kern_cpu.c revision 1.57 1 /* $NetBSD: kern_cpu.c,v 1.57 2012/08/29 17:13:21 drochner Exp $ */
2
3 /*-
4 * Copyright (c) 2007, 2008, 2009, 2010, 2012 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 1.57 2012/08/29 17:13:21 drochner Exp $");
60
61 #include "opt_cpu_ucode.h"
62 #include "opt_compat_netbsd.h"
63
64 #include <sys/param.h>
65 #include <sys/systm.h>
66 #include <sys/idle.h>
67 #include <sys/sched.h>
68 #include <sys/intr.h>
69 #include <sys/conf.h>
70 #include <sys/cpu.h>
71 #include <sys/cpuio.h>
72 #include <sys/proc.h>
73 #include <sys/percpu.h>
74 #include <sys/kernel.h>
75 #include <sys/kauth.h>
76 #include <sys/xcall.h>
77 #include <sys/pool.h>
78 #include <sys/kmem.h>
79 #include <sys/select.h>
80 #include <sys/namei.h>
81 #include <sys/callout.h>
82
83 #include <uvm/uvm_extern.h>
84
85 /*
86 * If the port has stated that cpu_data is the first thing in cpu_info,
87 * verify that the claim is true. This will prevent them from getting out
88 * of sync.
89 */
90 #ifdef __HAVE_CPU_DATA_FIRST
91 CTASSERT(offsetof(struct cpu_info, ci_data) == 0);
92 #else
93 CTASSERT(offsetof(struct cpu_info, ci_data) != 0);
94 #endif
95
96 void cpuctlattach(int);
97
98 static void cpu_xc_online(struct cpu_info *);
99 static void cpu_xc_offline(struct cpu_info *);
100
101 dev_type_ioctl(cpuctl_ioctl);
102
103 const struct cdevsw cpuctl_cdevsw = {
104 nullopen, nullclose, nullread, nullwrite, cpuctl_ioctl,
105 nullstop, notty, nopoll, nommap, nokqfilter,
106 D_OTHER | D_MPSAFE
107 };
108
109 kmutex_t cpu_lock __cacheline_aligned;
110 int ncpu __read_mostly;
111 int ncpuonline __read_mostly;
112 bool mp_online __read_mostly;
113
114 /* Note: set on mi_cpu_attach() and idle_loop(). */
115 kcpuset_t * kcpuset_attached __read_mostly = NULL;
116 kcpuset_t * kcpuset_running __read_mostly = NULL;
117
118 struct cpuqueue cpu_queue __cacheline_aligned
119 = CIRCLEQ_HEAD_INITIALIZER(cpu_queue);
120
121 static struct cpu_info **cpu_infos __read_mostly;
122
123 /*
124 * mi_cpu_init: early initialisation of MI CPU related structures.
125 *
126 * Note: may not block and memory allocator is not yet available.
127 */
128 void
129 mi_cpu_init(void)
130 {
131
132 mutex_init(&cpu_lock, MUTEX_DEFAULT, IPL_NONE);
133
134 kcpuset_create(&kcpuset_attached, true);
135 kcpuset_create(&kcpuset_running, true);
136 kcpuset_set(kcpuset_running, 0);
137 }
138
139 int
140 mi_cpu_attach(struct cpu_info *ci)
141 {
142 int error;
143
144 KASSERT(maxcpus > 0);
145
146 ci->ci_index = ncpu;
147 kcpuset_set(kcpuset_attached, cpu_index(ci));
148
149 CIRCLEQ_INSERT_TAIL(&cpu_queue, ci, ci_data.cpu_qchain);
150 TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
151 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
152
153 /* This is useful for eg, per-cpu evcnt */
154 snprintf(ci->ci_data.cpu_name, sizeof(ci->ci_data.cpu_name), "cpu%d",
155 cpu_index(ci));
156
157 if (__predict_false(cpu_infos == NULL)) {
158 cpu_infos =
159 kmem_zalloc(sizeof(cpu_infos[0]) * maxcpus, KM_SLEEP);
160 }
161 cpu_infos[cpu_index(ci)] = ci;
162
163 sched_cpuattach(ci);
164
165 error = create_idle_lwp(ci);
166 if (error != 0) {
167 /* XXX revert sched_cpuattach */
168 return error;
169 }
170
171 if (ci == curcpu())
172 ci->ci_data.cpu_onproc = curlwp;
173 else
174 ci->ci_data.cpu_onproc = ci->ci_data.cpu_idlelwp;
175
176 percpu_init_cpu(ci);
177 softint_init(ci);
178 callout_init_cpu(ci);
179 xc_init_cpu(ci);
180 pool_cache_cpu_init(ci);
