xref: /netbsd-src/sys/arch/xen/x86/cpu.c (revision 46f5119e40af2e51998f686b2fdcc76b5488f7f3)
1 /*	$NetBSD: cpu.c,v 1.56 2011/02/26 14:43:18 jruoho Exp $	*/
2 /* NetBSD: cpu.c,v 1.18 2004/02/20 17:35:01 yamt Exp  */
3 
4 /*-
5  * Copyright (c) 2000 The NetBSD Foundation, Inc.
6  * Copyright (c) 2002, 2006, 2007 YAMAMOTO Takashi,
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to The NetBSD Foundation
10  * by RedBack Networks Inc.
11  *
12  * Author: Bill Sommerfeld
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
25  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
26  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
27  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 /*
37  * Copyright (c) 1999 Stefan Grefen
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. All advertising materials mentioning features or use of this software
48  *    must display the following acknowledgement:
49  *      This product includes software developed by the NetBSD
50  *      Foundation, Inc. and its contributors.
51  * 4. Neither the name of The NetBSD Foundation nor the names of its
52  *    contributors may be used to endorse or promote products derived
53  *    from this software without specific prior written permission.
54  *
55  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
56  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
59  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65  * SUCH DAMAGE.
66  */
67 
68 #include <sys/cdefs.h>
69 __KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.56 2011/02/26 14:43:18 jruoho Exp $");
70 
71 #include "opt_ddb.h"
72 #include "opt_multiprocessor.h"
73 #include "opt_mpbios.h"		/* for MPDEBUG */
74 #include "opt_mtrr.h"
75 #include "opt_xen.h"
76 
77 #include "lapic.h"
78 #include "ioapic.h"
79 
80 #include <sys/param.h>
81 #include <sys/proc.h>
82 #include <sys/systm.h>
83 #include <sys/device.h>
84 #include <sys/kmem.h>
85 #include <sys/cpu.h>
86 #include <sys/atomic.h>
87 #include <sys/reboot.h>
88 
89 #include <uvm/uvm.h>
90 
91 #include <machine/cpufunc.h>
92 #include <machine/cpuvar.h>
93 #include <machine/pmap.h>
94 #include <machine/vmparam.h>
95 #include <machine/mpbiosvar.h>
96 #include <machine/pcb.h>
97 #include <machine/specialreg.h>
98 #include <machine/segments.h>
99 #include <machine/gdt.h>
100 #include <machine/mtrr.h>
101 #include <machine/pio.h>
102 
103 #include <xen/vcpuvar.h>
104 
105 #if NLAPIC > 0
106 #include <machine/apicvar.h>
107 #include <machine/i82489reg.h>
108 #include <machine/i82489var.h>
109 #endif
110 
111 #include <dev/ic/mc146818reg.h>
112 #include <dev/isa/isareg.h>
113 
114 #if MAXCPUS > 32
115 #error cpu_info contains 32bit bitmasks
116 #endif
117 
118 static int	cpu_match(device_t, cfdata_t, void *);
119 static void	cpu_attach(device_t, device_t, void *);
120 static void	cpu_defer(device_t);
121 static int	cpu_rescan(device_t, const char *, const int *);
122 static void	cpu_childdetached(device_t, device_t);
123 static int	vcpu_match(device_t, cfdata_t, void *);
124 static void	vcpu_attach(device_t, device_t, void *);
125 static void	cpu_attach_common(device_t, device_t, void *);
126 void		cpu_offline_md(void);
127 
128 struct cpu_softc {
129 	device_t sc_dev;		/* device tree glue */
130 	struct cpu_info *sc_info;	/* pointer to CPU info */
131 	bool sc_wasonline;
132 };
133 
134 int mp_cpu_start(struct cpu_info *, paddr_t);
135 void mp_cpu_start_cleanup(struct cpu_info *);
136 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL,
137 				      mp_cpu_start_cleanup };
138 
139 CFATTACH_DECL2_NEW(cpu, sizeof(struct cpu_softc),
140     cpu_match, cpu_attach, NULL, NULL, cpu_rescan, cpu_childdetached);
141 
142 CFATTACH_DECL_NEW(vcpu, sizeof(struct cpu_softc),
143     vcpu_match, vcpu_attach, NULL, NULL);
144 
145 /*
146  * Statically-allocated CPU info for the primary CPU (or the only
147  * CPU, on uniprocessors).  The CPU info list is initialized to
148  * point at it.
