Home | History | Annotate | Line # | Download | only in x86
cpu.c revision 1.11
      1 /*	$NetBSD: cpu.c,v 1.11 2008/04/13 22:29:38 cegger Exp $	*/
      2 /* NetBSD: cpu.c,v 1.18 2004/02/20 17:35:01 yamt Exp  */
      3 
      4 /*-
      5  * Copyright (c) 2000 The NetBSD Foundation, Inc.
      6  * All rights reserved.
      7  *
      8  * This code is derived from software contributed to The NetBSD Foundation
      9  * by RedBack Networks Inc.
     10  *
     11  * Author: Bill Sommerfeld
     12  *
     13  * Redistribution and use in source and binary forms, with or without
     14  * modification, are permitted provided that the following conditions
     15  * are met:
     16  * 1. Redistributions of source code must retain the above copyright
     17  *    notice, this list of conditions and the following disclaimer.
     18  * 2. Redistributions in binary form must reproduce the above copyright
     19  *    notice, this list of conditions and the following disclaimer in the
     20  *    documentation and/or other materials provided with the distribution.
     21  * 3. All advertising materials mentioning features or use of this software
     22  *    must display the following acknowledgement:
     23  *        This product includes software developed by the NetBSD
     24  *        Foundation, Inc. and its contributors.
     25  * 4. Neither the name of The NetBSD Foundation nor the names of its
     26  *    contributors may be used to endorse or promote products derived
     27  *    from this software without specific prior written permission.
     28  *
     29  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     30  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     31  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     32  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     33  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     34  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     35  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     36  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     37  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     38  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     39  * POSSIBILITY OF SUCH DAMAGE.
     40  */
     41 
     42 /*
     43  * Copyright (c) 1999 Stefan Grefen
     44  *
     45  * Redistribution and use in source and binary forms, with or without
     46  * modification, are permitted provided that the following conditions
     47  * are met:
     48  * 1. Redistributions of source code must retain the above copyright
     49  *    notice, this list of conditions and the following disclaimer.
     50  * 2. Redistributions in binary form must reproduce the above copyright
     51  *    notice, this list of conditions and the following disclaimer in the
     52  *    documentation and/or other materials provided with the distribution.
     53  * 3. All advertising materials mentioning features or use of this software
     54  *    must display the following acknowledgement:
     55  *      This product includes software developed by the NetBSD
     56  *      Foundation, Inc. and its contributors.
     57  * 4. Neither the name of The NetBSD Foundation nor the names of its
     58  *    contributors may be used to endorse or promote products derived
     59  *    from this software without specific prior written permission.
     60  *
     61  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
     62  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     63  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     64  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
     65  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     66  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     67  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     68  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     69  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     70  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     71  * SUCH DAMAGE.
     72  */
     73 
     74 #include <sys/cdefs.h>
     75 __KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.11 2008/04/13 22:29:38 cegger Exp $");
     76 
     77 #include "opt_ddb.h"
     78 #include "opt_multiprocessor.h"
     79 #include "opt_mpbios.h"		/* for MPDEBUG */
     80 #include "opt_mtrr.h"
     81 #include "opt_xen.h"
     82 
     83 #include "lapic.h"
     84 #include "ioapic.h"
     85 
     86 #include <sys/param.h>
     87 #include <sys/proc.h>
     88 #include <sys/user.h>
     89 #include <sys/systm.h>
     90 #include <sys/device.h>
     91 #include <sys/malloc.h>
     92 #include <sys/cpu.h>
     93 #include <sys/atomic.h>
     94 
     95 #include <uvm/uvm_extern.h>
     96 
     97 #include <machine/cpufunc.h>
     98 #include <machine/cpuvar.h>
     99 #include <machine/pmap.h>
    100 #include <machine/vmparam.h>
    101 #include <machine/mpbiosvar.h>
    102 #include <machine/pcb.h>
    103 #include <machine/specialreg.h>
    104 #include <machine/segments.h>
    105 #include <machine/gdt.h>
    106 #include <machine/mtrr.h>
    107 #include <machine/pio.h>
    108 
    109 #ifdef XEN3
    110 #include <xen/vcpuvar.h>
    111 #endif
    112 
    113 #if NLAPIC > 0
    114 #include <machine/apicvar.h>
    115 #include <machine/i82489reg.h>
    116 #include <machine/i82489var.h>
    117 #endif
    118 
    119 #if NIOAPIC > 0
    120 #include <machine/i82093var.h>
    121 #endif
    122 
    123 #include <dev/ic/mc146818reg.h>
    124 #include <dev/isa/isareg.h>
    125 
    126 int     cpu_match(device_t, cfdata_t, void *);
    127 void    cpu_attach(device_t, device_t, void *);
    128 #ifdef XEN3
    129 int     vcpu_match(device_t, cfdata_t, void *);
    130 void    vcpu_attach(device_t, device_t, void *);
    131 #endif
    132 void    cpu_attach_common(device_t, device_t, void *);
    133 void	cpu_offline_md(void);
    134 
    135 struct cpu_softc {
    136 	device_t sc_dev;		/* device tree glue */
    137 	struct cpu_info *sc_info;	/* pointer to CPU info */
    138 };
    139 
    140 int mp_cpu_start(struct cpu_info *, paddr_t);
    141 void mp_cpu_start_cleanup(struct cpu_info *);
    142 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL,
    143 				      mp_cpu_start_cleanup };
    144 
    145 CFATTACH_DECL_NEW(cpu, sizeof(struct cpu_softc),
    146     cpu_match, cpu_attach, NULL, NULL);
    147 #ifdef XEN3
    148 CFATTACH_DECL_NEW(vcpu, sizeof(struct cpu_softc),
    149     vcpu_match, vcpu_attach, NULL, NULL);
    150 #endif
    151 
    152 /*
    153  * Statically-allocated CPU info for the primary CPU (or the only
    154  * CPU, on uniprocessors).  The CPU info list is initialized to
    155  * point at it.
