xref: /netbsd-src/sys/arch/xen/x86/cpu.c (revision 10ad5ffa714ce1a679dcc9dd8159648df2d67b5a)
1 /*	$NetBSD: cpu.c,v 1.34 2009/07/30 13:56:57 cegger 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 1.34 2009/07/30 13:56:57 cegger 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/user.h>
83 #include <sys/systm.h>
84 #include <sys/device.h>
85 #include <sys/kmem.h>
86 #include <sys/cpu.h>
87 #include <sys/atomic.h>
88 #include <sys/reboot.h>
89 
90 #include <uvm/uvm_extern.h>
91 
92 #include <machine/cpufunc.h>
93 #include <machine/cpuvar.h>
94 #include <machine/pmap.h>
95 #include <machine/vmparam.h>
96 #include <machine/mpbiosvar.h>
97 #include <machine/pcb.h>
98 #include <machine/specialreg.h>
99 #include <machine/segments.h>
100 #include <machine/gdt.h>
101 #include <machine/mtrr.h>
102 #include <machine/pio.h>
103 
104 #include <xen/vcpuvar.h>
105 
106 #if NLAPIC > 0
107 #include <machine/apicvar.h>
108 #include <machine/i82489reg.h>
109 #include <machine/i82489var.h>
110 #endif
111 
112 #include <dev/ic/mc146818reg.h>
113 #include <dev/isa/isareg.h>
114 
115 #define	X86_MAXPROCS	32
116 
117 int     cpu_match(device_t, cfdata_t, void *);
118 void    cpu_attach(device_t, device_t, void *);
119 int     vcpu_match(device_t, cfdata_t, void *);
120 void    vcpu_attach(device_t, device_t, void *);
121 void    cpu_attach_common(device_t, device_t, void *);
122 void	cpu_offline_md(void);
123 
124 struct cpu_softc {
125 	device_t sc_dev;		/* device tree glue */
126 	struct cpu_info *sc_info;	/* pointer to CPU info */
127 	bool sc_wasonline;
128 };
129 
130 int mp_cpu_start(struct cpu_info *, paddr_t);
131 void mp_cpu_start_cleanup(struct cpu_info *);
132 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL,
133 				      mp_cpu_start_cleanup };
134 
135 CFATTACH_DECL_NEW(cpu, sizeof(struct cpu_softc),
136     cpu_match, cpu_attach, NULL, NULL);
137 CFATTACH_DECL_NEW(vcpu, sizeof(struct cpu_softc),
138     vcpu_match, vcpu_attach, NULL, NULL);
139 
140 /*
141  * Statically-allocated CPU info for the primary CPU (or the only
142  * CPU, on uniprocessors).  The CPU info list is initialized to
143  * point at it.
144  */
145 #ifdef TRAPLOG
146 #include <machine/tlog.h>
147 struct tlog tlog_primary;
148 #endif
149 struct cpu_info cpu_info_primary = {
150 	.ci_dev = 0,
151 	.ci_self = &cpu_info_primary,
152 	.ci_idepth = -1,
153 	.ci_curlwp = &lwp0,
154 	.ci_curldt = -1,
155 #ifdef TRAPLOG
156 	.ci_tlog = &tlog_primary,
157 #endif
158 
159 };
160 struct cpu_info phycpu_info_primary = {
161 	.ci_dev = 0,
162 	.ci_self = &phycpu_info_primary,
163 };
164 
165 struct cpu_info *cpu_info_list = &cpu_info_primary;
166 
167 static void	cpu_set_tss_gates(struct cpu_info *ci);
168 
169 uint32_t cpus_attached = 0;
170 uint32_t cpus_running = 0;
171 
172 bool x86_mp_online;
173 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE;
174 
175 struct cpu_info *phycpu_info[X86_MAXPROCS] = { &cpu_info_primary };
176 
177 #ifdef MULTIPROCESSOR
178 /*
179  * Array of CPU info structures.  Must be statically-allocated because
180  * curproc, etc. are used early.