149  */
150 #ifdef TRAPLOG
151 #include <machine/tlog.h>
152 struct tlog tlog_primary;
153 #endif
154 struct cpu_info cpu_info_primary __aligned(CACHE_LINE_SIZE) = {
155 	.ci_dev = 0,
156 	.ci_self = &cpu_info_primary,
157 	.ci_idepth = -1,
158 	.ci_curlwp = &lwp0,
159 	.ci_curldt = -1,
160 #ifdef TRAPLOG
161 	.ci_tlog = &tlog_primary,
162 #endif
163 
164 };
165 struct cpu_info phycpu_info_primary __aligned(CACHE_LINE_SIZE) = {
166 	.ci_dev = 0,
167 	.ci_self = &phycpu_info_primary,
168 };
169 
170 struct cpu_info *cpu_info_list = &cpu_info_primary;
171 struct cpu_info *phycpu_info_list = &phycpu_info_primary;
172 
173 static void	cpu_set_tss_gates(struct cpu_info *ci);
174 
175 uint32_t cpus_attached = 0;
176 uint32_t cpus_running = 0;
177 
178 uint32_t phycpus_attached = 0;
179 uint32_t phycpus_running = 0;
180 
181 uint32_t cpu_feature[5]; /* X86 CPUID feature bits
182 			  *	[0] basic features %edx
183 			  *	[1] basic features %ecx
184 			  *	[2] extended features %edx
185 			  *	[3] extended features %ecx
186 			  *	[4] VIA padlock features
187 			  */
188 
189 bool x86_mp_online;
190 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE;
191 
192 #if defined(MULTIPROCESSOR)
193 void    	cpu_hatch(void *);
194 static void    	cpu_boot_secondary(struct cpu_info *ci);
195 static void    	cpu_start_secondary(struct cpu_info *ci);
196 static void	cpu_copy_trampoline(void);
197 
198 /*
199  * Runs once per boot once multiprocessor goo has been detected and
200  * the local APIC on the boot processor has been mapped.
201  *
202  * Called from lapic_boot_init() (from mpbios_scan()).
203  */
204 void
205 cpu_init_first(void)
206 {
207 
208 	cpu_info_primary.ci_cpuid = lapic_cpu_number();
209 	cpu_copy_trampoline();
210 }
211 #endif	/* MULTIPROCESSOR */
212 
213 static int
214 cpu_match(device_t parent, cfdata_t match, void *aux)
215 {
216 
217 	return 1;
218 }
219 
220 static void
221 cpu_attach(device_t parent, device_t self, void *aux)
222 {
223 	struct cpu_softc *sc = device_private(self);
224 	struct cpu_attach_args *caa = aux;
225 	struct cpu_info *ci;
226 	uintptr_t ptr;
227 	static int nphycpu = 0;
228 
229 	sc->sc_dev = self;
230 
231 	if (phycpus_attached == ~0) {
232 		aprint_error(": increase MAXCPUS\n");
233 		return;
234 	}
235 
236 	/*
237 	 * If we're an Application Processor, allocate a cpu_info
238 	 * If we're the first attached CPU use the primary cpu_info,
239 	 * otherwise allocate a new one
240 	 */
241 	aprint_naive("\n");
242 	aprint_normal("\n");
243 	if (nphycpu > 0) {
244 		struct cpu_info *tmp;
245 		ptr = (uintptr_t)kmem_zalloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
246 		    KM_SLEEP);
247 		ci = (struct cpu_info *)roundup2(ptr, CACHE_LINE_SIZE);
248 		ci->ci_curldt = -1;
249 
250 		tmp = phycpu_info_list;
251 		while (tmp->ci_next)
252 			tmp = tmp->ci_next;
253 
254 		tmp->ci_next = ci;
255 	} else {
256 		ci = &phycpu_info_primary;
257 	}
258 
259 	ci->ci_self = ci;
260 	sc->sc_info = ci;
261 
262 	ci->ci_dev = self;
263 	ci->ci_acpiid = caa->cpu_id;
264 	ci->ci_cpuid = caa->cpu_number;
265 	ci->ci_vcpu = NULL;
266 	ci->ci_index = nphycpu++;
267 	ci->ci_cpumask = (1 << cpu_index(ci));
268 
269 	atomic_or_32(&phycpus_attached, ci->ci_cpumask);
270 
271 	if (!pmf_device_register(self, NULL, NULL))
272 		aprint_error_dev(self, "couldn't establish power handler\n");
273 
274 	(void)config_defer(self, cpu_defer);
275 }
276 
277 static void
278 cpu_defer(device_t self)
279 {
280 	cpu_rescan(self, NULL, NULL);
281 }
282 
283 static int
284 cpu_rescan(device_t self, const char *ifattr, const int *locators)
285 {
286 	struct cpu_softc *sc = device_private(self);
287 	struct cpufeature_attach_args cfaa;
288 	struct cpu_info *ci = sc->sc_info;
289 
290 	memset(&cfaa, 0, sizeof(cfaa));
291 	cfaa.ci = ci;
292 
293 	if (ifattr_match(ifattr, "cpufeaturebus")) {
294 
295 		if (ci->ci_frequency == NULL) {
296 			cfaa.name = "frequency";
297 			ci->ci_frequency = config_found_ia(self,
298 			    "cpufeaturebus", &cfaa, NULL);
