xref: /netbsd-src/sys/arch/xen/x86/cpu.c (revision 8ac07aec990b9d2e483062509d0a9fa5b4f57cf2)
1 /*	$NetBSD: cpu.c,v 1.17 2008/04/24 20:44:02 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  * All rights reserved.
7  *
8  * This code is derived from software contributed to The NetBSD Foundation
9  * by RedBack Networks Inc.
10  *
11  * Author: Bill Sommerfeld
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *        This product includes software developed by the NetBSD
24  *        Foundation, Inc. and its contributors.
25  * 4. Neither the name of The NetBSD Foundation nor the names of its
26  *    contributors may be used to endorse or promote products derived
27  *    from this software without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
30  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
31  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
32  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
33  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
34  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
35  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
36  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
37  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
38  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
39  * POSSIBILITY OF SUCH DAMAGE.
40  */
41 
42 /*
43  * Copyright (c) 1999 Stefan Grefen
44  *
45  * Redistribution and use in source and binary forms, with or without
46  * modification, are permitted provided that the following conditions
47  * are met:
48  * 1. Redistributions of source code must retain the above copyright
49  *    notice, this list of conditions and the following disclaimer.
50  * 2. Redistributions in binary form must reproduce the above copyright
51  *    notice, this list of conditions and the following disclaimer in the
52  *    documentation and/or other materials provided with the distribution.
53  * 3. All advertising materials mentioning features or use of this software
54  *    must display the following acknowledgement:
55  *      This product includes software developed by the NetBSD
56  *      Foundation, Inc. and its contributors.
57  * 4. Neither the name of The NetBSD Foundation nor the names of its
58  *    contributors may be used to endorse or promote products derived
59  *    from this software without specific prior written permission.
60  *
61  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
62  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
65  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71  * SUCH DAMAGE.
72  */
73 
74 #include <sys/cdefs.h>
75 __KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.17 2008/04/24 20:44:02 cegger Exp $");
76 
77 #include "opt_ddb.h"
78 #include "opt_multiprocessor.h"
79 #include "opt_mpbios.h"		/* for MPDEBUG */
80 #include "opt_mtrr.h"
81 #include "opt_xen.h"
82 
83 #include "lapic.h"
84 #include "ioapic.h"
85 
86 #include <sys/param.h>
87 #include <sys/proc.h>
88 #include <sys/user.h>
89 #include <sys/systm.h>
90 #include <sys/device.h>
91 #include <sys/malloc.h>
92 #include <sys/cpu.h>
93 #include <sys/atomic.h>
94 
95 #include <uvm/uvm_extern.h>
96 
97 #include <machine/cpufunc.h>
98 #include <machine/cpuvar.h>
99 #include <machine/pmap.h>
100 #include <machine/vmparam.h>
101 #include <machine/mpbiosvar.h>
102 #include <machine/pcb.h>
103 #include <machine/specialreg.h>
104 #include <machine/segments.h>
105 #include <machine/gdt.h>
106 #include <machine/mtrr.h>
107 #include <machine/pio.h>
108 
109 #ifdef XEN3
110 #include <xen/vcpuvar.h>
111 #endif
112 
113 #if NLAPIC > 0
114 #include <machine/apicvar.h>
115 #include <machine/i82489reg.h>
116 #include <machine/i82489var.h>
117 #endif
118 
119 #include <dev/ic/mc146818reg.h>
120 #include <dev/isa/isareg.h>
121 
122 int     cpu_match(device_t, cfdata_t, void *);
123 void    cpu_attach(device_t, device_t, void *);
124 #ifdef XEN3
125 int     vcpu_match(device_t, cfdata_t, void *);
126 void    vcpu_attach(device_t, device_t, void *);
127 #endif
128 void    cpu_attach_common(device_t, device_t, void *);
129 void	cpu_offline_md(void);
130 
131 struct cpu_softc {
132 	device_t sc_dev;		/* device tree glue */
133 	struct cpu_info *sc_info;	/* pointer to CPU info */
134 };
135 
136 int mp_cpu_start(struct cpu_info *, paddr_t);
137 void mp_cpu_start_cleanup(struct cpu_info *);
138 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL,
139 				      mp_cpu_start_cleanup };
140 
141 CFATTACH_DECL_NEW(cpu, sizeof(struct cpu_softc),
142     cpu_match, cpu_attach, NULL, NULL);
143 #ifdef XEN3
144 CFATTACH_DECL_NEW(vcpu, sizeof(struct cpu_softc),
145     vcpu_match, vcpu_attach, NULL, NULL);
146 #endif
147 
148 /*
149  * Statically-allocated CPU info for the primary CPU (or the only
150  * CPU, on uniprocessors).  The CPU info list is initialized to
151  * point at it.