181  */
182 struct cpu_info *cpu_info[X86_MAXPROCS] = { &cpu_info_primary };
183 
184 void    	cpu_hatch(void *);
185 static void    	cpu_boot_secondary(struct cpu_info *ci);
186 static void    	cpu_start_secondary(struct cpu_info *ci);
187 static void	cpu_copy_trampoline(void);
188 
189 /*
190  * Runs once per boot once multiprocessor goo has been detected and
191  * the local APIC on the boot processor has been mapped.
192  *
193  * Called from lapic_boot_init() (from mpbios_scan()).
194  */
195 void
196 cpu_init_first(void)
197 {
198 	int cpunum = lapic_cpu_number();
199 
200 	if (cpunum != 0) {
201 		cpu_info[0] = NULL;
202 		cpu_info[cpunum] = &cpu_info_primary;
203 	}
204 
205 	cpu_copy_trampoline();
206 }
207 #endif
208 
209 int
210 cpu_match(device_t parent, cfdata_t match, void *aux)
211 {
212 
213 	return 1;
214 }
215 
216 void
217 cpu_attach(device_t parent, device_t self, void *aux)
218 {
219 	struct cpu_softc *sc = device_private(self);
220 	struct cpu_attach_args *caa = aux;
221 	struct cpu_info *ci;
222 	uintptr_t ptr;
223 	int cpunum = caa->cpu_number;
224 
225 	sc->sc_dev = self;
226 
227 	if (cpus_attached == ~0) {
228 		aprint_error(": increase MAXCPUS\n");
229 		return;
230 	}
231 
232 	/*
233 	 * If we're an Application Processor, allocate a cpu_info
234 	 * structure, otherwise use the primary's.
235 	 */
236 	if (caa->cpu_role == CPU_ROLE_AP) {
237 		if ((boothowto & RB_MD1) != 0) {
238 			aprint_error(": multiprocessor boot disabled\n");
239 			if (!pmf_device_register(self, NULL, NULL))
240 				aprint_error_dev(self,
241 				   "couldn't establish power handler\n");
242 			return;
243 		}
244 		aprint_naive(": Application Processor\n");
245 		ptr = (uintptr_t)kmem_zalloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
246 		    KM_SLEEP);
247 		ci = (struct cpu_info *)((ptr + CACHE_LINE_SIZE - 1) &
248 		    ~(CACHE_LINE_SIZE - 1));
249 		ci->ci_curldt = -1;
250 		if (phycpu_info[cpunum] != NULL)
251 			panic("cpu at apic id %d already attached?", cpunum);
252 		phycpu_info[cpunum] = ci;
253 	} else {
254 		aprint_naive(": %s Processor\n",
255 		    caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
256 		ci = &phycpu_info_primary;
257 		if (cpunum != 0) {
258 			phycpu_info[0] = NULL;
259 			phycpu_info[cpunum] = ci;
260 		}
261 	}
262 
263 	ci->ci_self = ci;
264 	sc->sc_info = ci;
265 
266 	ci->ci_dev = self;
267 	ci->ci_cpuid = caa->cpu_number;
268 	ci->ci_vcpu = NULL;
269 
270 	printf(": ");
271 	switch (caa->cpu_role) {
272 	case CPU_ROLE_SP:
273 		printf("(uniprocessor)\n");
274 		atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_SP | CPUF_PRIMARY);
275 		break;
276 
277 	case CPU_ROLE_BP:
278 		printf("(boot processor)\n");
279 		atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_BSP | CPUF_PRIMARY);
280 		break;
281 
282 	case CPU_ROLE_AP:
283 		/*
284 		 * report on an AP
285 		 */
286 		printf("(application processor)\n");
287 		break;
288 
289 	default:
290 		panic("unknown processor type??\n");
291 	}
292 
293 	atomic_or_32(&cpus_attached, ci->ci_cpumask);
294 
295 	return;
296 }
297 
298 int
299 vcpu_match(device_t parent, cfdata_t match, void *aux)
300 {
301 	struct vcpu_attach_args *vcaa = aux;
302 
303 	if (strcmp(vcaa->vcaa_name, match->cf_name) == 0)
304 		return 1;
305 	return 0;
306 }
307 
308 void
309 vcpu_attach(device_t parent, device_t self, void *aux)
310 {
311 	struct vcpu_attach_args *vcaa = aux;
312 
313 	cpu_attach_common(parent, self, &vcaa->vcaa_caa);
314 }
315 
316 static void
317 cpu_vm_init(struct cpu_info *ci)
318 {
319 	int ncolors = 2, i;
320 
321 	for (i = CAI_ICACHE; i <= CAI_L2CACHE; i++) {
322 		struct x86_cache_info *cai;
323 		int tcolors;
324 
325 		cai = &ci->ci_cinfo[i];
326 
327 		tcolors = atop(cai->cai_totalsize);
328 		switch(cai->cai_associativity) {
329 		case 0xff:
330 			tcolors = 1; /* fully associative */
331 			break;
332 		case 0:
333 		case 1:
334 			break;
335 		default:
336 			tcolors /= cai->cai_associativity;
337 		}
338 		ncolors = max(ncolors, tcolors);
339 	}
340 
341 	/*
342 	 * Knowing the size of the largest cache on this CPU, re-color
343 	 * our pages.