    156  */
    157 #ifdef TRAPLOG
    158 #include <machine/tlog.h>
    159 struct tlog tlog_primary;
    160 #endif
    161 struct cpu_info cpu_info_primary = {
    162 	.ci_dev = 0,
    163 	.ci_self = &cpu_info_primary,
    164 	.ci_idepth = -1,
    165 	.ci_curlwp = &lwp0,
    166 #ifdef TRAPLOG
    167 	.ci_tlog = &tlog_primary,
    168 #endif
    169 
    170 };
    171 struct cpu_info phycpu_info_primary = {
    172 	.ci_dev = 0,
    173 	.ci_self = &phycpu_info_primary,
    174 };
    175 
    176 struct cpu_info *cpu_info_list = &cpu_info_primary;
    177 
    178 static void	cpu_set_tss_gates(struct cpu_info *ci);
    179 
    180 uint32_t cpus_attached = 0;
    181 uint32_t cpus_running = 0;
    182 
    183 bool x86_mp_online;
    184 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE;
    185 
    186 struct cpu_info *phycpu_info[X86_MAXPROCS] = { &cpu_info_primary };
    187 
    188 #ifdef MULTIPROCESSOR
    189 /*
    190  * Array of CPU info structures.  Must be statically-allocated because
    191  * curproc, etc. are used early.
    192  */
    193 struct cpu_info *cpu_info[X86_MAXPROCS] = { &cpu_info_primary };
    194 
    195 void    	cpu_hatch(void *);
    196 static void    	cpu_boot_secondary(struct cpu_info *ci);
    197 static void    	cpu_start_secondary(struct cpu_info *ci);
    198 static void	cpu_copy_trampoline(void);
    199 
    200 /*
    201  * Runs once per boot once multiprocessor goo has been detected and
    202  * the local APIC on the boot processor has been mapped.
    203  *
    204  * Called from lapic_boot_init() (from mpbios_scan()).
    205  */
    206 void
    207 cpu_init_first(void)
    208 {
    209 	int cpunum = lapic_cpu_number();
    210 
    211 	if (cpunum != 0) {
    212 		cpu_info[0] = NULL;
    213 		cpu_info[cpunum] = &cpu_info_primary;
    214 	}
    215 
    216 	cpu_copy_trampoline();
    217 }
    218 #endif
    219 
    220 int
    221 cpu_match(device_t parent, cfdata_t match, void *aux)
    222 {
    223 
    224 	return 1;
    225 }
    226 
    227 void
    228 cpu_attach(device_t parent, device_t self, void *aux)
    229 {
    230 #ifdef XEN3
    231 	struct cpu_softc *sc = device_private(self);
    232 	struct cpu_attach_args *caa = aux;
    233 	struct cpu_info *ci;
    234 	int cpunum = caa->cpu_number;
    235 
    236 	sc->sc_dev = self;
    237 
    238 	/*
    239 	 * If we're an Application Processor, allocate a cpu_info
    240 	 * structure, otherwise use the primary's.
