Home | History | Annotate | Line # | Download | only in x86
cpu.c revision 1.36
      1 /*	$NetBSD: cpu.c,v 1.36 2008/04/29 19:19:29 ad Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2000, 2006, 2007 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Bill Sommerfeld of RedBack Networks Inc, and by Andrew Doran.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 1999 Stefan Grefen
     34  *
     35  * Redistribution and use in source and binary forms, with or without
     36  * modification, are permitted provided that the following conditions
     37  * are met:
     38  * 1. Redistributions of source code must retain the above copyright
     39  *    notice, this list of conditions and the following disclaimer.
     40  * 2. Redistributions in binary form must reproduce the above copyright
     41  *    notice, this list of conditions and the following disclaimer in the
     42  *    documentation and/or other materials provided with the distribution.
     43  * 3. All advertising materials mentioning features or use of this software
     44  *    must display the following acknowledgement:
     45  *      This product includes software developed by the NetBSD
     46  *      Foundation, Inc. and its contributors.
     47  * 4. Neither the name of The NetBSD Foundation nor the names of its
     48  *    contributors may be used to endorse or promote products derived
     49  *    from this software without specific prior written permission.
     50  *
     51  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
     52  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
     55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     61  * SUCH DAMAGE.
     62  */
     63 
     64 #include <sys/cdefs.h>
     65 __KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.36 2008/04/29 19:19:29 ad Exp $");
     66 
     67 #include "opt_ddb.h"
     68 #include "opt_multiprocessor.h"
     69 #include "opt_mpbios.h"		/* for MPDEBUG */
     70 #include "opt_mtrr.h"
     71 
     72 #include "lapic.h"
     73 #include "ioapic.h"
     74 
     75 #include <sys/param.h>
     76 #include <sys/proc.h>
     77 #include <sys/user.h>
     78 #include <sys/systm.h>
     79 #include <sys/device.h>
     80 #include <sys/malloc.h>
     81 #include <sys/cpu.h>
     82 #include <sys/atomic.h>
     83 #include <sys/reboot.h>
     84 
     85 #include <uvm/uvm_extern.h>
     86 
     87 #include <machine/cpufunc.h>
     88 #include <machine/cpuvar.h>
     89 #include <machine/pmap.h>
     90 #include <machine/vmparam.h>
     91 #include <machine/mpbiosvar.h>
     92 #include <machine/pcb.h>
     93 #include <machine/specialreg.h>
     94 #include <machine/segments.h>
     95 #include <machine/gdt.h>
     96 #include <machine/mtrr.h>
     97 #include <machine/pio.h>
     98 
     99 #ifdef i386
    100 #include <machine/tlog.h>
    101 #endif
    102 
    103 #if NLAPIC > 0
    104 #include <machine/apicvar.h>
    105 #include <machine/i82489reg.h>
    106 #include <machine/i82489var.h>
    107 #endif
    108 
    109 #include <dev/ic/mc146818reg.h>
    110 #include <i386/isa/nvram.h>
    111 #include <dev/isa/isareg.h>
    112 
    113 int     cpu_match(device_t, cfdata_t, void *);
    114 void    cpu_attach(device_t, device_t, void *);
    115 
    116 static bool	cpu_suspend(device_t PMF_FN_PROTO);
    117 static bool	cpu_resume(device_t PMF_FN_PROTO);
    118 
    119 struct cpu_softc {
    120 	device_t sc_dev;		/* device tree glue */
    121 	struct cpu_info *sc_info;	/* pointer to CPU info */
    122 	bool sc_wasonline;
    123 };
    124 
    125 int mp_cpu_start(struct cpu_info *, paddr_t);
    126 void mp_cpu_start_cleanup(struct cpu_info *);
    127 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL,
    128 					    mp_cpu_start_cleanup };
    129 
    130 
    131 CFATTACH_DECL_NEW(cpu, sizeof(struct cpu_softc),
    132     cpu_match, cpu_attach, NULL, NULL);
    133 
    134 /*
    135  * Statically-allocated CPU info for the primary CPU (or the only
    136  * CPU, on uniprocessors).  The CPU info list is initialized to
    137  * point at it.
