xref: /netbsd-src/sys/kern/kern_cpu.c (revision aad9773e38ed2370a628a6416e098f9008fc10a7)
1 /*	$NetBSD: kern_cpu.c,v 1.66 2014/07/25 08:10:40 dholland 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 1.66 2014/07/25 08:10:40 dholland 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 #include <sys/pcu.h>
83 
84 #include <uvm/uvm_extern.h>
85 
86 /*
87  * If the port has stated that cpu_data is the first thing in cpu_info,
88  * verify that the claim is true. This will prevent them from getting out
89  * of sync.
90  */
91 #ifdef __HAVE_CPU_DATA_FIRST
92 CTASSERT(offsetof(struct cpu_info, ci_data) == 0);
93 #else
94 CTASSERT(offsetof(struct cpu_info, ci_data) != 0);
95 #endif
96 
97 void	cpuctlattach(int);
98 
99 static void	cpu_xc_online(struct cpu_info *);
100 static void	cpu_xc_offline(struct cpu_info *);
101 
102 dev_type_ioctl(cpuctl_ioctl);
103 
104 const struct cdevsw cpuctl_cdevsw = {
105 	.d_open = nullopen,
106 	.d_close = nullclose,
107 	.d_read = nullread,
108 	.d_write = nullwrite,
109 	.d_ioctl = cpuctl_ioctl,
110 	.d_stop = nullstop,
111 	.d_tty = notty,
112 	.d_poll = nopoll,
113 	.d_mmap = nommap,
114 	.d_kqfilter = nokqfilter,
115 	.d_discard = nodiscard,
116 	.d_flag = D_OTHER | D_MPSAFE
117 };
118 
119 kmutex_t	cpu_lock		__cacheline_aligned;
120 int		ncpu			__read_mostly;
121 int		ncpuonline		__read_mostly;
122 bool		mp_online		__read_mostly;
123 
124 /* An array of CPUs.  There are ncpu entries. */
125 struct cpu_info **cpu_infos		__read_mostly;
126 
127 /* Note: set on mi_cpu_attach() and idle_loop(). */
128 kcpuset_t *	kcpuset_attached	__read_mostly	= NULL;
129 kcpuset_t *	kcpuset_running		__read_mostly	= NULL;
130 
131 
132 static char cpu_model[128];
133 
134 /*
135  * mi_cpu_init: early initialisation of MI CPU related structures.
136  *
137  * Note: may not block and memory allocator is not yet available.
138  */
139 void
140 mi_cpu_init(void)
141 {
142 
143 	mutex_init(&cpu_lock, MUTEX_DEFAULT, IPL_NONE);
144 
145 	kcpuset_create(&kcpuset_attached, true);
146 	kcpuset_create(&kcpuset_running, true);
147 	kcpuset_set(kcpuset_running, 0);
148 }
149 
150 int
151 mi_cpu_attach(struct cpu_info *ci)
152 {
153 	int error;
154 
155 	KASSERT(maxcpus > 0);
156 
157 	ci->ci_index = ncpu;
158 	kcpuset_set(kcpuset_attached, cpu_index(ci));
159 
160 	/*
161 	 * Create a convenience cpuset of just ourselves.