    241 	 */
    242 	if (caa->cpu_role == CPU_ROLE_AP) {
    243 		ci = malloc(sizeof(*ci), M_DEVBUF, M_WAITOK | M_ZERO);
    244 		if (phycpu_info[cpunum] != NULL)
    245 			panic("cpu at apic id %d already attached?", cpunum);
    246 		phycpu_info[cpunum] = ci;
    247 	} else {
    248 		ci = &phycpu_info_primary;
    249 		if (cpunum != 0) {
    250 			phycpu_info[0] = NULL;
    251 			phycpu_info[cpunum] = ci;
    252 		}
    253 	}
    254 
    255 	ci->ci_self = ci;
    256 	sc->sc_info = ci;
    257 
    258 	ci->ci_dev = self;
    259 	ci->ci_apicid = caa->cpu_number;
    260 	ci->ci_cpuid = ci->ci_apicid;
    261 
    262 	printf(": ");
    263 	switch (caa->cpu_role) {
    264 	case CPU_ROLE_SP:
    265 		printf("(uniprocessor)\n");
    266 		ci->ci_flags |= CPUF_PRESENT | CPUF_SP | CPUF_PRIMARY;
    267 		break;
    268 
    269 	case CPU_ROLE_BP:
    270 		printf("(boot processor)\n");
    271 		ci->ci_flags |= CPUF_PRESENT | CPUF_BSP | CPUF_PRIMARY;
    272 #if NIOAPIC > 0
    273 		ioapic_bsp_id = caa->cpu_number;
    274 #endif
    275 		break;
    276 
    277 	case CPU_ROLE_AP:
    278 		/*
    279 		 * report on an AP
    280 		 */
    281 		printf("(application processor)\n");
    282 		break;
    283 
    284 	default:
    285 		panic("unknown processor type??\n");
    286 	}
    287 	return;
    288 #else
    289 	cpu_attach_common(parent, self, aux);
    290 #endif
    291 }
    292 
    293 #ifdef XEN3
    294 int
    295 vcpu_match(device_t parent, cfdata_t match, void *aux)
    296 {
    297 	struct vcpu_attach_args *vcaa = aux;
    298 
    299 	if (strcmp(vcaa->vcaa_name, match->cf_name) == 0)
    300 		return 1;
    301 	return 0;
    302 }
    303 
    304 void
    305 vcpu_attach(device_t parent, device_t self, void *aux)
    306 {
    307 	struct vcpu_attach_args *vcaa = aux;
    308 
    309 	cpu_attach_common(parent, self, &vcaa->vcaa_caa);
    310 }
    311 #endif
    312 
    313 static void
    314 cpu_vm_init(struct cpu_info *ci)
    315 {
    316 	int ncolors = 2, i;
    317 
    318 	for (i = CAI_ICACHE; i <= CAI_L2CACHE; i++) {
    319 		struct x86_cache_info *cai;
    320 		int tcolors;
    321 
    322 		cai = &ci->ci_cinfo[i];
    323 
    324 		tcolors = atop(cai->cai_totalsize);
    325 		switch(cai->cai_associativity) {
    326 		case 0xff:
    327 			tcolors = 1; /* fully associative */
    328 			break;
    329 		case 0:
    330 		case 1:
    331 			break;
    332 		default:
    333 			tcolors /= cai->cai_associativity;
    334 		}
    335 		ncolors = max(ncolors, tcolors);
    336 	}
    337 
    338 	/*
    339 	 * Knowing the size of the largest cache on this CPU, re-color
    340 	 * our pages.
    341 	 */
    342 	if (ncolors <= uvmexp.ncolors)
    343 		return;
    344 	printf("%s: %d page colors\n", device_xname(ci->ci_dev), ncolors);
    345 	uvm_page_recolor(ncolors);
    346 }
    347 
    348 void
    349 cpu_attach_common(device_t parent, device_t self, void *aux)
    350 {
    351 	struct cpu_softc *sc = device_private(self);
    352 	struct cpu_attach_args *caa = aux;
    353 	struct cpu_info *ci;
    354 #if defined(MULTIPROCESSOR)
    355 	int cpunum = caa->cpu_number;
    356 #endif
    357 
    358 	sc->sc_dev = self;
    359 
    360 	/*
    361 	 * If we're an Application Processor, allocate a cpu_info
    362 	 * structure, otherwise use the primary's.
