xref: /netbsd-src/sys/arch/xen/x86/cpu.c (revision bbde328be4e75ea9ad02e9715ea13ca54b797ada)
1 /*	$NetBSD: cpu.c,v 1.43 2010/04/18 23:47:52 jym 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 1.43 2010/04/18 23:47:52 jym 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_extern.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 int     cpu_match(device_t, cfdata_t, void *);
119 void    cpu_attach(device_t, device_t, void *);
120 int     vcpu_match(device_t, cfdata_t, void *);
121 void    vcpu_attach(device_t, device_t, void *);
122 void    cpu_attach_common(device_t, device_t, void *);
123 void	cpu_offline_md(void);
124 
125 struct cpu_softc {
126 	device_t sc_dev;		/* device tree glue */
127 	struct cpu_info *sc_info;	/* pointer to CPU info */
128 	bool sc_wasonline;
129 };
130 
131 int mp_cpu_start(struct cpu_info *, paddr_t);
132 void mp_cpu_start_cleanup(struct cpu_info *);
133 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL,
134 				      mp_cpu_start_cleanup };
135 
136 CFATTACH_DECL_NEW(cpu, sizeof(struct cpu_softc),
137     cpu_match, cpu_attach, NULL, NULL);
138 CFATTACH_DECL_NEW(vcpu, sizeof(struct cpu_softc),
139     vcpu_match, vcpu_attach, NULL, NULL);
140 
141 /*
142  * Statically-allocated CPU info for the primary CPU (or the only
143  * CPU, on uniprocessors).  The CPU info list is initialized to
144  * point at it.
145  */
146 #ifdef TRAPLOG
147 #include <machine/tlog.h>
148 struct tlog tlog_primary;
149 #endif
150 struct cpu_info cpu_info_primary __aligned(CACHE_LINE_SIZE) = {
151 	.ci_dev = 0,
152 	.ci_self = &cpu_info_primary,
153 	.ci_idepth = -1,
154 	.ci_curlwp = &lwp0,
155 	.ci_curldt = -1,
156 #ifdef TRAPLOG
157 	.ci_tlog = &tlog_primary,
158 #endif
159 
160 };
161 struct cpu_info phycpu_info_primary __aligned(CACHE_LINE_SIZE) = {
162 	.ci_dev = 0,
163 	.ci_self = &phycpu_info_primary,
164 };
165 
166 struct cpu_info *cpu_info_list = &cpu_info_primary;
167 struct cpu_info *phycpu_info_list = &phycpu_info_primary;
168 
169 static void	cpu_set_tss_gates(struct cpu_info *ci);
170 
171 uint32_t cpus_attached = 0;
172 uint32_t cpus_running = 0;
173 
174 uint32_t phycpus_attached = 0;
175 uint32_t phycpus_running = 0;
176 
177 uint32_t cpu_feature[5]; /* X86 CPUID feature bits
178 			  *	[0] basic features %edx
179 			  *	[1] basic features %ecx
180 			  *	[2] extended features %edx
181 			  *	[3] extended features %ecx
182 			  *	[4] VIA padlock features
183 			  */
184 
185 bool x86_mp_online;
186 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE;
187 
188 #if defined(MULTIPROCESSOR)
189 void    	cpu_hatch(void *);
190 static void    	cpu_boot_secondary(struct cpu_info *ci);
191 static void    	cpu_start_secondary(struct cpu_info *ci);
192 static void	cpu_copy_trampoline(void);
193 
194 /*
195  * Runs once per boot once multiprocessor goo has been detected and
196  * the local APIC on the boot processor has been mapped.
197  *
198  * Called from lapic_boot_init() (from mpbios_scan()).
