xref: /netbsd-src/sys/arch/xen/x86/cpu.c (revision 179b12252ecaf3553d9c2b7458ce62b6a2203d0c)
1 /*	$NetBSD: cpu.c,v 1.49 2010/08/20 06:34:33 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 1.49 2010/08/20 06:34:33 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_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 	pat_init(ci);
521 	atomic_or_32(&cpus_attached, ci->ci_cpumask);
522 
523 #if 0
524 	if (!pmf_device_register(self, cpu_suspend, cpu_resume))
525 		aprint_error_dev(self, "couldn't establish power handler\n");
526 #endif
527 
528 #if defined(MULTIPROCESSOR)
529 	if (mp_verbose) {
530 		struct lwp *l = ci->ci_data.cpu_idlelwp;
531 		struct pcb *pcb = lwp_getpcb(l);
532 
533 		aprint_verbose_dev(self,
534 		    "idle lwp at %p, idle sp at 0x%p\n",
535 		    l,
536 #ifdef i386
537 		    (void *)pcb->pcb_esp
538 #else
539 		    (void *)pcb->pcb_rsp
540 #endif
541 		);
542 
543 	}
544 #endif
545 }
546 
547 /*
548  * Initialize the processor appropriately.
549  */
550 
551 void
552 cpu_init(struct cpu_info *ci)
553 {
554 
555 	/*
556 	 * On a P6 or above, enable global TLB caching if the
557 	 * hardware supports it.
558 	 */
559 	if (cpu_feature[0] & CPUID_PGE)
560 		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
561 
562 #ifdef XXXMTRR
563 	/*
564 	 * On a P6 or above, initialize MTRR's if the hardware supports them.
565 	 */
566 	if (cpu_feature[0] & CPUID_MTRR) {
567 		if ((ci->ci_flags & CPUF_AP) == 0)
568 			i686_mtrr_init_first();
569 		mtrr_init_cpu(ci);
570 	}
571 #endif
572 	/*
573 	 * If we have FXSAVE/FXRESTOR, use them.
574 	 */
575 	if (cpu_feature[0] & CPUID_FXSR) {
576 		lcr4(rcr4() | CR4_OSFXSR);
577 
578 		/*
579 		 * If we have SSE/SSE2, enable XMM exceptions.
580 		 */
581 		if (cpu_feature[0] & (CPUID_SSE|CPUID_SSE2))
582 			lcr4(rcr4() | CR4_OSXMMEXCPT);
583 	}
584 
585 #ifdef __x86_64__
586 	/* No user PGD mapped for this CPU yet */
587 	ci->ci_xen_current_user_pgd = 0;
588 #endif
589 
590 	atomic_or_32(&cpus_running, ci->ci_cpumask);
591 	atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
592 }
593 
594 
595 #ifdef MULTIPROCESSOR
596 void
597 cpu_boot_secondary_processors(void)
598 {
599 	struct cpu_info *ci;
600 	u_long i;
601 
602 	for (i = 0; i < maxcpus; i++) {
603 		ci = cpu_lookup(i);
604 		if (ci == NULL)
605 			continue;
606 		if (ci->ci_data.cpu_idlelwp == NULL)
607 			continue;
608 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
609 			continue;
610 		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
611 			continue;
612 		cpu_boot_secondary(ci);
613 	}
614 
615 	x86_mp_online = true;
616 }
617 
618 static void
619 cpu_init_idle_lwp(struct cpu_info *ci)
620 {
621 	struct lwp *l = ci->ci_data.cpu_idlelwp;
622 	struct pcb *pcb = lwp_getpcb(l);
623 
624 	pcb->pcb_cr0 = rcr0();
625 }
626 
627 void
628 cpu_init_idle_lwps(void)
629 {
630 	struct cpu_info *ci;
631 	u_long i;
632 
633 	for (i = 0; i < maxcpus; i++) {
634 		ci = cpu_lookup(i);
635 		if (ci == NULL)
636 			continue;
637 		if (ci->ci_data.cpu_idlelwp == NULL)
638 			continue;
639 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
640 			continue;
641 		cpu_init_idle_lwp(ci);
642 	}
643 }
644 
645 void
646 cpu_start_secondary(struct cpu_info *ci)
647 {
648 	int i;
649 	struct pmap *kpm = pmap_kernel();
650 	extern uint32_t mp_pdirpa;
651 
652 	mp_pdirpa = kpm->pm_pdirpa; /* XXX move elsewhere, not per CPU. */
653 
654 	atomic_or_32(&ci->ci_flags, CPUF_AP);
655 