    363 	 */
    364 	if (caa->cpu_role == CPU_ROLE_AP) {
    365 		ci = malloc(sizeof(*ci), M_DEVBUF, M_WAITOK | M_ZERO);
    366 #if defined(MULTIPROCESSOR)
    367 		if (cpu_info[cpunum] != NULL)
    368 			panic("cpu at apic id %d already attached?", cpunum);
    369 		cpu_info[cpunum] = ci;
    370 #endif
    371 #ifdef TRAPLOG
    372 		ci->ci_tlog_base = malloc(sizeof(struct tlog),
    373 		    M_DEVBUF, M_WAITOK);
    374 #endif
    375 	} else {
    376 		ci = &cpu_info_primary;
    377 #if defined(MULTIPROCESSOR)
    378 		if (cpunum != lapic_cpu_number()) {
    379 			panic("%s: running CPU is at apic %d"
    380 			    " instead of at expected %d",
    381 			    device_xname(sc->sc_dev), lapic_cpu_number(), cpunum);
    382 		}
    383 #endif
    384 	}
    385 
    386 	ci->ci_self = ci;
    387 	sc->sc_info = ci;
    388 
    389 	ci->ci_dev = self;
    390 	ci->ci_apicid = caa->cpu_number;
    391 #ifdef MULTIPROCESSOR
    392 	ci->ci_cpuid = ci->ci_apicid;
    393 #else
    394 	ci->ci_cpuid = 0;	/* False for APs, but they're not used anyway */
    395 #endif
    396 	ci->ci_cpumask = (1 << ci->ci_cpuid);
    397 	ci->ci_func = caa->cpu_func;
    398 
    399 	if (caa->cpu_role == CPU_ROLE_AP) {
    400 #if defined(MULTIPROCESSOR)
    401 		int error;
    402 
    403 		error = mi_cpu_attach(ci);
    404 		if (error != 0) {
    405 			aprint_normal("\n");
    406 			aprint_error_dev(sc->sc_dev, "mi_cpu_attach failed with %d\n",
    407 			    error);
    408 			return;
    409 		}
    410 #endif
    411 	} else {
    412 		KASSERT(ci->ci_data.cpu_idlelwp != NULL);
    413 	}
    414 
    415 	pmap_reference(pmap_kernel());
    416 	ci->ci_pmap = pmap_kernel();
    417 	ci->ci_tlbstate = TLBSTATE_STALE;
    418 
    419 	/* further PCB init done later. */
    420 
    421 	printf(": ");
    422 
    423 	switch (caa->cpu_role) {
    424 	case CPU_ROLE_SP:
    425 		printf("(uniprocessor)\n");
    426 		ci->ci_flags |= CPUF_PRESENT | CPUF_SP | CPUF_PRIMARY;
    427 		cpu_intr_init(ci);
    428 		identifycpu(ci);
    429 		cpu_init(ci);
    430 		cpu_set_tss_gates(ci);
    431 		break;
    432 
    433 	case CPU_ROLE_BP:
    434 		printf("apid %d (boot processor)\n", caa->cpu_number);
    435 		ci->ci_flags |= CPUF_PRESENT | CPUF_BSP | CPUF_PRIMARY;
    436 		cpu_intr_init(ci);
    437 		identifycpu(ci);
    438 		cpu_init(ci);
    439 		cpu_set_tss_gates(ci);
    440 		break;
    441 
    442 	case CPU_ROLE_AP:
    443 		/*
    444 		 * report on an AP
    445 		 */
    446 		printf("apid %d (application processor)\n", caa->cpu_number);
    447 
    448 #if defined(MULTIPROCESSOR)
    449 		cpu_intr_init(ci);
    450 		gdt_alloc_cpu(ci);
    451 		cpu_set_tss_gates(ci);
    452 		cpu_start_secondary(ci);
    453 		if (ci->ci_flags & CPUF_PRESENT) {
    454 			identifycpu(ci);
    455 			ci->ci_next = cpu_info_list->ci_next;
    456 			cpu_info_list->ci_next = ci;
    457 		}
    458 #else
    459 		printf("%s: not started\n", device_xname(sc->sc_dev));
    460 #endif
    461 		break;
    462 
    463 	default:
    464 		panic("unknown processor type??\n");
    465 	}
    466 	cpu_vm_init(ci);
    467 
    468 	cpus_attached |= (1 << ci->ci_cpuid);
    469 
    470 #if defined(MULTIPROCESSOR)
    471 	if (mp_verbose) {
    472 		struct lwp *l = ci->ci_data.cpu_idlelwp;
    473 
    474 		aprint_verbose_dev(sc->sc_dev, "idle lwp at %p, idle sp at 0x%x\n",
    475 		    l, l->l_addr->u_pcb.pcb_esp);
    476 	}
    477 #endif
    478 }
    479 
    480 /*
    481  * Initialize the processor appropriately.
    482  */
    483 
    484 void
    485 cpu_init(struct cpu_info *ci)
    486 {
    487 	/* configure the CPU if needed */
    488 	if (ci->cpu_setup != NULL)
    489 		(*ci->cpu_setup)(ci);
    490 
    491 	/*
    492 	 * On a P6 or above, enable global TLB caching if the
    493 	 * hardware supports it.
