 i = 0;
200 for (CPU_INFO_FOREACH(cii, ci)) {
201 if (i++ == *(int *)data)
202 break;
203 }
204 if (ci == NULL)
205 error = ESRCH;
206 else
207 *(int *)data = cpu_index(ci);
208 break;
209
210 case IOC_CPU_GETCOUNT:
211 *(int *)data = ncpu;
212 break;
213
214 default:
215 error = ENOTTY;
216 break;
217 }
218 mutex_exit(&cpu_lock);
219
220 return error;
221 }
222
223 struct cpu_info *
224 cpu_lookup(u_int idx)
225 {
226 struct cpu_info *ci = cpu_infos[idx];
227
228 KASSERT(idx < __arraycount(cpu_infos));
229 KASSERT(ci == NULL || cpu_index(ci) == idx);
230
231 return ci;
232 }
233
234 static void
235 cpu_xc_offline(struct cpu_info *ci)
236 {
237 struct schedstate_percpu *spc, *mspc = NULL;
238 struct cpu_info *target_ci;
239 struct lwp *l;
240 CPU_INFO_ITERATOR cii;
241 int s;
242
243 /*
244 * Thread which sent unicast (separate context) is holding
245 * the cpu_lock for us.
246 */
247 spc = &ci->ci_schedstate;
248 s = splsched();
249 spc->spc_flags |= SPCF_OFFLINE;
250 splx(s);
251
252 /* Take the first available CPU for the migration */
253 for (CPU_INFO_FOREACH(cii, target_ci)) {
254 mspc = &target_ci->ci_schedstate;
255 if ((mspc->spc_flags & SPCF_OFFLINE) == 0)
256 break;
257 }
258 KASSERT(target_ci != NULL);
259
260 /*
261 * Migrate all non-bound threads to the other CPU. Note that this
262 * runs from the xcall thread, thus handling of LSONPROC is not needed.
263 */
264 mutex_enter(proc_lock);
265 LIST_FOREACH(l, &alllwp, l_list) {
266 struct cpu_info *mci;
267
268 lwp_lock(l);
269 if (l->l_cpu != ci || (l->l_pflag & (LP_BOUND | LP_INTR))) {
270 lwp_unlock(l);
271 continue;
272 }
273 /* Normal case - no affinity */
274 if ((l->l_flag & LW_AFFINITY) == 0) {
275 lwp_migrate(l, target_ci);
276 continue;
277 }
278 /* Affinity is set, find an online CPU in the set */
279 KASSERT(l->l_affinity != NULL);
280 for (CPU_INFO_FOREACH(cii, mci)) {
281 mspc = &mci->ci_schedstate;
282 if ((mspc->spc_flags & SPCF_OFFLINE) == 0 &&
283 kcpuset_isset(cpu_index(mci), l->l_affinity))
284 break;
285 }
286 if (mci == NULL) {
287 lwp_unlock(l);
288 mutex_exit(proc_lock);
289 goto fail;
290 }
291 lwp_migrate(l, mci);
292 }
293 mutex_exit(proc_lock);
294
295 #ifdef __HAVE_MD_CPU_OFFLINE
296 cpu_offline_md();
297 #endif
298 return;
299 fail:
300 /* Just unset the SPCF_OFFLINE flag, caller will check */
301 s = splsched();
302 spc->spc_flags &= ~SPCF_OFFLINE;
303 splx(s);
304 }
305
306 static void
307 cpu_xc_online(struct cpu_info *ci)
308 {
309 struct schedstate_percpu *spc;
310 int s;
311
312 spc = &ci->ci_schedstate;
313 s = splsched();
314 spc->spc_flags &= ~SPCF_OFFLINE;
315 splx(s);
316 }
317
318 int
319 cpu_setstate(struct cpu_info *ci, bool online)
320 {
321 struct schedstate_percpu *spc;
322 CPU_INFO_ITERATOR cii;
323 struct cpu_info *ci2;
324 uint64_t where;
325 xcfunc_t func;
326 int nonline;
327
328 spc = &ci->ci_schedstate;
329
330 KASSERT(mutex_owned(&cpu_lock));
331
332 if (online) {
333 if ((spc->spc_flags & SPCF_OFFLINE) == 0)
334 return 0;
335 func = (xcfunc_t)cpu_xc_online;
336 ncpuonline++;
337 } else {
338 if ((spc->spc_flags & SPCF_OFFLINE) != 0)
339 return 0;
340 nonline = 0;
341 /*
342 * Ensure that at least one CPU within the processor set
343 * stays online. Revisit this later.
344 */
345 for (CPU_INFO_FOREACH(cii, ci2)) {
346 if ((ci2->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
347 continue;
348 if (ci2->ci_schedstate.spc_psid != spc->spc_psid)
349 continue;
350 nonline++;
351 }
352 if (nonline == 1)
353 return EBUSY;
354 func = (xcfunc_t)cpu_xc_offline;
355 ncpuonline--;
356 }
357
358 where = xc_unicast(0, func, ci, NULL, ci);
359 xc_wait(where);
360 if (online) {
361 KASSERT((spc->spc_flags & SPCF_OFFLINE) == 0);
362 } else if ((spc->spc_flags & SPCF_OFFLINE) == 0) {
363 /* If was not set offline, then it is busy */
364 return EBUSY;
365 }
366
367 spc->spc_lastmod = time_second;
368 return 0;
369 }
370
371 bool
372 cpu_softintr_p(void)
373 {
374
375 return (curlwp->l_pflag & LP_INTR) != 0;
376 }
377