656 	aprint_debug_dev(ci->ci_dev, "starting\n");
657 
658 	ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
659 	if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0)
660 		return;
661 
662 	/*
663 	 * wait for it to become ready
664 	 */
665 	for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
666 #ifdef MPDEBUG
667 		extern int cpu_trace[3];
668 		static int otrace[3];
669 		if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) {
670 			aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n",
671 				cpu_trace[0], cpu_trace[1], cpu_trace[2]);
672 			memcpy(otrace, cpu_trace, sizeof(otrace));
673 		}
674 #endif
675 		delay(10);
676 	}
677 	if ((ci->ci_flags & CPUF_PRESENT) == 0) {
678 		aprint_error_dev(ci->ci_dev, "failed to become ready\n");
679 #if defined(MPDEBUG) && defined(DDB)
680 		printf("dropping into debugger; continue from here to resume boot\n");
681 		Debugger();
682 #endif
683 	}
684 
685 	CPU_START_CLEANUP(ci);
686 }
687 
688 void
689 cpu_boot_secondary(struct cpu_info *ci)
690 {
691 	int i;
692 
693 	atomic_or_32(&ci->ci_flags, CPUF_GO);
694 	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
695 		delay(10);
696 	}
697 	if ((ci->ci_flags & CPUF_RUNNING) == 0) {
698 		aprint_error_dev(ci->ci_dev, "CPU failed to start\n");
699 #if defined(MPDEBUG) && defined(DDB)
700 		printf("dropping into debugger; continue from here to resume boot\n");
701 		Debugger();
702 #endif
703 	}
704 }
705 
706 /*
707  * The CPU ends up here when its ready to run
708  * This is called from code in mptramp.s; at this point, we are running
709  * in the idle pcb/idle stack of the new CPU.  When this function returns,
710  * this processor will enter the idle loop and start looking for work.
711  *
712  * XXX should share some of this with init386 in machdep.c
713  */
714 void
715 cpu_hatch(void *v)
716 {
717 	struct cpu_info *ci = (struct cpu_info *)v;
718 	struct pcb *pcb;
719 	int s, i;
720 
721 	cpu_probe(ci);
722 
723 	cpu_feature[0] &= ~CPUID_FEAT_BLACKLIST;
724 	cpu_feature[2] &= ~CPUID_FEAT_EXT_BLACKLIST;
725 
726         cpu_init_msrs(ci, true);
727 
728 	KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
729 	atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
730 	while ((ci->ci_flags & CPUF_GO) == 0) {
731 		/* Don't use delay, boot CPU may be patching the text. */
732 		for (i = 10000; i != 0; i--)
733 			x86_pause();
734 	}
735 
736 	/* Because the text may have been patched in x86_patch(). */
737 	wbinvd();
738 	x86_flush();
739 
740 	KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
741 
742 	pcb = lwp_getpcb(curlwp);
743 	lcr3(pmap_kernel()->pm_pdirpa);
744 	pcb->pcb_cr3 = pmap_kernel()->pm_pdirpa;
745 	pcb = lwp_getpcb(ci->ci_data.cpu_idlelwp);
746 	lcr0(pcb->pcb_cr0);
747 
748 	cpu_init_idt();
749 	gdt_init_cpu(ci);
750 	lapic_enable();
751 	lapic_set_lvt();
752 	lapic_initclocks();
753 
754 #ifdef i386
755 	npxinit(ci);
756 #else
757 	fpuinit(ci);
758 #endif
759 
760 	lldt(GSEL(GLDT_SEL, SEL_KPL));
761 	ltr(ci->ci_tss_sel);
762 
763 	cpu_init(ci);
764 	cpu_get_tsc_freq(ci);
765 
766 	s = splhigh();
767 #ifdef i386
768 	lapic_tpr = 0;
769 #else
770 	lcr8(0);
771 #endif
772 	x86_enable_intr();
773 	splx(s);
774 #if 0
775 	x86_errata();
776 #endif
777 
778 	aprint_debug_dev(ci->ci_dev, "CPU %ld running\n",
779 		(long)ci->ci_cpuid);
780 }
781 
782 #if defined(DDB)
783 
784 #include <ddb/db_output.h>
785 #include <machine/db_machdep.h>
786 
787 /*
788  * Dump CPU information from ddb.