    138  */
    139 #ifdef TRAPLOG
    140 struct tlog tlog_primary;
    141 #endif
    142 struct cpu_info cpu_info_primary __aligned(CACHE_LINE_SIZE) = {
    143 	.ci_dev = 0,
    144 	.ci_self = &cpu_info_primary,
    145 	.ci_idepth = -1,
    146 	.ci_curlwp = &lwp0,
    147 #ifdef TRAPLOG
    148 	.ci_tlog_base = &tlog_primary,
    149 #endif /* !TRAPLOG */
    150 };
    151 
    152 struct cpu_info *cpu_info_list = &cpu_info_primary;
    153 
    154 static void	cpu_set_tss_gates(struct cpu_info *);
    155 
    156 #ifdef i386
    157 static void	tss_init(struct i386tss *, void *, void *);
    158 #endif
    159 
    160 #ifdef MULTIPROCESSOR
    161 static void	cpu_init_idle_lwp(struct cpu_info *);
    162 #endif
    163 
    164 uint32_t cpus_attached = 0;
    165 uint32_t cpus_running = 0;
    166 
    167 extern char x86_64_doubleflt_stack[];
    168 
    169 bool x86_mp_online;
    170 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE;
    171 
    172 static vaddr_t cmos_data_mapping;
    173 
    174 #ifdef MULTIPROCESSOR
    175 /*
    176  * Array of CPU info structures.  Must be statically-allocated because
    177  * curproc, etc. are used early.
    178  */
    179 struct cpu_info *cpu_info[X86_MAXPROCS] = { &cpu_info_primary, };
    180 
    181 void    	cpu_hatch(void *);
    182 static void    	cpu_boot_secondary(struct cpu_info *ci);
    183 static void    	cpu_start_secondary(struct cpu_info *ci);
    184 static void	cpu_copy_trampoline(void);
    185 
    186 /*
    187  * Runs once per boot once multiprocessor goo has been detected and
    188  * the local APIC on the boot processor has been mapped.
    189  *
    190  * Called from lapic_boot_init() (from mpbios_scan()).
    191  */
    192 void
    193 cpu_init_first(void)
    194 {
    195 	int cpunum = lapic_cpu_number();
    196 
    197 	if (cpunum != 0) {
    198 		cpu_info[0] = NULL;
    199 		cpu_info[cpunum] = &cpu_info_primary;
    200 	}
    201 
    202 	cpu_info_primary.ci_cpuid = cpunum;
    203 	cpu_copy_trampoline();
    204 
    205 	cmos_data_mapping = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, UVM_KMF_VAONLY);
    206 	if (cmos_data_mapping == 0)
    207 		panic("No KVA for page 0");
    208 	pmap_kenter_pa(cmos_data_mapping, 0, VM_PROT_READ|VM_PROT_WRITE);
    209 	pmap_update(pmap_kernel());
    210 }
    211 #endif
    212 
    213 int
    214 cpu_match(device_t parent, cfdata_t match, void *aux)
    215 {
    216 
    217 	return 1;
    218 }
    219 
    220 static void
    221 cpu_vm_init(struct cpu_info *ci)
    222 {
    223 	int ncolors = 2, i;
    224 
    225 	for (i = CAI_ICACHE; i <= CAI_L2CACHE; i++) {
    226 		struct x86_cache_info *cai;
    227 		int tcolors;
    228 
    229 		cai = &ci->ci_cinfo[i];
    230 
    231 		tcolors = atop(cai->cai_totalsize);
    232 		switch(cai->cai_associativity) {
    233 		case 0xff:
    234 			tcolors = 1; /* fully associative */
    235 			break;
    236 		case 0:
    237 		case 1:
    238 			break;
    239 		default:
    240 			tcolors /= cai->cai_associativity;
    241 		}
    242 		ncolors = max(ncolors, tcolors);
    243 		/*
    244 		 * If the desired number of colors is not a power of
    245 		 * two, it won't be good.  Find the greatest power of
    246 		 * two which is an even divisor of the number of colors,
    247 		 * to preserve even coloring of pages.
    248 		 */
    249 		if (ncolors & (ncolors - 1) ) {
    250 			int try, picked = 1;
    251 			for (try = 1; try < ncolors; try *= 2) {
    252 				if (ncolors % try == 0) picked = try;
    253 			}
    254 			if (picked == 1) {
    255 				panic("desired number of cache colors %d is "
    256 			      	" > 1, but not even!", ncolors);
    257 			}
    258 			ncolors = picked;
    259 		}
    260 	}
    261 
    262 	/*
    263 	 * Knowing the size of the largest cache on this CPU, re-color
    264 	 * our pages.