344 	 */
345 	if (ncolors <= uvmexp.ncolors)
346 		return;
347 	aprint_debug_dev(ci->ci_dev, "%d page colors\n", ncolors);
348 	uvm_page_recolor(ncolors);
349 }
350 
351 void
352 cpu_attach_common(device_t parent, device_t self, void *aux)
353 {
354 	struct cpu_softc *sc = device_private(self);
355 	struct cpu_attach_args *caa = aux;
356 	struct cpu_info *ci;
357 	uintptr_t ptr;
358 	int cpunum = caa->cpu_number;
359 
360 	sc->sc_dev = self;
361 
362 	/*
363 	 * If we're an Application Processor, allocate a cpu_info
364 	 * structure, otherwise use the primary's.
365 	 */
366 	if (caa->cpu_role == CPU_ROLE_AP) {
367 		if (cpunum >= X86_MAXPROCS) {
368 			aprint_error(": apic id %d ignored, "
369 				"please increase X86_MAXPROCS\n", cpunum);
370 		}
371 
372 		aprint_naive(": Application Processor\n");
373 		ptr = (uintptr_t)kmem_alloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
374 		    KM_SLEEP);
375 		ci = (struct cpu_info *)((ptr + CACHE_LINE_SIZE - 1) &
376 		    ~(CACHE_LINE_SIZE - 1));
377 		memset(ci, 0, sizeof(*ci));
378 #if defined(MULTIPROCESSOR)
379 		if (cpu_info[cpunum] != NULL)
380 			panic("cpu at apic id %d already attached?", cpunum);
381 		cpu_info[cpunum] = ci;
382 #endif
383 #ifdef TRAPLOG
384 		ci->ci_tlog_base = kmem_zalloc(sizeof(struct tlog), KM_SLEEP);
385 #endif
386 	} else {
387 		aprint_naive(": %s Processor\n",
388 		    caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
389 		ci = &cpu_info_primary;
390 #if defined(MULTIPROCESSOR)
391 		if (cpunum != lapic_cpu_number()) {
392 			panic("%s: running CPU is at apic %d"
393 			    " instead of at expected %d",
394 			    device_xname(sc->sc_dev), lapic_cpu_number(), cpunum);
395 		}
396 #endif
397 	}
398 
399 	ci->ci_self = ci;
400 	sc->sc_info = ci;
401 
402 	ci->ci_dev = self;
403 	ci->ci_cpuid = cpunum;
404 
405 	KASSERT(HYPERVISOR_shared_info != NULL);
406 	ci->ci_vcpu = &HYPERVISOR_shared_info->vcpu_info[cpunum];
407 
408 	ci->ci_func = caa->cpu_func;
409 
410 	if (caa->cpu_role == CPU_ROLE_AP) {
411 #if defined(MULTIPROCESSOR)
412 		int error;
413 
414 		error = mi_cpu_attach(ci);
415 		if (error != 0) {
416 			aprint_normal("\n");
417 			aprint_error_dev(sc->sc_dev, "mi_cpu_attach failed with %d\n",
418 			    error);
419 			return;
420 		}
421 #endif
422 	} else {
423 		KASSERT(ci->ci_data.cpu_idlelwp != NULL);
424 	}
425 
426 	ci->ci_cpumask = (1 << cpu_index(ci));
427 	pmap_reference(pmap_kernel());
428 	ci->ci_pmap = pmap_kernel();
429 	ci->ci_tlbstate = TLBSTATE_STALE;
430 
431 	/* further PCB init done later. */
432 