152  */
153 #ifdef TRAPLOG
154 #include <machine/tlog.h>
155 struct tlog tlog_primary;
156 #endif
157 struct cpu_info cpu_info_primary = {
158 	.ci_dev = 0,
159 	.ci_self = &cpu_info_primary,
160 	.ci_idepth = -1,
161 	.ci_curlwp = &lwp0,
162 #ifdef TRAPLOG
163 	.ci_tlog = &tlog_primary,
164 #endif
165 
166 };
167 struct cpu_info phycpu_info_primary = {
168 	.ci_dev = 0,
169 	.ci_self = &phycpu_info_primary,
170 };
171 
172 struct cpu_info *cpu_info_list = &cpu_info_primary;
173 
174 static void	cpu_set_tss_gates(struct cpu_info *ci);
175 
176 uint32_t cpus_attached = 0;
177 uint32_t cpus_running = 0;
178 
179 bool x86_mp_online;
180 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE;
181 
182 struct cpu_info *phycpu_info[X86_MAXPROCS] = { &cpu_info_primary };
183 
184 #ifdef MULTIPROCESSOR
185 /*
186  * Array of CPU info structures.  Must be statically-allocated because
187  * curproc, etc. are used early.
188  */
189 struct cpu_info *cpu_info[X86_MAXPROCS] = { &cpu_info_primary };
190 
191 void    	cpu_hatch(void *);
192 static void    	cpu_boot_secondary(struct cpu_info *ci);
193 static void    	cpu_start_secondary(struct cpu_info *ci);
194 static void	cpu_copy_trampoline(void);
195 
196 /*
197  * Runs once per boot once multiprocessor goo has been detected and
198  * the local APIC on the boot processor has been mapped.
199  *
200  * Called from lapic_boot_init() (from mpbios_scan()).
201  */
202 void
203 cpu_init_first(void)
204 {
205 	int cpunum = lapic_cpu_number();
206 
207 	if (cpunum != 0) {
208 		cpu_info[0] = NULL;
209 		cpu_info[cpunum] = &cpu_info_primary;
210 	}
211 
212 	cpu_copy_trampoline();
213 }
214 #endif
215 
216 int
217 cpu_match(device_t parent, cfdata_t match, void *aux)
218 {
219 
220 	return 1;
221 }
222 
223 void
224 cpu_attach(device_t parent, device_t self, void *aux)
225 {
226 #ifdef XEN3
227 	struct cpu_softc *sc = device_private(self);
228 	struct cpu_attach_args *caa = aux;
229 	struct cpu_info *ci;
230 	int cpunum = caa->cpu_number;
231 
232 	sc->sc_dev = self;
233 
234 	/*
235 	 * If we're an Application Processor, allocate a cpu_info
236 	 * structure, otherwise use the primary's.