    265 	 */
    266 	if (ncolors <= uvmexp.ncolors)
    267 		return;
    268 	aprint_verbose_dev(ci->ci_dev, "%d page colors\n", ncolors);
    269 	uvm_page_recolor(ncolors);
    270 }
    271 
    272 
    273 void
    274 cpu_attach(device_t parent, device_t self, void *aux)
    275 {
    276 	struct cpu_softc *sc = device_private(self);
    277 	struct cpu_attach_args *caa = aux;
    278 	struct cpu_info *ci;
    279 	uintptr_t ptr;
    280 	int cpunum = caa->cpu_number;
    281 
    282 	sc->sc_dev = self;
    283 
    284 	/*
    285 	 * If we're an Application Processor, allocate a cpu_info
    286 	 * structure, otherwise use the primary's.
    287 	 */
    288 	if (caa->cpu_role == CPU_ROLE_AP) {
    289 		if ((boothowto & RB_MD1) != 0) {
    290 			aprint_error(": multiprocessor boot disabled\n");
    291 			return;
    292 		}
    293 		if (cpunum >= X86_MAXPROCS) {
    294 			aprint_error(": apic id %d ignored, "
    295 			    "please increase X86_MAXPROCS\n", cpunum);
    296 			return;
    297 		}
    298 		aprint_naive(": Application Processor\n");
    299 		ptr = (uintptr_t)malloc(sizeof(*ci) + CACHE_LINE_SIZE - 1,
    300 		    M_DEVBUF, M_WAITOK);
    301 		ci = (struct cpu_info *)((ptr + CACHE_LINE_SIZE - 1) &
    302 		    ~(CACHE_LINE_SIZE - 1));
    303 		memset(ci, 0, sizeof(*ci));
    304 #if defined(MULTIPROCESSOR)
    305 		if (cpu_info[cpunum] != NULL) {
    306 			printf("\n");
    307 			panic("cpu at apic id %d already attached?", cpunum);
    308 		}
    309 		cpu_info[cpunum] = ci;
    310 #endif
    311 #ifdef TRAPLOG
    312 		ci->ci_tlog_base = malloc(sizeof(struct tlog),
    313 		    M_DEVBUF, M_WAITOK);
    314 #endif
    315 	} else {
    316 		aprint_naive(": %s Processor\n",
    317 		    caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
    318 		ci = &cpu_info_primary;
    319 #if defined(MULTIPROCESSOR)
    320 		if (cpunum != lapic_cpu_number()) {
    321 			printf("\n");
    322 			panic("%s: running CPU is at apic %d"
    323 			    " instead of at expected %d",
    324 			    device_xname(sc->sc_dev), lapic_cpu_number(),
    325 			    cpunum);
    326 		}
    327 #endif
    328 	}
    329 
    330 	ci->ci_self = ci;
    331 	sc->sc_info = ci;
    332 
    333 	ci->ci_dev = self;
    334 	ci->ci_apicid = caa->cpu_number;
    335 #ifdef MULTIPROCESSOR
    336 	ci->ci_cpuid = ci->ci_apicid;
    337 #else
    338 	ci->ci_cpuid = 0;	/* False for APs, but they're not used anyway */
    339 #endif
    340 	ci->ci_cpumask = (1 << ci->ci_cpuid);
    341 	ci->ci_func = caa->cpu_func;
    342 
    343 	if (caa->cpu_role == CPU_ROLE_AP) {
    344 #ifdef MULTIPROCESSOR
    345 		int error;
    346 
    347 		error = mi_cpu_attach(ci);
    348 		if (error != 0) {
    349 			aprint_normal("\n");
    350 			aprint_error_dev(sc->sc_dev,
    351 			    "mi_cpu_attach failed with %d\n", error);
    352 			return;
    353 		}
    354 #endif
    355 		cpu_init_tss(ci);
    356 	} else {
    357 		KASSERT(ci->ci_data.cpu_idlelwp != NULL);
    358 	}
    359 
    360 	pmap_reference(pmap_kernel());
    361 	ci->ci_pmap = pmap_kernel();
    362 	ci->ci_tlbstate = TLBSTATE_STALE;
    363 
    364 	/* further PCB init done later. */
    365 
    366 	switch (caa->cpu_role) {
    367 	case CPU_ROLE_SP:
    368 		aprint_normal(": (uniprocessor)\n");
    369 		atomic_or_32(&ci->ci_flags,
    370 		    CPUF_PRESENT | CPUF_SP | CPUF_PRIMARY);