433 	switch (caa->cpu_role) {
434 	case CPU_ROLE_SP:
435 		atomic_or_32(&ci->ci_flags,
436 		     CPUF_PRESENT | CPUF_SP | CPUF_PRIMARY);
437 		cpu_intr_init(ci);
438 		cpu_get_tsc_freq(ci);
439 		cpu_identify(ci);
440 		cpu_init(ci);
441 		cpu_set_tss_gates(ci);
442 		pmap_cpu_init_late(ci);
443 		x86_cpu_idle_init();
444 #if 0
445 		x86_errata();
446 #endif
447 		break;
448 
449 	case CPU_ROLE_BP:
450 		atomic_or_32(&ci->ci_flags,
451 		    CPUF_PRESENT | CPUF_BSP | CPUF_PRIMARY);
452 		cpu_intr_init(ci);
453 		cpu_get_tsc_freq(ci);
454 		cpu_identify(ci);
455 		cpu_init(ci);
456 		cpu_set_tss_gates(ci);
457 		pmap_cpu_init_late(ci);
458 		x86_cpu_idle_init();
459 #if NLAPIC > 0
460 		/*
461 		 * Enable local apic
462 		 */
463 		lapic_enable();
464 		lapic_set_lvt();
465 		lapic_calibrate_timer(ci);
466 #endif
467 #if 0
468 		x86_errata();
469 #endif
470 		break;
471 
472 	case CPU_ROLE_AP:
473 		/*
474 		 * report on an AP
475 		 */
476 
477 #if defined(MULTIPROCESSOR)
478 		cpu_intr_init(ci);
479 		gdt_alloc_cpu(ci);
480 		cpu_set_tss_gates(ci);
481 		pmap_cpu_init_early(ci);
482 		pmap_cpu_init_late(ci);
483 		cpu_start_secondary(ci);
484 		if (ci->ci_flags & CPUF_PRESENT) {
485 			struct cpu_info *tmp;
486 
487 			identifycpu(ci);
488 			tmp = cpu_info_list;
489 			while (tmp->ci_next)
490 				tmp = tmp->ci_next;
491 
492 			tmp->ci_next = ci;
493 		}
494 #else
495 		aprint_normal_dev(sc->sc_dev, "not started\n");
496 #endif
497 		break;
498 
499 	default:
500 		aprint_normal("\n");
501 		panic("unknown processor type??\n");
502 	}
503 	cpu_vm_init(ci);
504 
505 	atomic_or_32(&cpus_attached, ci->ci_cpumask);
506 
507 #if 0
508 	if (!pmf_device_register(self, cpu_suspend, cpu_resume))
509 		aprint_error_dev(self, "couldn't establish power handler\n");
510 #endif
511 
512 #if defined(MULTIPROCESSOR)
513 	if (mp_verbose) {
514 		struct lwp *l = ci->ci_data.cpu_idlelwp;
515 
516 		aprint_verbose_dev(sc->sc_dev, "idle lwp at %p, idle sp at 0x%p\n",
517 		    l,
518 #ifdef i386
519 		    (void *)l->l_addr->u_pcb.pcb_esp
520 #else
521 		    (void *)l->l_addr->u_pcb.pcb_rsp
522 #endif
523 		);
524 
525 	}
526 #endif
527 }
528 
529 /*
530  * Initialize the processor appropriately.
531  */
532 
533 void
534 cpu_init(struct cpu_info *ci)
535 {
536 
537 	/*
538 	 * On a P6 or above, enable global TLB caching if the
539 	 * hardware supports it.
540 	 */
541 	if (cpu_feature & CPUID_PGE)
542 		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
543 
544 #ifdef XXXMTRR
545 	/*
546 	 * On a P6 or above, initialize MTRR's if the hardware supports them.