    494 	 */
    495 	if (cpu_feature & CPUID_PGE)
    496 		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
    497 
    498 #ifdef XXXMTRR
    499 	/*
    500 	 * On a P6 or above, initialize MTRR's if the hardware supports them.
    501 	 */
    502 	if (cpu_feature & CPUID_MTRR) {
    503 		if ((ci->ci_flags & CPUF_AP) == 0)
    504 			i686_mtrr_init_first();
    505 		mtrr_init_cpu(ci);
    506 	}
    507 #endif
    508 	/*
    509 	 * If we have FXSAVE/FXRESTOR, use them.
    510 	 */
    511 	if (cpu_feature & CPUID_FXSR) {
    512 		lcr4(rcr4() | CR4_OSFXSR);
    513 
    514 		/*
    515 		 * If we have SSE/SSE2, enable XMM exceptions.
    516 		 */
    517 		if (cpu_feature & (CPUID_SSE|CPUID_SSE2))
    518 			lcr4(rcr4() | CR4_OSXMMEXCPT);
    519 	}
    520 
    521 #ifdef MULTIPROCESSOR
    522 	atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
    523 	atomic_or_32(&cpus_running, ci->ci_cpumask);
    524 #endif
    525 }
    526 
    527 
    528 #ifdef MULTIPROCESSOR
    529 void
    530 cpu_boot_secondary_processors(void)
    531 {
    532 	struct cpu_info *ci;
    533 	u_long i;
    534 
    535 	for (i = 0; i < X86_MAXPROCS; i++) {
    536 		ci = cpu_info[i];
    537 		if (ci == NULL)
    538 			continue;
    539 		if (ci->ci_data.cpu_idlelwp == NULL)
    540 			continue;
    541 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
    542 			continue;
    543 		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
    544 			continue;
    545 		cpu_boot_secondary(ci);
    546 	}
    547 
    548 	x86_mp_online = true;
    549 }
    550 
    551 static void
    552 cpu_init_idle_lwp(struct cpu_info *ci)
    553 {
    554 	struct lwp *l = ci->ci_data.cpu_idlelwp;
    555 	struct pcb *pcb = &l->l_addr->u_pcb;
    556 
    557 	pcb->pcb_cr0 = rcr0();
    558 }
    559 
    560 void
    561 cpu_init_idle_lwps(void)
    562 {
    563 	struct cpu_info *ci;
    564 	u_long i;
    565 
    566 	for (i = 0; i < X86_MAXPROCS; i++) {
    567 		ci = cpu_info[i];
    568 		if (ci == NULL)
    569 			continue;
    570 		if (ci->ci_data.cpu_idlelwp == NULL)
    571 			continue;
    572 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
    573 			continue;
    574 		cpu_init_idle_lwp(ci);
    575 	}
    576 }
    577 
    578 void
    579 cpu_start_secondary(struct cpu_info *ci)
    580 {
    581 	int i;
    582 	struct pmap *kpm = pmap_kernel();
    583 	extern uint32_t mp_pdirpa;
    584 
    585 	mp_pdirpa = kpm->pm_pdirpa; /* XXX move elsewhere, not per CPU. */
    586 
    587 	atomic_or_32(&ci->ci_flags, CPUF_AP);
    588 
    589 	aprint_debug_dev(ci->ci_dev, "starting\n");
    590 
    591 	ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
    592 	if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0)
    593 		return;
    594 
    595 	/*
    596 	 * wait for it to become ready
    597 	 */
    598 	for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
    599 #ifdef MPDEBUG
    600 		extern int cpu_trace[3];
    601 		static int otrace[3];
    602 		if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) {
    603 			aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n",
    604 				cpu_trace[0], cpu_trace[1], cpu_trace[2]);
    605 			memcpy(otrace, cpu_trace, sizeof(otrace));
    606 		}
    607 #endif
    608 		delay(10);
    609 	}
    610 	if ((ci->ci_flags & CPUF_PRESENT) == 0) {
    611 		aprint_error_dev(ci->ci_dev, "failed to become ready\n");
    612 #if defined(MPDEBUG) && defined(DDB)
    613 		printf("dropping into debugger; continue from here to resume boot\n");
    614 		Debugger();
    615 #endif
    616 	}
    617 
    618 	CPU_START_CLEANUP(ci);
    619 }
    620 
    621 void
    622 cpu_boot_secondary(struct cpu_info *ci)
    623 {
    624 	int i;
    625 
    626 	atomic_or_32(&ci->ci_flags, CPUF_GO);
    627 	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
    628 		delay(10);
    629 	}
    630 	if ((ci->ci_flags & CPUF_RUNNING) == 0) {
    631 		aprint_error_dev(ci->ci_dev, "CPU failed to start\n");
    632 #if defined(MPDEBUG) && defined(DDB)
    633 		printf("dropping into debugger; continue from here to resume boot\n");
    634 		Debugger();
    635 #endif
    636 	}
    637 }
    638 
    639 /*
    640  * The CPU ends up here when its ready to run
    641  * This is called from code in mptramp.s; at this point, we are running
    642  * in the idle pcb/idle stack of the new CPU.  When this function returns,
    643  * this processor will enter the idle loop and start looking for work.