237 	 */
238 	if (caa->cpu_role == CPU_ROLE_AP) {
239 		ci = malloc(sizeof(*ci), M_DEVBUF, M_WAITOK | M_ZERO);
240 		if (phycpu_info[cpunum] != NULL)
241 			panic("cpu at apic id %d already attached?", cpunum);
242 		phycpu_info[cpunum] = ci;
243 	} else {
244 		ci = &phycpu_info_primary;
245 		if (cpunum != 0) {
246 			phycpu_info[0] = NULL;
247 			phycpu_info[cpunum] = ci;
248 		}
249 	}
250 
251 	ci->ci_self = ci;
252 	sc->sc_info = ci;
253 
254 	ci->ci_dev = self;
255 	ci->ci_apicid = caa->cpu_number;
256 	ci->ci_cpuid = ci->ci_apicid;
257 	ci->ci_vcpu = NULL;
258 
259 	printf(": ");
260 	switch (caa->cpu_role) {
261 	case CPU_ROLE_SP:
262 		printf("(uniprocessor)\n");
263 		ci->ci_flags |= CPUF_PRESENT | CPUF_SP | CPUF_PRIMARY;
264 		break;
265 
266 	case CPU_ROLE_BP:
267 		printf("(boot processor)\n");
268 		ci->ci_flags |= CPUF_PRESENT | CPUF_BSP | CPUF_PRIMARY;
269 		break;
270 
271 	case CPU_ROLE_AP:
272 		/*
273 		 * report on an AP
274 		 */
275 		printf("(application processor)\n");
276 		break;
277 
278 	default:
279 		panic("unknown processor type??\n");
280 	}
281 	return;
282 #else
283 	cpu_attach_common(parent, self, aux);
284 #endif
285 }
286 
287 #ifdef XEN3
288 int
289 vcpu_match(device_t parent, cfdata_t match, void *aux)
290 {
291 	struct vcpu_attach_args *vcaa = aux;
292 
293 	if (strcmp(vcaa->vcaa_name, match->cf_name) == 0)
294 		return 1;
295 	return 0;
296 }
297 
298 void
299 vcpu_attach(device_t parent, device_t self, void *aux)
300 {
301 	struct vcpu_attach_args *vcaa = aux;
302 
303 	cpu_attach_common(parent, self, &vcaa->vcaa_caa);
304 }
305 #endif
306 
307 static void
308 cpu_vm_init(struct cpu_info *ci)
309 {
310 	int ncolors = 2, i;
311 
312 	for (i = CAI_ICACHE; i <= CAI_L2CACHE; i++) {
313 		struct x86_cache_info *cai;
314 		int tcolors;
315 
316 		cai = &ci->ci_cinfo[i];
317 
318 		tcolors = atop(cai->cai_totalsize);
319 		switch(cai->cai_associativity) {
320 		case 0xff:
321 			tcolors = 1; /* fully associative */
322 			break;
323 		case 0:
324 		case 1:
325 			break;
326 		default:
327 			tcolors /= cai->cai_associativity;
328 		}
329 		ncolors = max(ncolors, tcolors);
330 	}
331 
332 	/*
333 	 * Knowing the size of the largest cache on this CPU, re-color
334 	 * our pages.
335 	 */
336 	if (ncolors <= uvmexp.ncolors)
337 		return;
338 	printf("%s: %d page colors\n", device_xname(ci->ci_dev), ncolors);
339 	uvm_page_recolor(ncolors);
340 }
341 
342 void
343 cpu_attach_common(device_t parent, device_t self, void *aux)
344 {
345 	struct cpu_softc *sc = device_private(self);
346 	struct cpu_attach_args *caa = aux;
347 	struct cpu_info *ci;
348 	uintptr_t ptr;
349 	int cpunum = caa->cpu_number;
350 
351 	sc->sc_dev = self;
352 
353 	/*
354 	 * If we're an Application Processor, allocate a cpu_info
355 	 * structure, otherwise use the primary's.
356 	 */
357 	if (caa->cpu_role == CPU_ROLE_AP) {
358 		if (cpunum >= X86_MAXPROCS) {
359 			aprint_error(": apic id %d ignored, "
360 				"please increase X86_MAXPROCS\n", cpunum);
361 		}
362 
363 		aprint_naive(": Application Processor\n");
364 		ptr = (uintptr_t)malloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
365 		    M_DEVBUF, M_WAITOK);
366 		ci = (struct cpu_info *)((ptr + CACHE_LINE_SIZE - 1) &
367 		    ~(CACHE_LINE_SIZE - 1));
368 		memset(ci, 0, sizeof(*ci));
369 #if defined(MULTIPROCESSOR)
370 		if (cpu_info[cpunum] != NULL)
371 			panic("cpu at apic id %d already attached?", cpunum);
372 		cpu_info[cpunum] = ci;
373 #endif
374 #ifdef TRAPLOG
375 		ci->ci_tlog_base = malloc(sizeof(struct tlog),
376 		    M_DEVBUF, M_WAITOK);
377 #endif
378 	} else {