162 	 */
163 	kcpuset_create(&ci->ci_data.cpu_kcpuset, true);
164 	kcpuset_set(ci->ci_data.cpu_kcpuset, cpu_index(ci));
165 
166 	TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
167 	__cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
168 
169 	/* This is useful for eg, per-cpu evcnt */
170 	snprintf(ci->ci_data.cpu_name, sizeof(ci->ci_data.cpu_name), "cpu%d",
171 	    cpu_index(ci));
172 
173 	if (__predict_false(cpu_infos == NULL)) {
174 		size_t ci_bufsize = (maxcpus + 1) * sizeof(struct cpu_info *);
175 		cpu_infos = kmem_zalloc(ci_bufsize, KM_SLEEP);
176 	}
177 	cpu_infos[cpu_index(ci)] = ci;
178 
179 	sched_cpuattach(ci);
180 
181 	error = create_idle_lwp(ci);
182 	if (error != 0) {
183 		/* XXX revert sched_cpuattach */
184 		return error;
185 	}
186 
187 	if (ci == curcpu())
188 		ci->ci_data.cpu_onproc = curlwp;
189 	else
190 		ci->ci_data.cpu_onproc = ci->ci_data.cpu_idlelwp;
191 
192 	percpu_init_cpu(ci);
193 	softint_init(ci);
194 	callout_init_cpu(ci);
195 	xc_init_cpu(ci);
196 	pool_cache_cpu_init(ci);
197 	selsysinit(ci);
198 	cache_cpu_init(ci);
199 	TAILQ_INIT(&ci->ci_data.cpu_biodone);
200 	ncpu++;
201 	ncpuonline++;
202 
203 	return 0;
204 }
205 
206 void
207 cpuctlattach(int dummy)
208 {
209 
210 	KASSERT(cpu_infos != NULL);
211 }
212 
213 int
214 cpuctl_ioctl(dev_t dev, u_long cmd, void *data, int flag, lwp_t *l)
215 {
216 	CPU_INFO_ITERATOR cii;
217 	cpustate_t *cs;
218 	struct cpu_info *ci;
219 	int error, i;
220 	u_int id;
221 
222 	error = 0;
223 
224 	mutex_enter(&cpu_lock);
225 	switch (cmd) {
226 	case IOC_CPU_SETSTATE:
227 		cs = data;
228 		error = kauth_authorize_system(l->l_cred,
229 		    KAUTH_SYSTEM_CPU, KAUTH_REQ_SYSTEM_CPU_SETSTATE, cs, NULL,
230 		    NULL);
231 		if (error != 0)
232 			break;
233 		if (cs->cs_id >= maxcpus ||
234 		    (ci = cpu_lookup(cs->cs_id)) == NULL) {
235 			error = ESRCH;
236 			break;
237 		}
238 		cpu_setintr(ci, cs->cs_intr);
239 		error = cpu_setstate(ci, cs->cs_online);
240 		break;
241 
242 	case IOC_CPU_GETSTATE:
243 		cs = data;
244 		id = cs->cs_id;
245 		memset(cs, 0, sizeof(*cs));
246 		cs->cs_id = id;
247 		if (cs->cs_id >= maxcpus ||
248 		    (ci = cpu_lookup(id)) == NULL) {
249 			error = ESRCH;
250 			break;
251 		}
252 		if ((ci->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
253 			cs->cs_online = false;
254 		else
255 			cs->cs_online = true;
256 		if ((ci->ci_schedstate.spc_flags & SPCF_NOINTR) != 0)
257 			cs->cs_intr = false;
258 		else
259 			cs->cs_intr = true;
260 		cs->cs_lastmod = (int32_t)ci->ci_schedstate.spc_lastmod;
261 		cs->cs_lastmodhi = (int32_t)
262 		    (ci->ci_schedstate.spc_lastmod >> 32);