299 		}
300 	}
301 
302 	return 0;
303 }
304 
305 static void
306 cpu_childdetached(device_t self, device_t child)
307 {
308 	struct cpu_softc *sc = device_private(self);
309 	struct cpu_info *ci = sc->sc_info;
310 
311 	if (ci->ci_frequency == child)
312 		ci->ci_frequency = NULL;
313 }
314 
315 static int
316 vcpu_match(device_t parent, cfdata_t match, void *aux)
317 {
318 	struct vcpu_attach_args *vcaa = aux;
319 
320 	if (strcmp(vcaa->vcaa_name, match->cf_name) == 0)
321 		return 1;
322 	return 0;
323 }
324 
325 static void
326 vcpu_attach(device_t parent, device_t self, void *aux)
327 {
328 	struct vcpu_attach_args *vcaa = aux;
329 
330 	cpu_attach_common(parent, self, &vcaa->vcaa_caa);
331 }
332 
333 static void
334 cpu_vm_init(struct cpu_info *ci)
335 {
336 	int ncolors = 2, i;
337 
338 	for (i = CAI_ICACHE; i <= CAI_L2CACHE; i++) {
339 		struct x86_cache_info *cai;
340 		int tcolors;
341 
342 		cai = &ci->ci_cinfo[i];
343 
344 		tcolors = atop(cai->cai_totalsize);
345 		switch(cai->cai_associativity) {
346 		case 0xff:
347 			tcolors = 1; /* fully associative */
348 			break;
349 		case 0:
350 		case 1:
351 			break;
352 		default:
353 			tcolors /= cai->cai_associativity;
354 		}
355 		ncolors = max(ncolors, tcolors);
356 	}
357 
358 	/*
359 	 * Knowing the size of the largest cache on this CPU, re-color
360 	 * our pages.
361 	 */
362 	if (ncolors <= uvmexp.ncolors)
363 		return;
364 	aprint_debug_dev(ci->ci_dev, "%d page colors\n", ncolors);
365 	uvm_page_recolor(ncolors);
366 }
367 
368 static void
369 cpu_attach_common(device_t parent, device_t self, void *aux)
370 {
371 	struct cpu_softc *sc = device_private(self);
372 	struct cpu_attach_args *caa = aux;
373 	struct cpu_info *ci;
374 	uintptr_t ptr;
375 	int cpunum = caa->cpu_number;
376 	static bool again = false;
377 
378 	sc->sc_dev = self;
379 
380 	/*
381 	 * If we're an Application Processor, allocate a cpu_info
382 	 * structure, otherwise use the primary's.
383 	 */
384 	if (caa->cpu_role == CPU_ROLE_AP) {
385 		aprint_naive(": Application Processor\n");
386 		ptr = (uintptr_t)kmem_alloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
387 		    KM_SLEEP);
388 		ci = (struct cpu_info *)roundup2(ptr, CACHE_LINE_SIZE);
389 		memset(ci, 0, sizeof(*ci));
390 #ifdef TRAPLOG
391 		ci->ci_tlog_base = kmem_zalloc(sizeof(struct tlog), KM_SLEEP);
392 #endif
393 	} else {
394 		aprint_naive(": %s Processor\n",
395 		    caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
396 		ci = &cpu_info_primary;
397 #if NLAPIC > 0
398 		if (cpunum != lapic_cpu_number()) {
399 			/* XXX should be done earlier */
400 			uint32_t reg;
401 			aprint_verbose("\n");
402 			aprint_verbose_dev(self, "running CPU at apic %d"
403 			    " instead of at expected %d", lapic_cpu_number(),
404 			    cpunum);
405 			reg = i82489_readreg(LAPIC_ID);
406 			i82489_writereg(LAPIC_ID, (reg & ~LAPIC_ID_MASK) |
407 			    (cpunum << LAPIC_ID_SHIFT));
408 		}
409 		if (cpunum != lapic_cpu_number()) {
410 			aprint_error_dev(self, "unable to reset apic id\n");
411 		}
412 #endif
413 	}
414 
415 	ci->ci_self = ci;
416 	sc->sc_info = ci;
417 	ci->ci_dev = self;
418 	ci->ci_cpuid = cpunum;
419 
420 	KASSERT(HYPERVISOR_shared_info != NULL);
421 	ci->ci_vcpu = &HYPERVISOR_shared_info->vcpu_info[cpunum];
422 
423 	ci->ci_func = caa->cpu_func;
424 
425 	/* Must be called before mi_cpu_attach(). */
426 	cpu_vm_init(ci);
427 
428 	if (caa->cpu_role == CPU_ROLE_AP) {
429 		int error;
430 
431 		error = mi_cpu_attach(ci);
432 		if (error != 0) {
433 			aprint_normal("\n");
434 			aprint_error_dev(self,
435 			    "mi_cpu_attach failed with %d\n", error);
436 			return;
437 		}
438 	} else {
439 		KASSERT(ci->ci_data.cpu_idlelwp != NULL);
440 	}
441 
442 	ci->ci_cpumask = (1 << cpu_index(ci));
443 	pmap_reference(pmap_kernel());
444 	ci->ci_pmap = pmap_kernel();
445 	ci->ci_tlbstate = TLBSTATE_STALE;
446 
447 	/*
448 	 * Boot processor may not be attached first, but the below
449 	 * must be done to allow booting other processors.