199  */
200 void
201 cpu_init_first(void)
202 {
203 
204 	cpu_info_primary.ci_cpuid = lapic_cpu_number();
205 	cpu_copy_trampoline();
206 }
207 #endif	/* MULTIPROCESSOR */
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 	static bool again = false;
224 
225 	sc->sc_dev = self;
226 
227 	if (phycpus_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 *)roundup2(ptr, CACHE_LINE_SIZE);
248 		ci->ci_curldt = -1;
249 	} else {
250 		aprint_naive(": %s Processor\n",
251 		    caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
252 		ci = &phycpu_info_primary;
253 	}
254 
255 	ci->ci_self = ci;
256 	sc->sc_info = ci;
257 
258 	ci->ci_dev = self;
259 	ci->ci_cpuid = caa->cpu_number;
260 	ci->ci_vcpu = NULL;
261 
262 	/*
263 	 * Boot processor may not be attached first, but the below
264 	 * must be done to allow booting other processors.
265 	 */
266 	if (!again) {
267 		atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_PRIMARY);
268 		/* Basic init */
269 		again = true;
270 	}
271 
272 	printf(": ");
273 	switch (caa->cpu_role) {
274 	case CPU_ROLE_SP:
275 		printf("(uniprocessor)\n");
276 		atomic_or_32(&ci->ci_flags, CPUF_SP);
277 		break;
278 
279 	case CPU_ROLE_BP:
280 		printf("(boot processor)\n");
281 		atomic_or_32(&ci->ci_flags, CPUF_BSP);
282 		break;
283 
284 	case CPU_ROLE_AP:
285 		/*
286 		 * report on an AP
287 		 */
288 		printf("(application processor)\n");
289 		if (ci->ci_flags & CPUF_PRESENT) {
290 			struct cpu_info *tmp;
291 
292 			tmp = phycpu_info_list;
293 			while (tmp->ci_next)
294 				tmp = tmp->ci_next;
295 
296 			tmp->ci_next = ci;
297 		}
298 		break;
299 
300 	default:
301 		panic("unknown processor type??\n");
302 	}
303 
304 	atomic_or_32(&phycpus_attached, ci->ci_cpumask);
305 
306 	return;
307 }
308 
309 int
310 vcpu_match(device_t parent, cfdata_t match, void *aux)
311 {
312 	struct vcpu_attach_args *vcaa = aux;
313 
314 	if (strcmp(vcaa->vcaa_name, match->cf_name) == 0)
315 		return 1;
316 	return 0;
317 }
318 
319 void
320 vcpu_attach(device_t parent, device_t self, void *aux)
321 {
322 	struct vcpu_attach_args *vcaa = aux;
323 
324 	cpu_attach_common(parent, self, &vcaa->vcaa_caa);
325 }
326 
327 static void
328 cpu_vm_init(struct cpu_info *ci)
329 {
330 	int ncolors = 2, i;
331 
332 	for (i = CAI_ICACHE; i <= CAI_L2CACHE; i++) {
333 		struct x86_cache_info *cai;
334 		int tcolors;
335 
336 		cai = &ci->ci_cinfo[i];
337 
338 		tcolors = atop(cai->cai_totalsize);
339 		switch(cai->cai_associativity) {
340 		case 0xff:
341 			tcolors = 1; /* fully associative */
342 			break;
343 		case 0:
344 		case 1:
345 			break;
346 		default:
347 			tcolors /= cai->cai_associativity;
348 		}
349 		ncolors = max(ncolors, tcolors);
350 	}
351 
352 	/*
353 	 * Knowing the size of the largest cache on this CPU, re-color
354 	 * our pages.
355 	 */
356 	if (ncolors <= uvmexp.ncolors)
357 		return;
358 	aprint_debug_dev(ci->ci_dev, "%d page colors\n", ncolors);
359 	uvm_page_recolor(ncolors);
360 }
361 
362 void
363 cpu_attach_common(device_t parent, device_t self, void *aux)
364 {
365 	struct cpu_softc *sc = device_private(self);
366 	struct cpu_attach_args *caa = aux;
367 	struct cpu_info *ci;
368 	uintptr_t ptr;
369 	int cpunum = caa->cpu_number;
370 	static bool again = false;
371 
372 	sc->sc_dev = self;
373 
374 	/*
375 	 * If we're an Application Processor, allocate a cpu_info
376 	 * structure, otherwise use the primary's.