181 selsysinit(ci);
182 cache_cpu_init(ci);
183 TAILQ_INIT(&ci->ci_data.cpu_biodone);
184 ncpu++;
185 ncpuonline++;
186
187 return 0;
188 }
189
190 void
191 cpuctlattach(int dummy)
192 {
193
194 KASSERT(cpu_infos != NULL);
195 }
196
197 int
198 cpuctl_ioctl(dev_t dev, u_long cmd, void *data, int flag, lwp_t *l)
199 {
200 CPU_INFO_ITERATOR cii;
201 cpustate_t *cs;
202 struct cpu_info *ci;
203 int error, i;
204 u_int id;
205
206 error = 0;
207
208 mutex_enter(&cpu_lock);
209 switch (cmd) {
210 case IOC_CPU_SETSTATE:
211 cs = data;
212 error = kauth_authorize_system(l->l_cred,
213 KAUTH_SYSTEM_CPU, KAUTH_REQ_SYSTEM_CPU_SETSTATE, cs, NULL,
214 NULL);
215 if (error != 0)
216 break;
217 if (cs->cs_id >= maxcpus ||
218 (ci = cpu_lookup(cs->cs_id)) == NULL) {
219 error = ESRCH;
220 break;
221 }
222 cpu_setintr(ci, cs->cs_intr);
223 error = cpu_setstate(ci, cs->cs_online);
224 break;
225
226 case IOC_CPU_GETSTATE:
227 cs = data;
228 id = cs->cs_id;
229 memset(cs, 0, sizeof(*cs));
230 cs->cs_id = id;
231 if (cs->cs_id >= maxcpus ||
232 (ci = cpu_lookup(id)) == NULL) {
233 error = ESRCH;
234 break;
235 }
236 if ((ci->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
237 cs->cs_online = false;
238 else
239 cs->cs_online = true;
240 if ((ci->ci_schedstate.spc_flags & SPCF_NOINTR) != 0)
241 cs->cs_intr = false;
242 else
243 cs->cs_intr = true;
244 cs->cs_lastmod = (int32_t)ci->ci_schedstate.spc_lastmod;
245 cs->cs_lastmodhi = (int32_t)
246 (ci->ci_schedstate.spc_lastmod >> 32);
247 cs->cs_intrcnt = cpu_intr_count(ci) + 1;
248 cs->cs_hwid = ci->ci_cpuid;
249 break;
250
251 case IOC_CPU_MAPID:
252 i = 0;
253 for (CPU_INFO_FOREACH(cii, ci)) {
254 if (i++ == *(int *)data)
255 break;
256 }
257 if (ci == NULL)
258 error = ESRCH;
259 else
260 *(int *)data = cpu_index(ci);
261 break;
262
263 case IOC_CPU_GETCOUNT:
264 *(int *)data = ncpu;
265 break;
266
267 #ifdef CPU_UCODE
268 case IOC_CPU_UCODE_GET_VERSION:
269 error = cpu_ucode_get_version((struct cpu_ucode_version *)data);
270 break;
271
272 /* XXX ifdef COMPAT */
273 case OIOC_CPU_UCODE_GET_VERSION:
274 error = compat6_cpu_ucode_get_version((struct compat6_cpu_ucode *)data);
275 break;
276
277 case IOC_CPU_UCODE_APPLY:
278 error = kauth_authorize_machdep(l->l_cred,
279 KAUTH_MACHDEP_CPU_UCODE_APPLY,
280 NULL, NULL, NULL, NULL);
281 if (error != 0)
282 break;
283 error = cpu_ucode_apply((const struct cpu_ucode *)data);
284 break;
285
286 /* XXX ifdef COMPAT */
287 case OIOC_CPU_UCODE_APPLY:
288 error = kauth_authorize_machdep(l->l_cred,
289 KAUTH_MACHDEP_CPU_UCODE_APPLY,
290 NULL, NULL, NULL, NULL);
291 if (error != 0)
292 break;
293 error = compat6_cpu_ucode_apply((const struct compat6_cpu_ucode *)data);
294 break;
295 #endif
296
297 default:
298 error = ENOTTY;
299 break;
300 }
301 mutex_exit(&cpu_lock);
302
303 return error;
304 }
305
306 struct cpu_info *
307 cpu_lookup(u_int idx)
308 {
309 struct cpu_info *ci;
310
311 KASSERT(idx < maxcpus);
312
313 if (__predict_false(cpu_infos == NULL)) {
314 KASSERT(idx == 0);
315 return curcpu();
316 }
317
318 ci = cpu_infos[idx];
319 KASSERT(ci == NULL || cpu_index(ci) == idx);
320
321 return ci;
322 }
323
324 static void
325 cpu_xc_offline(struct cpu_info *ci)
326 {
327 struct schedstate_percpu *spc, *mspc = NULL;
328 struct cpu_info *target_ci;
329 struct lwp *l;
330 CPU_INFO_ITERATOR cii;
331 int s;
332
333 /*
334 * Thread that made the cross call (separate context) holds
335 * cpu_lock on our behalf.