547 	 */
548 	if (cpu_feature & CPUID_MTRR) {
549 		if ((ci->ci_flags & CPUF_AP) == 0)
550 			i686_mtrr_init_first();
551 		mtrr_init_cpu(ci);
552 	}
553 #endif
554 	/*
555 	 * If we have FXSAVE/FXRESTOR, use them.
556 	 */
557 	if (cpu_feature & CPUID_FXSR) {
558 		lcr4(rcr4() | CR4_OSFXSR);
559 
560 		/*
561 		 * If we have SSE/SSE2, enable XMM exceptions.
562 		 */
563 		if (cpu_feature & (CPUID_SSE|CPUID_SSE2))
564 			lcr4(rcr4() | CR4_OSXMMEXCPT);
565 	}
566 
567 	atomic_or_32(&cpus_running, ci->ci_cpumask);
568 
569 	wbinvd();
570 	atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
571 }
572 
573 
574 #ifdef MULTIPROCESSOR
575 void
576 cpu_boot_secondary_processors(void)
577 {
578 	struct cpu_info *ci;
579 	u_long i;
580 
581 	for (i = 0; i < X86_MAXPROCS; i++) {
582 		ci = cpu_info[i];
583 		if (ci == NULL)
584 			continue;
585 		if (ci->ci_data.cpu_idlelwp == NULL)
586 			continue;
587 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
588 			continue;
589 		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
590 			continue;
591 		cpu_boot_secondary(ci);
592 	}
593 
594 	x86_mp_online = true;
595 }
596 
597 static void
598 cpu_init_idle_lwp(struct cpu_info *ci)
599 {
600 	struct lwp *l = ci->ci_data.cpu_idlelwp;
601 	struct pcb *pcb = &l->l_addr->u_pcb;
602 
603 	pcb->pcb_cr0 = rcr0();
604 }
605 
606 void
607 cpu_init_idle_lwps(void)
608 {
609 	struct cpu_info *ci;
610 	u_long i;
611 
612 	for (i = 0; i < X86_MAXPROCS; i++) {
613 		ci = cpu_info[i];
614 		if (ci == NULL)
615 			continue;
616 		if (ci->ci_data.cpu_idlelwp == NULL)
617 			continue;
618 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
619 			continue;
620 		cpu_init_idle_lwp(ci);
621 	}
622 }
623 
624 void
625 cpu_start_secondary(struct cpu_info *ci)
626 {
627 	int i;
628 	struct pmap *kpm = pmap_kernel();
629 	extern uint32_t mp_pdirpa;
630 
631 	mp_pdirpa = kpm->pm_pdirpa; /* XXX move elsewhere, not per CPU. */
632 
633 	atomic_or_32(&ci->ci_flags, CPUF_AP);
634 
635 	aprint_debug_dev(ci->ci_dev, "starting\n");
636 
637 	ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
638 	if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0)
639 		return;
640 
641 	/*
642 	 * wait for it to become ready
643 	 */
644 	for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
645 #ifdef MPDEBUG
646 		extern int cpu_trace[3];
647 		static int otrace[3];
648 		if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) {
649 			aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n",
650 				cpu_trace[0], cpu_trace[1], cpu_trace[2]);
651 			memcpy(otrace, cpu_trace, sizeof(otrace));
652 		}
653 #endif
654 		delay(10);
655 	}
656 	if ((ci->ci_flags & CPUF_PRESENT) == 0) {
657 		aprint_error_dev(ci->ci_dev, "failed to become ready\n");
658 #if defined(MPDEBUG) && defined(DDB)
659 		printf("dropping into debugger; continue from here to resume boot\n");
660 		Debugger();
661 #endif
662 	}
663 
664 	CPU_START_CLEANUP(ci);
665 }
666 
667 void
668 cpu_boot_secondary(struct cpu_info *ci)
669 {
670 	int i;
671 
672 	atomic_or_32(&ci->ci_flags, CPUF_GO);
673 	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
674 		delay(10);
675 	}
676 	if ((ci->ci_flags & CPUF_RUNNING) == 0) {
677 		aprint_error_dev(ci->ci_dev, "CPU failed to start\n");
678 #if defined(MPDEBUG) && defined(DDB)
679 		printf("dropping into debugger; continue from here to resume boot\n");
680 		Debugger();
681 #endif
682 	}
683 }
684 
685 /*
686  * The CPU ends up here when its ready to run
687  * This is called from code in mptramp.s; at this point, we are running
688  * in the idle pcb/idle stack of the new CPU.  When this function returns,
689  * this processor will enter the idle loop and start looking for work.