377 	 */
378 	if (caa->cpu_role == CPU_ROLE_AP) {
379 		aprint_naive(": Application Processor\n");
380 		ptr = (uintptr_t)kmem_alloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
381 		    KM_SLEEP);
382 		ci = (struct cpu_info *)roundup2(ptr, CACHE_LINE_SIZE);
383 		memset(ci, 0, sizeof(*ci));
384 #ifdef TRAPLOG
385 		ci->ci_tlog_base = kmem_zalloc(sizeof(struct tlog), KM_SLEEP);
386 #endif
387 	} else {
388 		aprint_naive(": %s Processor\n",
389 		    caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
390 		ci = &cpu_info_primary;
391 #if NLAPIC > 0
392 		if (cpunum != lapic_cpu_number()) {
393 			/* XXX should be done earlier */
394 			uint32_t reg;
395 			aprint_verbose("\n");
396 			aprint_verbose_dev(self, "running CPU at apic %d"
397 			    " instead of at expected %d", lapic_cpu_number(),
398 			    cpunum);
399 			reg = i82489_readreg(LAPIC_ID);
400 			i82489_writereg(LAPIC_ID, (reg & ~LAPIC_ID_MASK) |
401 			    (cpunum << LAPIC_ID_SHIFT));
402 		}
403 		if (cpunum != lapic_cpu_number()) {
404 			aprint_error_dev(self, "unable to reset apic id\n");
405 		}
406 #endif
407 	}
408 
409 	ci->ci_self = ci;
410 	sc->sc_info = ci;
411 	ci->ci_dev = self;
412 	ci->ci_cpuid = cpunum;
413 
414 	KASSERT(HYPERVISOR_shared_info != NULL);
415 	ci->ci_vcpu = &HYPERVISOR_shared_info->vcpu_info[cpunum];
416 
417 	ci->ci_func = caa->cpu_func;
418 
419 	/* Must be called before mi_cpu_attach(). */
420 	cpu_vm_init(ci);
421 
422 	if (caa->cpu_role == CPU_ROLE_AP) {
423 		int error;
424 
425 		error = mi_cpu_attach(ci);
426 		if (error != 0) {
427 			aprint_normal("\n");
428 			aprint_error_dev(self,
429 			    "mi_cpu_attach failed with %d\n", error);
430 			return;
431 		}
432 	} else {
433 		KASSERT(ci->ci_data.cpu_idlelwp != NULL);
434 	}
435 
436 	ci->ci_cpumask = (1 << cpu_index(ci));
437 	pmap_reference(pmap_kernel());
438 	ci->ci_pmap = pmap_kernel();
439 	ci->ci_tlbstate = TLBSTATE_STALE;
440 
441 	/*
442 	 * Boot processor may not be attached first, but the below
443 	 * must be done to allow booting other processors.