    371 		cpu_intr_init(ci);
    372 		identifycpu(ci);
    373 		cpu_init(ci);
    374 		cpu_set_tss_gates(ci);
    375 		pmap_cpu_init_late(ci);
    376 		x86_errata();
    377 		break;
    378 
    379 	case CPU_ROLE_BP:
    380 		aprint_normal(": (boot processor)\n");
    381 		atomic_or_32(&ci->ci_flags,
    382 		    CPUF_PRESENT | CPUF_BSP | CPUF_PRIMARY);
    383 		cpu_intr_init(ci);
    384 		identifycpu(ci);
    385 		cpu_init(ci);
    386 		cpu_set_tss_gates(ci);
    387 		pmap_cpu_init_late(ci);
    388 #if NLAPIC > 0
    389 		/*
    390 		 * Enable local apic
    391 		 */
    392 		lapic_enable();
    393 		lapic_set_lvt();
    394 		lapic_calibrate_timer(ci);
    395 #endif
    396 		x86_errata();
    397 		break;
    398 
    399 	case CPU_ROLE_AP:
    400 		/*
    401 		 * report on an AP
    402 		 */
    403 		aprint_normal(": (application processor)\n");
    404 
    405 #if defined(MULTIPROCESSOR)
    406 		cpu_intr_init(ci);
    407 		gdt_alloc_cpu(ci);
    408 		cpu_set_tss_gates(ci);
    409 		pmap_cpu_init_early(ci);
    410 		pmap_cpu_init_late(ci);
    411 		cpu_start_secondary(ci);
    412 		if (ci->ci_flags & CPUF_PRESENT) {
    413 			identifycpu(ci);
    414 			ci->ci_next = cpu_info_list->ci_next;
    415 			cpu_info_list->ci_next = ci;
    416 		}
    417 #else
    418 		aprint_normal_dev(sc->sc_dev, "not started\n");
    419 #endif
    420 		break;
    421 
    422 	default:
    423 		aprint_normal("\n");
    424 		panic("unknown processor type??\n");
    425 	}
    426 	cpu_vm_init(ci);
    427 
    428 	cpus_attached |= ci->ci_cpumask;
    429 
    430 	if (!pmf_device_register(self, cpu_suspend, cpu_resume))
    431 		aprint_error_dev(self, "couldn't establish power handler\n");
    432 
    433 #if defined(MULTIPROCESSOR)
    434 	if (mp_verbose) {
    435 		struct lwp *l = ci->ci_data.cpu_idlelwp;
    436 
    437 		aprint_verbose_dev(sc->sc_dev,
    438 		    "idle lwp at %p, idle sp at %p\n",
    439 		    l,
    440 #ifdef i386
    441 		    (void *)l->l_addr->u_pcb.pcb_esp
    442 #else
    443 		    (void *)l->l_addr->u_pcb.pcb_rsp
    444 #endif
    445 		);
    446 	}
    447 #endif
    448 }
    449 
    450 /*
    451  * Initialize the processor appropriately.
    452  */
    453 
    454 void
    455 cpu_init(struct cpu_info *ci)
    456 {
    457 	/* configure the CPU if needed */
    458 	if (ci->cpu_setup != NULL)
    459 		(*ci->cpu_setup)(ci);
    460 
    461 #ifdef i386
    462 	/*
    463 	 * On a 486 or above, enable ring 0 write protection.
    464 	 */
    465 	if (ci->ci_cpu_class >= CPUCLASS_486)
    466 		lcr0(rcr0() | CR0_WP);
    467 #else
    468 	lcr0(rcr0() | CR0_WP);
    469 #endif
    470 
    471 	/*
    472 	 * On a P6 or above, enable global TLB caching if the
    473 	 * hardware supports it.
    474 	 */
    475 	if (cpu_feature & CPUID_PGE)
    476 		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
    477 
    478 	/*
    479 	 * If we have FXSAVE/FXRESTOR, use them.
    480 	 */
    481 	if (cpu_feature & CPUID_FXSR) {
    482 		lcr4(rcr4() | CR4_OSFXSR);
    483 
    484 		/*
    485 		 * If we have SSE/SSE2, enable XMM exceptions.
    486 		 */
    487 		if (cpu_feature & (CPUID_SSE|CPUID_SSE2))
    488 			lcr4(rcr4() | CR4_OSXMMEXCPT);
    489 	}
    490 
    491 #ifdef MTRR
    492 	/*
    493 	 * On a P6 or above, initialize MTRR's if the hardware supports them.