379 		aprint_naive(": %s Processor\n",
380 		    caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
381 		ci = &cpu_info_primary;
382 #if defined(MULTIPROCESSOR)
383 		if (cpunum != lapic_cpu_number()) {
384 			panic("%s: running CPU is at apic %d"
385 			    " instead of at expected %d",
386 			    device_xname(sc->sc_dev), lapic_cpu_number(), cpunum);
387 		}
388 #endif
389 	}
390 
391 	ci->ci_self = ci;
392 	sc->sc_info = ci;
393 
394 	ci->ci_dev = self;
395 	ci->ci_apicid = cpunum;
396 
397 	KASSERT(HYPERVISOR_shared_info != NULL);
398 	ci->ci_vcpu = &HYPERVISOR_shared_info->vcpu_info[cpunum];
399 
400 #ifdef MULTIPROCESSOR
401 	ci->ci_cpuid = ci->ci_apicid;
402 #else
403 	ci->ci_cpuid = 0;	/* False for APs, but they're not used anyway */
404 #endif
405 	ci->ci_cpumask = (1 << ci->ci_cpuid);
406 	ci->ci_func = caa->cpu_func;
407 
408 	if (caa->cpu_role == CPU_ROLE_AP) {
409 #if defined(MULTIPROCESSOR)
410 		int error;
411 
412 		error = mi_cpu_attach(ci);
413 		if (error != 0) {
414 			aprint_normal("\n");
415 			aprint_error_dev(sc->sc_dev, "mi_cpu_attach failed with %d\n",
416 			    error);
417 			return;
418 		}
419 #endif
420 	} else {
421 		KASSERT(ci->ci_data.cpu_idlelwp != NULL);
422 	}
423 
424 	pmap_reference(pmap_kernel());
425 	ci->ci_pmap = pmap_kernel();
426 	ci->ci_tlbstate = TLBSTATE_STALE;
427 
428 	/* further PCB init done later. */
429 
430 	switch (caa->cpu_role) {
431 	case CPU_ROLE_SP:
432 		aprint_normal(": (uniprocessor)\n");
433 		atomic_or_32(&ci->ci_flags,
434 		     CPUF_PRESENT | CPUF_SP | CPUF_PRIMARY);
435 		cpu_intr_init(ci);
436 		identifycpu(ci);
437 		cpu_init(ci);
438 		cpu_set_tss_gates(ci);
439 		pmap_cpu_init_late(ci);
440 #if 0
441 		x86_errata();
442 #endif
443 		break;
444 
445 	case CPU_ROLE_BP:
446 		aprint_normal("apid %d (boot processor)\n", caa->cpu_number);
447 		atomic_or_32(&ci->ci_flags,
448 		    CPUF_PRESENT | CPUF_BSP | CPUF_PRIMARY);
449 		cpu_intr_init(ci);
450 		identifycpu(ci);
451 		cpu_init(ci);
452 		cpu_set_tss_gates(ci);
453 		pmap_cpu_init_late(ci);
454 #if NLAPIC > 0
455 		/*
456 		 * Enable local apic
457 		 */
458 		lapic_enable();
459 		lapic_set_lvt();
460 		lapic_calibrate_timer(ci);
461 #endif
462 #if 0
463 		x86_errata();
464 #endif
465 		break;
466 
467 	case CPU_ROLE_AP:
468 		/*
469 		 * report on an AP
470 		 */
471 		aprint_normal("apid %d (application processor)\n", caa->cpu_number);
472 
473 #if defined(MULTIPROCESSOR)
474 		cpu_intr_init(ci);
475 		gdt_alloc_cpu(ci);
476 		cpu_set_tss_gates(ci);
477 		pmap_cpu_init_early(ci);
478 		pmap_cpu_init_late(ci);
479 		cpu_start_secondary(ci);
480 		if (ci->ci_flags & CPUF_PRESENT) {
481 			identifycpu(ci);
482 			ci->ci_next = cpu_info_list->ci_next;
483 			cpu_info_list->ci_next = ci;
484 		}
485 #else
486 		aprint_normal_dev(sc->sc_dev, "not started\n");
487 #endif
488 		break;
489 
490 	default:
491 		aprint_normal("\n");
492 		panic("unknown processor type??\n");
493 	}
494 	cpu_vm_init(ci);
495 
496 	cpus_attached |= (1 << ci->ci_cpuid);
497 
498 #if 0
499 	if (!pmf_device_register(self, cpu_suspend, cpu_resume))
500 		aprint_error_dev(self, "couldn't establish power handler\n");
501 #endif
502 
503 #if defined(MULTIPROCESSOR)
504 	if (mp_verbose) {
505 		struct lwp *l = ci->ci_data.cpu_idlelwp;
506 
507 		aprint_verbose_dev(sc->sc_dev, "idle lwp at %p, idle sp at 0x%p\n",
508 		    l,
509 #ifdef i386
510 		    (void *)l->l_addr->u_pcb.pcb_esp
511 #else
512 		    (void *)l->l_addr->u_pcb.pcb_rsp
513 #endif
514 		);
515 
516 	}
517 #endif
518 }
519 
520 /*
521  * Initialize the processor appropriately.