336 */
337 spc = &ci->ci_schedstate;
338 s = splsched();
339 spc->spc_flags |= SPCF_OFFLINE;
340 splx(s);
341
342 /* Take the first available CPU for the migration. */
343 for (CPU_INFO_FOREACH(cii, target_ci)) {
344 mspc = &target_ci->ci_schedstate;
345 if ((mspc->spc_flags & SPCF_OFFLINE) == 0)
346 break;
347 }
348 KASSERT(target_ci != NULL);
349
350 /*
351 * Migrate all non-bound threads to the other CPU. Note that this
352 * runs from the xcall thread, thus handling of LSONPROC is not needed.
353 */
354 mutex_enter(proc_lock);
355 LIST_FOREACH(l, &alllwp, l_list) {
356 struct cpu_info *mci;
357
358 lwp_lock(l);
359 if (l->l_cpu != ci || (l->l_pflag & (LP_BOUND | LP_INTR))) {
360 lwp_unlock(l);
361 continue;
362 }
363 /* Regular case - no affinity. */
364 if (l->l_affinity == NULL) {
365 lwp_migrate(l, target_ci);
366 continue;
367 }
368 /* Affinity is set, find an online CPU in the set. */
369 for (CPU_INFO_FOREACH(cii, mci)) {
370 mspc = &mci->ci_schedstate;
371 if ((mspc->spc_flags & SPCF_OFFLINE) == 0 &&
372 kcpuset_isset(l->l_affinity, cpu_index(mci)))
373 break;
374 }
375 if (mci == NULL) {
376 lwp_unlock(l);
377 mutex_exit(proc_lock);
378 goto fail;
379 }
380 lwp_migrate(l, mci);
381 }
382 mutex_exit(proc_lock);
383
384 #ifdef __HAVE_MD_CPU_OFFLINE
385 cpu_offline_md();
386 #endif
387 return;
388 fail:
389 /* Just unset the SPCF_OFFLINE flag, caller will check */
390 s = splsched();
391 spc->spc_flags &= ~SPCF_OFFLINE;
392 splx(s);
393 }
394
395 static void
396 cpu_xc_online(struct cpu_info *ci)
397 {
398 struct schedstate_percpu *spc;
399 int s;
400
401 spc = &ci->ci_schedstate;
402 s = splsched();
403 spc->spc_flags &= ~SPCF_OFFLINE;
404 splx(s);
405 }
406
407 int
408 cpu_setstate(struct cpu_info *ci, bool online)
409 {
410 struct schedstate_percpu *spc;
411 CPU_INFO_ITERATOR cii;
412 struct cpu_info *ci2;
413 uint64_t where;
414 xcfunc_t func;
415 int nonline;
416
417 spc = &ci->ci_schedstate;
418
419 KASSERT(mutex_owned(&cpu_lock));
420
421 if (online) {
422 if ((spc->spc_flags & SPCF_OFFLINE) == 0)
423 return 0;
424 func = (xcfunc_t)cpu_xc_online;
425 ncpuonline++;
426 } else {
427 if ((spc->spc_flags & SPCF_OFFLINE) != 0)
428 return 0;
429 nonline = 0;
430 /*
431 * Ensure that at least one CPU within the processor set
432 * stays online. Revisit this later.
