xref: /netbsd-src/sys/arch/xen/x86/cpu.c (revision 1b9578b8c2c1f848eeb16dabbfd7d1f0d9fdefbd)
1 /*	$NetBSD: cpu.c,v 1.59 2011/06/15 20:50:02 rmind 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 1.59 2011/06/15 20:50:02 rmind Exp $");
70 
71 #include "opt_ddb.h"
72 #include "opt_multiprocessor.h"
73 #include "opt_mpbios.h"		/* for MPDEBUG */
74 #include "opt_mtrr.h"
75 #include "opt_xen.h"
76 
77 #include "lapic.h"
78 #include "ioapic.h"
79 
80 #include <sys/param.h>
81 #include <sys/proc.h>
82 #include <sys/systm.h>
83 #include <sys/device.h>
84 #include <sys/kmem.h>
85 #include <sys/cpu.h>
86 #include <sys/atomic.h>
87 #include <sys/reboot.h>
88 
89 #include <uvm/uvm.h>
90 
91 #include <machine/cpufunc.h>
92 #include <machine/cpuvar.h>
93 #include <machine/pmap.h>
94 #include <machine/vmparam.h>
95 #include <machine/mpbiosvar.h>
96 #include <machine/pcb.h>
97 #include <machine/specialreg.h>
98 #include <machine/segments.h>
99 #include <machine/gdt.h>
100 #include <machine/mtrr.h>
101 #include <machine/pio.h>
102 
103 #include <xen/vcpuvar.h>
104 
105 #if NLAPIC > 0
106 #include <machine/apicvar.h>
107 #include <machine/i82489reg.h>
108 #include <machine/i82489var.h>
109 #endif
110 
111 #include <dev/ic/mc146818reg.h>
112 #include <dev/isa/isareg.h>
113 
114 #if MAXCPUS > 32
115 #error cpu_info contains 32bit bitmasks
116 #endif
117 
118 static int	cpu_match(device_t, cfdata_t, void *);
119 static void	cpu_attach(device_t, device_t, void *);
120 static void	cpu_defer(device_t);
121 static int	cpu_rescan(device_t, const char *, const int *);
122 static void	cpu_childdetached(device_t, device_t);
123 static int	vcpu_match(device_t, cfdata_t, void *);
124 static void	vcpu_attach(device_t, device_t, void *);
125 static void	cpu_attach_common(device_t, device_t, void *);
126 void		cpu_offline_md(void);
127 
128 struct cpu_softc {
129 	device_t sc_dev;		/* device tree glue */
130 	struct cpu_info *sc_info;	/* pointer to CPU info */
131 	bool sc_wasonline;
132 };
133 
134 int mp_cpu_start(struct cpu_info *, paddr_t);
135 void mp_cpu_start_cleanup(struct cpu_info *);
136 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL,
137 				      mp_cpu_start_cleanup };
138 
139 CFATTACH_DECL2_NEW(cpu, sizeof(struct cpu_softc),
140     cpu_match, cpu_attach, NULL, NULL, cpu_rescan, cpu_childdetached);
141 
142 CFATTACH_DECL_NEW(vcpu, sizeof(struct cpu_softc),
143     vcpu_match, vcpu_attach, NULL, NULL);
144 
145 /*
146  * Statically-allocated CPU info for the primary CPU (or the only
147  * CPU, on uniprocessors).  The CPU info list is initialized to
148  * point at it.
149  */
150 #ifdef TRAPLOG
151 #include <machine/tlog.h>
152 struct tlog tlog_primary;
153 #endif
154 struct cpu_info cpu_info_primary __aligned(CACHE_LINE_SIZE) = {
155 	.ci_dev = 0,
156 	.ci_self = &cpu_info_primary,
157 	.ci_idepth = -1,
158 	.ci_curlwp = &lwp0,
159 	.ci_curldt = -1,
160 	.ci_cpumask = 1,
161 #ifdef TRAPLOG
162 	.ci_tlog = &tlog_primary,
163 #endif
164 
165 };
166 struct cpu_info phycpu_info_primary __aligned(CACHE_LINE_SIZE) = {
167 	.ci_dev = 0,
168 	.ci_self = &phycpu_info_primary,
169 };
170 
171 struct cpu_info *cpu_info_list = &cpu_info_primary;
172 struct cpu_info *phycpu_info_list = &phycpu_info_primary;
173 
174 static void	cpu_set_tss_gates(struct cpu_info *ci);
175 
176 uint32_t cpus_attached = 1;
177 uint32_t cpus_running = 0;
178 
179 uint32_t phycpus_attached = 0;
180 uint32_t phycpus_running = 0;
181 
182 uint32_t cpu_feature[5]; /* X86 CPUID feature bits
183 			  *	[0] basic features %edx
184 			  *	[1] basic features %ecx
185 			  *	[2] extended features %edx
186 			  *	[3] extended features %ecx
187 			  *	[4] VIA padlock features
188 			  */
189 
190 bool x86_mp_online;
191 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE;
192 
193 #if defined(MULTIPROCESSOR)
194 void    	cpu_hatch(void *);
195 static void    	cpu_boot_secondary(struct cpu_info *ci);
196 static void    	cpu_start_secondary(struct cpu_info *ci);
197 static void	cpu_copy_trampoline(void);
198 
199 /*
200  * Runs once per boot once multiprocessor goo has been detected and
201  * the local APIC on the boot processor has been mapped.