444 	 */
445 	if (!again) {
446 		atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_PRIMARY);
447 		/* Basic init. */
448 		cpu_intr_init(ci);
449 		cpu_get_tsc_freq(ci);
450 		cpu_init(ci);
451 		cpu_set_tss_gates(ci);
452 		pmap_cpu_init_late(ci);
453 #if NLAPIC > 0
454 		if (caa->cpu_role != CPU_ROLE_SP) {
455 			/* Enable lapic. */
456 			lapic_enable();
457 			lapic_set_lvt();
458 			lapic_calibrate_timer();
459 		}
460 #endif
461 		/* Make sure DELAY() is initialized. */
462 		DELAY(1);
463 		again = true;
464 	}
465 
466 	/* further PCB init done later. */
467 
468 	switch (caa->cpu_role) {
469 	case CPU_ROLE_SP:
470 		atomic_or_32(&ci->ci_flags, CPUF_SP);
471 		cpu_identify(ci);
472 #if 0
473 		x86_errata();
474 #endif
475 		x86_cpu_idle_init();
476 		break;
477 
478 	case CPU_ROLE_BP:
479 		atomic_or_32(&ci->ci_flags, CPUF_BSP);
480 		cpu_identify(ci);
481 		cpu_init(ci);
482 #if 0
483 		x86_errata();
484 #endif
485 		x86_cpu_idle_init();
486 		break;
487 
488 	case CPU_ROLE_AP:
489 		/*
490 		 * report on an AP
491 		 */
492 
493 #if defined(MULTIPROCESSOR)
494 		cpu_intr_init(ci);
495 		gdt_alloc_cpu(ci);
496 		cpu_set_tss_gates(ci);
497 		pmap_cpu_init_early(ci);
498 		pmap_cpu_init_late(ci);
499 		cpu_start_secondary(ci);
500 		if (ci->ci_flags & CPUF_PRESENT) {
501 			struct cpu_info *tmp;
502 
503 			identifycpu(ci);
504 			tmp = cpu_info_list;
505 			while (tmp->ci_next)
506 				tmp = tmp->ci_next;
507 
508 			tmp->ci_next = ci;
509 		}
510 #else
511 		aprint_error_dev(self, "not started\n");
512 #endif
513 		break;
514 
515 	default:
516 		aprint_normal("\n");
517 		panic("unknown processor type??\n");
518 	}
519 
520 	atomic_or_32(&cpus_attached, ci->ci_cpumask);
521 
522 #if 0
523 	if (!pmf_device_register(self, cpu_suspend, cpu_resume))
524 		aprint_error_dev(self, "couldn't establish power handler\n");
525 #endif
526 
527 #if defined(MULTIPROCESSOR)
528 	if (mp_verbose) {
529 		struct lwp *l = ci->ci_data.cpu_idlelwp;
530 		struct pcb *pcb = lwp_getpcb(l);
531 
532 		aprint_verbose_dev(self,
533 		    "idle lwp at %p, idle sp at 0x%p\n",
534 		    l,
535 #ifdef i386
536 		    (void *)pcb->pcb_esp
537 #else
538 		    (void *)pcb->pcb_rsp
539 #endif
540 		);
541 
542 	}
543 #endif
544 }
545 
546 /*
547  * Initialize the processor appropriately.
548  */
549 
550 void
551 cpu_init(struct cpu_info *ci)
552 {
553 
554 	/*
555 	 * On a P6 or above, enable global TLB caching if the
556 	 * hardware supports it.
557 	 */
558 	if (cpu_feature[0] & CPUID_PGE)
559 		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
560 
561 #ifdef XXXMTRR
562 	/*
563 	 * On a P6 or above, initialize MTRR's if the hardware supports them.
564 	 */
565 	if (cpu_feature[0] & CPUID_MTRR) {
566 		if ((ci->ci_flags & CPUF_AP) == 0)
567 			i686_mtrr_init_first();
568 		mtrr_init_cpu(ci);
569 	}
570 #endif
571 	/*
572 	 * If we have FXSAVE/FXRESTOR, use them.
573 	 */
574 	if (cpu_feature[0] & CPUID_FXSR) {
575 		lcr4(rcr4() | CR4_OSFXSR);
576 
577 		/*
578 		 * If we have SSE/SSE2, enable XMM exceptions.