    494 	 */
    495 	if (cpu_feature & CPUID_MTRR) {
    496 		if ((ci->ci_flags & CPUF_AP) == 0)
    497 			i686_mtrr_init_first();
    498 		mtrr_init_cpu(ci);
    499 	}
    500 
    501 #ifdef i386
    502 	if (strcmp((char *)(ci->ci_vendor), "AuthenticAMD") == 0) {
    503 		/*
    504 		 * Must be a K6-2 Step >= 7 or a K6-III.
    505 		 */
    506 		if (CPUID2FAMILY(ci->ci_signature) == 5) {
    507 			if (CPUID2MODEL(ci->ci_signature) > 8 ||
    508 			    (CPUID2MODEL(ci->ci_signature) == 8 &&
    509 			     CPUID2STEPPING(ci->ci_signature) >= 7)) {
    510 				mtrr_funcs = &k6_mtrr_funcs;
    511 				k6_mtrr_init_first();
    512 				mtrr_init_cpu(ci);
    513 			}
    514 		}
    515 	}
    516 #endif	/* i386 */
    517 #endif /* MTRR */
    518 
    519 	atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
    520 	atomic_or_32(&cpus_running, ci->ci_cpumask);
    521 
    522 #ifndef MULTIPROCESSOR
    523 	/* XXX */
    524 	x86_patch();
    525 #endif
    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 	/* Now that we know the number of CPUs, patch the text segment. */
    536 	x86_patch();
    537 
    538 	for (i=0; i < X86_MAXPROCS; i++) {
    539 		ci = cpu_info[i];
    540 		if (ci == NULL)
    541 			continue;
    542 		if (ci->ci_data.cpu_idlelwp == NULL)
    543 			continue;
    544 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
    545 			continue;
    546 		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
    547 			continue;
    548 		cpu_boot_secondary(ci);
    549 	}
    550 
    551 	x86_mp_online = true;
    552 }
    553 
    554 static void
    555 cpu_init_idle_lwp(struct cpu_info *ci)
    556 {
    557 	struct lwp *l = ci->ci_data.cpu_idlelwp;
    558 	struct pcb *pcb = &l->l_addr->u_pcb;
    559 
    560 	pcb->pcb_cr0 = rcr0();
    561 }
    562 
    563 void
    564 cpu_init_idle_lwps(void)
    565 {
    566 	struct cpu_info *ci;
    567 	u_long i;
    568 
    569 	for (i = 0; i < X86_MAXPROCS; i++) {
    570 		ci = cpu_info[i];
    571 		if (ci == NULL)
    572 			continue;
    573 		if (ci->ci_data.cpu_idlelwp == NULL)
    574 			continue;
    575 		if ((ci->ci_flags & CPUF_PRESENT) == 0)
    576 			continue;
    577 		cpu_init_idle_lwp(ci);
    578 	}
    579 }
    580 
    581 void
    582 cpu_start_secondary(struct cpu_info *ci)
    583 {
    584 	int i;
    585 	extern paddr_t mp_pdirpa;
    586 
    587 	mp_pdirpa = pmap_init_tmp_pgtbl(mp_trampoline_paddr);
    588 
    589 	atomic_or_32(&ci->ci_flags, CPUF_AP);
    590 
    591 	aprint_debug_dev(ci->ci_dev, "starting\n");
    592 
    593 	ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
    594 	if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0)
    595 		return;
    596 
    597 	/*
    598 	 * wait for it to become ready
    599 	 */
    600 	for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
    601 #ifdef MPDEBUG
    602 		extern int cpu_trace[3];
    603 		static int otrace[3];
    604 		if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) {
    605 			aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n",
    606 			    cpu_trace[0], cpu_trace[1], cpu_trace[2]);
    607 			memcpy(otrace, cpu_trace, sizeof(otrace));
    608 		}
    609 #endif
    610 		i8254_delay(10);
    611 	}
    612 	if ((ci->ci_flags & CPUF_PRESENT) == 0) {
    613 		aprint_error_dev(ci->ci_dev, "failed to become ready\n");
    614 #if defined(MPDEBUG) && defined(DDB)
    615 		printf("dropping into debugger; continue from here to resume boot\n");
    616 		Debugger();
    617 #endif
    618 	}
    619 
    620 	CPU_START_CLEANUP(ci);
    621 }
    622 
    623 void
    624 cpu_boot_secondary(struct cpu_info *ci)
    625 {
    626 	int i;
    627 
    628 	atomic_or_32(&ci->ci_flags, CPUF_GO);
    629 	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
    630 		i8254_delay(10);
    631 	}
    632 	if ((ci->ci_flags & CPUF_RUNNING) == 0) {
    633 		aprint_error_dev(ci->ci_dev, "failed to start\n");
    634 #if defined(MPDEBUG) && defined(DDB)
    635 		printf("dropping into debugger; continue from here to resume boot\n");
    636 		Debugger();
    637 #endif
    638 	}
    639 }
    640 
    641 /*
    642  * The CPU ends up here when its ready to run
    643  * This is called from code in mptramp.s; at this point, we are running
    644  * in the idle pcb/idle stack of the new CPU.  When this function returns,
    645  * this processor will enter the idle loop and start looking for work.