202  *
203  * Called from lapic_boot_init() (from mpbios_scan()).
204  */
205 void
206 cpu_init_first(void)
207 {
208 
209 	cpu_info_primary.ci_cpuid = lapic_cpu_number();
210 	cpu_copy_trampoline();
211 }
212 #endif	/* MULTIPROCESSOR */
213 
214 static int
215 cpu_match(device_t parent, cfdata_t match, void *aux)
216 {
217 
218 	return 1;
219 }
220 
221 static void
222 cpu_attach(device_t parent, device_t self, void *aux)
223 {
224 	struct cpu_softc *sc = device_private(self);
225 	struct cpu_attach_args *caa = aux;
226 	struct cpu_info *ci;
227 	uintptr_t ptr;
228 	static int nphycpu = 0;
229 
230 	sc->sc_dev = self;
231 
232 	if (phycpus_attached == ~0) {
233 		aprint_error(": increase MAXCPUS\n");
234 		return;
235 	}
236 
237 	/*
238 	 * If we're an Application Processor, allocate a cpu_info
239 	 * If we're the first attached CPU use the primary cpu_info,
240 	 * otherwise allocate a new one
241 	 */
242 	aprint_naive("\n");
243 	aprint_normal("\n");
244 	if (nphycpu > 0) {
245 		struct cpu_info *tmp;
246 		ptr = (uintptr_t)kmem_zalloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
247 		    KM_SLEEP);
248 		ci = (struct cpu_info *)roundup2(ptr, CACHE_LINE_SIZE);
249 		ci->ci_curldt = -1;
250 
251 		tmp = phycpu_info_list;
252 		while (tmp->ci_next)
253 			tmp = tmp->ci_next;
254 
255 		tmp->ci_next = ci;
256 	} else {
257 		ci = &phycpu_info_primary;
258 	}
259 
260 	ci->ci_self = ci;
261 	sc->sc_info = ci;
262 
263 	ci->ci_dev = self;
264 	ci->ci_acpiid = caa->cpu_id;
265 	ci->ci_cpuid = caa->cpu_number;
266 	ci->ci_vcpu = NULL;
267 	ci->ci_index = nphycpu++;
268 	ci->ci_cpumask = (1 << cpu_index(ci));
269 
270 	atomic_or_32(&phycpus_attached, ci->ci_cpumask);
271 
272 	if (!pmf_device_register(self, NULL, NULL))
273 		aprint_error_dev(self, "couldn't establish power handler\n");
274 
275 	(void)config_defer(self, cpu_defer);
276 }
277 
278 static void
279 cpu_defer(device_t self)
280 {
281 	cpu_rescan(self, NULL, NULL);
282 }
283 
284 static int
285 cpu_rescan(device_t self, const char *ifattr, const int *locators)
286 {
287 	struct cpu_softc *sc = device_private(self);
288 	struct cpufeature_attach_args cfaa;
289 	struct cpu_info *ci = sc->sc_info;
290 
291 	memset(&cfaa, 0, sizeof(cfaa));
292 	cfaa.ci = ci;
293 
294 	if (ifattr_match(ifattr, "cpufeaturebus")) {
295 
296 		if (ci->ci_frequency == NULL) {
297 			cfaa.name = "frequency";
298 			ci->ci_frequency = config_found_ia(self,
299 			    "cpufeaturebus", &cfaa, NULL);
300 		}
301 	}
302 
303 	return 0;
304 }
305 
306 static void
307 cpu_childdetached(device_t self, device_t child)
308 {
309 	struct cpu_softc *sc = device_private(self);
310 	struct cpu_info *ci = sc->sc_info;
311 
312 	if (ci->ci_frequency == child)
313 		ci->ci_frequency = NULL;
314 }
315 
316 static int
317 vcpu_match(device_t parent, cfdata_t match, void *aux)
318 {
319 	struct vcpu_attach_args *vcaa = aux;
320 
321 	if (strcmp(vcaa->vcaa_name, match->cf_name) == 0)
322 		return 1;
323 	return 0;
324 }
325 
326 static void
327 vcpu_attach(device_t parent, device_t self, void *aux)
328 {
329 	struct vcpu_attach_args *vcaa = aux;
330 
331 	cpu_attach_common(parent, self, &vcaa->vcaa_caa);
332 }
333 
334 static void
335 cpu_vm_init(struct cpu_info *ci)
336 {
337 	int ncolors = 2, i;
338 
339 	for (i = CAI_ICACHE; i <= CAI_L2CACHE; i++) {
340 		struct x86_cache_info *cai;
341 		int tcolors;
342 
343 		cai = &ci->ci_cinfo[i];
344 
345 		tcolors = atop(cai->cai_totalsize);
346 		switch(cai->cai_associativity) {
347 		case 0xff:
348 			tcolors = 1; /* fully associative */
349 			break;
350 		case 0:
351 		case 1:
352 			break;
353 		default:
354 			tcolors /= cai->cai_associativity;
355 		}
356 		ncolors = max(ncolors, tcolors);
357 	}
358 
359 	/*
360 	 * Knowing the size of the largest cache on this CPU, re-color
361 	 * our pages.