    644  *
    645  * XXX should share some of this with init386 in machdep.c
    646  */
    647 void
    648 cpu_hatch(void *v)
    649 {
    650 	struct cpu_info *ci = (struct cpu_info *)v;
    651 	int s, i;
    652 	uint32_t blacklist_features;
    653 
    654 #ifdef __x86_64__
    655         cpu_init_msrs(ci, true);
    656 #endif
    657 
    658 	cpu_probe_features(ci);
    659 	cpu_feature &= ci->ci_feature_flags;
    660 	cpu_feature2 &= ci->ci_feature2_flags;
    661 
    662 	/* not on Xen... */
    663 	blacklist_features = ~(CPUID_PGE|CPUID_PSE|CPUID_MTRR|CPUID_FXSR|CPUID_NOX); /* XXX add CPUID_SVM */
    664 
    665 	cpu_feature &= blacklist_features;
    666 
    667 	KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
    668 	atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
    669 	while ((ci->ci_flags & CPUF_GO) == 0) {
    670 		/* Don't use delay, boot CPU may be patching the text. */
    671 		for (i = 10000; i != 0; i--)
    672 			x86_pause();
    673 	}
    674 
    675 	/* Because the text may have been patched in x86_patch(). */
    676 	wbinvd();
    677 	x86_flush();
    678 
    679 	KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
    680 
    681 	lcr0(ci->ci_data.cpu_idlelwp->l_addr->u_pcb.pcb_cr0);
    682 	cpu_init_idt();
    683 	gdt_init_cpu(ci);
    684 	lapic_enable();
    685 	lapic_set_lvt();
    686 	lapic_initclocks();
    687 
    688 	npxinit(ci);
    689 
    690 	lldt(GSEL(GLDT_SEL, SEL_KPL));
    691 	/* ltr(ci->ci_tss_sel); */ /* XXX */
    692 
    693 	cpu_init(ci);
    694 	cpu_get_tsc_freq(ci);
    695 
    696 	s = splhigh();
    697 #ifdef i386
    698 	lapic_tpr = 0;
    699 #else
    700 	lcr8(0);
    701 #endif
    702 	x86_enable_intr();
    703 	splx(s);
    704 #ifdef DOM0OPS
    705 	x86_errata();
    706 #endif
    707 
    708 	aprint_debug_dev(ci->ci_dev, "CPU %ld running\n",
    709 		(long)ci->ci_cpuid);
    710 }
    711 
    712 #if defined(DDB)
    713 
    714 #include <ddb/db_output.h>
    715 #include <machine/db_machdep.h>
    716 
    717 /*
    718  * Dump CPU information from ddb.
    719  */
    720 void
    721 cpu_debug_dump(void)
    722 {
    723 	struct cpu_info *ci;
    724 	CPU_INFO_ITERATOR cii;
    725 
    726 	db_printf("addr		dev	id	flags	ipis	curproc		fpcurproc\n");
    727 	for (CPU_INFO_FOREACH(cii, ci)) {
    728 		db_printf("%p	%s	%ld	%x	%x	%10p	%10p\n",
    729 		    ci,
    730 		    ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
    731 		    ci->ci_cpuid,
    732 		    ci->ci_flags, ci->ci_ipis,
    733 		    ci->ci_curlwp,
    734 		    ci->ci_fpcurlwp);
    735 	}
    736 }
    737 #endif
    738 
    739 static void
    740 cpu_copy_trampoline(void)
    741 {
    742 	/*
    743 	 * Copy boot code.