    646  */
    647 void
    648 cpu_hatch(void *v)
    649 {
    650 	struct cpu_info *ci = (struct cpu_info *)v;
    651 	int s, i;
    652 
    653 #ifdef __x86_64__
    654 	cpu_init_msrs(ci, true);
    655 #endif
    656 	cpu_probe_features(ci);
    657 	cpu_feature &= ci->ci_feature_flags;
    658 	cpu_feature2 &= ci->ci_feature2_flags;
    659 
    660 	KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
    661 	atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
    662 	while ((ci->ci_flags & CPUF_GO) == 0) {
    663 		/* Don't use delay, boot CPU may be patching the text. */
    664 		for (i = 10000; i != 0; i--)
    665 			x86_pause();
    666 	}
    667 
    668 	/* Because the text may have been patched in x86_patch(). */
    669 	wbinvd();
    670 	x86_flush();
    671 
    672 	KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
    673 
    674 	lcr3(pmap_kernel()->pm_pdirpa);
    675 	curlwp->l_addr->u_pcb.pcb_cr3 = pmap_kernel()->pm_pdirpa;
    676 	lcr0(ci->ci_data.cpu_idlelwp->l_addr->u_pcb.pcb_cr0);
    677 	cpu_init_idt();
    678 	gdt_init_cpu(ci);
    679 	lapic_enable();
    680 	lapic_set_lvt();
    681 	lapic_initclocks();
    682 
    683 #ifdef i386
    684 	npxinit(ci);
    685 #else
    686 	fpuinit(ci);
    687 #endif
    688 	lldt(GSYSSEL(GLDT_SEL, SEL_KPL));
    689 	ltr(ci->ci_tss_sel);
    690 
    691 	cpu_init(ci);
    692 	cpu_get_tsc_freq(ci);
    693 
    694 	s = splhigh();
    695 #ifdef i386
    696 	lapic_tpr = 0;
    697 #else
    698 	lcr8(0);
    699 #endif
    700 	x86_enable_intr();
    701 	splx(s);
    702 	x86_errata();
    703 
    704 	aprint_debug_dev(ci->ci_dev, "CPU %ld running\n",
    705 	    (long)ci->ci_cpuid);
    706 }
    707 
    708 #if defined(DDB)
    709 
    710 #include <ddb/db_output.h>
    711 #include <machine/db_machdep.h>
    712 
    713 /*
    714  * Dump CPU information from ddb.
    715  */
    716 void
    717 cpu_debug_dump(void)
    718 {
    719 	struct cpu_info *ci;
    720 	CPU_INFO_ITERATOR cii;
    721 
    722 	db_printf("addr		dev	id	flags	ipis	curlwp 		fpcurlwp\n");
    723 	for (CPU_INFO_FOREACH(cii, ci)) {
    724 		db_printf("%p	%s	%ld	%x	%x	%10p	%10p\n",
    725 		    ci,
    726 		    ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
    727 		    (long)ci->ci_cpuid,
    728 		    ci->ci_flags, ci->ci_ipis,
    729 		    ci->ci_curlwp,
    730 		    ci->ci_fpcurlwp);
    731 	}
    732 }
    733 #endif
    734 
    735 static void
    736 cpu_copy_trampoline(void)
    737 {
    738 	/*
    739 	 * Copy boot code.