690  *
691  * XXX should share some of this with init386 in machdep.c
692  */
693 void
694 cpu_hatch(void *v)
695 {
696 	struct cpu_info *ci = (struct cpu_info *)v;
697 	int s, i;
698 	uint32_t blacklist_features;
699 
700 #ifdef __x86_64__
701         cpu_init_msrs(ci, true);
702 #endif
703 
704 	cpu_probe(ci);
705 
706 	/* not on Xen... */
707 	blacklist_features = ~(CPUID_PGE|CPUID_PSE|CPUID_MTRR|CPUID_FXSR|CPUID_NOX); /* XXX add CPUID_SVM */
708 
709 	cpu_feature &= blacklist_features;
710 
711 	KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
712 	atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
713 	while ((ci->ci_flags & CPUF_GO) == 0) {
714 		/* Don't use delay, boot CPU may be patching the text. */
715 		for (i = 10000; i != 0; i--)
716 			x86_pause();
717 	}
718 
719 	/* Because the text may have been patched in x86_patch(). */
720 	wbinvd();
721 	x86_flush();
722 
723 	KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
724 
725 	lcr3(pmap_kernel()->pm_pdirpa);
726 	curlwp->l_addr->u_pcb.pcb_cr3 = pmap_kernel()->pm_pdirpa;
727 	lcr0(ci->ci_data.cpu_idlelwp->l_addr->u_pcb.pcb_cr0);
728 	cpu_init_idt();
729 	gdt_init_cpu(ci);
730 	lapic_enable();
731 	lapic_set_lvt();
732 	lapic_initclocks();
733 
734 #ifdef i386
735 	npxinit(ci);
736 #else
737 	fpuinit(ci);
738 #endif
739 
740 	lldt(GSEL(GLDT_SEL, SEL_KPL));
741 	ltr(ci->ci_tss_sel);
742 
743 	cpu_init(ci);
744 	cpu_get_tsc_freq(ci);
745 
746 	s = splhigh();
747 #ifdef i386
748 	lapic_tpr = 0;
749 #else
750 	lcr8(0);
751 #endif
752 	x86_enable_intr();
753 	splx(s);
754 #if 0
755 	x86_errata();
756 #endif
757 
758 	aprint_debug_dev(ci->ci_dev, "CPU %ld running\n",
759 		(long)ci->ci_cpuid);
760 }
761 
762 #if defined(DDB)
763 
764 #include <ddb/db_output.h>
765 #include <machine/db_machdep.h>
766 
767 /*
768  * Dump CPU information from ddb.
769  */
770 void
771 cpu_debug_dump(void)
772 {
773 	struct cpu_info *ci;
774 	CPU_INFO_ITERATOR cii;
775 
776 	db_printf("addr		dev	id	flags	ipis	curlwp 		fpcurlwp\n");
777 	for (CPU_INFO_FOREACH(cii, ci)) {
778 		db_printf("%p	%s	%ld	%x	%x	%10p	%10p\n",
779 		    ci,
780 		    ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
781 		    (long)ci->ci_cpuid,
782 		    ci->ci_flags, ci->ci_ipis,
783 		    ci->ci_curlwp,
784 		    ci->ci_fpcurlwp);
785 	}
786 }
787 #endif
788 
789 static void
790 cpu_copy_trampoline(void)
791 {
792 	/*
793 	 * Copy boot code.