450 	 */
451 	if (!again) {
452 		atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_PRIMARY);
453 		/* Basic init. */
454 		cpu_intr_init(ci);
455 		cpu_get_tsc_freq(ci);
456 		cpu_init(ci);
457 		cpu_set_tss_gates(ci);
458 		pmap_cpu_init_late(ci);
459 #if NLAPIC > 0
460 		if (caa->cpu_role != CPU_ROLE_SP) {
461 			/* Enable lapic. */
462 			lapic_enable();
463 			lapic_set_lvt();
464 			lapic_calibrate_timer();
465 		}
466 #endif
467 		/* Make sure DELAY() is initialized. */
468 		DELAY(1);
469 		again = true;
470 	}
471 
472 	/* further PCB init done later. */
473 
474 	switch (caa->cpu_role) {
475 	case CPU_ROLE_SP:
476 		atomic_or_32(&ci->ci_flags, CPUF_SP);
477 		cpu_identify(ci);
478 #if 0
479 		x86_errata();
480 #endif
481 		x86_cpu_idle_init();
482 		break;
483 
484 	case CPU_ROLE_BP:
485 		atomic_or_32(&ci->ci_flags, CPUF_BSP);
486 		cpu_identify(ci);
487 		cpu_init(ci);
488 #if 0
489 		x86_errata();
490 #endif
491 		x86_cpu_idle_init();
492 		break;
493 
494 	case CPU_ROLE_AP:
495 		/*
496 		 * report on an AP
497 		 */
498 
499 #if defined(MULTIPROCESSOR)
500 		cpu_intr_init(ci);
501 		gdt_alloc_cpu(ci);
502 		cpu_set_tss_gates(ci);
503 		pmap_cpu_init_early(ci);
504 		pmap_cpu_init_late(ci);
505 		cpu_start_secondary(ci);
506 		if (ci->ci_flags & CPUF_PRESENT) {
507 			struct cpu_info *tmp;
508 
509 			identifycpu(ci);
510 			tmp = cpu_info_list;
511 			while (tmp->ci_next)
512 				tmp = tmp->ci_next;
513 
514 			tmp->ci_next = ci;
515 		}
516 #else
517 		aprint_error_dev(self, "not started\n");
518 #endif
519 		break;
520 
521 	default:
522 		aprint_normal("\n");
523 		panic("unknown processor type??\n");
524 	}
525 
526 	pat_init(ci);
527 	atomic_or_32(&cpus_attached, ci->ci_cpumask);
528 
529 #if 0
530 	if (!pmf_device_register(self, cpu_suspend, cpu_resume))
531 		aprint_error_dev(self, "couldn't establish power handler\n");
532 #endif
533 
534 #if defined(MULTIPROCESSOR)
535 	if (mp_verbose) {
536 		struct lwp *l = ci->ci_data.cpu_idlelwp;
537 		struct pcb *pcb = lwp_getpcb(l);
538 
539 		aprint_verbose_dev(self,
540 		    "idle lwp at %p, idle sp at 0x%p\n",
541 		    l,
542 #ifdef i386
543 		    (void *)pcb->pcb_esp
544 #else
545 		    (void *)pcb->pcb_rsp
546 #endif
547 		);
548 
549 	}
550 #endif
551 }
552 
553 /*
554  * Initialize the processor appropriately.
555  */
556 
557 void
558 cpu_init(struct cpu_info *ci)
559 {
560 
561 	/*
562 	 * On a P6 or above, enable global TLB caching if the
563 	 * hardware supports it.
564 	 */
565 	if (cpu_feature[0] & CPUID_PGE)
566 		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
567 
568 #ifdef XXXMTRR
569 	/*
570 	 * On a P6 or above, initialize MTRR's if the hardware supports them.
571 	 */
572 	if (cpu_feature[0] & CPUID_MTRR) {
573 		if ((ci->ci_flags & CPUF_AP) == 0)
574 			i686_mtrr_init_first();
575 		mtrr_init_cpu(ci);
576 	}
577 #endif
578 	/*
579 	 * If we have FXSAVE/FXRESTOR, use them.
580 	 */
581 	if (cpu_feature[0] & CPUID_FXSR) {
582 		lcr4(rcr4() | CR4_OSFXSR);
583 
584 		/*
585 		 * If we have SSE/SSE2, enable XMM exceptions.