522  */
523 
524 void
525 cpu_init(struct cpu_info *ci)
526 {
527 	/* configure the CPU if needed */
528 	if (ci->cpu_setup != NULL)
529 		(*ci->cpu_setup)(ci);
530 
531 	/*
532 	 * On a P6 or above, enable global TLB caching if the
533 	 * hardware supports it.
534 	 */
535 	if (cpu_feature & CPUID_PGE)
536 		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
537 
538 #ifdef XXXMTRR
539 	/*
540 	 * On a P6 or above, initialize MTRR's if the hardware supports them.
541 	 */
542 	if (cpu_feature & CPUID_MTRR) {
543 		if ((ci->ci_flags & CPUF_AP) == 0)
544 			i686_mtrr_init_first();
545 		mtrr_init_cpu(ci);
546 	}
547 #endif
548 	/*
549 	 * If we have FXSAVE/FXRESTOR, use them.
550 	 */
551 	if (cpu_feature & CPUID_FXSR) {
552 		lcr4(rcr4() | CR4_OSFXSR);
553 
554 		/*
555 		 * If we have SSE/SSE2, enable XMM exceptions.
556 		 */
557 		if (cpu_feature & (CPUID_SSE|CPUID_SSE2))
558 			lcr4(rcr4() | CR4_OSXMMEXCPT);
559 	}
560 
561 #ifdef MULTIPROCESSOR
562 	atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
563 	atomic_or_32(&cpus_running, ci->ci_cpumask);
564 #endif
565 }
566 
567 
568 #ifdef MULTIPROCESSOR
569 void
570 cpu_boot_secondary_processors(void)
571 {
572 	struct cpu_info *ci;
573 	u_long i;
574 
575 	for (i = 0; i < X86_MAXPROCS; i++) {
576 		ci = cpu_info[i];
577 		if (ci == NULL)
578 			continue;
579 		if (ci->ci_data.cpu_idlelwp == NULL)
580 			continue;
581 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
582 			continue;
583 		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
584 			continue;
585 		cpu_boot_secondary(ci);
586 	}
587 
588 	x86_mp_online = true;
589 }
590 
591 static void
592 cpu_init_idle_lwp(struct cpu_info *ci)
593 {
594 	struct lwp *l = ci->ci_data.cpu_idlelwp;
595 	struct pcb *pcb = &l->l_addr->u_pcb;
596 
597 	pcb->pcb_cr0 = rcr0();
598 }
599 
600 void
601 cpu_init_idle_lwps(void)
602 {
603 	struct cpu_info *ci;
604 	u_long i;
605 
606 	for (i = 0; i < X86_MAXPROCS; i++) {
607 		ci = cpu_info[i];
608 		if (ci == NULL)
609 			continue;
610 		if (ci->ci_data.cpu_idlelwp == NULL)
611 			continue;
612 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
613 			continue;
614 		cpu_init_idle_lwp(ci);
615 	}
616 }
617 
618 void
619 cpu_start_secondary(struct cpu_info *ci)
620 {
621 	int i;
622 	struct pmap *kpm = pmap_kernel();
623 	extern uint32_t mp_pdirpa;
624 
625 	mp_pdirpa = kpm->pm_pdirpa; /* XXX move elsewhere, not per CPU. */
626 
627 	atomic_or_32(&ci->ci_flags, CPUF_AP);
628 
629 	aprint_debug_dev(ci->ci_dev, "starting\n");
630 
631 	ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
632 	if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0)
633 		return;
634 
635 	/*
636 	 * wait for it to become ready
637 	 */
638 	for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
639 #ifdef MPDEBUG
640 		extern int cpu_trace[3];
641 		static int otrace[3];
642 		if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) {
643 			aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n",
644 				cpu_trace[0], cpu_trace[1], cpu_trace[2]);
645 			memcpy(otrace, cpu_trace, sizeof(otrace));
646 		}
647 #endif
648 		delay(10);
649 	}
650 	if ((ci->ci_flags & CPUF_PRESENT) == 0) {
651 		aprint_error_dev(ci->ci_dev, "failed to become ready\n");
652 #if defined(MPDEBUG) && defined(DDB)
653 		printf("dropping into debugger; continue from here to resume boot\n");
654 		Debugger();
655 #endif
656 	}
657 
658 	CPU_START_CLEANUP(ci);
659 }
660 
661 void
662 cpu_boot_secondary(struct cpu_info *ci)
663 {
664 	int i;
665 
666 	atomic_or_32(&ci->ci_flags, CPUF_GO);
667 	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
668 		delay(10);
669 	}
670 	if ((ci->ci_flags & CPUF_RUNNING) == 0) {
671 		aprint_error_dev(ci->ci_dev, "CPU failed to start\n");
672 #if defined(MPDEBUG) && defined(DDB)
673 		printf("dropping into debugger; continue from here to resume boot\n");
674 		Debugger();
675 #endif
676 	}
677 }
678 
679 /*
680  * The CPU ends up here when its ready to run
681  * This is called from code in mptramp.s; at this point, we are running
682  * in the idle pcb/idle stack of the new CPU.  When this function returns,
683  * this processor will enter the idle loop and start looking for work.