579 		 */
580 		if (cpu_feature[0] & (CPUID_SSE|CPUID_SSE2))
581 			lcr4(rcr4() | CR4_OSXMMEXCPT);
582 	}
583 
584 	atomic_or_32(&cpus_running, ci->ci_cpumask);
585 	atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
586 }
587 
588 
589 #ifdef MULTIPROCESSOR
590 void
591 cpu_boot_secondary_processors(void)
592 {
593 	struct cpu_info *ci;
594 	u_long i;
595 
596 	for (i = 0; i < maxcpus; i++) {
597 		ci = cpu_lookup(i);
598 		if (ci == NULL)
599 			continue;
600 		if (ci->ci_data.cpu_idlelwp == NULL)
601 			continue;
602 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
603 			continue;
604 		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
605 			continue;
606 		cpu_boot_secondary(ci);
607 	}
608 
609 	x86_mp_online = true;
610 }
611 
612 static void
613 cpu_init_idle_lwp(struct cpu_info *ci)
614 {
615 	struct lwp *l = ci->ci_data.cpu_idlelwp;
616 	struct pcb *pcb = lwp_getpcb(l);
617 
618 	pcb->pcb_cr0 = rcr0();
619 }
620 
621 void
622 cpu_init_idle_lwps(void)
623 {
624 	struct cpu_info *ci;
625 	u_long i;
626 
627 	for (i = 0; i < maxcpus; i++) {
628 		ci = cpu_lookup(i);
629 		if (ci == NULL)
630 			continue;
631 		if (ci->ci_data.cpu_idlelwp == NULL)
632 			continue;
633 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
634 			continue;
635 		cpu_init_idle_lwp(ci);
636 	}
637 }
638 
639 void
640 cpu_start_secondary(struct cpu_info *ci)
641 {
642 	int i;
643 	struct pmap *kpm = pmap_kernel();
644 	extern uint32_t mp_pdirpa;
645 
646 	mp_pdirpa = kpm->pm_pdirpa; /* XXX move elsewhere, not per CPU. */
647 
648 	atomic_or_32(&ci->ci_flags, CPUF_AP);
649 
650 	aprint_debug_dev(ci->ci_dev, "starting\n");
651 
652 	ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
653 	if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0)
654 		return;
655 
656 	/*
657 	 * wait for it to become ready
658 	 */
659 	for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
660 #ifdef MPDEBUG
661 		extern int cpu_trace[3];
662 		static int otrace[3];
663 		if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) {
664 			aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n",
665 				cpu_trace[0], cpu_trace[1], cpu_trace[2]);
666 			memcpy(otrace, cpu_trace, sizeof(otrace));
667 		}
668 #endif
669 		delay(10);
670 	}
671 	if ((ci->ci_flags & CPUF_PRESENT) == 0) {
672 		aprint_error_dev(ci->ci_dev, "failed to become ready\n");
673 #if defined(MPDEBUG) && defined(DDB)
674 		printf("dropping into debugger; continue from here to resume boot\n");
675 		Debugger();
676 #endif
677 	}
678 
679 	CPU_START_CLEANUP(ci);
680 }
681 
682 void
683 cpu_boot_secondary(struct cpu_info *ci)
684 {
685 	int i;
686 
687 	atomic_or_32(&ci->ci_flags, CPUF_GO);
688 	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
689 		delay(10);
690 	}
691 	if ((ci->ci_flags & CPUF_RUNNING) == 0) {
692 		aprint_error_dev(ci->ci_dev, "CPU failed to start\n");
693 #if defined(MPDEBUG) && defined(DDB)
694 		printf("dropping into debugger; continue from here to resume boot\n");
695 		Debugger();
696 #endif
697 	}
698 }
699 
700 /*
701  * The CPU ends up here when its ready to run
702  * This is called from code in mptramp.s; at this point, we are running
703  * in the idle pcb/idle stack of the new CPU.  When this function returns,
704  * this processor will enter the idle loop and start looking for work.