362 	 */
363 	if (ncolors <= uvmexp.ncolors)
364 		return;
365 	aprint_debug_dev(ci->ci_dev, "%d page colors\n", ncolors);
366 	uvm_page_recolor(ncolors);
367 }
368 
369 static void
370 cpu_attach_common(device_t parent, device_t self, void *aux)
371 {
372 	struct cpu_softc *sc = device_private(self);
373 	struct cpu_attach_args *caa = aux;
374 	struct cpu_info *ci;
375 	uintptr_t ptr;
376 	int cpunum = caa->cpu_number;
377 	static bool again = false;
378 
379 	sc->sc_dev = self;
380 
381 	/*
382 	 * If we're an Application Processor, allocate a cpu_info
383 	 * structure, otherwise use the primary's.
384 	 */
385 	if (caa->cpu_role == CPU_ROLE_AP) {
386 		aprint_naive(": Application Processor\n");
387 		ptr = (uintptr_t)kmem_alloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
388 		    KM_SLEEP);
389 		ci = (struct cpu_info *)roundup2(ptr, CACHE_LINE_SIZE);
390 		memset(ci, 0, sizeof(*ci));
391 #ifdef TRAPLOG
392 		ci->ci_tlog_base = kmem_zalloc(sizeof(struct tlog), KM_SLEEP);
393 #endif
394 	} else {
395 		aprint_naive(": %s Processor\n",
396 		    caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
397 		ci = &cpu_info_primary;
398 #if NLAPIC > 0
399 		if (cpunum != lapic_cpu_number()) {
400 			/* XXX should be done earlier */
401 			uint32_t reg;
402 			aprint_verbose("\n");
403 			aprint_verbose_dev(self, "running CPU at apic %d"
404 			    " instead of at expected %d", lapic_cpu_number(),
405 			    cpunum);
406 			reg = i82489_readreg(LAPIC_ID);
407 			i82489_writereg(LAPIC_ID, (reg & ~LAPIC_ID_MASK) |
408 			    (cpunum << LAPIC_ID_SHIFT));
409 		}
410 		if (cpunum != lapic_cpu_number()) {
411 			aprint_error_dev(self, "unable to reset apic id\n");
412 		}
413 #endif
414 	}
415 
416 	ci->ci_self = ci;
417 	sc->sc_info = ci;
418 	ci->ci_dev = self;
419 	ci->ci_cpuid = cpunum;
420 
421 	KASSERT(HYPERVISOR_shared_info != NULL);
422 	ci->ci_vcpu = &HYPERVISOR_shared_info->vcpu_info[cpunum];
423 
424 	ci->ci_func = caa->cpu_func;
425 
426 	/* Must be called before mi_cpu_attach(). */
427 	cpu_vm_init(ci);
428 
429 	if (caa->cpu_role == CPU_ROLE_AP) {
430 		int error;
431 
432 		error = mi_cpu_attach(ci);
433 		if (error != 0) {
434 			aprint_normal("\n");
435 			aprint_error_dev(self,
436 			    "mi_cpu_attach failed with %d\n", error);
437 			return;
438 		}
439 	} else {
440 		KASSERT(ci->ci_data.cpu_idlelwp != NULL);
441 	}
442 
443 	ci->ci_cpumask = (1 << cpu_index(ci));
444 	pmap_reference(pmap_kernel());
445 	ci->ci_pmap = pmap_kernel();
446 	ci->ci_tlbstate = TLBSTATE_STALE;
447 
448 	/*
449 	 * Boot processor may not be attached first, but the below
450 	 * must be done to allow booting other processors.