433 */
434 for (CPU_INFO_FOREACH(cii, ci2)) {
435 if ((ci2->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
436 continue;
437 if (ci2->ci_schedstate.spc_psid != spc->spc_psid)
438 continue;
439 nonline++;
440 }
441 if (nonline == 1)
442 return EBUSY;
443 func = (xcfunc_t)cpu_xc_offline;
444 ncpuonline--;
445 }
446
447 where = xc_unicast(0, func, ci, NULL, ci);
448 xc_wait(where);
449 if (online) {
450 KASSERT((spc->spc_flags & SPCF_OFFLINE) == 0);
451 } else if ((spc->spc_flags & SPCF_OFFLINE) == 0) {
452 /* If was not set offline, then it is busy */
453 return EBUSY;
454 }
455
456 spc->spc_lastmod = time_second;
457 return 0;
458 }
459
460 #ifdef __HAVE_INTR_CONTROL
461 static void
462 cpu_xc_intr(struct cpu_info *ci)
463 {
464 struct schedstate_percpu *spc;
465 int s;
466
467 spc = &ci->ci_schedstate;
468 s = splsched();
469 spc->spc_flags &= ~SPCF_NOINTR;
470 splx(s);
471 }
472
473 static void
474 cpu_xc_nointr(struct cpu_info *ci)
475 {
476 struct schedstate_percpu *spc;
477 int s;
478
479 spc = &ci->ci_schedstate;
480 s = splsched();
481 spc->spc_flags |= SPCF_NOINTR;
482 splx(s);
483 }
484
485 int
486 cpu_setintr(struct cpu_info *ci, bool intr)
487 {
488 struct schedstate_percpu *spc;
489 CPU_INFO_ITERATOR cii;
490 struct cpu_info *ci2;
491 uint64_t where;
492 xcfunc_t func;
493 int nintr;
494
495 spc = &ci->ci_schedstate;
496
497 KASSERT(mutex_owned(&cpu_lock));
498
499 if (intr) {
500 if ((spc->spc_flags & SPCF_NOINTR) == 0)
501 return 0;
502 func = (xcfunc_t)cpu_xc_intr;
503 } else {
504 if ((spc->spc_flags & SPCF_NOINTR) != 0)
505 return 0;
506 /*
507 * Ensure that at least one CPU within the system
508 * is handing device interrupts.
509 */
510 nintr = 0;
511 for (CPU_INFO_FOREACH(cii, ci2)) {
512 if ((ci2->ci_schedstate.spc_flags & SPCF_NOINTR) != 0)
513 continue;
514 if (ci2 == ci)
515 continue;
516 nintr++;
517 }
518 if (nintr == 0)
519 return EBUSY;
520 func = (xcfunc_t)cpu_xc_nointr;
521 }
522
523 where = xc_unicast(0, func, ci, NULL, ci);
524 xc_wait(where);
525 if (intr) {
526 KASSERT((spc->spc_flags & SPCF_NOINTR) == 0);
527 } else if ((spc->spc_flags & SPCF_NOINTR) == 0) {
528 /* If was not set offline, then it is busy */
529 return EBUSY;
530 }
531
532 /* Direct interrupts away from the CPU and record the change. */
533 cpu_intr_redistribute();
534 spc->spc_lastmod = time_second;
535 return 0;
536 }
537 #else /* __HAVE_INTR_CONTROL */
538 int
539 cpu_setintr(struct cpu_info *ci, bool intr)
540 {
541
542 return EOPNOTSUPP;
543 }
544
545 u_int
546 cpu_intr_count(struct cpu_info *ci)
547 {
548
549 return 0; /* 0 == "don't know" */
550 }
551 #endif /* __HAVE_INTR_CONTROL */
552
553 bool
554 cpu_softintr_p(void)
555 {
556
557 return (curlwp->l_pflag & LP_INTR) != 0;
558 }
559
560 #ifdef CPU_UCODE
561 int
562 cpu_ucode_load(struct cpu_ucode_softc *sc, const char *fwname)
563 {
564 firmware_handle_t fwh;
565 int error;
566
567 if (sc->sc_blob != NULL) {
568 firmware_free(sc->sc_blob, 0);
569 sc->sc_blob = NULL;
570 sc->sc_blobsize = 0;
571 }
572
573 error = cpu_ucode_md_open(&fwh, sc->loader_version, fwname);
574 if (error != 0) {
575 aprint_error("ucode: firmware_open failed: %i\n", error);
576 goto err0;
577 }
578
579 sc->sc_blobsize = firmware_get_size(fwh);
580 sc->sc_blob = firmware_malloc(sc->sc_blobsize);
581 if (sc->sc_blob == NULL) {
582 error = ENOMEM;
583 firmware_close(fwh);
584 goto err0;
585 }
586
587 error = firmware_read(fwh, 0, sc->sc_blob, sc->sc_blobsize);
588 firmware_close(fwh);
589 if (error != 0)
590 goto err1;
591
592 return 0;
593
594 err1:
595 firmware_free(sc->sc_blob, 0);
596 sc->sc_blob = NULL;
597 sc->sc_blobsize = 0;
598 err0:
599 return error;
600 }
601 #endif
602