794 	 */
795 	extern u_char cpu_spinup_trampoline[];
796 	extern u_char cpu_spinup_trampoline_end[];
797 
798 	vaddr_t mp_trampoline_vaddr;
799 
800 	mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
801 		UVM_KMF_VAONLY);
802 
803 	pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
804 		VM_PROT_READ | VM_PROT_WRITE);
805 	pmap_update(pmap_kernel());
806 	memcpy((void *)mp_trampoline_vaddr,
807 		cpu_spinup_trampoline,
808 		cpu_spinup_trampoline_end - cpu_spinup_trampoline);
809 
810 	pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
811 	pmap_update(pmap_kernel());
812 	uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
813 }
814 
815 #endif
816 
817 #ifdef i386
818 #if 0
819 static void
820 tss_init(struct i386tss *tss, void *stack, void *func)
821 {
822 	memset(tss, 0, sizeof *tss);
823 	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
824 	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
825 	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
826 	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
827 	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
828 	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
829 	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
830 	tss->tss_esp = (int)((char *)stack + USPACE - 16);
831 	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
832 	tss->__tss_eflags = PSL_MBO | PSL_NT;   /* XXX not needed? */
833 	tss->__tss_eip = (int)func;
834 }
835 #endif
836 
837 /* XXX */
838 #define IDTVEC(name)	__CONCAT(X, name)
839 typedef void (vector)(void);
840 extern vector IDTVEC(tss_trap08);
841 #ifdef DDB
842 extern vector Xintrddbipi;
843 extern int ddb_vec;
844 #endif
845 
846 static void
847 cpu_set_tss_gates(struct cpu_info *ci)
848 {
849 #if 0
850 	struct segment_descriptor sd;
851 
852 	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
853 	    UVM_KMF_WIRED);
854 	tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
855 	    IDTVEC(tss_trap08));
856 	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
857 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
858 	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
859 	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
860 	    GSEL(GTRAPTSS_SEL, SEL_KPL));
861 #endif
862 
863 #if defined(DDB) && defined(MULTIPROCESSOR)
864 	/*
865 	 * Set up separate handler for the DDB IPI, so that it doesn't
866 	 * stomp on a possibly corrupted stack.
867 	 *
868 	 * XXX overwriting the gate set in db_machine_init.
869 	 * Should rearrange the code so that it's set only once.
870 	 */
871 	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
872 	    UVM_KMF_WIRED);
873 	tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
874 	    Xintrddbipi);
875 
876 	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
877 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
878 	ci->ci_gdt[GIPITSS_SEL].sd = sd;
879 
880 	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
881 	    GSEL(GIPITSS_SEL, SEL_KPL));
882 #endif
883 }
884 #else
885 static void
886 cpu_set_tss_gates(struct cpu_info *ci)
887 {
888 
889 }
890 #endif	/* i386 */
891 
892 int
893 mp_cpu_start(struct cpu_info *ci, paddr_t target)
894 {
895 #if 0
896 #if NLAPIC > 0
897 	int error;
898 #endif
899 	unsigned short dwordptr[2];
900 
901 	/*
902 	 * Bootstrap code must be addressable in real mode
903 	 * and it must be page aligned.
904 	 */
905 	KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
906 
907 	/*
908 	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
909 	 */
910 
911 	outb(IO_RTC, NVRAM_RESET);
912 	outb(IO_RTC+1, NVRAM_RESET_JUMP);
913 
914 	/*
915 	 * "and the warm reset vector (DWORD based at 40:67) to point
916 	 * to the AP startup code ..."
917 	 */
918 
919 	dwordptr[0] = 0;
920 	dwordptr[1] = target >> 4;
921 
922 	pmap_kenter_pa (0, 0, VM_PROT_READ|VM_PROT_WRITE);
923 	memcpy ((uint8_t *) 0x467, dwordptr, 4);
924 	pmap_kremove (0, PAGE_SIZE);
925 
926 #if NLAPIC > 0
927 	/*
928 	 * ... prior to executing the following sequence:"
929 	 */
930 
931 	if (ci->ci_flags & CPUF_AP) {
932 		if ((error = x86_ipi_init(ci->ci_cpuid)) != 0)
933 			return error;
934 
935 		delay(10000);
936 
937 		if (cpu_feature & CPUID_APIC) {
938 			error = x86_ipi_init(ci->ci_cpuid);
939 			if (error != 0) {
940 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
941 						__func__);
942 				return error;
943 			}
944 
945 			delay(10000);
946 
947 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
948 					LAPIC_DLMODE_STARTUP);
949 			if (error != 0) {
950 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
951 						__func__);
952 				return error;
953 			}
954 			delay(200);
955 
956 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
957 					LAPIC_DLMODE_STARTUP);
958 			if (error != 0) {
959 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken ((3)\n",
960 						__func__);
961 				return error;
962 			}
963 			delay(200);
964 		}
965 	}
966 #endif
967 #endif /* 0 */
968 	return 0;
969 }
970 
971 void
972 mp_cpu_start_cleanup(struct cpu_info *ci)
973 {
974 #if 0
975 	/*
976 	 * Ensure the NVRAM reset byte contains something vaguely sane.