451 	 */
452 	if (!again) {
453 		atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_PRIMARY);
454 		/* Basic init. */
455 		cpu_intr_init(ci);
456 		cpu_get_tsc_freq(ci);
457 		cpu_init(ci);
458 		cpu_set_tss_gates(ci);
459 #if NLAPIC > 0
460 		if (caa->cpu_role != CPU_ROLE_SP) {
461 			/* Enable lapic. */
462 			lapic_enable();
463 			lapic_set_lvt();
464 			lapic_calibrate_timer();
465 		}
466 #endif
467 		/* Make sure DELAY() is initialized. */
468 		DELAY(1);
469 		again = true;
470 	}
471 
472 	/* further PCB init done later. */
473 
474 	switch (caa->cpu_role) {
475 	case CPU_ROLE_SP:
476 		atomic_or_32(&ci->ci_flags, CPUF_SP);
477 		cpu_identify(ci);
478 #if 0
479 		x86_errata();
480 #endif
481 		x86_cpu_idle_init();
482 		break;
483 
484 	case CPU_ROLE_BP:
485 		atomic_or_32(&ci->ci_flags, CPUF_BSP);
486 		cpu_identify(ci);
487 		cpu_init(ci);
488 #if 0
489 		x86_errata();
490 #endif
491 		x86_cpu_idle_init();
492 		break;
493 
494 	case CPU_ROLE_AP:
495 		/*
496 		 * report on an AP
497 		 */
498 
499 #if defined(MULTIPROCESSOR)
500 		cpu_intr_init(ci);
501 		gdt_alloc_cpu(ci);
502 		cpu_set_tss_gates(ci);
503 		cpu_start_secondary(ci);
504 		if (ci->ci_flags & CPUF_PRESENT) {
505 			struct cpu_info *tmp;
506 
507 			identifycpu(ci);
508 			tmp = cpu_info_list;
509 			while (tmp->ci_next)
510 				tmp = tmp->ci_next;
511 
512 			tmp->ci_next = ci;
513 		}
514 #else
515 		aprint_error_dev(self, "not started\n");
516 #endif
517 		break;
518 
519 	default:
520 		aprint_normal("\n");
521 		panic("unknown processor type??\n");
522 	}
523 
524 	pat_init(ci);
525 	atomic_or_32(&cpus_attached, ci->ci_cpumask);
526 
527 #if 0
528 	if (!pmf_device_register(self, cpu_suspend, cpu_resume))
529 		aprint_error_dev(self, "couldn't establish power handler\n");
530 #endif
531 
532 #if defined(MULTIPROCESSOR)
533 	if (mp_verbose) {
534 		struct lwp *l = ci->ci_data.cpu_idlelwp;
535 		struct pcb *pcb = lwp_getpcb(l);
536 
537 		aprint_verbose_dev(self,
538 		    "idle lwp at %p, idle sp at 0x%p\n",
539 		    l,
540 #ifdef i386
541 		    (void *)pcb->pcb_esp
542 #else
543 		    (void *)pcb->pcb_rsp
544 #endif
545 		);
546 
547 	}
548 #endif
549 }
550 
551 /*
552  * Initialize the processor appropriately.
553  */
554 
555 void
556 cpu_init(struct cpu_info *ci)
557 {
558 
559 	/*
560 	 * On a P6 or above, enable global TLB caching if the
561 	 * hardware supports it.
562 	 */
563 	if (cpu_feature[0] & CPUID_PGE)
564 		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
565 
566 #ifdef XXXMTRR
567 	/*
568 	 * On a P6 or above, initialize MTRR's if the hardware supports them.
569 	 */
570 	if (cpu_feature[0] & CPUID_MTRR) {
571 		if ((ci->ci_flags & CPUF_AP) == 0)
572 			i686_mtrr_init_first();
573 		mtrr_init_cpu(ci);
574 	}
575 #endif
576 	/*
577 	 * If we have FXSAVE/FXRESTOR, use them.
578 	 */
579 	if (cpu_feature[0] & CPUID_FXSR) {
580 		lcr4(rcr4() | CR4_OSFXSR);
581 
582 		/*
583 		 * If we have SSE/SSE2, enable XMM exceptions.