263 		cs->cs_intrcnt = cpu_intr_count(ci) + 1;
264 		cs->cs_hwid = ci->ci_cpuid;
265 		break;
266 
267 	case IOC_CPU_MAPID:
268 		i = 0;
269 		for (CPU_INFO_FOREACH(cii, ci)) {
270 			if (i++ == *(int *)data)
271 				break;
272 		}
273 		if (ci == NULL)
274 			error = ESRCH;
275 		else
276 			*(int *)data = cpu_index(ci);
277 		break;
278 
279 	case IOC_CPU_GETCOUNT:
280 		*(int *)data = ncpu;
281 		break;
282 
283 #ifdef CPU_UCODE
284 	case IOC_CPU_UCODE_GET_VERSION:
285 		error = cpu_ucode_get_version((struct cpu_ucode_version *)data);
286 		break;
287 
288 #ifdef COMPAT_60
289 	case OIOC_CPU_UCODE_GET_VERSION:
290 		error = compat6_cpu_ucode_get_version((struct compat6_cpu_ucode *)data);
291 		break;
292 #endif
293 
294 	case IOC_CPU_UCODE_APPLY:
295 		error = kauth_authorize_machdep(l->l_cred,
296 		    KAUTH_MACHDEP_CPU_UCODE_APPLY,
297 		    NULL, NULL, NULL, NULL);
298 		if (error != 0)
299 			break;
300 		error = cpu_ucode_apply((const struct cpu_ucode *)data);
301 		break;
302 
303 #ifdef COMPAT_60
304 	case OIOC_CPU_UCODE_APPLY:
305 		error = kauth_authorize_machdep(l->l_cred,
306 		    KAUTH_MACHDEP_CPU_UCODE_APPLY,
307 		    NULL, NULL, NULL, NULL);
308 		if (error != 0)
309 			break;
310 		error = compat6_cpu_ucode_apply((const struct compat6_cpu_ucode *)data);
311 		break;
312 #endif
313 #endif
314 
315 	default:
316 		error = ENOTTY;
317 		break;
318 	}
319 	mutex_exit(&cpu_lock);
320 
321 	return error;
322 }
323 
324 struct cpu_info *
325 cpu_lookup(u_int idx)
326 {
327 	struct cpu_info *ci;
328 
329 	KASSERT(idx < maxcpus);
330 
331 	if (__predict_false(cpu_infos == NULL)) {
332 		KASSERT(idx == 0);
333 		return curcpu();
334 	}
335 
336 	ci = cpu_infos[idx];
337 	KASSERT(ci == NULL || cpu_index(ci) == idx);
338 
339 	return ci;
340 }
341 
342 static void
343 cpu_xc_offline(struct cpu_info *ci)
344 {
345 	struct schedstate_percpu *spc, *mspc = NULL;
346 	struct cpu_info *target_ci;
347 	struct lwp *l;
348 	CPU_INFO_ITERATOR cii;
349 	int s;
350 
351 	/*
352 	 * Thread that made the cross call (separate context) holds
353 	 * cpu_lock on our behalf.
354 	 */
355 	spc = &ci->ci_schedstate;
356 	s = splsched();
357 	spc->spc_flags |= SPCF_OFFLINE;
358 	splx(s);
359 
360 	/* Take the first available CPU for the migration. */
361 	for (CPU_INFO_FOREACH(cii, target_ci)) {
362 		mspc = &target_ci->ci_schedstate;
363 		if ((mspc->spc_flags & SPCF_OFFLINE) == 0)
364 			break;
365 	}
366 	KASSERT(target_ci != NULL);
367 
368 	/*
369 	 * Migrate all non-bound threads to the other CPU.  Note that this
370 	 * runs from the xcall thread, thus handling of LSONPROC is not needed.