    740 	 */
    741 	extern u_char cpu_spinup_trampoline[];
    742 	extern u_char cpu_spinup_trampoline_end[];
    743 
    744 	vaddr_t mp_trampoline_vaddr;
    745 
    746 	mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
    747 	    UVM_KMF_VAONLY);
    748 
    749 	pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
    750 	    VM_PROT_READ | VM_PROT_WRITE);
    751 	pmap_update(pmap_kernel());
    752 	memcpy((void *)mp_trampoline_vaddr,
    753 	    cpu_spinup_trampoline,
    754 	    cpu_spinup_trampoline_end - cpu_spinup_trampoline);
    755 
    756 	pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
    757 	pmap_update(pmap_kernel());
    758 	uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
    759 }
    760 
    761 #endif
    762 
    763 #ifdef i386
    764 static void
    765 tss_init(struct i386tss *tss, void *stack, void *func)
    766 {
    767 	memset(tss, 0, sizeof *tss);
    768 	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
    769 	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
    770 	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
    771 	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
    772 	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
    773 	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
    774 	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
    775 	tss->tss_esp = (int)((char *)stack + USPACE - 16);
    776 	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
    777 	tss->__tss_eflags = PSL_MBO | PSL_NT;	/* XXX not needed? */
    778 	tss->__tss_eip = (int)func;
    779 }
    780 
    781 /* XXX */
    782 #define IDTVEC(name)	__CONCAT(X, name)
    783 typedef void (vector)(void);
    784 extern vector IDTVEC(tss_trap08);
    785 #ifdef DDB
    786 extern vector Xintrddbipi;
    787 extern int ddb_vec;
    788 #endif
    789 
    790 static void
    791 cpu_set_tss_gates(struct cpu_info *ci)
    792 {
    793 	struct segment_descriptor sd;
    794 
    795 	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
    796 	    UVM_KMF_WIRED);
    797 	tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
    798 	    IDTVEC(tss_trap08));
    799 	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
    800 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
    801 	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
    802 	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
    803 	    GSEL(GTRAPTSS_SEL, SEL_KPL));
    804 
    805 #if defined(DDB) && defined(MULTIPROCESSOR)
    806 	/*
    807 	 * Set up separate handler for the DDB IPI, so that it doesn't
    808 	 * stomp on a possibly corrupted stack.
    809 	 *
    810 	 * XXX overwriting the gate set in db_machine_init.
    811 	 * Should rearrange the code so that it's set only once.
    812 	 */
    813 	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0,
    814 	    UVM_KMF_WIRED);
    815 	tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack, Xintrddbipi);
    816 
    817 	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
    818 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
    819 	ci->ci_gdt[GIPITSS_SEL].sd = sd;
    820 
    821 	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
    822 	    GSEL(GIPITSS_SEL, SEL_KPL));
    823 #endif
    824 }
    825 #else
    826 static void
    827 cpu_set_tss_gates(struct cpu_info *ci)
    828 {
    829 
    830 }
    831 #endif	/* i386 */
    832 
    833 int
    834 mp_cpu_start(struct cpu_info *ci, paddr_t target)
    835 {
    836 #if NLAPIC > 0
    837 	int error;
    838 #endif
    839 	unsigned short dwordptr[2];
    840 
    841 	/*
    842 	 * Bootstrap code must be addressable in real mode
    843 	 * and it must be page aligned.
    844 	 */
    845 	KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
    846 
    847 	/*
    848 	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
    849 	 */
    850 
    851 	outb(IO_RTC, NVRAM_RESET);
    852 	outb(IO_RTC+1, NVRAM_RESET_JUMP);
    853 
    854 	/*
    855 	 * "and the warm reset vector (DWORD based at 40:67) to point
    856 	 * to the AP startup code ..."
    857 	 */
    858 
    859 	dwordptr[0] = 0;
    860 	dwordptr[1] = target >> 4;
    861 
    862 	memcpy((uint8_t *)cmos_data_mapping + 0x467, dwordptr, 4);
    863 
    864 #if NLAPIC > 0
    865 	if ((cpu_feature & CPUID_APIC) == 0) {
    866 		aprint_error("mp_cpu_start: CPU does not have APIC\n");
    867 		return ENODEV;
    868 	}
    869 
    870 	/*
    871 	 * ... prior to executing the following sequence:"
    872 	 */
    873 
    874 	if (ci->ci_flags & CPUF_AP) {
    875 		error = x86_ipi_init(ci->ci_apicid);
    876 		if (error != 0) {
    877 			aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
    878 					__func__);
    879 			return error;
    880 		}
    881 
    882 		i8254_delay(10000);
    883 
    884 		error = x86_ipi(target / PAGE_SIZE, ci->ci_apicid,
    885 		    LAPIC_DLMODE_STARTUP);
    886 		if (error != 0) {
    887 			aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
    888 					__func__);
    889 			return error;
    890 		}
    891 		i8254_delay(200);
    892 
    893 		error = x86_ipi(target / PAGE_SIZE, ci->ci_apicid,
    894 		    LAPIC_DLMODE_STARTUP);
    895 		if (error != 0) {
    896 			aprint_error_dev(ci->ci_dev, "%s: IPI not taken (3)\n",
    897 					__func__);
    898 			return error;
    899 		}
    900 		i8254_delay(200);
    901 	}
    902 #endif
    903 	return 0;
    904 }
    905 
    906 void
    907 mp_cpu_start_cleanup(struct cpu_info *ci)
    908 {
    909 	/*
    910 	 * Ensure the NVRAM reset byte contains something vaguely sane.