584 		 */
585 		if (cpu_feature[0] & (CPUID_SSE|CPUID_SSE2))
586 			lcr4(rcr4() | CR4_OSXMMEXCPT);
587 	}
588 
589 #ifdef __x86_64__
590 	/* No user PGD mapped for this CPU yet */
591 	ci->ci_xen_current_user_pgd = 0;
592 #endif
593 
594 	atomic_or_32(&cpus_running, ci->ci_cpumask);
595 	atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
596 }
597 
598 
599 #ifdef MULTIPROCESSOR
600 void
601 cpu_boot_secondary_processors(void)
602 {
603 	struct cpu_info *ci;
604 	u_long i;
605 
606 	for (i = 0; i < maxcpus; i++) {
607 		ci = cpu_lookup(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 		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
615 			continue;
616 		cpu_boot_secondary(ci);
617 	}
618 
619 	x86_mp_online = true;
620 }
621 
622 static void
623 cpu_init_idle_lwp(struct cpu_info *ci)
624 {
625 	struct lwp *l = ci->ci_data.cpu_idlelwp;
626 	struct pcb *pcb = lwp_getpcb(l);
627 
628 	pcb->pcb_cr0 = rcr0();
629 }
630 
631 void
632 cpu_init_idle_lwps(void)
633 {
634 	struct cpu_info *ci;
635 	u_long i;
636 
637 	for (i = 0; i < maxcpus; i++) {
638 		ci = cpu_lookup(i);
639 		if (ci == NULL)
640 			continue;
641 		if (ci->ci_data.cpu_idlelwp == NULL)
642 			continue;
643 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
644 			continue;
645 		cpu_init_idle_lwp(ci);
646 	}
647 }
648 
649 void
650 cpu_start_secondary(struct cpu_info *ci)
651 {
652 	int i;
653 	struct pmap *kpm = pmap_kernel();
654 	extern uint32_t mp_pdirpa;
655 
656 	mp_pdirpa = kpm->pm_pdirpa; /* XXX move elsewhere, not per CPU. */
657 
658 	atomic_or_32(&ci->ci_flags, CPUF_AP);
659 
660 	aprint_debug_dev(ci->ci_dev, "starting\n");
661 
662 	ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
663 	if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0)
664 		return;
665 
666 	/*
667 	 * wait for it to become ready
668 	 */
669 	for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
670 #ifdef MPDEBUG
671 		extern int cpu_trace[3];
672 		static int otrace[3];
673 		if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) {
674 			aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n",
675 				cpu_trace[0], cpu_trace[1], cpu_trace[2]);
676 			memcpy(otrace, cpu_trace, sizeof(otrace));
677 		}
678 #endif
679 		delay(10);
680 	}
681 	if ((ci->ci_flags & CPUF_PRESENT) == 0) {
682 		aprint_error_dev(ci->ci_dev, "failed to become ready\n");
683 #if defined(MPDEBUG) && defined(DDB)
684 		printf("dropping into debugger; continue from here to resume boot\n");
685 		Debugger();
686 #endif
687 	}
688 
689 	CPU_START_CLEANUP(ci);
690 }
691 
692 void
693 cpu_boot_secondary(struct cpu_info *ci)
694 {
695 	int i;
696 
697 	atomic_or_32(&ci->ci_flags, CPUF_GO);
698 	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
699 		delay(10);
700 	}
701 	if ((ci->ci_flags & CPUF_RUNNING) == 0) {
702 		aprint_error_dev(ci->ci_dev, "CPU failed to start\n");
703 #if defined(MPDEBUG) && defined(DDB)
704 		printf("dropping into debugger; continue from here to resume boot\n");
705 		Debugger();
706 #endif
707 	}
708 }
709 
710 /*
711  * The CPU ends up here when its ready to run
712  * This is called from code in mptramp.s; at this point, we are running
713  * in the idle pcb/idle stack of the new CPU.  When this function returns,
714  * this processor will enter the idle loop and start looking for work.