586 		 */
587 		if (cpu_feature[0] & (CPUID_SSE|CPUID_SSE2))
588 			lcr4(rcr4() | CR4_OSXMMEXCPT);
589 	}
590 
591 #ifdef __x86_64__
592 	/* No user PGD mapped for this CPU yet */
593 	ci->ci_xen_current_user_pgd = 0;
594 #endif
595 
596 	atomic_or_32(&cpus_running, ci->ci_cpumask);
597 	atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
598 }
599 
600 
601 #ifdef MULTIPROCESSOR
602 void
603 cpu_boot_secondary_processors(void)
604 {
605 	struct cpu_info *ci;
606 	u_long i;
607 
608 	for (i = 0; i < maxcpus; i++) {
609 		ci = cpu_lookup(i);
610 		if (ci == NULL)
611 			continue;
612 		if (ci->ci_data.cpu_idlelwp == NULL)
613 			continue;
614 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
615 			continue;
616 		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
617 			continue;
618 		cpu_boot_secondary(ci);
619 	}
620 
621 	x86_mp_online = true;
622 }
623 
624 static void
625 cpu_init_idle_lwp(struct cpu_info *ci)
626 {
627 	struct lwp *l = ci->ci_data.cpu_idlelwp;
628 	struct pcb *pcb = lwp_getpcb(l);
629 
630 	pcb->pcb_cr0 = rcr0();
631 }
632 
633 void
634 cpu_init_idle_lwps(void)
635 {
636 	struct cpu_info *ci;
637 	u_long i;
638 
639 	for (i = 0; i < maxcpus; i++) {
640 		ci = cpu_lookup(i);
641 		if (ci == NULL)
642 			continue;
643 		if (ci->ci_data.cpu_idlelwp == NULL)
644 			continue;
645 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
646 			continue;
647 		cpu_init_idle_lwp(ci);
648 	}
649 }
650 
651 void
652 cpu_start_secondary(struct cpu_info *ci)
653 {
654 	int i;
655 	struct pmap *kpm = pmap_kernel();
656 	extern uint32_t mp_pdirpa;
657 
658 	mp_pdirpa = kpm->pm_pdirpa; /* XXX move elsewhere, not per CPU. */
659 
660 	atomic_or_32(&ci->ci_flags, CPUF_AP);
661 
662 	aprint_debug_dev(ci->ci_dev, "starting\n");
663 
664 	ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
665 	if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0)
666 		return;
667 
668 	/*
669 	 * wait for it to become ready
670 	 */
671 	for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
672 #ifdef MPDEBUG
673 		extern int cpu_trace[3];
674 		static int otrace[3];
675 		if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) {
676 			aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n",
677 				cpu_trace[0], cpu_trace[1], cpu_trace[2]);
678 			memcpy(otrace, cpu_trace, sizeof(otrace));
679 		}
680 #endif
681 		delay(10);
682 	}
683 	if ((ci->ci_flags & CPUF_PRESENT) == 0) {
684 		aprint_error_dev(ci->ci_dev, "failed to become ready\n");
685 #if defined(MPDEBUG) && defined(DDB)
686 		printf("dropping into debugger; continue from here to resume boot\n");
687 		Debugger();
688 #endif
689 	}
690 
691 	CPU_START_CLEANUP(ci);
692 }
693 
694 void
695 cpu_boot_secondary(struct cpu_info *ci)
696 {
697 	int i;
698 
699 	atomic_or_32(&ci->ci_flags, CPUF_GO);
700 	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
701 		delay(10);
702 	}
703 	if ((ci->ci_flags & CPUF_RUNNING) == 0) {
704 		aprint_error_dev(ci->ci_dev, "CPU failed to start\n");
705 #if defined(MPDEBUG) && defined(DDB)
706 		printf("dropping into debugger; continue from here to resume boot\n");
707 		Debugger();
708 #endif
709 	}
710 }
711 
712 /*
713  * The CPU ends up here when its ready to run
714  * This is called from code in mptramp.s; at this point, we are running
715  * in the idle pcb/idle stack of the new CPU.  When this function returns,
716  * this processor will enter the idle loop and start looking for work.