    744 	 */
    745 	extern u_char cpu_spinup_trampoline[];
    746 	extern u_char cpu_spinup_trampoline_end[];
    747 
    748 	vaddr_t mp_trampoline_vaddr;
    749 
    750 	mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
    751 		UVM_KMF_VAONLY);
    752 
    753 	pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
    754 		VM_PROT_READ | VM_PROT_WRITE);
    755 	pmap_update(pmap_kernel());
    756 	memcpy((void *)mp_trampoline_vaddr,
    757 		cpu_spinup_trampoline,
    758 		cpu_spinup_trampoline_end - cpu_spinup_trampoline);
    759 
    760 	pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
    761 	pmap_update(pmap_kernel());
    762 	uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
    763 }
    764 
    765 #endif
    766 
    767 #ifdef i386
    768 #if 0
    769 static void
    770 tss_init(struct i386tss *tss, void *stack, void *func)
    771 {
    772 	memset(tss, 0, sizeof *tss);
    773 	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
    774 	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
    775 	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
    776 	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
    777 	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
    778 	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
    779 	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
    780 	tss->tss_esp = (int)((char *)stack + USPACE - 16);
    781 	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
    782 	tss->__tss_eflags = PSL_MBO | PSL_NT;   /* XXX not needed? */
    783 	tss->__tss_eip = (int)func;
    784 }
    785 #endif
    786 
    787 /* XXX */
    788 #define IDTVEC(name)	__CONCAT(X, name)
    789 typedef void (vector)(void);
    790 extern vector IDTVEC(tss_trap08);
    791 #ifdef DDB
    792 extern vector Xintrddbipi;
    793 extern int ddb_vec;
    794 #endif
    795 
    796 static void
    797 cpu_set_tss_gates(struct cpu_info *ci)
    798 {
    799 #if 0
    800 	struct segment_descriptor sd;
    801 
    802 	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
    803 	    UVM_KMF_WIRED);
    804 	tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
    805 	    IDTVEC(tss_trap08));
    806 	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
    807 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
    808 	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
    809 	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
    810 	    GSEL(GTRAPTSS_SEL, SEL_KPL));
    811 #endif
    812 
    813 #if defined(DDB) && defined(MULTIPROCESSOR)
    814 	/*
    815 	 * Set up separate handler for the DDB IPI, so that it doesn't
    816 	 * stomp on a possibly corrupted stack.
    817 	 *
    818 	 * XXX overwriting the gate set in db_machine_init.
    819 	 * Should rearrange the code so that it's set only once.
    820 	 */
    821 	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
    822 	    UVM_KMF_WIRED);
    823 	tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
    824 	    Xintrddbipi);
    825 
    826 	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
    827 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
    828 	ci->ci_gdt[GIPITSS_SEL].sd = sd;
    829 
    830 	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
    831 	    GSEL(GIPITSS_SEL, SEL_KPL));
    832 #endif
    833 }
    834 #else
    835 static void
    836 cpu_set_tss_gates(struct cpu_info *ci)
    837 {
    838 
    839 }
    840 #endif	/* i386 */
    841 
    842 int
    843 mp_cpu_start(struct cpu_info *ci, paddr_t target)
    844 {
    845 #if 0
    846 #if NLAPIC > 0
    847 	int error;
    848 #endif
    849 	unsigned short dwordptr[2];
    850 
    851 	/*
    852 	 * Bootstrap code must be addressable in real mode
    853 	 * and it must be page aligned.
    854 	 */
    855 	KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
    856 
    857 	/*
    858 	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
    859 	 */
    860 
    861 	outb(IO_RTC, NVRAM_RESET);
    862 	outb(IO_RTC+1, NVRAM_RESET_JUMP);
    863 
    864 	/*
    865 	 * "and the warm reset vector (DWORD based at 40:67) to point
    866 	 * to the AP startup code ..."
    867 	 */
    868 
    869 	dwordptr[0] = 0;
    870 	dwordptr[1] = target >> 4;
    871 
    872 	pmap_kenter_pa (0, 0, VM_PROT_READ|VM_PROT_WRITE);
    873 	memcpy ((uint8_t *) 0x467, dwordptr, 4);
    874 	pmap_kremove (0, PAGE_SIZE);
    875 
    876 #if NLAPIC > 0
    877 	/*
    878 	 * ... prior to executing the following sequence:"
    879 	 */
    880 
    881 	if (ci->ci_flags & CPUF_AP) {
    882 		if ((error = x86_ipi_init(ci->ci_apicid)) != 0)
    883 			return error;
    884 
    885 		delay(10000);
    886 
    887 		if (cpu_feature & CPUID_APIC) {
    888 			error = x86_ipi_init(ci->ci_apicid);
    889 			if (error != 0) {
    890 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
    891 						__func__);
    892 				return error;
    893 			}
    894 
    895 			delay(10000);
    896 
    897 			error = x86_ipi(target / PAGE_SIZE, ci->ci_apicid,
    898 					LAPIC_DLMODE_STARTUP);
    899 			if (error != 0) {
    900 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
    901 						__func__);
    902 				return error;
    903 			}
    904 			delay(200);
    905 
    906 			error = x86_ipi(target / PAGE_SIZE, ci->ci_apicid,
    907 					LAPIC_DLMODE_STARTUP);
    908 			if (error != 0) {
    909 				aprint_error_dev(ci->ci_dev, "%s: IPI not taken ((3)\n",
    910 						__func__);
    911 				return error;
    912 			}
    913 			delay(200);
    914 		}
    915 	}
    916 #endif
    917 #endif /* 0 */
    918 	return 0;
    919 }
    920 
    921 void
    922 mp_cpu_start_cleanup(struct cpu_info *ci)
    923 {
    924 #if 0
    925 	/*
    926 	 * Ensure the NVRAM reset byte contains something vaguely sane.