715  *
716  * XXX should share some of this with init386 in machdep.c
717  */
718 void
719 cpu_hatch(void *v)
720 {
721 	struct cpu_info *ci = (struct cpu_info *)v;
722 	struct pcb *pcb;
723 	int s, i;
724 
725 	cpu_probe(ci);
726 
727 	cpu_feature[0] &= ~CPUID_FEAT_BLACKLIST;
728 	cpu_feature[2] &= ~CPUID_FEAT_EXT_BLACKLIST;
729 
730         cpu_init_msrs(ci, true);
731 
732 	KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
733 	atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
734 	while ((ci->ci_flags & CPUF_GO) == 0) {
735 		/* Don't use delay, boot CPU may be patching the text. */
736 		for (i = 10000; i != 0; i--)
737 			x86_pause();
738 	}
739 
740 	/* Because the text may have been patched in x86_patch(). */
741 	wbinvd();
742 	x86_flush();
743 	tlbflushg();
744 
745 	KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
746 
747 	pcb = lwp_getpcb(curlwp);
748 	lcr3(pmap_kernel()->pm_pdirpa);
749 	pcb->pcb_cr3 = pmap_kernel()->pm_pdirpa;
750 	pcb = lwp_getpcb(ci->ci_data.cpu_idlelwp);
751 	lcr0(pcb->pcb_cr0);
752 
753 	cpu_init_idt();
754 	gdt_init_cpu(ci);
755 	lapic_enable();
756 	lapic_set_lvt();
757 	lapic_initclocks();
758 
759 #ifdef i386
760 	npxinit(ci);
761 #else
762 	fpuinit(ci);
763 #endif
764 
765 	lldt(GSEL(GLDT_SEL, SEL_KPL));
766 	ltr(ci->ci_tss_sel);
767 
768 	cpu_init(ci);
769 	cpu_get_tsc_freq(ci);
770 
771 	s = splhigh();
772 #ifdef i386
773 	lapic_tpr = 0;
774 #else
775 	lcr8(0);
776 #endif
777 	x86_enable_intr();
778 	splx(s);
779 #if 0
780 	x86_errata();
781 #endif
782 
783 	aprint_debug_dev(ci->ci_dev, "CPU %ld running\n",
784 		(long)ci->ci_cpuid);
785 }
786 
787 #if defined(DDB)
788 
789 #include <ddb/db_output.h>
790 #include <machine/db_machdep.h>
791 
792 /*
793  * Dump CPU information from ddb.
794  */
795 void
796 cpu_debug_dump(void)
797 {
798 	struct cpu_info *ci;
799 	CPU_INFO_ITERATOR cii;
800 
801 	db_printf("addr		dev	id	flags	ipis	curlwp 		fpcurlwp\n");
802 	for (CPU_INFO_FOREACH(cii, ci)) {
803 		db_printf("%p	%s	%ld	%x	%x	%10p	%10p\n",
804 		    ci,
805 		    ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
806 		    (long)ci->ci_cpuid,
807 		    ci->ci_flags, ci->ci_ipis,
808 		    ci->ci_curlwp,
809 		    ci->ci_fpcurlwp);
810 	}
811 }
812 #endif /* DDB */
813 
814 static void
815 cpu_copy_trampoline(void)
816 {
817 	/*
818 	 * Copy boot code.
819 	 */
820 	extern u_char cpu_spinup_trampoline[];
821 	extern u_char cpu_spinup_trampoline_end[];
822 
823 	vaddr_t mp_trampoline_vaddr;
824 
825 	mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
826 		UVM_KMF_VAONLY);
827 
828 	pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
829 		VM_PROT_READ | VM_PROT_WRITE, 0);
830 	pmap_update(pmap_kernel());
831 	memcpy((void *)mp_trampoline_vaddr,
832 		cpu_spinup_trampoline,
833 		cpu_spinup_trampoline_end - cpu_spinup_trampoline);
834 
835 	pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
836 	pmap_update(pmap_kernel());
837 	uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
838 }
839 
840 #endif /* MULTIPROCESSOR */
841 
842 #ifdef i386
843 #if 0
844 static void
845 tss_init(struct i386tss *tss, void *stack, void *func)
846 {
847 	memset(tss, 0, sizeof *tss);
848 	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
849 	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
850 	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
851 	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
852 	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
853 	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
854 	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
855 	tss->tss_esp = (int)((char *)stack + USPACE - 16);
856 	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
857 	tss->__tss_eflags = PSL_MBO | PSL_NT;   /* XXX not needed? */
858 	tss->__tss_eip = (int)func;
859 }
860 #endif
861 
862 /* XXX */
863 #define IDTVEC(name)	__CONCAT(X, name)
864 typedef void (vector)(void);
865 extern vector IDTVEC(tss_trap08);
866 #ifdef DDB
867 extern vector Xintrddbipi;
868 extern int ddb_vec;
869 #endif
870 
871 static void
872 cpu_set_tss_gates(struct cpu_info *ci)
873 {
874 #if 0
875 	struct segment_descriptor sd;
876 
877 	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
878 	    UVM_KMF_WIRED);
879 	tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
880 	    IDTVEC(tss_trap08));
881 	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
882 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
883 	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
884 	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
885 	    GSEL(GTRAPTSS_SEL, SEL_KPL));
886 #endif
887 
888 #if defined(DDB) && defined(MULTIPROCESSOR)
889 	/*
890 	 * Set up separate handler for the DDB IPI, so that it doesn't
891 	 * stomp on a possibly corrupted stack.
892 	 *
893 	 * XXX overwriting the gate set in db_machine_init.
894 	 * Should rearrange the code so that it's set only once.