684  *
685  * XXX should share some of this with init386 in machdep.c
686  */
687 void
688 cpu_hatch(void *v)
689 {
690 	struct cpu_info *ci = (struct cpu_info *)v;
691 	int s, i;
692 	uint32_t blacklist_features;
693 
694 #ifdef __x86_64__
695         cpu_init_msrs(ci, true);
696 #endif
697 
698 	cpu_probe_features(ci);
699 	cpu_feature &= ci->ci_feature_flags;
700 	cpu_feature2 &= ci->ci_feature2_flags;
701 
702 	/* not on Xen... */
703 	blacklist_features = ~(CPUID_PGE|CPUID_PSE|CPUID_MTRR|CPUID_FXSR|CPUID_NOX); /* XXX add CPUID_SVM */
704 
705 	cpu_feature &= blacklist_features;
706 
707 	KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
708 	atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
709 	while ((ci->ci_flags & CPUF_GO) == 0) {
710 		/* Don't use delay, boot CPU may be patching the text. */
711 		for (i = 10000; i != 0; i--)
712 			x86_pause();
713 	}
714 
715 	/* Because the text may have been patched in x86_patch(). */
716 	wbinvd();
717 	x86_flush();
718 
719 	KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
720 
721 	lcr3(pmap_kernel()->pm_pdirpa);
722 	curlwp->l_addr->u_pcb.pcb_cr3 = pmap_kernel()->pm_pdirpa;
723 	lcr0(ci->ci_data.cpu_idlelwp->l_addr->u_pcb.pcb_cr0);
724 	cpu_init_idt();
725 	gdt_init_cpu(ci);
726 	lapic_enable();
727 	lapic_set_lvt();
728 	lapic_initclocks();
729 
730 #ifdef i386
731 	npxinit(ci);
732 #else
733 	fpuinit(ci);
734 #endif
735 
736 	lldt(GSEL(GLDT_SEL, SEL_KPL));
737 	ltr(ci->ci_tss_sel);
738 
739 	cpu_init(ci);
740 	cpu_get_tsc_freq(ci);
741 
742 	s = splhigh();
743 #ifdef i386
744 	lapic_tpr = 0;
745 #else
746 	lcr8(0);
747 #endif
748 	x86_enable_intr();
749 	splx(s);
750 #if 0
751 	x86_errata();
752 #endif
753 
754 	aprint_debug_dev(ci->ci_dev, "CPU %ld running\n",
755 		(long)ci->ci_cpuid);
756 }
757 
758 #if defined(DDB)
759 
760 #include <ddb/db_output.h>
761 #include <machine/db_machdep.h>
762 
763 /*
764  * Dump CPU information from ddb.
765  */
766 void
767 cpu_debug_dump(void)
768 {
769 	struct cpu_info *ci;
770 	CPU_INFO_ITERATOR cii;
771 
772 	db_printf("addr		dev	id	flags	ipis	curlwp 		fpcurlwp\n");
773 	for (CPU_INFO_FOREACH(cii, ci)) {
774 		db_printf("%p	%s	%ld	%x	%x	%10p	%10p\n",
775 		    ci,
776 		    ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
777 		    (long)ci->ci_cpuid,
778 		    ci->ci_flags, ci->ci_ipis,
779 		    ci->ci_curlwp,
780 		    ci->ci_fpcurlwp);
781 	}
782 }
783 #endif
784 
785 static void
786 cpu_copy_trampoline(void)
787 {
788 	/*
789 	 * Copy boot code.