789  */
790 void
791 cpu_debug_dump(void)
792 {
793 	struct cpu_info *ci;
794 	CPU_INFO_ITERATOR cii;
795 
796 	db_printf("addr		dev	id	flags	ipis	curlwp 		fpcurlwp\n");
797 	for (CPU_INFO_FOREACH(cii, ci)) {
798 		db_printf("%p	%s	%ld	%x	%x	%10p	%10p\n",
799 		    ci,
800 		    ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
801 		    (long)ci->ci_cpuid,
802 		    ci->ci_flags, ci->ci_ipis,
803 		    ci->ci_curlwp,
804 		    ci->ci_fpcurlwp);
805 	}
806 }
807 #endif /* DDB */
808 
809 static void
810 cpu_copy_trampoline(void)
811 {
812 	/*
813 	 * Copy boot code.
814 	 */
815 	extern u_char cpu_spinup_trampoline[];
816 	extern u_char cpu_spinup_trampoline_end[];
817 
818 	vaddr_t mp_trampoline_vaddr;
819 
820 	mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
821 		UVM_KMF_VAONLY);
822 
823 	pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
824 		VM_PROT_READ | VM_PROT_WRITE, 0);
825 	pmap_update(pmap_kernel());
826 	memcpy((void *)mp_trampoline_vaddr,
827 		cpu_spinup_trampoline,
828 		cpu_spinup_trampoline_end - cpu_spinup_trampoline);
829 
830 	pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
831 	pmap_update(pmap_kernel());
832 	uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
833 }
834 
835 #endif /* MULTIPROCESSOR */
836 
837 #ifdef i386
838 #if 0
839 static void
840 tss_init(struct i386tss *tss, void *stack, void *func)
841 {
842 	memset(tss, 0, sizeof *tss);
843 	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
844 	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
845 	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
846 	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
847 	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
848 	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
849 	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
850 	tss->tss_esp = (int)((char *)stack + USPACE - 16);
851 	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
852 	tss->__tss_eflags = PSL_MBO | PSL_NT;   /* XXX not needed? */
853 	tss->__tss_eip = (int)func;
854 }
855 #endif
856 
857 /* XXX */
858 #define IDTVEC(name)	__CONCAT(X, name)
859 typedef void (vector)(void);
860 extern vector IDTVEC(tss_trap08);
861 #ifdef DDB
862 extern vector Xintrddbipi;
863 extern int ddb_vec;
864 #endif
865 
866 static void
867 cpu_set_tss_gates(struct cpu_info *ci)
868 {
869 #if 0
870 	struct segment_descriptor sd;
871 
872 	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
873 	    UVM_KMF_WIRED);
874 	tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
875 	    IDTVEC(tss_trap08));
876 	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
877 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
878 	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
879 	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
880 	    GSEL(GTRAPTSS_SEL, SEL_KPL));
881 #endif
882 
883 #if defined(DDB) && defined(MULTIPROCESSOR)
884 	/*
885 	 * Set up separate handler for the DDB IPI, so that it doesn't
886 	 * stomp on a possibly corrupted stack.
887 	 *
888 	 * XXX overwriting the gate set in db_machine_init.
889 	 * Should rearrange the code so that it's set only once.
890 	 */
891 	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
892 	    UVM_KMF_WIRED);
893 	tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
894 	    Xintrddbipi);
895 
896 	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
897 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
898 	ci->ci_gdt[GIPITSS_SEL].sd = sd;
899 
900 	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
901 	    GSEL(GIPITSS_SEL, SEL_KPL));
902 #endif
903 }
904 #else
905 static void
906 cpu_set_tss_gates(struct cpu_info *ci)
907 {
908 
909 }
910 #endif	/* i386 */
911 
912 int
913 mp_cpu_start(struct cpu_info *ci, paddr_t target)
914 {
915 #if 0
916 #if NLAPIC > 0
917 	int error;
918 #endif
919 	unsigned short dwordptr[2];
920 
921 	/*
922 	 * Bootstrap code must be addressable in real mode
923 	 * and it must be page aligned.