717  *
718  * XXX should share some of this with init386 in machdep.c
719  */
720 void
721 cpu_hatch(void *v)
722 {
723 	struct cpu_info *ci = (struct cpu_info *)v;
724 	struct pcb *pcb;
725 	int s, i;
726 
727 	cpu_probe(ci);
728 
729 	cpu_feature[0] &= ~CPUID_FEAT_BLACKLIST;
730 	cpu_feature[2] &= ~CPUID_FEAT_EXT_BLACKLIST;
731 
732         cpu_init_msrs(ci, true);
733 
734 	KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
735 	atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
736 	while ((ci->ci_flags & CPUF_GO) == 0) {
737 		/* Don't use delay, boot CPU may be patching the text. */
738 		for (i = 10000; i != 0; i--)
739 			x86_pause();
740 	}
741 
742 	/* Because the text may have been patched in x86_patch(). */
743 	wbinvd();
744 	x86_flush();
745 
746 	KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
747 
748 	pcb = lwp_getpcb(curlwp);
749 	lcr3(pmap_kernel()->pm_pdirpa);
750 	pcb->pcb_cr3 = pmap_kernel()->pm_pdirpa;
751 	pcb = lwp_getpcb(ci->ci_data.cpu_idlelwp);
752 	lcr0(pcb->pcb_cr0);
753 
754 	cpu_init_idt();
755 	gdt_init_cpu(ci);
756 	lapic_enable();
757 	lapic_set_lvt();
758 	lapic_initclocks();
759 
760 #ifdef i386
761 	npxinit(ci);
762 #else
763 	fpuinit(ci);
764 #endif
765 
766 	lldt(GSEL(GLDT_SEL, SEL_KPL));
767 	ltr(ci->ci_tss_sel);
768 
769 	cpu_init(ci);
770 	cpu_get_tsc_freq(ci);
771 
772 	s = splhigh();
773 #ifdef i386
774 	lapic_tpr = 0;
775 #else
776 	lcr8(0);
777 #endif
778 	x86_enable_intr();
779 	splx(s);
780 #if 0
781 	x86_errata();
782 #endif
783 
784 	aprint_debug_dev(ci->ci_dev, "CPU %ld running\n",
785 		(long)ci->ci_cpuid);
786 }
787 
788 #if defined(DDB)
789 
790 #include <ddb/db_output.h>
791 #include <machine/db_machdep.h>
792 
793 /*
794  * Dump CPU information from ddb.
795  */
796 void
797 cpu_debug_dump(void)
798 {
799 	struct cpu_info *ci;
800 	CPU_INFO_ITERATOR cii;
801 
802 	db_printf("addr		dev	id	flags	ipis	curlwp 		fpcurlwp\n");
803 	for (CPU_INFO_FOREACH(cii, ci)) {
804 		db_printf("%p	%s	%ld	%x	%x	%10p	%10p\n",
805 		    ci,
806 		    ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
807 		    (long)ci->ci_cpuid,
808 		    ci->ci_flags, ci->ci_ipis,
809 		    ci->ci_curlwp,
810 		    ci->ci_fpcurlwp);
811 	}
812 }
813 #endif /* DDB */
814 
815 static void
816 cpu_copy_trampoline(void)
817 {
818 	/*
819 	 * Copy boot code.
820 	 */
821 	extern u_char cpu_spinup_trampoline[];
822 	extern u_char cpu_spinup_trampoline_end[];
823 
824 	vaddr_t mp_trampoline_vaddr;
825 
826 	mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
827 		UVM_KMF_VAONLY);
828 
829 	pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
830 		VM_PROT_READ | VM_PROT_WRITE, 0);
831 	pmap_update(pmap_kernel());
832 	memcpy((void *)mp_trampoline_vaddr,
833 		cpu_spinup_trampoline,
834 		cpu_spinup_trampoline_end - cpu_spinup_trampoline);
835 
836 	pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
837 	pmap_update(pmap_kernel());
838 	uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
839 }
840 
841 #endif /* MULTIPROCESSOR */
842 
843 #ifdef i386
844 #if 0
845 static void
846 tss_init(struct i386tss *tss, void *stack, void *func)
847 {
848 	memset(tss, 0, sizeof *tss);
849 	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
850 	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
851 	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
852 	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
853 	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
854 	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
855 	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
856 	tss->tss_esp = (int)((char *)stack + USPACE - 16);
857 	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
858 	tss->__tss_eflags = PSL_MBO | PSL_NT;   /* XXX not needed? */
859 	tss->__tss_eip = (int)func;
860 }
861 #endif
862 
863 /* XXX */
864 #define IDTVEC(name)	__CONCAT(X, name)
865 typedef void (vector)(void);
866 extern vector IDTVEC(tss_trap08);
867 #ifdef DDB
868 extern vector Xintrddbipi;
869 extern int ddb_vec;
870 #endif
871 
872 static void
873 cpu_set_tss_gates(struct cpu_info *ci)
874 {
875 #if 0
876 	struct segment_descriptor sd;
877 
878 	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
879 	    UVM_KMF_WIRED);
880 	tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
881 	    IDTVEC(tss_trap08));
882 	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
883 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
884 	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
885 	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
886 	    GSEL(GTRAPTSS_SEL, SEL_KPL));
887 #endif
888 
889 #if defined(DDB) && defined(MULTIPROCESSOR)
890 	/*
891 	 * Set up separate handler for the DDB IPI, so that it doesn't
892 	 * stomp on a possibly corrupted stack.