    911 	 */
    912 
    913 	outb(IO_RTC, NVRAM_RESET);
    914 	outb(IO_RTC+1, NVRAM_RESET_RST);
    915 }
    916 
    917 #ifdef __x86_64__
    918 typedef void (vector)(void);
    919 extern vector Xsyscall, Xsyscall32;
    920 
    921 void
    922 cpu_init_msrs(struct cpu_info *ci, bool full)
    923 {
    924 	wrmsr(MSR_STAR,
    925 	    ((uint64_t)GSEL(GCODE_SEL, SEL_KPL) << 32) |
    926 	    ((uint64_t)LSEL(LSYSRETBASE_SEL, SEL_UPL) << 48));
    927 	wrmsr(MSR_LSTAR, (uint64_t)Xsyscall);
    928 	wrmsr(MSR_CSTAR, (uint64_t)Xsyscall32);
    929 	wrmsr(MSR_SFMASK, PSL_NT|PSL_T|PSL_I|PSL_C);
    930 
    931 	if (full) {
    932 		wrmsr(MSR_FSBASE, 0);
    933 		wrmsr(MSR_GSBASE, (uint64_t)ci);
    934 		wrmsr(MSR_KERNELGSBASE, 0);
    935 	}
    936 
    937 	if (cpu_feature & CPUID_NOX)
    938 		wrmsr(MSR_EFER, rdmsr(MSR_EFER) | EFER_NXE);
    939 }
    940 #endif	/* __x86_64__ */
    941 
    942 void
    943 cpu_offline_md(void)
    944 {
    945 	int s;
    946 
    947 	s = splhigh();
    948 #ifdef __i386__
    949 	npxsave_cpu(true);
    950 #else
    951 	fpusave_cpu(true);
    952 #endif
    953 	splx(s);
    954 }
    955 
    956 /* XXX joerg restructure and restart CPUs individually */
    957 static bool
    958 cpu_suspend(device_t dv PMF_FN_ARGS)
    959 {
    960 	struct cpu_softc *sc = device_private(dv);
    961 	struct cpu_info *ci = sc->sc_info;
    962 	int err;
    963 
    964 	if (ci->ci_flags & CPUF_PRIMARY)
    965 		return true;
    966 	if (ci->ci_data.cpu_idlelwp == NULL)
    967 		return true;
    968 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
    969 		return true;
    970 
    971 	sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
    972 
    973 	if (sc->sc_wasonline) {
    974 		mutex_enter(&cpu_lock);
    975 		err = cpu_setonline(ci, false);
    976 		mutex_exit(&cpu_lock);
    977 
    978 		if (err)
    979 			return false;
    980 	}
    981 
    982 	return true;
    983 }
    984 
    985 static bool
    986 cpu_resume(device_t dv PMF_FN_ARGS)
    987 {
    988 	struct cpu_softc *sc = device_private(dv);
    989 	struct cpu_info *ci = sc->sc_info;
    990 	int err = 0;
    991 
    992 	if (ci->ci_flags & CPUF_PRIMARY)
    993 		return true;
    994 	if (ci->ci_data.cpu_idlelwp == NULL)
    995 		return true;
    996 	if ((ci->ci_flags & CPUF_PRESENT) == 0)
    997 		return true;
    998 
    999 	if (sc->sc_wasonline) {
   1000 		mutex_enter(&cpu_lock);
   1001 		err = cpu_setonline(ci, true);
   1002 		mutex_exit(&cpu_lock);
   1003 	}
   1004 
   1005 	return err == 0;
   1006 }
   1007 
   1008 void
   1009 cpu_get_tsc_freq(struct cpu_info *ci)
   1010 {
   1011 	uint64_t last_tsc;
   1012 	u_int junk[4];
   1013 
   1014 	if (ci->ci_feature_flags & CPUID_TSC) {
   1015 		/* Serialize. */
   1016 		x86_cpuid(0, junk);
   1017 		last_tsc = rdtsc();
   1018 		i8254_delay(100000);
   1019 		ci->ci_tsc_freq = (rdtsc() - last_tsc) * 10;
   1020 	}
   1021 }
   1022