790 	 */
791 	extern u_char cpu_spinup_trampoline[];
792 	extern u_char cpu_spinup_trampoline_end[];
793 
794 	vaddr_t mp_trampoline_vaddr;
795 
796 	mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
797 		UVM_KMF_VAONLY);
798 
799 	pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
800 		VM_PROT_READ | VM_PROT_WRITE);
801 	pmap_update(pmap_kernel());
802 	memcpy((void *)mp_trampoline_vaddr,
803 		cpu_spinup_trampoline,
804 		cpu_spinup_trampoline_end - cpu_spinup_trampoline);
805 
806 	pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
807 	pmap_update(pmap_kernel());
808 	uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
809 }
810 
811 #endif
812 
813 #ifdef i386
814 #if 0
815 static void
816 tss_init(struct i386tss *tss, void *stack, void *func)
817 {
818 	memset(tss, 0, sizeof *tss);
819 	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
820 	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
821 	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
822 	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
823 	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
824 	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
825 	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
826 	tss->tss_esp = (int)((char *)stack + USPACE - 16);
827 	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
828 	tss->__tss_eflags = PSL_MBO | PSL_NT;   /* XXX not needed? */
829 	tss->__tss_eip = (int)func;
830 }
831 #endif
832 
833 /* XXX */
834 #define IDTVEC(name)	__CONCAT(X, name)
835 typedef void (vector)(void);
836 extern vector IDTVEC(tss_trap08);
837 #ifdef DDB
838 extern vector Xintrddbipi;
839 extern int ddb_vec;
840 #endif
841 
842 static void
843 cpu_set_tss_gates(struct cpu_info *ci)
844 {
845 #if 0
846 	struct segment_descriptor sd;
847 
848 	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
849 	    UVM_KMF_WIRED);
850 	tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
851 	    IDTVEC(tss_trap08));
852 	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
853 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
854 	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
855 	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
856 	    GSEL(GTRAPTSS_SEL, SEL_KPL));
857 #endif
858 
859 #if defined(DDB) && defined(MULTIPROCESSOR)
860 	/*
861 	 * Set up separate handler for the DDB IPI, so that it doesn't
862 	 * stomp on a possibly corrupted stack.
863 	 *
864 	 * XXX overwriting the gate set in db_machine_init.
865 	 * Should rearrange the code so that it's set only once.
866 	 */
867 	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
868 	    UVM_KMF_WIRED);
869 	tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
870 	    Xintrddbipi);
871 
872 	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
873 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
874 	ci->ci_gdt[GIPITSS_SEL].sd = sd;
875 
876 	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
877 	    GSEL(GIPITSS_SEL, SEL_KPL));
878 #endif
879 }
880 #else
881 static void
882 cpu_set_tss_gates(struct cpu_info *ci)
883 {
884 
885 }
886 #endif	/* i386 */
887 
888 int
889 mp_cpu_start(struct cpu_info *ci, paddr_t target)
890 {
891 #if 0
892 #if NLAPIC > 0
893 	int error;
894 #endif
895 	unsigned short dwordptr[2];
896 
897 	/*
898 	 * Bootstrap code must be addressable in real mode
899 	 * and it must be page aligned.
900 	 */
901 	KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
902 
903 	/*
904 	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
905 	 */
906 
907 	outb(IO_RTC, NVRAM_RESET);
908 	outb(IO_RTC+1, NVRAM_RESET_JUMP);
909 
910 	/*
911 	 * "and the warm reset vector (DWORD based at 40:67) to point
912 	 * to the AP startup code ..."
913 	 */
914 
915 	dwordptr[0] = 0;
916 	dwordptr[1] = target >> 4;
917 
918 	pmap_kenter_pa (0, 0, VM_PROT_READ|VM_PROT_WRITE);
919 	memcpy ((uint8_t *) 0x467, dwordptr, 4);
920 	pmap_kremove (0, PAGE_SIZE);
921 
922 #if NLAPIC > 0
923 	/*
924 	 * ... prior to executing the following sequence:"
925 	 */
926 
927 	if (ci->ci_flags & CPUF_AP) {
928 		if ((error = x86_ipi_init(ci->ci_apicid)) != 0)
929 			return error;
930 
931 		delay(10000);
932 
933 		if (cpu_feature & CPUID_APIC) {
934 			error = x86_ipi_init(ci->ci_apicid);
935 			if (error != 0) {
936 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
937 						__func__);
938 				return error;
939 			}
940 
941 			delay(10000);
942 
943 			error = x86_ipi(target / PAGE_SIZE, ci->ci_apicid,
944 					LAPIC_DLMODE_STARTUP);
945 			if (error != 0) {
946 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
947 						__func__);
948 				return error;
949 			}
950 			delay(200);
951 
952 			error = x86_ipi(target / PAGE_SIZE, ci->ci_apicid,
953 					LAPIC_DLMODE_STARTUP);
954 			if (error != 0) {
955 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken ((3)\n",
956 						__func__);
957 				return error;
958 			}
959 			delay(200);
960 		}
961 	}
962 #endif
963 #endif /* 0 */
964 	return 0;
965 }
966 
967 void
968 mp_cpu_start_cleanup(struct cpu_info *ci)
969 {
970 #if 0
971 	/*
972 	 * Ensure the NVRAM reset byte contains something vaguely sane.