893 	 *
894 	 * XXX overwriting the gate set in db_machine_init.
895 	 * Should rearrange the code so that it's set only once.
896 	 */
897 	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
898 	    UVM_KMF_WIRED);
899 	tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
900 	    Xintrddbipi);
901 
902 	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
903 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
904 	ci->ci_gdt[GIPITSS_SEL].sd = sd;
905 
906 	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
907 	    GSEL(GIPITSS_SEL, SEL_KPL));
908 #endif
909 }
910 #else
911 static void
912 cpu_set_tss_gates(struct cpu_info *ci)
913 {
914 
915 }
916 #endif	/* i386 */
917 
918 int
919 mp_cpu_start(struct cpu_info *ci, paddr_t target)
920 {
921 #if 0
922 #if NLAPIC > 0
923 	int error;
924 #endif
925 	unsigned short dwordptr[2];
926 
927 	/*
928 	 * Bootstrap code must be addressable in real mode
929 	 * and it must be page aligned.
930 	 */
931 	KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
932 
933 	/*
934 	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
935 	 */
936 
937 	outb(IO_RTC, NVRAM_RESET);
938 	outb(IO_RTC+1, NVRAM_RESET_JUMP);
939 
940 	/*
941 	 * "and the warm reset vector (DWORD based at 40:67) to point
942 	 * to the AP startup code ..."
943 	 */
944 
945 	dwordptr[0] = 0;
946 	dwordptr[1] = target >> 4;
947 
948 	pmap_kenter_pa (0, 0, VM_PROT_READ|VM_PROT_WRITE, 0);
949 	pmap_update(pmap_kernel());
950 
951 	memcpy ((uint8_t *) 0x467, dwordptr, 4);
952 
953 	pmap_kremove (0, PAGE_SIZE);
954 	pmap_update(pmap_kernel());
955 
956 #if NLAPIC > 0
957 	/*
958 	 * ... prior to executing the following sequence:"
959 	 */
960 
961 	if (ci->ci_flags & CPUF_AP) {
962 		if ((error = x86_ipi_init(ci->ci_cpuid)) != 0)
963 			return error;
964 
965 		delay(10000);
966 
967 		if (cpu_feature & CPUID_APIC) {
968 			error = x86_ipi_init(ci->ci_cpuid);
969 			if (error != 0) {
970 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
971 						__func__);
972 				return error;
973 			}
974 
975 			delay(10000);
976 
977 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
978 					LAPIC_DLMODE_STARTUP);
979 			if (error != 0) {
980 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
981 						__func__);
982 				return error;
983 			}
984 			delay(200);
985 
986 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
987 					LAPIC_DLMODE_STARTUP);
988 			if (error != 0) {
989 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken ((3)\n",
990 						__func__);
991 				return error;
992 			}
993 			delay(200);
994 		}
995 	}
996 #endif
997 #endif /* 0 */
998 	return 0;
999 }
1000 
1001 void
1002 mp_cpu_start_cleanup(struct cpu_info *ci)
1003 {
1004 #if 0
1005 	/*
1006 	 * Ensure the NVRAM reset byte contains something vaguely sane.
1007 	 */
1008 
1009 	outb(IO_RTC, NVRAM_RESET);
1010 	outb(IO_RTC+1, NVRAM_RESET_RST);
1011 #endif
1012 }
1013 
1014 void
1015 cpu_init_msrs(struct cpu_info *ci, bool full)
1016 {
1017 #ifdef __x86_64__
1018 	if (full) {
1019 		HYPERVISOR_set_segment_base (SEGBASE_FS, 0);
1020 		HYPERVISOR_set_segment_base (SEGBASE_GS_KERNEL, (uint64_t) ci);
1021 		HYPERVISOR_set_segment_base (SEGBASE_GS_USER, 0);
1022 	}
1023 #endif	/* __x86_64__ */
1024 
1025 	if (cpu_feature[2] & CPUID_NOX)
1026 		wrmsr(MSR_EFER, rdmsr(MSR_EFER) | EFER_NXE);
1027 }
1028 
1029 void
1030 cpu_offline_md(void)
1031 {
1032         int s;
1033 
1034         s = splhigh();
1035 #ifdef __i386__
1036         npxsave_cpu(true);
1037 #else
1038         fpusave_cpu(true);
1039 #endif
1040         splx(s);
1041 }
1042 
1043 #if 0