705  *
706  * XXX should share some of this with init386 in machdep.c
707  */
708 void
709 cpu_hatch(void *v)
710 {
711 	struct cpu_info *ci = (struct cpu_info *)v;
712 	struct pcb *pcb;
713 	int s, i;
714 
715 	cpu_probe(ci);
716 
717 	cpu_feature[0] &= ~CPUID_FEAT_BLACKLIST;
718 	cpu_feature[2] &= ~CPUID_FEAT_EXT_BLACKLIST;
719 
720         cpu_init_msrs(ci, true);
721 
722 	KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
723 	atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
724 	while ((ci->ci_flags & CPUF_GO) == 0) {
725 		/* Don't use delay, boot CPU may be patching the text. */
726 		for (i = 10000; i != 0; i--)
727 			x86_pause();
728 	}
729 
730 	/* Because the text may have been patched in x86_patch(). */
731 	wbinvd();
732 	x86_flush();
733 
734 	KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
735 
736 	pcb = lwp_getpcb(curlwp);
737 	lcr3(pmap_kernel()->pm_pdirpa);
738 	pcb->pcb_cr3 = pmap_kernel()->pm_pdirpa;
739 	pcb = lwp_getpcb(ci->ci_data.cpu_idlelwp);
740 	lcr0(pcb->pcb_cr0);
741 
742 	cpu_init_idt();
743 	gdt_init_cpu(ci);
744 	lapic_enable();
745 	lapic_set_lvt();
746 	lapic_initclocks();
747 
748 #ifdef i386
749 	npxinit(ci);
750 #else
751 	fpuinit(ci);
752 #endif
753 
754 	lldt(GSEL(GLDT_SEL, SEL_KPL));
755 	ltr(ci->ci_tss_sel);
756 
757 	cpu_init(ci);
758 	cpu_get_tsc_freq(ci);
759 
760 	s = splhigh();
761 #ifdef i386
762 	lapic_tpr = 0;
763 #else
764 	lcr8(0);
765 #endif
766 	x86_enable_intr();
767 	splx(s);
768 #if 0
769 	x86_errata();
770 #endif
771 
772 	aprint_debug_dev(ci->ci_dev, "CPU %ld running\n",
773 		(long)ci->ci_cpuid);
774 }
775 
776 #if defined(DDB)
777 
778 #include <ddb/db_output.h>
779 #include <machine/db_machdep.h>
780 
781 /*
782  * Dump CPU information from ddb.
783  */
784 void
785 cpu_debug_dump(void)
786 {
787 	struct cpu_info *ci;
788 	CPU_INFO_ITERATOR cii;
789 
790 	db_printf("addr		dev	id	flags	ipis	curlwp 		fpcurlwp\n");
791 	for (CPU_INFO_FOREACH(cii, ci)) {
792 		db_printf("%p	%s	%ld	%x	%x	%10p	%10p\n",
793 		    ci,
794 		    ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
795 		    (long)ci->ci_cpuid,
796 		    ci->ci_flags, ci->ci_ipis,
797 		    ci->ci_curlwp,
798 		    ci->ci_fpcurlwp);
799 	}
800 }
801 #endif /* DDB */
802 
803 static void
804 cpu_copy_trampoline(void)
805 {
806 	/*
807 	 * Copy boot code.
808 	 */
809 	extern u_char cpu_spinup_trampoline[];
810 	extern u_char cpu_spinup_trampoline_end[];
811 
812 	vaddr_t mp_trampoline_vaddr;
813 
814 	mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
815 		UVM_KMF_VAONLY);
816 
817 	pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
818 		VM_PROT_READ | VM_PROT_WRITE, 0);
819 	pmap_update(pmap_kernel());
820 	memcpy((void *)mp_trampoline_vaddr,
821 		cpu_spinup_trampoline,
822 		cpu_spinup_trampoline_end - cpu_spinup_trampoline);
823 
824 	pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
825 	pmap_update(pmap_kernel());
826 	uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
827 }
828 
829 #endif /* MULTIPROCESSOR */
830 
831 #ifdef i386
832 #if 0
833 static void
834 tss_init(struct i386tss *tss, void *stack, void *func)
835 {
836 	memset(tss, 0, sizeof *tss);
837 	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
838 	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
839 	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
840 	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
841 	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
842 	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
843 	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
844 	tss->tss_esp = (int)((char *)stack + USPACE - 16);
845 	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
846 	tss->__tss_eflags = PSL_MBO | PSL_NT;   /* XXX not needed? */
847 	tss->__tss_eip = (int)func;
848 }
849 #endif
850 
851 /* XXX */
852 #define IDTVEC(name)	__CONCAT(X, name)
853 typedef void (vector)(void);
854 extern vector IDTVEC(tss_trap08);
855 #ifdef DDB
856 extern vector Xintrddbipi;
857 extern int ddb_vec;
858 #endif
859 
860 static void
861 cpu_set_tss_gates(struct cpu_info *ci)
862 {
863 #if 0
864 	struct segment_descriptor sd;
865 
866 	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
867 	    UVM_KMF_WIRED);
868 	tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
869 	    IDTVEC(tss_trap08));
870 	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
871 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
872 	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
873 	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
874 	    GSEL(GTRAPTSS_SEL, SEL_KPL));
875 #endif
876 
877 #if defined(DDB) && defined(MULTIPROCESSOR)
878 	/*
879 	 * Set up separate handler for the DDB IPI, so that it doesn't
880 	 * stomp on a possibly corrupted stack.