973 	 */
974 
975 	outb(IO_RTC, NVRAM_RESET);
976 	outb(IO_RTC+1, NVRAM_RESET_RST);
977 #endif
978 }
979 
980 #ifdef __x86_64__
981 
982 void
983 cpu_init_msrs(struct cpu_info *ci, bool full)
984 {
985 	if (full) {
986 		HYPERVISOR_set_segment_base (SEGBASE_FS, 0);
987 		HYPERVISOR_set_segment_base (SEGBASE_GS_KERNEL, (uint64_t) ci);
988 		HYPERVISOR_set_segment_base (SEGBASE_GS_USER, 0);
989 	}
990 }
991 #endif	/* __x86_64__ */
992 
993 void
994 cpu_offline_md(void)
995 {
996         int s;
997 
998         s = splhigh();
999 #ifdef __i386__
1000         npxsave_cpu(true);
1001 #else
1002         fpusave_cpu(true);
1003 #endif
1004         splx(s);
1005 }
1006 
1007 #if 0
1008 /* XXX joerg restructure and restart CPUs individually */
1009 static bool
1010 cpu_suspend(device_t dv PMF_FN_ARGS)
1011 {
1012 	struct cpu_softc *sc = device_private(dv);
1013 	struct cpu_info *ci = sc->sc_info;
1014 	int err;
1015 
1016 	if (ci->ci_flags & CPUF_PRIMARY)
1017 		return true;
1018 	if (ci->ci_data.cpu_idlelwp == NULL)
1019 		return true;
1020 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1021 		return true;
1022 
1023 	sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
1024 
1025 	if (sc->sc_wasonline) {
1026 		mutex_enter(&cpu_lock);
1027 		err = cpu_setonline(ci, false);
1028 		mutex_exit(&cpu_lock);
1029 
1030 		if (err)
1031 			return false;
1032 	}
1033 
1034 	return true;
1035 }
1036 
1037 static bool
1038 cpu_resume(device_t dv PMF_FN_ARGS)
1039 {
1040 	struct cpu_softc *sc = device_private(dv);
1041 	struct cpu_info *ci = sc->sc_info;
1042 	int err = 0;
1043 
1044 	if (ci->ci_flags & CPUF_PRIMARY)
1045 		return true;
1046 	if (ci->ci_data.cpu_idlelwp == NULL)
1047 		return true;
1048 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1049 		return true;
1050 
1051 	if (sc->sc_wasonline) {
1052 		mutex_enter(&cpu_lock);
1053 		err = cpu_setonline(ci, true);
1054 		mutex_exit(&cpu_lock);
1055 	}
1056 
1057 	return err == 0;
1058 }
1059 #endif
1060 
1061 void
1062 cpu_get_tsc_freq(struct cpu_info *ci)
1063 {
1064 #ifdef XEN3
1065 	const volatile vcpu_time_info_t *tinfo = &ci->ci_vcpu->time;
1066 	delay(1000000);
1067 	uint64_t freq = 1000000000ULL << 32;
1068 	freq = freq / (uint64_t)tinfo->tsc_to_system_mul;
1069 	if ( tinfo->tsc_shift < 0 )
1070 		freq = freq << -tinfo->tsc_shift;
1071 	else
1072 		freq = freq >> tinfo->tsc_shift;
1073 	ci->ci_tsc_freq = freq;
1074 #else
1075 	/* Xen2 */
1076 	/* XXX this needs to read the shared_info of the CPU being probed.. */
1077 	ci->ci_tsc_freq = HYPERVISOR_shared_info->cpu_freq;
1078 #endif /* XEN3 */
1079 }
1080