924 	 */
925 	KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
926 
927 	/*
928 	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
929 	 */
930 
931 	outb(IO_RTC, NVRAM_RESET);
932 	outb(IO_RTC+1, NVRAM_RESET_JUMP);
933 
934 	/*
935 	 * "and the warm reset vector (DWORD based at 40:67) to point
936 	 * to the AP startup code ..."
937 	 */
938 
939 	dwordptr[0] = 0;
940 	dwordptr[1] = target >> 4;
941 
942 	pmap_kenter_pa (0, 0, VM_PROT_READ|VM_PROT_WRITE, 0);
943 	pmap_update(pmap_kernel());
944 
945 	memcpy ((uint8_t *) 0x467, dwordptr, 4);
946 
947 	pmap_kremove (0, PAGE_SIZE);
948 	pmap_update(pmap_kernel());
949 
950 #if NLAPIC > 0
951 	/*
952 	 * ... prior to executing the following sequence:"
953 	 */
954 
955 	if (ci->ci_flags & CPUF_AP) {
956 		if ((error = x86_ipi_init(ci->ci_cpuid)) != 0)
957 			return error;
958 
959 		delay(10000);
960 
961 		if (cpu_feature & CPUID_APIC) {
962 			error = x86_ipi_init(ci->ci_cpuid);
963 			if (error != 0) {
964 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
965 						__func__);
966 				return error;
967 			}
968 
969 			delay(10000);
970 
971 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
972 					LAPIC_DLMODE_STARTUP);
973 			if (error != 0) {
974 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
975 						__func__);
976 				return error;
977 			}
978 			delay(200);
979 
980 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
981 					LAPIC_DLMODE_STARTUP);
982 			if (error != 0) {
983 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken ((3)\n",
984 						__func__);
985 				return error;
986 			}
987 			delay(200);
988 		}
989 	}
990 #endif
991 #endif /* 0 */
992 	return 0;
993 }
994 
995 void
996 mp_cpu_start_cleanup(struct cpu_info *ci)
997 {
998 #if 0
999 	/*
1000 	 * Ensure the NVRAM reset byte contains something vaguely sane.
1001 	 */
1002 
1003 	outb(IO_RTC, NVRAM_RESET);
1004 	outb(IO_RTC+1, NVRAM_RESET_RST);
1005 #endif
1006 }
1007 
1008 void
1009 cpu_init_msrs(struct cpu_info *ci, bool full)
1010 {
1011 #ifdef __x86_64__
1012 	if (full) {
1013 		HYPERVISOR_set_segment_base (SEGBASE_FS, 0);
1014 		HYPERVISOR_set_segment_base (SEGBASE_GS_KERNEL, (uint64_t) ci);
1015 		HYPERVISOR_set_segment_base (SEGBASE_GS_USER, 0);
1016 	}
1017 #endif	/* __x86_64__ */
1018 
1019 	if (cpu_feature[2] & CPUID_NOX)
1020 		wrmsr(MSR_EFER, rdmsr(MSR_EFER) | EFER_NXE);
1021 }
1022 
1023 void
1024 cpu_offline_md(void)
1025 {
1026         int s;
1027 
1028         s = splhigh();
1029 #ifdef __i386__
1030         npxsave_cpu(true);
1031 #else
1032         fpusave_cpu(true);
1033 #endif
1034         splx(s);
1035 }
1036 
1037 #if 0
1038 /* XXX joerg restructure and restart CPUs individually */
1039 static bool
1040 cpu_suspend(device_t dv, const pmf_qual_t *qual)
1041 {
1042 	struct cpu_softc *sc = device_private(dv);
1043 	struct cpu_info *ci = sc->sc_info;
1044 	int err;
1045 
1046 	if (ci->ci_flags & CPUF_PRIMARY)
1047 		return true;
1048 	if (ci->ci_data.cpu_idlelwp == NULL)
1049 		return true;
1050 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1051 		return true;
1052 
1053 	sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