881 	 *
882 	 * XXX overwriting the gate set in db_machine_init.
883 	 * Should rearrange the code so that it's set only once.
884 	 */
885 	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
886 	    UVM_KMF_WIRED);
887 	tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
888 	    Xintrddbipi);
889 
890 	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
891 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
892 	ci->ci_gdt[GIPITSS_SEL].sd = sd;
893 
894 	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
895 	    GSEL(GIPITSS_SEL, SEL_KPL));
896 #endif
897 }
898 #else
899 static void
900 cpu_set_tss_gates(struct cpu_info *ci)
901 {
902 
903 }
904 #endif	/* i386 */
905 
906 int
907 mp_cpu_start(struct cpu_info *ci, paddr_t target)
908 {
909 #if 0
910 #if NLAPIC > 0
911 	int error;
912 #endif
913 	unsigned short dwordptr[2];
914 
915 	/*
916 	 * Bootstrap code must be addressable in real mode
917 	 * and it must be page aligned.
918 	 */
919 	KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
920 
921 	/*
922 	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
923 	 */
924 
925 	outb(IO_RTC, NVRAM_RESET);
926 	outb(IO_RTC+1, NVRAM_RESET_JUMP);
927 
928 	/*
929 	 * "and the warm reset vector (DWORD based at 40:67) to point
930 	 * to the AP startup code ..."
931 	 */
932 
933 	dwordptr[0] = 0;
934 	dwordptr[1] = target >> 4;
935 
936 	pmap_kenter_pa (0, 0, VM_PROT_READ|VM_PROT_WRITE, 0);
937 	memcpy ((uint8_t *) 0x467, dwordptr, 4);
938 	pmap_kremove (0, PAGE_SIZE);
939 
940 #if NLAPIC > 0
941 	/*
942 	 * ... prior to executing the following sequence:"
943 	 */
944 
945 	if (ci->ci_flags & CPUF_AP) {
946 		if ((error = x86_ipi_init(ci->ci_cpuid)) != 0)
947 			return error;
948 
949 		delay(10000);
950 
951 		if (cpu_feature & CPUID_APIC) {
952 			error = x86_ipi_init(ci->ci_cpuid);
953 			if (error != 0) {
954 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
955 						__func__);
956 				return error;
957 			}
958 
959 			delay(10000);
960 
961 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
962 					LAPIC_DLMODE_STARTUP);
963 			if (error != 0) {
964 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
965 						__func__);
966 				return error;
967 			}
968 			delay(200);
969 
970 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
971 					LAPIC_DLMODE_STARTUP);
972 			if (error != 0) {
973 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken ((3)\n",
974 						__func__);
975 				return error;
976 			}
977 			delay(200);
978 		}
979 	}
980 #endif
981 #endif /* 0 */
982 	return 0;
983 }
984 
985 void
986 mp_cpu_start_cleanup(struct cpu_info *ci)
987 {
988 #if 0
989 	/*
990 	 * Ensure the NVRAM reset byte contains something vaguely sane.