1044 /* XXX joerg restructure and restart CPUs individually */
1045 static bool
1046 cpu_suspend(device_t dv, const pmf_qual_t *qual)
1047 {
1048 	struct cpu_softc *sc = device_private(dv);
1049 	struct cpu_info *ci = sc->sc_info;
1050 	int err;
1051 
1052 	if (ci->ci_flags & CPUF_PRIMARY)
1053 		return true;
1054 	if (ci->ci_data.cpu_idlelwp == NULL)
1055 		return true;
1056 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1057 		return true;
1058 
1059 	sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
1060 
1061 	if (sc->sc_wasonline) {
1062 		mutex_enter(&cpu_lock);
1063 		err = cpu_setstate(ci, false);
1064 		mutex_exit(&cpu_lock);
1065 
1066 		if (err)
1067 			return false;
1068 	}
1069 
1070 	return true;
1071 }
1072 
1073 static bool
1074 cpu_resume(device_t dv, const pmf_qual_t *qual)
1075 {
1076 	struct cpu_softc *sc = device_private(dv);
1077 	struct cpu_info *ci = sc->sc_info;
1078 	int err = 0;
1079 
1080 	if (ci->ci_flags & CPUF_PRIMARY)
1081 		return true;
1082 	if (ci->ci_data.cpu_idlelwp == NULL)
1083 		return true;
1084 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1085 		return true;
1086 
1087 	if (sc->sc_wasonline) {
1088 		mutex_enter(&cpu_lock);
1089 		err = cpu_setstate(ci, true);
1090 		mutex_exit(&cpu_lock);
1091 	}
1092 
1093 	return err == 0;
1094 }
1095 #endif
1096 
1097 void
1098 cpu_get_tsc_freq(struct cpu_info *ci)
1099 {
1100 	const volatile vcpu_time_info_t *tinfo = &ci->ci_vcpu->time;
1101 	delay(1000000);
1102 	uint64_t freq = 1000000000ULL << 32;
1103 	freq = freq / (uint64_t)tinfo->tsc_to_system_mul;
1104 	if ( tinfo->tsc_shift < 0 )
1105 		freq = freq << -tinfo->tsc_shift;
1106 	else
1107 		freq = freq >> tinfo->tsc_shift;
1108 	ci->ci_data.cpu_cc_freq = freq;
1109 }
1110 
1111 void
1112 x86_cpu_idle_xen(void)
1113 {
1114 	struct cpu_info *ci = curcpu();
1115 
1116 	KASSERT(ci->ci_ilevel == IPL_NONE);
1117 
1118 	x86_disable_intr();
1119 	if (!__predict_false(ci->ci_want_resched)) {
1120 		idle_block();
1121 	} else {
1122 		x86_enable_intr();
1123 	}
1124 }
1125 
1126 /*
1127  * Loads pmap for the current CPU.
1128  */
1129 void
1130 cpu_load_pmap(struct pmap *pmap)
1131 {
1132 #ifdef i386
1133 #ifdef PAE
1134 	int i, s;
1135 	struct cpu_info *ci;
1136 
1137 	s = splvm(); /* just to be safe */
1138 	ci = curcpu();
1139 	paddr_t l3_pd = xpmap_ptom_masked(ci->ci_pae_l3_pdirpa);
1140 	/* don't update the kernel L3 slot */
1141 	for (i = 0 ; i < PDP_SIZE - 1; i++) {
1142 		xpq_queue_pte_update(l3_pd + i * sizeof(pd_entry_t),
1143 		    xpmap_ptom(pmap->pm_pdirpa[i]) | PG_V);
1144 	}
1145 	splx(s);
1146 	tlbflush();
1147 #else /* PAE */
1148 	lcr3(pmap_pdirpa(pmap, 0));
1149 #endif /* PAE */
1150 #endif /* i386 */
1151 
1152 #ifdef __x86_64__
1153 	int i, s;
1154 	pd_entry_t *old_pgd, *new_pgd;
1155 	paddr_t addr;
1156 	struct cpu_info *ci;
1157 
1158 	/* kernel pmap always in cr3 and should never go in user cr3 */
1159 	if (pmap_pdirpa(pmap, 0) != pmap_pdirpa(pmap_kernel(), 0)) {
1160 		ci = curcpu();
1161 		/*
1162 		 * Map user space address in kernel space and load
1163 		 * user cr3
1164 		 */
1165 		s = splvm();
1166 		new_pgd = pmap->pm_pdir;
1167 		old_pgd = pmap_kernel()->pm_pdir;
1168 		addr = xpmap_ptom(pmap_pdirpa(pmap_kernel(), 0));
1169 		for (i = 0; i < PDIR_SLOT_PTE;
1170 		    i++, addr += sizeof(pd_entry_t)) {
1171 			if ((new_pgd[i] & PG_V) || (old_pgd[i] & PG_V))
1172 				xpq_queue_pte_update(addr, new_pgd[i]);
1173 		}
1174 		tlbflush();
1175 		xen_set_user_pgd(pmap_pdirpa(pmap, 0));
1176 		ci->ci_xen_current_user_pgd = pmap_pdirpa(pmap, 0);
1177 		splx(s);
1178 	}
1179 #endif /* __x86_64__ */
1180 }
1181