371 	 */
372 	mutex_enter(proc_lock);
373 	LIST_FOREACH(l, &alllwp, l_list) {
374 		struct cpu_info *mci;
375 
376 		lwp_lock(l);
377 		if (l->l_cpu != ci || (l->l_pflag & (LP_BOUND | LP_INTR))) {
378 			lwp_unlock(l);
379 			continue;
380 		}
381 		/* Regular case - no affinity. */
382 		if (l->l_affinity == NULL) {
383 			lwp_migrate(l, target_ci);
384 			continue;
385 		}
386 		/* Affinity is set, find an online CPU in the set. */
387 		for (CPU_INFO_FOREACH(cii, mci)) {
388 			mspc = &mci->ci_schedstate;
389 			if ((mspc->spc_flags & SPCF_OFFLINE) == 0 &&
390 			    kcpuset_isset(l->l_affinity, cpu_index(mci)))
391 				break;
392 		}
393 		if (mci == NULL) {
394 			lwp_unlock(l);
395 			mutex_exit(proc_lock);
396 			goto fail;
397 		}
398 		lwp_migrate(l, mci);
399 	}
400 	mutex_exit(proc_lock);
401 
402 #if PCU_UNIT_COUNT > 0
403 	pcu_save_all_on_cpu();
404 #endif
405 
406 #ifdef __HAVE_MD_CPU_OFFLINE
407 	cpu_offline_md();
408 #endif
409 	return;
410 fail:
411 	/* Just unset the SPCF_OFFLINE flag, caller will check */
412 	s = splsched();
413 	spc->spc_flags &= ~SPCF_OFFLINE;
414 	splx(s);
415 }
416 
417 static void
418 cpu_xc_online(struct cpu_info *ci)
419 {
420 	struct schedstate_percpu *spc;
421 	int s;
422 
423 	spc = &ci->ci_schedstate;
424 	s = splsched();
425 	spc->spc_flags &= ~SPCF_OFFLINE;
426 	splx(s);
427 }
428 
429 int
430 cpu_setstate(struct cpu_info *ci, bool online)
431 {
432 	struct schedstate_percpu *spc;
433 	CPU_INFO_ITERATOR cii;
434 	struct cpu_info *ci2;
435 	uint64_t where;
436 	xcfunc_t func;
437 	int nonline;
438 
439 	spc = &ci->ci_schedstate;
440 
441 	KASSERT(mutex_owned(&cpu_lock));
442 
443 	if (online) {
444 		if ((spc->spc_flags & SPCF_OFFLINE) == 0)
445 			return 0;
446 		func = (xcfunc_t)cpu_xc_online;
447 		ncpuonline++;
448 	} else {
449 		if ((spc->spc_flags & SPCF_OFFLINE) != 0)
450 			return 0;
451 		nonline = 0;
452 		/*
453 		 * Ensure that at least one CPU within the processor set
454 		 * stays online.  Revisit this later.
455 		 */
456 		for (CPU_INFO_FOREACH(cii, ci2)) {
457 			if ((ci2->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
458 				continue;
459 			if (ci2->ci_schedstate.spc_psid != spc->spc_psid)
460 				continue;
461 			nonline++;
462 		}
463 		if (nonline == 1)
464 			return EBUSY;
465 		func = (xcfunc_t)cpu_xc_offline;
466 		ncpuonline--;
467 	}
468 
469 	where = xc_unicast(0, func, ci, NULL, ci);
470 	xc_wait(where);
471 	if (online) {
472 		KASSERT((spc->spc_flags & SPCF_OFFLINE) == 0);
473 	} else if ((spc->spc_flags & SPCF_OFFLINE) == 0) {
474 		/* If was not set offline, then it is busy */
475 		return EBUSY;
476 	}
477 
478 	spc->spc_lastmod = time_second;
479 	return 0;
480 }
481 
482 int
483 cpu_setmodel(const char *fmt, ...)
484 {
485 	int len;
486 	va_list ap;
487 
488 	va_start(ap, fmt);
489 	len = vsnprintf(cpu_model, sizeof(cpu_model), fmt, ap);
490 	va_end(ap);
491 	return len;
492 }
493 
494 const char *
495 cpu_getmodel(void)
496 {
497 	return cpu_model;
498 }
499 
500 #ifdef __HAVE_INTR_CONTROL
501 static void
502 cpu_xc_intr(struct cpu_info *ci)
503 {
504 	struct schedstate_percpu *spc;
505 	int s;
506 
507 	spc = &ci->ci_schedstate;
508 	s = splsched();
509 	spc->spc_flags &= ~SPCF_NOINTR;
510 	splx(s);
511 }
512 
513 static void
514 cpu_xc_nointr(struct cpu_info *ci)
515 {
516 	struct schedstate_percpu *spc;
517 	int s;
518 
519 	spc = &ci->ci_schedstate;
520 	s = splsched();
521 	spc->spc_flags |= SPCF_NOINTR;
522 	splx(s);
523 }
524 
525 int
526 cpu_setintr(struct cpu_info *ci, bool intr)
527 {
528 	struct schedstate_percpu *spc;
529 	CPU_INFO_ITERATOR cii;
530 	struct cpu_info *ci2;
531 	uint64_t where;
532 	xcfunc_t func;
533 	int nintr;
534 
535 	spc = &ci->ci_schedstate;
536 
537 	KASSERT(mutex_owned(&cpu_lock));
538 
539 	if (intr) {
540 		if ((spc->spc_flags & SPCF_NOINTR) == 0)
541 			return 0;
542 		func = (xcfunc_t)cpu_xc_intr;
543 	} else {
544 		if ((spc->spc_flags & SPCF_NOINTR) != 0)
545 			return 0;
546 		/*
547 		 * Ensure that at least one CPU within the system
548 		 * is handing device interrupts.