991 	 */
992 
993 	outb(IO_RTC, NVRAM_RESET);
994 	outb(IO_RTC+1, NVRAM_RESET_RST);
995 #endif
996 }
997 
998 void
999 cpu_init_msrs(struct cpu_info *ci, bool full)
1000 {
1001 #ifdef __x86_64__
1002 	if (full) {
1003 		HYPERVISOR_set_segment_base (SEGBASE_FS, 0);
1004 		HYPERVISOR_set_segment_base (SEGBASE_GS_KERNEL, (uint64_t) ci);
1005 		HYPERVISOR_set_segment_base (SEGBASE_GS_USER, 0);
1006 	}
1007 #endif	/* __x86_64__ */
1008 }
1009 
1010 void
1011 cpu_offline_md(void)
1012 {
1013         int s;
1014 
1015         s = splhigh();
1016 #ifdef __i386__
1017         npxsave_cpu(true);
1018 #else
1019         fpusave_cpu(true);
1020 #endif
1021         splx(s);
1022 }
1023 
1024 #if 0
1025 /* XXX joerg restructure and restart CPUs individually */
1026 static bool
1027 cpu_suspend(device_t dv, const pmf_qual_t *qual)
1028 {
1029 	struct cpu_softc *sc = device_private(dv);
1030 	struct cpu_info *ci = sc->sc_info;
1031 	int err;
1032 
1033 	if (ci->ci_flags & CPUF_PRIMARY)
1034 		return true;
1035 	if (ci->ci_data.cpu_idlelwp == NULL)
1036 		return true;
1037 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1038 		return true;
1039 
1040 	sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
1041 
1042 	if (sc->sc_wasonline) {
1043 		mutex_enter(&cpu_lock);
1044 		err = cpu_setstate(ci, false);
1045 		mutex_exit(&cpu_lock);
1046 
1047 		if (err)
1048 			return false;
1049 	}
1050 
1051 	return true;
1052 }
1053 
1054 static bool
1055 cpu_resume(device_t dv, const pmf_qual_t *qual)
1056 {
1057 	struct cpu_softc *sc = device_private(dv);
1058 	struct cpu_info *ci = sc->sc_info;
1059 	int err = 0;
1060 
1061 	if (ci->ci_flags & CPUF_PRIMARY)
1062 		return true;
1063 	if (ci->ci_data.cpu_idlelwp == NULL)
1064 		return true;
1065 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1066 		return true;
1067 
1068 	if (sc->sc_wasonline) {
1069 		mutex_enter(&cpu_lock);
1070 		err = cpu_setstate(ci, true);
1071 		mutex_exit(&cpu_lock);
1072 	}
1073 
1074 	return err == 0;
1075 }
1076 #endif
1077 
1078 void
1079 cpu_get_tsc_freq(struct cpu_info *ci)
1080 {
1081 	const volatile vcpu_time_info_t *tinfo = &ci->ci_vcpu->time;
1082 	delay(1000000);
1083 	uint64_t freq = 1000000000ULL << 32;
1084 	freq = freq / (uint64_t)tinfo->tsc_to_system_mul;
1085 	if ( tinfo->tsc_shift < 0 )
1086 		freq = freq << -tinfo->tsc_shift;
1087 	else
1088 		freq = freq >> tinfo->tsc_shift;
1089 	ci->ci_data.cpu_cc_freq = freq;
1090 }
1091 
1092 void
1093 x86_cpu_idle_xen(void)
1094 {
1095 	struct cpu_info *ci = curcpu();
1096 
1097 	KASSERT(ci->ci_ilevel == IPL_NONE);
1098 
1099 	x86_disable_intr();
1100 	if (!__predict_false(ci->ci_want_resched)) {
1101 		idle_block();
1102 	} else {
1103 		x86_enable_intr();
1104 	}
1105 }
1106