    927 	 */
    928 
    929 	outb(IO_RTC, NVRAM_RESET);
    930 	outb(IO_RTC+1, NVRAM_RESET_RST);
    931 #endif
    932 }
    933 
    934 #ifdef __x86_64__
    935 
    936 void
    937 cpu_init_msrs(struct cpu_info *ci, bool full)
    938 {
    939 	if (full) {
    940 		HYPERVISOR_set_segment_base (SEGBASE_FS, 0);
    941 		HYPERVISOR_set_segment_base (SEGBASE_GS_KERNEL, (uint64_t) ci);
    942 		HYPERVISOR_set_segment_base (SEGBASE_GS_USER, 0);
    943 	}
    944 }
    945 #endif	/* __x86_64__ */
    946 
    947 void
    948 cpu_offline_md(void)
    949 {
    950         int s;
    951 
    952         s = splhigh();
    953 #ifdef __i386__
    954         npxsave_cpu(true);
    955 #else
    956         fpusave_cpu(true);
    957 #endif
    958         splx(s);
    959 }
    960 
    961 #if 0
    962 /* XXX joerg restructure and restart CPUs individually */
    963 static bool
    964 cpu_suspend(device_t dv PMF_FN_ARGS)
    965 {
    966 	struct cpu_softc *sc = device_private(dv);
    967 	struct cpu_info *ci = sc->sc_info;
    968 	int err;
    969 
    970 	if (ci->ci_flags & CPUF_PRIMARY)
    971 		return true;
    972 	if (ci->ci_data.cpu_idlelwp == NULL)
    973 		return true;
    974 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
    975 		return true;
    976 
    977 	sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
    978 
    979 	if (sc->sc_wasonline) {
    980 		mutex_enter(&cpu_lock);
    981 		err = cpu_setonline(ci, false);
    982 		mutex_exit(&cpu_lock);
    983 
    984 		if (err)
    985 			return false;
    986 	}
    987 
    988 	return true;
    989 }
    990 
    991 static bool
    992 cpu_resume(device_t dv PMF_FN_ARGS)
    993 {
    994 	struct cpu_softc *sc = device_private(dv);
    995 	struct cpu_info *ci = sc->sc_info;
    996 	int err = 0;
    997 
    998 	if (ci->ci_flags & CPUF_PRIMARY)
    999 		return true;
   1000 	if (ci->ci_data.cpu_idlelwp == NULL)
   1001 		return true;
   1002 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
   1003 		return true;
   1004 
   1005 	if (sc->sc_wasonline) {
   1006 		mutex_enter(&cpu_lock);
   1007 		err = cpu_setonline(ci, true);
   1008 		mutex_exit(&cpu_lock);
   1009 	}
   1010 
   1011 	return err == 0;
   1012 }
   1013 #endif
   1014 
   1015 void
   1016 cpu_get_tsc_freq(struct cpu_info *ci)
   1017 {
   1018 #ifdef XEN3
   1019 	const volatile vcpu_time_info_t *tinfo =
   1020 		   &HYPERVISOR_shared_info->vcpu_info[0].time;
   1021 	delay(1000000);
   1022 	uint64_t freq = 1000000000ULL << 32;
   1023 	freq = freq / (uint64_t)tinfo->tsc_to_system_mul;
   1024 	if ( tinfo->tsc_shift < 0 )
   1025 		freq = freq << -tinfo->tsc_shift;
   1026 	else
   1027 		freq = freq >> tinfo->tsc_shift;
   1028 	ci->ci_tsc_freq = freq;
   1029 #else
   1030 	/* XXX this needs to read the shared_info of the CPU being probed.. */
   1031 	ci->ci_tsc_freq = HYPERVISOR_shared_info->cpu_freq;
   1032 #endif /* XEN3 */
   1033 }
   1034