895 	 */
896 	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
897 	    UVM_KMF_WIRED);
898 	tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
899 	    Xintrddbipi);
900 
901 	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
902 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
903 	ci->ci_gdt[GIPITSS_SEL].sd = sd;
904 
905 	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
906 	    GSEL(GIPITSS_SEL, SEL_KPL));
907 #endif
908 }
909 #else
910 static void
911 cpu_set_tss_gates(struct cpu_info *ci)
912 {
913 
914 }
915 #endif	/* i386 */
916 
917 int
918 mp_cpu_start(struct cpu_info *ci, paddr_t target)
919 {
920 #if 0
921 #if NLAPIC > 0
922 	int error;
923 #endif
924 	unsigned short dwordptr[2];
925 
926 	/*
927 	 * Bootstrap code must be addressable in real mode
928 	 * and it must be page aligned.
929 	 */
930 	KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
931 
932 	/*
933 	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
934 	 */
935 
936 	outb(IO_RTC, NVRAM_RESET);
937 	outb(IO_RTC+1, NVRAM_RESET_JUMP);
938 
939 	/*
940 	 * "and the warm reset vector (DWORD based at 40:67) to point
941 	 * to the AP startup code ..."
942 	 */
943 
944 	dwordptr[0] = 0;
945 	dwordptr[1] = target >> 4;
946 
947 	pmap_kenter_pa (0, 0, VM_PROT_READ|VM_PROT_WRITE, 0);
948 	pmap_update(pmap_kernel());
949 
950 	memcpy ((uint8_t *) 0x467, dwordptr, 4);
951 
952 	pmap_kremove (0, PAGE_SIZE);
953 	pmap_update(pmap_kernel());
954 
955 #if NLAPIC > 0
956 	/*
957 	 * ... prior to executing the following sequence:"
958 	 */
959 
960 	if (ci->ci_flags & CPUF_AP) {
961 		if ((error = x86_ipi_init(ci->ci_cpuid)) != 0)
962 			return error;
963 
964 		delay(10000);
965 
966 		if (cpu_feature & CPUID_APIC) {
967 			error = x86_ipi_init(ci->ci_cpuid);
968 			if (error != 0) {
969 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
970 						__func__);
971 				return error;
972 			}
973 
974 			delay(10000);
975 
976 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
977 					LAPIC_DLMODE_STARTUP);
978 			if (error != 0) {
979 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
980 						__func__);
981 				return error;
982 			}
983 			delay(200);
984 
985 			error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid,
986 					LAPIC_DLMODE_STARTUP);
987 			if (error != 0) {
988 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken ((3)\n",
989 						__func__);
990 				return error;
991 			}
992 			delay(200);
993 		}
994 	}
995 #endif
996 #endif /* 0 */
997 	return 0;
998 }
999 
1000 void
1001 mp_cpu_start_cleanup(struct cpu_info *ci)
1002 {
1003 #if 0
1004 	/*
1005 	 * Ensure the NVRAM reset byte contains something vaguely sane.
1006 	 */
1007 
1008 	outb(IO_RTC, NVRAM_RESET);
1009 	outb(IO_RTC+1, NVRAM_RESET_RST);
1010 #endif
1011 }
1012 
1013 void
1014 cpu_init_msrs(struct cpu_info *ci, bool full)
1015 {
1016 #ifdef __x86_64__
1017 	if (full) {
1018 		HYPERVISOR_set_segment_base (SEGBASE_FS, 0);
1019 		HYPERVISOR_set_segment_base (SEGBASE_GS_KERNEL, (uint64_t) ci);
1020 		HYPERVISOR_set_segment_base (SEGBASE_GS_USER, 0);
1021 	}
1022 #endif	/* __x86_64__ */
1023 
1024 	if (cpu_feature[2] & CPUID_NOX)
1025 		wrmsr(MSR_EFER, rdmsr(MSR_EFER) | EFER_NXE);
1026 }
1027 
1028 void
1029 cpu_offline_md(void)
1030 {
1031         int s;
1032 
1033         s = splhigh();
1034 #ifdef __i386__
1035         npxsave_cpu(true);
1036 #else
1037         fpusave_cpu(true);
1038 #endif
1039         splx(s);
1040 }
1041 
1042 #if 0
1043 /* XXX joerg restructure and restart CPUs individually */
1044 static bool