1054 
1055 	if (sc->sc_wasonline) {
1056 		mutex_enter(&cpu_lock);
1057 		err = cpu_setstate(ci, false);
1058 		mutex_exit(&cpu_lock);
1059 
1060 		if (err)
1061 			return false;
1062 	}
1063 
1064 	return true;
1065 }
1066 
1067 static bool
1068 cpu_resume(device_t dv, const pmf_qual_t *qual)
1069 {
1070 	struct cpu_softc *sc = device_private(dv);
1071 	struct cpu_info *ci = sc->sc_info;
1072 	int err = 0;
1073 
1074 	if (ci->ci_flags & CPUF_PRIMARY)
1075 		return true;
1076 	if (ci->ci_data.cpu_idlelwp == NULL)
1077 		return true;
1078 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1079 		return true;
1080 
1081 	if (sc->sc_wasonline) {
1082 		mutex_enter(&cpu_lock);
1083 		err = cpu_setstate(ci, true);
1084 		mutex_exit(&cpu_lock);
1085 	}
1086 
1087 	return err == 0;
1088 }
1089 #endif
1090 
1091 void
1092 cpu_get_tsc_freq(struct cpu_info *ci)
1093 {
1094 	const volatile vcpu_time_info_t *tinfo = &ci->ci_vcpu->time;
1095 	delay(1000000);
1096 	uint64_t freq = 1000000000ULL << 32;
1097 	freq = freq / (uint64_t)tinfo->tsc_to_system_mul;
1098 	if ( tinfo->tsc_shift < 0 )
1099 		freq = freq << -tinfo->tsc_shift;
1100 	else
1101 		freq = freq >> tinfo->tsc_shift;
1102 	ci->ci_data.cpu_cc_freq = freq;
1103 }
1104 
1105 void
1106 x86_cpu_idle_xen(void)
1107 {
1108 	struct cpu_info *ci = curcpu();
1109 
1110 	KASSERT(ci->ci_ilevel == IPL_NONE);
1111 
1112 	x86_disable_intr();
1113 	if (!__predict_false(ci->ci_want_resched)) {
1114 		idle_block();
1115 	} else {
1116 		x86_enable_intr();
1117 	}
1118 }
1119 
1120 /*
1121  * Loads pmap for the current CPU.
1122  */
1123 void
1124 cpu_load_pmap(struct pmap *pmap)
1125 {
1126 #ifdef i386
1127 #ifdef PAE
1128 	int i, s;
1129 	struct cpu_info *ci;
1130 
1131 	s = splvm(); /* just to be safe */
1132 	ci = curcpu();
1133 	paddr_t l3_pd = xpmap_ptom_masked(ci->ci_pae_l3_pdirpa);
1134 	/* don't update the kernel L3 slot */
1135 	for (i = 0 ; i < PDP_SIZE - 1; i++) {
1136 		xpq_queue_pte_update(l3_pd + i * sizeof(pd_entry_t),
1137 		    xpmap_ptom(pmap->pm_pdirpa[i]) | PG_V);
1138 	}
1139 	splx(s);
1140 	tlbflush();
1141 #else /* PAE */
1142 	lcr3(pmap_pdirpa(pmap, 0));
1143 #endif /* PAE */
1144 #endif /* i386 */
1145 
1146 #ifdef __x86_64__
1147 	int i, s;
1148 	pd_entry_t *old_pgd, *new_pgd;
1149 	paddr_t addr;
1150 	struct cpu_info *ci;
1151 
1152 	/* kernel pmap always in cr3 and should never go in user cr3 */
1153 	if (pmap_pdirpa(pmap, 0) != pmap_pdirpa(pmap_kernel(), 0)) {
1154 		ci = curcpu();
1155 		/*
1156 		 * Map user space address in kernel space and load
1157 		 * user cr3
1158 		 */
1159 		s = splvm();
1160 		new_pgd = pmap->pm_pdir;
1161 		old_pgd = pmap_kernel()->pm_pdir;
1162 		addr = xpmap_ptom(pmap_pdirpa(pmap_kernel(), 0));
1163 		for (i = 0; i < PDIR_SLOT_PTE;
1164 		    i++, addr += sizeof(pd_entry_t)) {
1165 			if ((new_pgd[i] & PG_V) || (old_pgd[i] & PG_V))
1166 				xpq_queue_pte_update(addr, new_pgd[i]);
1167 		}
1168 		tlbflush();
1169 		xen_set_user_pgd(pmap_pdirpa(pmap, 0));
1170 		ci->ci_xen_current_user_pgd = pmap_pdirpa(pmap, 0);
1171 		splx(s);
1172 	}
1173 #endif /* __x86_64__ */
1174 }
1175