977 	 */
978 
979 	outb(IO_RTC, NVRAM_RESET);
980 	outb(IO_RTC+1, NVRAM_RESET_RST);
981 #endif
982 }
983 
984 #ifdef __x86_64__
985 
986 void
987 cpu_init_msrs(struct cpu_info *ci, bool full)
988 {
989 	if (full) {
990 		HYPERVISOR_set_segment_base (SEGBASE_FS, 0);
991 		HYPERVISOR_set_segment_base (SEGBASE_GS_KERNEL, (uint64_t) ci);
992 		HYPERVISOR_set_segment_base (SEGBASE_GS_USER, 0);
993 	}
994 }
995 #endif	/* __x86_64__ */
996 
997 void
998 cpu_offline_md(void)
999 {
1000         int s;
1001 
1002         s = splhigh();
1003 #ifdef __i386__
1004         npxsave_cpu(true);
1005 #else
1006         fpusave_cpu(true);
1007 #endif
1008         splx(s);
1009 }
1010 
1011 #if 0
1012 /* XXX joerg restructure and restart CPUs individually */
1013 static bool
1014 cpu_suspend(device_t dv PMF_FN_ARGS)
1015 {
1016 	struct cpu_softc *sc = device_private(dv);
1017 	struct cpu_info *ci = sc->sc_info;
1018 	int err;
1019 
1020 	if (ci->ci_flags & CPUF_PRIMARY)
1021 		return true;
1022 	if (ci->ci_data.cpu_idlelwp == NULL)
1023 		return true;
1024 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1025 		return true;
1026 
1027 	sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
1028 
1029 	if (sc->sc_wasonline) {
1030 		mutex_enter(&cpu_lock);
1031 		err = cpu_setstate(ci, false);
1032 		mutex_exit(&cpu_lock);
1033 
1034 		if (err)
1035 			return false;
1036 	}
1037 
1038 	return true;
1039 }
1040 
1041 static bool
1042 cpu_resume(device_t dv PMF_FN_ARGS)
1043 {
1044 	struct cpu_softc *sc = device_private(dv);
1045 	struct cpu_info *ci = sc->sc_info;
1046 	int err = 0;
1047 
1048 	if (ci->ci_flags & CPUF_PRIMARY)
1049 		return true;
1050 	if (ci->ci_data.cpu_idlelwp == NULL)
1051 		return true;
1052 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1053 		return true;
1054 
1055 	if (sc->sc_wasonline) {
1056 		mutex_enter(&cpu_lock);
1057 		err = cpu_setstate(ci, true);
1058 		mutex_exit(&cpu_lock);
1059 	}
1060 
1061 	return err == 0;
1062 }
1063 #endif
1064 
1065 void
1066 cpu_get_tsc_freq(struct cpu_info *ci)
1067 {
1068 	const volatile vcpu_time_info_t *tinfo = &ci->ci_vcpu->time;
1069 	delay(1000000);
1070 	uint64_t freq = 1000000000ULL << 32;
1071 	freq = freq / (uint64_t)tinfo->tsc_to_system_mul;
1072 	if ( tinfo->tsc_shift < 0 )
1073 		freq = freq << -tinfo->tsc_shift;
1074 	else
1075 		freq = freq >> tinfo->tsc_shift;
1076 	ci->ci_data.cpu_cc_freq = freq;
1077 }
1078 
1079 void
1080 x86_cpu_idle_xen(void)
1081 {
1082 	struct cpu_info *ci = curcpu();
1083 
1084 	KASSERT(ci->ci_ilevel == IPL_NONE);
1085 
1086 	x86_disable_intr();
1087 	if (!__predict_false(ci->ci_want_resched)) {
1088 		idle_block();
1089 	} else {
1090 		x86_enable_intr();
1091 	}
1092 }
1093