1045 cpu_suspend(device_t dv, const pmf_qual_t *qual)
1046 {
1047 	struct cpu_softc *sc = device_private(dv);
1048 	struct cpu_info *ci = sc->sc_info;
1049 	int err;
1050 
1051 	if (ci->ci_flags & CPUF_PRIMARY)
1052 		return true;
1053 	if (ci->ci_data.cpu_idlelwp == NULL)
1054 		return true;
1055 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1056 		return true;
1057 
1058 	sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
1059 
1060 	if (sc->sc_wasonline) {
1061 		mutex_enter(&cpu_lock);
1062 		err = cpu_setstate(ci, false);
1063 		mutex_exit(&cpu_lock);
1064 
1065 		if (err)
1066 			return false;
1067 	}
1068 
1069 	return true;
1070 }
1071 
1072 static bool
1073 cpu_resume(device_t dv, const pmf_qual_t *qual)
1074 {
1075 	struct cpu_softc *sc = device_private(dv);
1076 	struct cpu_info *ci = sc->sc_info;
1077 	int err = 0;
1078 
1079 	if (ci->ci_flags & CPUF_PRIMARY)
1080 		return true;
1081 	if (ci->ci_data.cpu_idlelwp == NULL)
1082 		return true;
1083 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
1084 		return true;
1085 
1086 	if (sc->sc_wasonline) {
1087 		mutex_enter(&cpu_lock);
1088 		err = cpu_setstate(ci, true);
1089 		mutex_exit(&cpu_lock);
1090 	}
1091 
1092 	return err == 0;
1093 }
1094 #endif
1095 
1096 void
1097 cpu_get_tsc_freq(struct cpu_info *ci)
1098 {
1099 	const volatile vcpu_time_info_t *tinfo = &ci->ci_vcpu->time;
1100 	delay(1000000);
1101 	uint64_t freq = 1000000000ULL << 32;
1102 	freq = freq / (uint64_t)tinfo->tsc_to_system_mul;
1103 	if ( tinfo->tsc_shift < 0 )
1104 		freq = freq << -tinfo->tsc_shift;
1105 	else
1106 		freq = freq >> tinfo->tsc_shift;
1107 	ci->ci_data.cpu_cc_freq = freq;
1108 }
1109 
1110 void
1111 x86_cpu_idle_xen(void)
1112 {
1113 	struct cpu_info *ci = curcpu();
1114 
1115 	KASSERT(ci->ci_ilevel == IPL_NONE);
1116 
1117 	x86_disable_intr();
1118 	if (!__predict_false(ci->ci_want_resched)) {
1119 		idle_block();
1120 	} else {
1121 		x86_enable_intr();
1122 	}
1123 }
1124 
1125 /*
1126  * Loads pmap for the current CPU.
1127  */
1128 void
1129 cpu_load_pmap(struct pmap *pmap)
1130 {
1131 #ifdef i386
1132 #ifdef PAE
1133 	int i, s;
1134 	struct cpu_info *ci;
1135 
1136 	s = splvm(); /* just to be safe */
1137 	ci = curcpu();
1138 	paddr_t l3_pd = xpmap_ptom_masked(ci->ci_pae_l3_pdirpa);
1139 	/* don't update the kernel L3 slot */
1140 	for (i = 0 ; i < PDP_SIZE - 1; i++) {
1141 		xpq_queue_pte_update(l3_pd + i * sizeof(pd_entry_t),
1142 		    xpmap_ptom(pmap->pm_pdirpa[i]) | PG_V);
1143 	}
1144 	splx(s);
1145 	tlbflush();
1146 #else /* PAE */
1147 	lcr3(pmap_pdirpa(pmap, 0));
1148 #endif /* PAE */
1149 #endif /* i386 */
1150 
1151 #ifdef __x86_64__
1152 	int i, s;
1153 	pd_entry_t *old_pgd, *new_pgd;
1154 	paddr_t addr;
1155 	struct cpu_info *ci;
1156 
1157 	/* kernel pmap always in cr3 and should never go in user cr3 */
1158 	if (pmap_pdirpa(pmap, 0) != pmap_pdirpa(pmap_kernel(), 0)) {
1159 		ci = curcpu();
1160 		/*
1161 		 * Map user space address in kernel space and load
1162 		 * user cr3
1163 		 */
1164 		s = splvm();
1165 		new_pgd = pmap->pm_pdir;
1166 		old_pgd = pmap_kernel()->pm_pdir;
1167 		addr = xpmap_ptom(pmap_pdirpa(pmap_kernel(), 0));
1168 		for (i = 0; i < PDIR_SLOT_PTE;
1169 		    i++, addr += sizeof(pd_entry_t)) {
1170 			if ((new_pgd[i] & PG_V) || (old_pgd[i] & PG_V))
1171 				xpq_queue_pte_update(addr, new_pgd[i]);
1172 		}
1173 		xen_set_user_pgd(pmap_pdirpa(pmap, 0));
1174 		ci->ci_xen_current_user_pgd = pmap_pdirpa(pmap, 0);
1175 		tlbflush();
1176 		splx(s);
1177 	}
1178 #endif /* __x86_64__ */
1179 }
1180