549 		 */
550 		nintr = 0;
551 		for (CPU_INFO_FOREACH(cii, ci2)) {
552 			if ((ci2->ci_schedstate.spc_flags & SPCF_NOINTR) != 0)
553 				continue;
554 			if (ci2 == ci)
555 				continue;
556 			nintr++;
557 		}
558 		if (nintr == 0)
559 			return EBUSY;
560 		func = (xcfunc_t)cpu_xc_nointr;
561 	}
562 
563 	where = xc_unicast(0, func, ci, NULL, ci);
564 	xc_wait(where);
565 	if (intr) {
566 		KASSERT((spc->spc_flags & SPCF_NOINTR) == 0);
567 	} else if ((spc->spc_flags & SPCF_NOINTR) == 0) {
568 		/* If was not set offline, then it is busy */
569 		return EBUSY;
570 	}
571 
572 	/* Direct interrupts away from the CPU and record the change. */
573 	cpu_intr_redistribute();
574 	spc->spc_lastmod = time_second;
575 	return 0;
576 }
577 #else	/* __HAVE_INTR_CONTROL */
578 int
579 cpu_setintr(struct cpu_info *ci, bool intr)
580 {
581 
582 	return EOPNOTSUPP;
583 }
584 
585 u_int
586 cpu_intr_count(struct cpu_info *ci)
587 {
588 
589 	return 0;	/* 0 == "don't know" */
590 }
591 #endif	/* __HAVE_INTR_CONTROL */
592 
593 bool
594 cpu_softintr_p(void)
595 {
596 
597 	return (curlwp->l_pflag & LP_INTR) != 0;
598 }
599 
600 #ifdef CPU_UCODE
601 int
602 cpu_ucode_load(struct cpu_ucode_softc *sc, const char *fwname)
603 {
604 	firmware_handle_t fwh;
605 	int error;
606 
607 	if (sc->sc_blob != NULL) {
608 		firmware_free(sc->sc_blob, 0);
609 		sc->sc_blob = NULL;
610 		sc->sc_blobsize = 0;
611 	}
612 
613 	error = cpu_ucode_md_open(&fwh, sc->loader_version, fwname);
614 	if (error != 0) {
615 		aprint_error("ucode: firmware_open failed: %i\n", error);
616 		goto err0;
617 	}
618 
619 	sc->sc_blobsize = firmware_get_size(fwh);
620 	sc->sc_blob = firmware_malloc(sc->sc_blobsize);
621 	if (sc->sc_blob == NULL) {
622 		error = ENOMEM;
623 		firmware_close(fwh);
624 		goto err0;
625 	}
626 
627 	error = firmware_read(fwh, 0, sc->sc_blob, sc->sc_blobsize);
628 	firmware_close(fwh);
629 	if (error != 0)
630 		goto err1;
631 
632 	return 0;
633 
634 err1:
635 	firmware_free(sc->sc_blob, 0);
636 	sc->sc_blob = NULL;
637 	sc->sc_blobsize = 0;
638 err0:
639 	return error;
640 }
641 #endif
642